class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
InstrItinClass rr = arg_rr;
InstrItinClass rm = arg_rm;
+ // InstrSchedModel info.
+ X86FoldableSchedWrite Sched = WriteFAdd;
}
class SizeItins<OpndItins arg_s, OpndItins arg_d> {
// scalar
+let Sched = WriteFAdd in {
def SSE_ALU_F32S : OpndItins<
IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
>;
def SSE_ALU_F64S : OpndItins<
IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
>;
+}
def SSE_ALU_ITINS_S : SizeItins<
SSE_ALU_F32S, SSE_ALU_F64S
>;
+let Sched = WriteFMul in {
def SSE_MUL_F32S : OpndItins<
IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
>;
def SSE_MUL_F64S : OpndItins<
IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
>;
+}
def SSE_MUL_ITINS_S : SizeItins<
SSE_MUL_F32S, SSE_MUL_F64S
>;
+let Sched = WriteFDiv in {
def SSE_DIV_F32S : OpndItins<
IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
>;
def SSE_DIV_F64S : OpndItins<
IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
>;
+}
def SSE_DIV_ITINS_S : SizeItins<
SSE_DIV_F32S, SSE_DIV_F64S
>;
// parallel
+let Sched = WriteFAdd in {
def SSE_ALU_F32P : OpndItins<
IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
>;
def SSE_ALU_F64P : OpndItins<
IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
>;
+}
def SSE_ALU_ITINS_P : SizeItins<
SSE_ALU_F32P, SSE_ALU_F64P
>;
+let Sched = WriteFMul in {
def SSE_MUL_F32P : OpndItins<
IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
>;
def SSE_MUL_F64P : OpndItins<
IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
>;
+}
def SSE_MUL_ITINS_P : SizeItins<
SSE_MUL_F32P, SSE_MUL_F64P
>;
+let Sched = WriteFDiv in {
def SSE_DIV_F32P : OpndItins<
IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
>;
def SSE_DIV_F64P : OpndItins<
IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
>;
+}
def SSE_DIV_ITINS_P : SizeItins<
SSE_DIV_F32P, SSE_DIV_F64P
>;
+let Sched = WriteVecLogic in
+def SSE_VEC_BIT_ITINS_P : OpndItins<
+ IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
+>;
+
def SSE_BIT_ITINS_P : OpndItins<
IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
>;
+let Sched = WriteVecALU in {
def SSE_INTALU_ITINS_P : OpndItins<
IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
>;
def SSE_INTALUQ_ITINS_P : OpndItins<
IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
>;
+}
+let Sched = WriteVecIMul in
def SSE_INTMUL_ITINS_P : OpndItins<
IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
>;
IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
>;
+def SSE_DPPD_ITINS : OpndItins<
+ IIC_SSE_DPPD_RR, IIC_SSE_DPPD_RM
+>;
+
+def SSE_DPPS_ITINS : OpndItins<
+ IIC_SSE_DPPS_RR, IIC_SSE_DPPD_RM
+>;
+
+def DEFAULT_ITINS : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+def SSE_EXTRACT_ITINS : OpndItins<
+ IIC_SSE_EXTRACTPS_RR, IIC_SSE_EXTRACTPS_RM
+>;
+
+def SSE_INSERT_ITINS : OpndItins<
+ IIC_SSE_INSERTPS_RR, IIC_SSE_INSERTPS_RM
+>;
+
+let Sched = WriteMPSAD in
+def SSE_MPSADBW_ITINS : OpndItins<
+ IIC_SSE_MPSADBW_RR, IIC_SSE_MPSADBW_RM
+>;
+
+let Sched = WriteVecIMul in
+def SSE_PMULLD_ITINS : OpndItins<
+ IIC_SSE_PMULLD_RR, IIC_SSE_PMULLD_RM
+>;
+
+// Definitions for backward compatibility.
+// The instructions mapped on these definitions uses a different itinerary
+// than the actual scheduling model.
+let Sched = WriteShuffle in
+def DEFAULT_ITINS_SHUFFLESCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteVecIMul in
+def DEFAULT_ITINS_VECIMULSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteShuffle in
+def SSE_INTALU_ITINS_SHUFF_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
+let Sched = WriteMPSAD in
+def DEFAULT_ITINS_MPSADSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteFBlend in
+def DEFAULT_ITINS_FBLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteBlend in
+def DEFAULT_ITINS_BLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteVarBlend in
+def DEFAULT_ITINS_VARBLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteFBlend in
+def SSE_INTALU_ITINS_FBLEND_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
+let Sched = WriteBlend in
+def SSE_INTALU_ITINS_BLEND_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 Instructions Classes
//===----------------------------------------------------------------------===//
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
+ [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>,
+ Sched<[itins.Sched]>;
}
def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
+ [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
/// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
Operand memopr, ComplexPattern mem_cpat,
OpndItins itins,
bit Is2Addr = 1> {
+let isCodeGenOnly = 1 in {
def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (!cast<Intrinsic>(
!strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
- RC:$src1, RC:$src2))], itins.rr>;
+ RC:$src1, RC:$src2))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
SSEVer, "_", OpcodeStr, FPSizeStr))
- RC:$src1, mem_cpat:$src2))], itins.rm>;
+ RC:$src1, mem_cpat:$src2))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
}
/// sse12_fp_packed - SSE 1 & 2 packed instructions class
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
+ [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
+ Sched<[itins.Sched]>;
let mayLoad = 1 in
def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
- itins.rm, d>;
+ itins.rm, d>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
/// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
string OpcodeStr, X86MemOperand x86memop,
list<dag> pat_rr, list<dag> pat_rm,
- bit Is2Addr = 1,
- bit rr_hasSideEffects = 0> {
- let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
+ bit Is2Addr = 1> {
+ let isCommutable = 1, hasSideEffects = 0 in
def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- pat_rr, IIC_DEFAULT, d>;
+ pat_rr, NoItinerary, d>,
+ Sched<[WriteVecLogic]>;
def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- pat_rm, IIC_DEFAULT, d>;
-}
-
-/// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
-multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
- string asm, string SSEVer, string FPSizeStr,
- X86MemOperand x86memop, PatFrag mem_frag,
- Domain d, OpndItins itins, bit Is2Addr = 1> {
- def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
- !if(Is2Addr,
- !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
- !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!cast<Intrinsic>(
- !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
- RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
- def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
- !if(Is2Addr,
- !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
- !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (!cast<Intrinsic>(
- !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
- RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
+ pat_rm, NoItinerary, d>,
+ Sched<[WriteVecLogicLd, ReadAfterLd]>;
}
//===----------------------------------------------------------------------===//
// Alias instructions that map fld0 to xorps for sse or vxorps for avx.
// This is expanded by ExpandPostRAPseudos.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isPseudo = 1 in {
+ isPseudo = 1, SchedRW = [WriteZero] in {
def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
[(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isPseudo = 1 in {
+ isPseudo = 1, SchedRW = [WriteZero] in {
def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4f32 immAllZerosV))]>;
}
// at the rename stage without using any execution unit, so SET0PSY
// and SET0PDY can be used for vector int instructions without penalty
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isPseudo = 1, Predicates = [HasAVX] in {
+ isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in {
def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
[(set VR256:$dst, (v8f32 immAllZerosV))]>;
}
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-ones value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isPseudo = 1 in {
+ isPseudo = 1, SchedRW = [WriteZero] in {
def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllOnesV))]>;
let Predicates = [HasAVX2] in
// SSE 1 & 2 - Move FP Scalar Instructions
//
// Move Instructions. Register-to-register movss/movsd is not used for FR32/64
-// register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
-// is used instead. Register-to-register movss/movsd is not modeled as an
-// INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
-// in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
+// register copies because it's a partial register update; Register-to-register
+// movss/movsd is not modeled as an INSERT_SUBREG because INSERT_SUBREG requires
+// that the insert be implementable in terms of a copy, and just mentioned, we
+// don't use movss/movsd for copies.
//===----------------------------------------------------------------------===//
-class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
- SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
- [(set VR128:$dst, (vt (OpNode VR128:$src1,
- (scalar_to_vector RC:$src2))))],
- IIC_SSE_MOV_S_RR>;
+multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
+ X86MemOperand x86memop, string base_opc,
+ string asm_opr> {
+ def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, RC:$src2),
+ !strconcat(base_opc, asm_opr),
+ [(set VR128:$dst, (vt (OpNode VR128:$src1,
+ (scalar_to_vector RC:$src2))))],
+ IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>;
-// Loading from memory automatically zeroing upper bits.
-class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
- PatFrag mem_pat, string OpcodeStr> :
- SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (mem_pat addr:$src))],
- IIC_SSE_MOV_S_RM>;
-
-// AVX
-def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
- "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
- VEX_LIG;
-def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
- "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
- VEX_LIG;
-
-// For the disassembler
-let isCodeGenOnly = 1 in {
- def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR32:$src2),
- "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
- IIC_SSE_MOV_S_RR>,
- XS, VEX_4V, VEX_LIG;
- def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR64:$src2),
- "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
- IIC_SSE_MOV_S_RR>,
- XD, VEX_4V, VEX_LIG;
+ // For the disassembler
+ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
+ def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
+ (ins VR128:$src1, RC:$src2),
+ !strconcat(base_opc, asm_opr),
+ [], IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>;
}
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
- def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
- VEX_LIG;
- let AddedComplexity = 20 in
- def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
- VEX_LIG;
-}
+multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
+ X86MemOperand x86memop, string OpcodeStr> {
+ // AVX
+ defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
+ VEX_4V, VEX_LIG;
-def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
- XS, VEX, VEX_LIG;
-def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
- XD, VEX, VEX_LIG;
+ def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+ VEX, VEX_LIG, Sched<[WriteStore]>;
+ // SSE1 & 2
+ let Constraints = "$src1 = $dst" in {
+ defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+ "\t{$src2, $dst|$dst, $src2}">;
+ }
-// SSE1 & 2
-let Constraints = "$src1 = $dst" in {
- def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
- "movss\t{$src2, $dst|$dst, $src2}">, XS;
- def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
- "movsd\t{$src2, $dst|$dst, $src2}">, XD;
+ def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+ Sched<[WriteStore]>;
+}
- // For the disassembler
- let isCodeGenOnly = 1 in {
- def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR32:$src2),
- "movss\t{$src2, $dst|$dst, $src2}", [],
- IIC_SSE_MOV_S_RR>, XS;
- def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR64:$src2),
- "movsd\t{$src2, $dst|$dst, $src2}", [],
- IIC_SSE_MOV_S_RR>, XD;
- }
+// Loading from memory automatically zeroing upper bits.
+multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
+ PatFrag mem_pat, string OpcodeStr> {
+ def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (mem_pat addr:$src))],
+ IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>;
+ def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (mem_pat addr:$src))],
+ IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>;
}
+defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
+defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
+
let canFoldAsLoad = 1, isReMaterializable = 1 in {
- def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
+ defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
let AddedComplexity = 20 in
- def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
+ defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
}
-def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-
// Patterns
-let Predicates = [HasAVX] in {
- let AddedComplexity = 15 in {
- // Move scalar to XMM zero-extended, zeroing a VR128 then do a
- // MOVS{S,D} to the lower bits.
- def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
- (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
- def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
- (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
- def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
- (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
- def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
- (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
-
- // Move low f32 and clear high bits.
- def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (v4f32 (V_SET0)),
- (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
- def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (v4i32 (V_SET0)),
- (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
- }
-
+let Predicates = [UseAVX] in {
let AddedComplexity = 20 in {
// MOVSSrm zeros the high parts of the register; represent this
// with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
(v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
(SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
}
- def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
- (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0),
- (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
- sub_xmm)>;
- def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
- (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
- (SUBREG_TO_REG (i64 0),
- (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
- sub_xmm)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
(SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
- // Move low f64 and clear high bits.
- def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (v2f64 (V_SET0)),
- (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
-
- def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (v2i64 (V_SET0)),
- (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
-
// Extract and store.
def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
addr:$dst),
(EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
sub_xmm)>;
-
// FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
// is during lowering, where it's not possible to recognize the fold cause
// it has two uses through a bitcast. One use disappears at isel time and the
}
let Predicates = [UseSSE1] in {
- let AddedComplexity = 15 in {
+ let Predicates = [NoSSE41], AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSS to the lower bits.
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
}
let Predicates = [UseSSE2] in {
- let AddedComplexity = 15 in {
+ let Predicates = [NoSSE41], AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSD to the lower bits.
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
bit IsReMaterializable = 1> {
let neverHasSideEffects = 1 in
def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
+ Sched<[WriteFShuffle]>;
let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(asm, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
+ [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>,
+ Sched<[WriteLoad]>;
}
+let Predicates = [HasAVX, NoVLX] in {
defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB, VEX;
+ PS, VEX;
defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
"movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
- TB, OpSize, VEX;
+ PD, VEX;
defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
"movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB, VEX;
+ PS, VEX;
defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
- TB, OpSize, VEX;
+ PD, VEX;
defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB, VEX;
+ PS, VEX, VEX_L;
defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
"movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
- TB, OpSize, VEX;
+ PD, VEX, VEX_L;
defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
"movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB, VEX;
+ PS, VEX, VEX_L;
defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
- TB, OpSize, VEX;
+ PD, VEX, VEX_L;
+}
+
+let Predicates = [UseSSE1] in {
defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB;
-defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
- "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
- TB, OpSize;
+ PS;
defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
"movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB;
+ PS;
+}
+let Predicates = [UseSSE2] in {
+defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
+ "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
+ PD;
defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
- TB, OpSize;
+ PD;
+}
+let SchedRW = [WriteStore], Predicates = [HasAVX, NoVLX] in {
def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore (v4f32 VR128:$src), addr:$dst)],
def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
- IIC_SSE_MOVA_P_MR>, VEX;
+ IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
"movapd\t{$src, $dst|$dst, $src}",
[(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
- IIC_SSE_MOVA_P_MR>, VEX;
+ IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
"movups\t{$src, $dst|$dst, $src}",
[(store (v8f32 VR256:$src), addr:$dst)],
- IIC_SSE_MOVU_P_MR>, VEX;
+ IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
"movupd\t{$src, $dst|$dst, $src}",
[(store (v4f64 VR256:$src), addr:$dst)],
- IIC_SSE_MOVU_P_MR>, VEX;
+ IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
+} // SchedRW
// For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
+ SchedRW = [WriteFShuffle] in {
def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
(ins VR128:$src),
"movaps\t{$src, $dst|$dst, $src}", [],
def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
(ins VR256:$src),
"movaps\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>, VEX;
+ IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
(ins VR256:$src),
"movapd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>, VEX;
+ IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
(ins VR256:$src),
"movups\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVU_P_RR>, VEX;
+ IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
(ins VR256:$src),
"movupd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVU_P_RR>, VEX;
+ IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
}
let Predicates = [HasAVX] in {
def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
(VMOVUPDYmr addr:$dst, VR256:$src)>;
+let SchedRW = [WriteStore] in {
def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore (v4f32 VR128:$src), addr:$dst)],
"movupd\t{$src, $dst|$dst, $src}",
[(store (v2f64 VR128:$src), addr:$dst)],
IIC_SSE_MOVU_P_MR>;
+} // SchedRW
// For disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
+ SchedRW = [WriteFShuffle] in {
def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movaps\t{$src, $dst|$dst, $src}", [],
IIC_SSE_MOVA_P_RR>;
(MOVUPDmr addr:$dst, VR128:$src)>;
// Use vmovaps/vmovups for AVX integer load/store.
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
// 128-bit load/store
def : Pat<(alignedloadv2i64 addr:$src),
(VMOVAPSrm addr:$src)>;
(VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
def : Pat<(store (v8i16 (extract_subvector
(v16i16 VR256:$src), (iPTR 0))), addr:$dst),
- (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+ (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
def : Pat<(store (v16i8 (extract_subvector
(v32i8 VR256:$src), (iPTR 0))), addr:$dst),
(VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
(MOVUPSmr addr:$dst, VR128:$src)>;
}
-// Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
-// bits are disregarded. FIXME: Set encoding to pseudo!
-let neverHasSideEffects = 1 in {
-def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- "movaps\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>, VEX;
-def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
- "movapd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>, VEX;
-def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- "movaps\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>;
-def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
- "movapd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>;
-}
-
// Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
// bits are disregarded. FIXME: Set encoding to pseudo!
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
+let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in {
let isCodeGenOnly = 1 in {
def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
"movaps\t{$src, $dst|$dst, $src}",
"movapd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (alignedloadfsf64 addr:$src))],
IIC_SSE_MOVA_P_RM>, VEX;
+ def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
+ "movaps\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
+ IIC_SSE_MOVA_P_RM>;
+ def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
+ "movapd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
+ IIC_SSE_MOVA_P_RM>;
}
-def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
- "movaps\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
- IIC_SSE_MOVA_P_RM>;
-def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
- "movapd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
- IIC_SSE_MOVA_P_RM>;
}
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Move Low packed FP Instructions
//===----------------------------------------------------------------------===//
-multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
- SDNode psnode, SDNode pdnode, string base_opc,
- string asm_opr, InstrItinClass itin> {
+multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
+ string base_opc, string asm_opr,
+ InstrItinClass itin> {
def PSrm : PI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
!strconcat(base_opc, "s", asm_opr),
- [(set RC:$dst,
- (psnode RC:$src1,
+ [(set VR128:$dst,
+ (psnode VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
- itin, SSEPackedSingle>, TB;
+ itin, SSEPackedSingle>, PS,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
def PDrm : PI<opc, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
+ (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
!strconcat(base_opc, "d", asm_opr),
- [(set RC:$dst, (v2f64 (pdnode RC:$src1,
+ [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))],
- itin, SSEPackedDouble>, TB, OpSize;
+ itin, SSEPackedDouble>, PD,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
+
}
-let AddedComplexity = 20 in {
- defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- IIC_SSE_MOV_LH>, VEX_4V;
+multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
+ string base_opc, InstrItinClass itin> {
+ defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ itin>, VEX_4V;
+
+let Constraints = "$src1 = $dst" in
+ defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ "\t{$src2, $dst|$dst, $src2}",
+ itin>;
}
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
- defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
- "\t{$src2, $dst|$dst, $src2}",
- IIC_SSE_MOV_LH>;
+
+let AddedComplexity = 20 in {
+ defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
+ IIC_SSE_MOV_LH>;
}
+let SchedRW = [WriteStore] in {
def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
[(store (f64 (vector_extract (v2f64 VR128:$src),
(iPTR 0))), addr:$dst)],
IIC_SSE_MOV_LH>;
+} // SchedRW
let Predicates = [HasAVX] in {
// Shuffle with VMOVLPS
(VMOVLPDrm VR128:$src1, addr:$src2)>;
def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
(VMOVLPDrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v2f64 (X86Movsd VR128:$src1,
+ (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (VMOVLPDrm VR128:$src1, addr:$src2)>;
// Store patterns
def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
(MOVLPDrm VR128:$src1, addr:$src2)>;
def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
(MOVLPDrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v2f64 (X86Movsd VR128:$src1,
+ (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
// Store patterns
def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
//===----------------------------------------------------------------------===//
let AddedComplexity = 20 in {
- defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- IIC_SSE_MOV_LH>, VEX_4V;
-}
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
- defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
- "\t{$src2, $dst|$dst, $src2}",
- IIC_SSE_MOV_LH>;
+ defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
+ IIC_SSE_MOV_LH>;
}
+let SchedRW = [WriteStore] in {
// v2f64 extract element 1 is always custom lowered to unpack high to low
// and extract element 0 so the non-store version isn't too horrible.
def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
[(store (f64 (vector_extract
(v2f64 (X86Unpckh VR128:$src, VR128:$src)),
(iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
+} // SchedRW
let Predicates = [HasAVX] in {
// VMOVHPS patterns
(VMOVHPSrm VR128:$src1, addr:$src2)>;
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
- // is during lowering, where it's not possible to recognize the load fold
+ // is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
// and the fold opportunity reappears.
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))),
(VMOVHPDrm VR128:$src1, addr:$src2)>;
+ // Also handle an i64 load because that may get selected as a faster way to
+ // load the data.
+ def : Pat<(v2f64 (X86Unpckl VR128:$src1,
+ (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
+ (VMOVHPDrm VR128:$src1, addr:$src2)>;
}
let Predicates = [UseSSE1] in {
let Predicates = [UseSSE2] in {
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
- // is during lowering, where it's not possible to recognize the load fold
+ // is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
// and the fold opportunity reappears.
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))),
(MOVHPDrm VR128:$src1, addr:$src2)>;
+ // Also handle an i64 load because that may get selected as a faster way to
+ // load the data.
+ def : Pat<(v2f64 (X86Unpckl VR128:$src1,
+ (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
+ (MOVHPDrm VR128:$src1, addr:$src2)>;
}
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
//===----------------------------------------------------------------------===//
-let AddedComplexity = 20 in {
+let AddedComplexity = 20, Predicates = [UseAVX] in {
def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
IIC_SSE_MOV_LH>,
- VEX_4V;
+ VEX_4V, Sched<[WriteFShuffle]>;
def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
IIC_SSE_MOV_LH>,
- VEX_4V;
+ VEX_4V, Sched<[WriteFShuffle]>;
}
let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
"movlhps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
- IIC_SSE_MOV_LH>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movhlps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
- IIC_SSE_MOV_LH>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
}
-let Predicates = [HasAVX] in {
+let Predicates = [UseAVX] in {
// MOVLHPS patterns
def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
(VMOVLHPSrr VR128:$src1, VR128:$src2)>;
IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
>;
+let Sched = WriteCvtI2F in
def SSE_CVT_PS : OpndItins<
IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
>;
+let Sched = WriteCvtI2F in
def SSE_CVT_Scalar : OpndItins<
IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
>;
+let Sched = WriteCvtF2I in
def SSE_CVT_SS2SI_32 : OpndItins<
IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
>;
+let Sched = WriteCvtF2I in
def SSE_CVT_SS2SI_64 : OpndItins<
IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
>;
+let Sched = WriteCvtF2I in
def SSE_CVT_SD2SI : OpndItins<
IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
>;
string asm, OpndItins itins> {
def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
[(set DstRC:$dst, (OpNode SrcRC:$src))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
[(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded]>;
}
multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
OpndItins itins> {
let neverHasSideEffects = 1 in {
def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
- [], itins.rr, d>;
+ [], itins.rr, d>, Sched<[itins.Sched]>;
let mayLoad = 1 in
def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
- [], itins.rm, d>;
+ [], itins.rm, d>, Sched<[itins.Sched.Folded]>;
}
}
multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
X86MemOperand x86memop, string asm> {
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, Predicates = [UseAVX] in {
def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
- !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
+ !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+ Sched<[WriteCvtI2F]>;
let mayLoad = 1 in
def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
(ins DstRC:$src1, x86memop:$src),
- !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
+ !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+ Sched<[WriteCvtI2FLd, ReadAfterLd]>;
} // neverHasSideEffects = 1
}
+let Predicates = [UseAVX] in {
defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
"cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_32>,
XS, VEX, VEX_LIG;
defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
- "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_64>,
XS, VEX, VEX_W, VEX_LIG;
defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
SSE_CVT_SD2SI>,
XD, VEX, VEX_LIG;
defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
- "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>,
XD, VEX, VEX_W, VEX_LIG;
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+}
// The assembler can recognize rr 64-bit instructions by seeing a rxx
// register, but the same isn't true when only using memory operands,
// provide other assembly "l" and "q" forms to address this explicitly
// where appropriate to do so.
-defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
+defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
XS, VEX_4V, VEX_LIG;
defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
XS, VEX_4V, VEX_W, VEX_LIG;
-defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
+defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
XD, VEX_4V, VEX_LIG;
defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
XD, VEX_4V, VEX_W, VEX_LIG;
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
- (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
- (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
+let Predicates = [UseAVX] in {
+ def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
+ (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src), 0>;
+ def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
+ (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src), 0>;
-let Predicates = [HasAVX] in {
def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
(VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
"cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_32>, XS;
defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
- "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_64>, XS, REX_W;
defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
"cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>, XD;
defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
- "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>, XD, REX_W;
defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
- "cvtsi2ss\t{$src, $dst|$dst, $src}",
+ "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XS;
defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
"cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XS, REX_W;
defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
- "cvtsi2sd\t{$src, $dst|$dst, $src}",
+ "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XD;
defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
"cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XD, REX_W;
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+
+def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
+ (CVTSI2SSrm FR64:$dst, i32mem:$src), 0>;
+def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
+ (CVTSI2SDrm FR64:$dst, i32mem:$src), 0>;
+
// Conversion Instructions Intrinsics - Match intrinsics which expect MM
// and/or XMM operand(s).
string asm, OpndItins itins> {
def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
!strconcat(asm, "\t{$src, $dst|$dst, $src}"),
- [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
+ [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
!strconcat(asm, "\t{$src, $dst|$dst, $src}"),
- [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
+ [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>,
+ Sched<[itins.Sched.Folded]>;
}
multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
(ins DstRC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
+let Predicates = [UseAVX] in {
defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
- int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
+ int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
- int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
+ int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
-
+}
defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
- sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
+ sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
- sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
-
-
-defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
- SSE_CVT_Scalar, 0>, XS, VEX_4V;
-defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
- SSE_CVT_Scalar, 0>, XS, VEX_4V,
- VEX_W;
-defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
- SSE_CVT_Scalar, 0>, XD, VEX_4V;
-defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
- SSE_CVT_Scalar, 0>, XD,
- VEX_4V, VEX_W;
+ sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
-let Constraints = "$src1 = $dst" in {
- defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse_cvtsi2ss, i32mem, loadi32,
- "cvtsi2ss", SSE_CVT_Scalar>, XS;
- defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse_cvtsi642ss, i64mem, loadi64,
- "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
- defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse2_cvtsi2sd, i32mem, loadi32,
- "cvtsi2sd", SSE_CVT_Scalar>, XD;
- defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
- int_x86_sse2_cvtsi642sd, i64mem, loadi64,
- "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
-}
+
+let isCodeGenOnly = 1 in {
+ let Predicates = [UseAVX] in {
+ defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
+ SSE_CVT_Scalar, 0>, XS, VEX_4V;
+ defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
+ SSE_CVT_Scalar, 0>, XS, VEX_4V,
+ VEX_W;
+ defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
+ SSE_CVT_Scalar, 0>, XD, VEX_4V;
+ defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
+ SSE_CVT_Scalar, 0>, XD,
+ VEX_4V, VEX_W;
+ }
+ let Constraints = "$src1 = $dst" in {
+ defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse_cvtsi2ss, i32mem, loadi32,
+ "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
+ defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse_cvtsi642ss, i64mem, loadi64,
+ "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
+ defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
+ int_x86_sse2_cvtsi2sd, i32mem, loadi32,
+ "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
+ defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
+ int_x86_sse2_cvtsi642sd, i64mem, loadi64,
+ "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
+ }
+} // isCodeGenOnly = 1
/// SSE 1 Only
// Aliases for intrinsics
+let isCodeGenOnly = 1 in {
+let Predicates = [UseAVX] in {
defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
ssmem, sse_load_f32, "cvttss2si",
SSE_CVT_SS2SI_32>, XS, VEX;
defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
- "cvttss2si{q}", SSE_CVT_SS2SI_64>,
+ "cvttss2si", SSE_CVT_SS2SI_64>,
XS, VEX, VEX_W;
defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
sdmem, sse_load_f64, "cvttsd2si",
SSE_CVT_SD2SI>, XD, VEX;
defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
- "cvttsd2si{q}", SSE_CVT_SD2SI>,
+ "cvttsd2si", SSE_CVT_SD2SI>,
XD, VEX, VEX_W;
+}
defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
ssmem, sse_load_f32, "cvttss2si",
SSE_CVT_SS2SI_32>, XS;
defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
- "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
+ "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
sdmem, sse_load_f64, "cvttsd2si",
SSE_CVT_SD2SI>, XD;
defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
- "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
+ "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
+} // isCodeGenOnly = 1
+let Predicates = [UseAVX] in {
defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- ssmem, sse_load_f32, "cvtss2si{l}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
- ssmem, sse_load_f32, "cvtss2si{q}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
-
+}
defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- ssmem, sse_load_f32, "cvtss2si{l}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_32>, XS;
defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
- ssmem, sse_load_f32, "cvtss2si{q}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_64>, XS, REX_W;
defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
"vcvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, VEX, Requires<[HasAVX]>;
+ PS, VEX, Requires<[HasAVX]>;
defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
"vcvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, VEX, Requires<[HasAVX]>;
+ PS, VEX, VEX_L, Requires<[HasAVX]>;
defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
"cvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, Requires<[UseSSE2]>;
+ PS, Requires<[UseSSE2]>;
+
+let Predicates = [UseAVX] in {
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
+}
+
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
/// SSE 2 Only
// Convert scalar double to scalar single
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, Predicates = [UseAVX] in {
def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
(ins FR64:$src1, FR64:$src2),
"cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
- IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
+ IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG,
+ Sched<[WriteCvtF2F]>;
let mayLoad = 1 in
def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
(ins FR64:$src1, f64mem:$src2),
"vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[], IIC_SSE_CVT_Scalar_RM>,
- XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
+ XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
}
def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
- Requires<[HasAVX]>;
+ Requires<[UseAVX]>;
def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
"cvtsd2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (fround FR64:$src))],
- IIC_SSE_CVT_Scalar_RR>;
+ IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>;
def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
"cvtsd2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (fround (loadf64 addr:$src)))],
IIC_SSE_CVT_Scalar_RM>,
XD,
- Requires<[UseSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
+let isCodeGenOnly = 1 in {
def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
+ IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[UseAVX]>,
+ Sched<[WriteCvtF2F]>;
def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
"vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (int_x86_sse2_cvtsd2ss
VR128:$src1, sse_load_f64:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
+ IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[UseAVX]>,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
let Constraints = "$src1 = $dst" in {
def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ "cvtsd2ss\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
+ IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>,
+ Sched<[WriteCvtF2F]>;
def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
- "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ "cvtsd2ss\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst, (int_x86_sse2_cvtsd2ss
VR128:$src1, sse_load_f64:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
+ IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
}
+} // isCodeGenOnly = 1
// Convert scalar single to scalar double
// SSE2 instructions with XS prefix
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, Predicates = [UseAVX] in {
def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
(ins FR32:$src1, FR32:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[], IIC_SSE_CVT_Scalar_RR>,
- XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
+ XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG,
+ Sched<[WriteCvtF2F]>;
let mayLoad = 1 in
def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
(ins FR32:$src1, f32mem:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[], IIC_SSE_CVT_Scalar_RM>,
- XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
+ XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
}
def : Pat<(f64 (fextend FR32:$src)),
- (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
+ (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[UseAVX]>;
def : Pat<(fextend (loadf32 addr:$src)),
- (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
+ (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[UseAVX]>;
def : Pat<(extloadf32 addr:$src),
(VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
+ Requires<[UseAVX, OptForSize]>;
def : Pat<(extloadf32 addr:$src),
(VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
- Requires<[HasAVX, OptForSpeed]>;
+ Requires<[UseAVX, OptForSpeed]>;
def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (fextend FR32:$src))],
IIC_SSE_CVT_Scalar_RR>, XS,
- Requires<[UseSSE2]>;
+ Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>;
def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (extloadf32 addr:$src))],
IIC_SSE_CVT_Scalar_RM>, XS,
- Requires<[UseSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>;
// extload f32 -> f64. This matches load+fextend because we have a hack in
// the isel (PreprocessForFPConvert) that can introduce loads after dag
def : Pat<(extloadf32 addr:$src),
(CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
+let isCodeGenOnly = 1 in {
def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
+ IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[UseAVX]>,
+ Sched<[WriteCvtF2F]>;
def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
+ IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[UseAVX]>,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"cvtss2sd\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
- IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
+ IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>,
+ Sched<[WriteCvtF2F]>;
def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
"cvtss2sd\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
- IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
+ IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>,
+ Sched<[WriteCvtF2FLd, ReadAfterLd]>;
}
+} // isCodeGenOnly = 1
// Convert packed single/double fp to doubleword
def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
- IIC_SSE_CVT_PS_RR>, VEX;
+ IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>, VEX;
+ (int_x86_sse2_cvtps2dq (loadv4f32 addr:$src)))],
+ IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvt_ps2dq_256 VR256:$src))],
- IIC_SSE_CVT_PS_RR>, VEX;
+ IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
- (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>, VEX;
+ (int_x86_avx_cvt_ps2dq_256 (loadv8f32 addr:$src)))],
+ IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
- IIC_SSE_CVT_PS_RR>;
+ IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>;
+ IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
// Convert Packed Double FP to Packed DW Integers
def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
- VEX;
+ VEX, Sched<[WriteCvtF2I]>;
// XMM only
def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
- (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
+ (VCVTPD2DQrr VR128:$dst, VR128:$src), 0>;
def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
+ (int_x86_sse2_cvtpd2dq (loadv2f64 addr:$src)))]>, VEX,
+ Sched<[WriteCvtF2ILd]>;
// YMM only
def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
"vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX;
+ (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L,
+ Sched<[WriteCvtF2I]>;
def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
"vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
- VEX, VEX_L;
+ (int_x86_avx_cvt_pd2dq_256 (loadv4f64 addr:$src)))]>,
+ VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
- (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
+ (VCVTPD2DQYrr VR128:$dst, VR256:$src), 0>;
}
def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>;
+ IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>;
def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
- IIC_SSE_CVT_PD_RR>;
+ IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
// Convert with truncation packed single/double fp to doubleword
// SSE2 packed instructions with XS prefix
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvttps2dq VR128:$src))],
- IIC_SSE_CVT_PS_RR>, VEX;
+ IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>;
def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttps2dq
- (memopv4f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>, VEX;
+ (loadv4f32 addr:$src)))],
+ IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>;
def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
- IIC_SSE_CVT_PS_RR>, VEX;
+ IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
- (memopv8f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>, VEX;
+ (loadv8f32 addr:$src)))],
+ IIC_SSE_CVT_PS_RM>, VEX, VEX_L,
+ Sched<[WriteCvtF2ILd]>;
def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
- IIC_SSE_CVT_PS_RR>;
+ IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>;
def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
- IIC_SSE_CVT_PS_RM>;
+ IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>;
let Predicates = [HasAVX] in {
def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
(VCVTDQ2PSrr VR128:$src)>;
- def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
+ def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (loadv2i64 addr:$src)))),
(VCVTDQ2PSrm addr:$src)>;
def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
(VCVTDQ2PSrr VR128:$src)>;
- def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
+ def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (loadv2i64 addr:$src))),
(VCVTDQ2PSrm addr:$src)>;
def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
(VCVTTPS2DQrr VR128:$src)>;
- def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
+ def : Pat<(v4i32 (fp_to_sint (loadv4f32 addr:$src))),
(VCVTTPS2DQrm addr:$src)>;
def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
(VCVTDQ2PSYrr VR256:$src)>;
- def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
+ def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (loadv4i64 addr:$src)))),
(VCVTDQ2PSYrm addr:$src)>;
def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
(VCVTTPS2DQYrr VR256:$src)>;
- def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
+ def : Pat<(v8i32 (fp_to_sint (loadv8f32 addr:$src))),
(VCVTTPS2DQYrm addr:$src)>;
}
"cvttpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvttpd2dq VR128:$src))],
- IIC_SSE_CVT_PD_RR>, VEX;
+ IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>;
// The assembler can recognize rr 256-bit instructions by seeing a ymm
// register, but the same isn't true when using memory operands instead.
// XMM only
def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
- (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
+ (VCVTTPD2DQrr VR128:$dst, VR128:$src), 0>;
def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttpd2dqx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttpd2dq
- (memopv2f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, VEX;
+ (loadv2f64 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>;
// YMM only
def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
"cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
- IIC_SSE_CVT_PD_RR>, VEX;
+ IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>;
def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
"cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
+ (int_x86_avx_cvtt_pd2dq_256 (loadv4f64 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
- (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
+ (VCVTTPD2DQYrr VR128:$dst, VR256:$src), 0>;
let Predicates = [HasAVX] in {
def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
(VCVTTPD2DQYrr VR256:$src)>;
- def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
+ def : Pat<(v4i32 (fp_to_sint (loadv4f64 addr:$src))),
(VCVTTPD2DQYrm addr:$src)>;
} // Predicates = [HasAVX]
def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvttpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
- IIC_SSE_CVT_PD_RR>;
+ IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>;
def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
"cvttpd2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttpd2dq
(memopv2f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>;
+ IIC_SSE_CVT_PD_RM>,
+ Sched<[WriteCvtF2ILd]>;
// Convert packed single to packed double
let Predicates = [HasAVX] in {
def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
- IIC_SSE_CVT_PD_RR>, TB, VEX;
-let neverHasSideEffects = 1, mayLoad = 1 in
+ IIC_SSE_CVT_PD_RR>, PS, VEX, Sched<[WriteCvtF2F]>;
def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
- "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_CVT_PD_RM>, TB, VEX;
+ "vcvtps2pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, PS, VEX, Sched<[WriteCvtF2FLd]>;
def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
- IIC_SSE_CVT_PD_RR>, TB, VEX;
+ IIC_SSE_CVT_PD_RR>, PS, VEX, VEX_L, Sched<[WriteCvtF2F]>;
def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
- (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, TB, VEX;
+ (int_x86_avx_cvt_ps2_pd_256 (loadv4f32 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, PS, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
}
let Predicates = [UseSSE2] in {
def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
- IIC_SSE_CVT_PD_RR>, TB;
-let neverHasSideEffects = 1, mayLoad = 1 in
+ IIC_SSE_CVT_PD_RR>, PS, Sched<[WriteCvtF2F]>;
def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
- "cvtps2pd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_CVT_PD_RM>, TB;
+ "cvtps2pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, PS, Sched<[WriteCvtF2FLd]>;
}
// Convert Packed DW Integers to Packed Double FP
let neverHasSideEffects = 1, mayLoad = 1 in
def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
- []>, VEX;
+ []>, VEX, Sched<[WriteCvtI2FLd]>;
def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
+ (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX,
+ Sched<[WriteCvtI2F]>;
def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvtdq2_pd_256
- (bitconvert (memopv2i64 addr:$src))))]>, VEX;
+ (bitconvert (loadv2i64 addr:$src))))]>, VEX, VEX_L,
+ Sched<[WriteCvtI2FLd]>;
def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
- (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX;
+ (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L,
+ Sched<[WriteCvtI2F]>;
}
let neverHasSideEffects = 1, mayLoad = 1 in
def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"cvtdq2pd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_CVT_PD_RR>;
+ IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtdq2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
- IIC_SSE_CVT_PD_RM>;
+ IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>;
// AVX 256-bit register conversion intrinsics
let Predicates = [HasAVX] in {
def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
(VCVTDQ2PDYrr VR128:$src)>;
- def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
+ def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (loadv2i64 addr:$src)))),
(VCVTDQ2PDYrm addr:$src)>;
} // Predicates = [HasAVX]
def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
- IIC_SSE_CVT_PD_RR>, VEX;
+ IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>;
// XMM only
def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
- (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
+ (VCVTPD2PSrr VR128:$dst, VR128:$src), 0>;
def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtpd2psx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, VEX;
+ (int_x86_sse2_cvtpd2ps (loadv2f64 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>;
// YMM only
def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
"cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
- IIC_SSE_CVT_PD_RR>, VEX;
+ IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>;
def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
"cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
+ (int_x86_avx_cvt_pd2_ps_256 (loadv4f64 addr:$src)))],
+ IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
- (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
+ (VCVTPD2PSYrr VR128:$dst, VR256:$src), 0>;
def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
- IIC_SSE_CVT_PD_RR>;
+ IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>;
def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
- IIC_SSE_CVT_PD_RM>;
+ IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>;
// AVX 256-bit register conversion intrinsics
let Predicates = [HasAVX] in {
def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
(VCVTDQ2PSYrr VR256:$src)>;
- def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
+ def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (loadv4i64 addr:$src))),
(VCVTDQ2PSYrm addr:$src)>;
// Match fround and fextend for 128/256-bit conversions
+ def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
+ (VCVTPD2PSrr VR128:$src)>;
+ def : Pat<(v4f32 (X86vfpround (loadv2f64 addr:$src))),
+ (VCVTPD2PSXrm addr:$src)>;
def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
(VCVTPD2PSYrr VR256:$src)>;
def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
(VCVTPS2PDrr VR128:$src)>;
def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
(VCVTPS2PDYrr VR128:$src)>;
- def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
+ def : Pat<(v4f64 (extloadv4f32 addr:$src)),
(VCVTPS2PDYrm addr:$src)>;
}
let Predicates = [UseSSE2] in {
- // Match fextend for 128 conversions
+ // Match fround and fextend for 128 conversions
+ def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
+ (CVTPD2PSrr VR128:$src)>;
+ def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
+ (CVTPD2PSrm addr:$src)>;
+
def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
(CVTPS2PDrr VR128:$src)>;
}
// sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
- Operand CC, SDNode OpNode, ValueType VT,
+ Operand CC, SDNode OpNode, ValueType VT,
PatFrag ld_frag, string asm, string asm_alt,
OpndItins itins> {
def rr : SIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
[(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def rm : SIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
[(set RC:$dst, (OpNode (VT RC:$src1),
(ld_frag addr:$src2), imm:$cc))],
- itins.rm>;
+ itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
// Accept explicit immediate argument form instead of comparison code.
- let neverHasSideEffects = 1 in {
+ let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
- IIC_SSE_ALU_F32S_RR>;
+ IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>;
let mayLoad = 1 in
def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
- IIC_SSE_ALU_F32S_RM>;
+ IIC_SSE_ALU_F32S_RM>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
-defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
+defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmps, f32, loadf32,
"cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
SSE_ALU_F32S>,
XS, VEX_4V, VEX_LIG;
-defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
+defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmps, f64, loadf64,
"cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
SSE_ALU_F32S>, // same latency as 32 bit compare
XD, VEX_4V, VEX_LIG;
let Constraints = "$src1 = $dst" in {
- defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
+ defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmps, f32, loadf32,
"cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
"cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
XS;
- defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
+ defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmps, f64, loadf64,
"cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
"cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSE_ALU_F32S>, // same latency as 32 bit compare
+ SSE_ALU_F64S>,
XD;
}
(ins VR128:$src1, VR128:$src, CC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
VR128:$src, imm:$cc))],
- itins.rr>;
+ itins.rr>,
+ Sched<[itins.Sched]>;
def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, x86memop:$src, CC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
(load addr:$src), imm:$cc))],
- itins.rm>;
-}
-
-// Aliases to match intrinsics which expect XMM operand(s).
-defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
- "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
- SSE_ALU_F32S>,
- XS, VEX_4V;
-defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
- "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
- SSE_ALU_F32S>, // same latency as f32
- XD, VEX_4V;
-let Constraints = "$src1 = $dst" in {
- defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
- "cmp${cc}ss\t{$src, $dst|$dst, $src}",
- SSE_ALU_F32S>, XS;
- defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
- "cmp${cc}sd\t{$src, $dst|$dst, $src}",
+ itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+let isCodeGenOnly = 1 in {
+ // Aliases to match intrinsics which expect XMM operand(s).
+ defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
+ "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
+ SSE_ALU_F32S>,
+ XS, VEX_4V;
+ defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
+ "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
SSE_ALU_F32S>, // same latency as f32
- XD;
+ XD, VEX_4V;
+ let Constraints = "$src1 = $dst" in {
+ defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
+ "cmp${cc}ss\t{$src, $dst|$dst, $src}",
+ SSE_ALU_F32S>, XS;
+ defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
+ "cmp${cc}sd\t{$src, $dst|$dst, $src}",
+ SSE_ALU_F64S>,
+ XD;
+}
}
// sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
ValueType vt, X86MemOperand x86memop,
- PatFrag ld_frag, string OpcodeStr, Domain d> {
- def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
+ PatFrag ld_frag, string OpcodeStr> {
+ def rr: SI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
[(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
- IIC_SSE_COMIS_RR, d>;
- def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
+ IIC_SSE_COMIS_RR>,
+ Sched<[WriteFAdd]>;
+ def rm: SI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
[(set EFLAGS, (OpNode (vt RC:$src1),
(ld_frag addr:$src2)))],
- IIC_SSE_COMIS_RM, d>;
+ IIC_SSE_COMIS_RM>,
+ Sched<[WriteFAddLd, ReadAfterLd]>;
}
let Defs = [EFLAGS] in {
defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
- "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
+ "ucomiss">, PS, VEX, VEX_LIG;
defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
- "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
- VEX_LIG;
+ "ucomisd">, PD, VEX, VEX_LIG;
let Pattern = []<dag> in {
defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
- "comiss", SSEPackedSingle>, TB, VEX,
- VEX_LIG;
+ "comiss">, PS, VEX, VEX_LIG;
defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
- "comisd", SSEPackedDouble>, TB, OpSize, VEX,
- VEX_LIG;
+ "comisd">, PD, VEX, VEX_LIG;
}
- defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
- load, "ucomiss", SSEPackedSingle>, TB, VEX;
- defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
- load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
-
- defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
- load, "comiss", SSEPackedSingle>, TB, VEX;
- defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
- load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
+ let isCodeGenOnly = 1 in {
+ defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
+ load, "ucomiss">, PS, VEX;
+ defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
+ load, "ucomisd">, PD, VEX;
+
+ defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
+ load, "comiss">, PS, VEX;
+ defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
+ load, "comisd">, PD, VEX;
+ }
defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
- "ucomiss", SSEPackedSingle>, TB;
+ "ucomiss">, PS;
defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
- "ucomisd", SSEPackedDouble>, TB, OpSize;
+ "ucomisd">, PD;
let Pattern = []<dag> in {
defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
- "comiss", SSEPackedSingle>, TB;
+ "comiss">, PS;
defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
- "comisd", SSEPackedDouble>, TB, OpSize;
+ "comisd">, PD;
}
- defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
- load, "ucomiss", SSEPackedSingle>, TB;
- defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
- load, "ucomisd", SSEPackedDouble>, TB, OpSize;
-
- defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
- "comiss", SSEPackedSingle>, TB;
- defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
- "comisd", SSEPackedDouble>, TB, OpSize;
+ let isCodeGenOnly = 1 in {
+ defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
+ load, "ucomiss">, PS;
+ defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
+ load, "ucomisd">, PD;
+
+ defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
+ "comiss">, PS;
+ defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
+ "comisd">, PD;
+ }
} // Defs = [EFLAGS]
// sse12_cmp_packed - sse 1 & 2 compare packed instructions
multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
- Operand CC, Intrinsic Int, string asm,
- string asm_alt, Domain d> {
+ Operand CC, Intrinsic Int, string asm,
+ string asm_alt, Domain d,
+ OpndItins itins = SSE_ALU_F32P> {
def rri : PIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
[(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
- IIC_SSE_CMPP_RR, d>;
+ itins.rr, d>,
+ Sched<[WriteFAdd]>;
def rmi : PIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
[(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
- IIC_SSE_CMPP_RM, d>;
+ itins.rm, d>,
+ Sched<[WriteFAddLd, ReadAfterLd]>;
// Accept explicit immediate argument form instead of comparison code.
- let neverHasSideEffects = 1 in {
+ let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : PIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
- asm_alt, [], IIC_SSE_CMPP_RR, d>;
+ asm_alt, [], itins.rr, d>, Sched<[WriteFAdd]>;
def rmi_alt : PIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
- asm_alt, [], IIC_SSE_CMPP_RM, d>;
+ asm_alt, [], itins.rm, d>,
+ Sched<[WriteFAddLd, ReadAfterLd]>;
}
}
defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
"cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedSingle>, TB, VEX_4V;
+ SSEPackedSingle>, PS, VEX_4V;
defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
"cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
+ SSEPackedDouble>, PD, VEX_4V;
defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
"cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedSingle>, TB, VEX_4V;
+ SSEPackedSingle>, PS, VEX_4V, VEX_L;
defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
"cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
+ SSEPackedDouble>, PD, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in {
defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
"cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
"cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSEPackedSingle>, TB;
+ SSEPackedSingle, SSE_ALU_F32P>, PS;
defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
"cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
"cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSEPackedDouble>, TB, OpSize;
+ SSEPackedDouble, SSE_ALU_F64P>, PD;
}
let Predicates = [HasAVX] in {
// SSE 1 & 2 - Shuffle Instructions
//===----------------------------------------------------------------------===//
-/// sse12_shuffle - sse 1 & 2 shuffle instructions
+/// sse12_shuffle - sse 1 & 2 fp shuffle instructions
multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
ValueType vt, string asm, PatFrag mem_frag,
- Domain d, bit IsConvertibleToThreeAddress = 0> {
+ Domain d> {
def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
[(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
- (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
- let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
- def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
- [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
- (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
+ (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
+ def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
+ [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
+ (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
+ Sched<[WriteFShuffle]>;
}
defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
"shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- memopv4f32, SSEPackedSingle>, TB, VEX_4V;
+ loadv4f32, SSEPackedSingle>, PS, VEX_4V;
defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
"shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- memopv8f32, SSEPackedSingle>, TB, VEX_4V;
+ loadv8f32, SSEPackedSingle>, PS, VEX_4V, VEX_L;
defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
- "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
- memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
+ "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ loadv2f64, SSEPackedDouble>, PD, VEX_4V;
defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
- "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
- memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
+ "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
+ loadv4f64, SSEPackedDouble>, PD, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in {
defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
"shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
- TB;
+ memopv4f32, SSEPackedSingle>, PS;
defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
"shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
- TB, OpSize;
+ memopv2f64, SSEPackedDouble>, PD;
}
let Predicates = [HasAVX] in {
def : Pat<(v4i32 (X86Shufp VR128:$src1,
- (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v4i32 (loadv2i64 addr:$src2)), (i8 imm:$imm))),
(VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v2i64 (X86Shufp VR128:$src1,
- (memopv2i64 addr:$src2), (i8 imm:$imm))),
+ (loadv2i64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v8i32 (X86Shufp VR256:$src1,
- (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v8i32 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
(VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86Shufp VR256:$src1,
- (memopv4i64 addr:$src2), (i8 imm:$imm))),
+ (loadv4i64 addr:$src2), (i8 imm:$imm))),
(VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
}
}
//===----------------------------------------------------------------------===//
-// SSE 1 & 2 - Unpack Instructions
+// SSE 1 & 2 - Unpack FP Instructions
//===----------------------------------------------------------------------===//
-/// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
+/// sse12_unpack_interleave - sse 1 & 2 fp unpack and interleave
multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
PatFrag mem_frag, RegisterClass RC,
X86MemOperand x86memop, string asm,
(outs RC:$dst), (ins RC:$src1, RC:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1, RC:$src2)))],
- IIC_SSE_UNPCK, d>;
+ IIC_SSE_UNPCK, d>, Sched<[WriteFShuffle]>;
def rm : PI<opc, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1,
(mem_frag addr:$src2))))],
- IIC_SSE_UNPCK, d>;
+ IIC_SSE_UNPCK, d>,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
-defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
+defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, loadv4f32,
VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
-defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
+ SSEPackedSingle>, PS, VEX_4V;
+defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, loadv2f64,
VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
-defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
+ SSEPackedDouble>, PD, VEX_4V;
+defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, loadv4f32,
VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
-defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
+ SSEPackedSingle>, PS, VEX_4V;
+defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, loadv2f64,
VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
+ SSEPackedDouble>, PD, VEX_4V;
-defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
+defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, loadv8f32,
VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
-defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
+ SSEPackedSingle>, PS, VEX_4V, VEX_L;
+defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, loadv4f64,
VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
-defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
+ SSEPackedDouble>, PD, VEX_4V, VEX_L;
+defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, loadv8f32,
VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
-defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
+ SSEPackedSingle>, PS, VEX_4V, VEX_L;
+defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, loadv4f64,
VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedDouble>, TB, OpSize, VEX_4V;
+ SSEPackedDouble>, PD, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in {
defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
- SSEPackedDouble>, TB, OpSize;
+ SSEPackedDouble>, PD;
defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
- SSEPackedDouble>, TB, OpSize;
+ SSEPackedDouble>, PD;
} // Constraints = "$src1 = $dst"
let Predicates = [HasAVX1Only] in {
- def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
+ def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
(VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
(VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
- def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
+ def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
(VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
(VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
- def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
+ def : Pat<(v4i64 (X86Unpckl VR256:$src1, (loadv4i64 addr:$src2))),
(VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
(VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
- def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
+ def : Pat<(v4i64 (X86Unpckh VR256:$src1, (loadv4i64 addr:$src2))),
(VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
(VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
/// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
Domain d> {
- def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
- [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
- def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
- IIC_SSE_MOVMSK, d>, REX_W;
+ def rr : PI<0x50, MRMSrcReg, (outs GR32orGR64:$dst), (ins RC:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
+ [(set GR32orGR64:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>,
+ Sched<[WriteVecLogic]>;
}
let Predicates = [HasAVX] in {
defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
- "movmskps", SSEPackedSingle>, TB, VEX;
+ "movmskps", SSEPackedSingle>, PS, VEX;
defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
- "movmskpd", SSEPackedDouble>, TB,
- OpSize, VEX;
+ "movmskpd", SSEPackedDouble>, PD, VEX;
defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
- "movmskps", SSEPackedSingle>, TB, VEX;
+ "movmskps", SSEPackedSingle>, PS,
+ VEX, VEX_L;
defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
- "movmskpd", SSEPackedDouble>, TB,
- OpSize, VEX;
+ "movmskpd", SSEPackedDouble>, PD,
+ VEX, VEX_L;
def : Pat<(i32 (X86fgetsign FR32:$src)),
- (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
def : Pat<(i64 (X86fgetsign FR32:$src)),
- (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ (SUBREG_TO_REG (i64 0),
+ (VMOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>;
def : Pat<(i32 (X86fgetsign FR64:$src)),
- (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
def : Pat<(i64 (X86fgetsign FR64:$src)),
- (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
-
- // Assembler Only
- def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
- "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
- SSEPackedSingle>, TB, VEX;
- def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
- "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
- SSEPackedDouble>, TB,
- OpSize, VEX;
- def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
- "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
- SSEPackedSingle>, TB, VEX;
- def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
- "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
- SSEPackedDouble>, TB,
- OpSize, VEX;
+ (SUBREG_TO_REG (i64 0),
+ (VMOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>;
}
defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
- SSEPackedDouble>, TB, OpSize;
+ SSEPackedDouble>, PD;
def : Pat<(i32 (X86fgetsign FR32:$src)),
- (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
+ (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128))>,
Requires<[UseSSE1]>;
def : Pat<(i64 (X86fgetsign FR32:$src)),
- (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
+ (SUBREG_TO_REG (i64 0),
+ (MOVMSKPSrr (COPY_TO_REGCLASS FR32:$src, VR128)), sub_32bit)>,
Requires<[UseSSE1]>;
def : Pat<(i32 (X86fgetsign FR64:$src)),
- (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
+ (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128))>,
Requires<[UseSSE2]>;
def : Pat<(i64 (X86fgetsign FR64:$src)),
- (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
+ (SUBREG_TO_REG (i64 0),
+ (MOVMSKPDrr (COPY_TO_REGCLASS FR64:$src, VR128)), sub_32bit)>,
Requires<[UseSSE2]>;
//===---------------------------------------------------------------------===//
/// PDI_binop_rm - Simple SSE2 binary operator.
multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
- X86MemOperand x86memop,
- OpndItins itins,
- bit IsCommutable = 0,
- bit Is2Addr = 1> {
+ X86MemOperand x86memop, OpndItins itins,
+ bit IsCommutable, bit Is2Addr> {
let isCommutable = IsCommutable in
def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
+ [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (OpVT (OpNode RC:$src1,
(bitconvert (memop_frag addr:$src2)))))],
- itins.rm>;
+ itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
} // ExeDomain = SSEPackedInt
-// These are ordered here for pattern ordering requirements with the fp versions
+multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
+ ValueType OpVT128, ValueType OpVT256,
+ OpndItins itins, bit IsCommutable = 0> {
+let Predicates = [HasAVX] in
+ defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
+ VR128, loadv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
-let Predicates = [HasAVX] in {
-defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
+let Constraints = "$src1 = $dst" in
+ defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
+ memopv2i64, i128mem, itins, IsCommutable, 1>;
+
+let Predicates = [HasAVX2] in
+ defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
+ OpVT256, VR256, loadv4i64, i256mem, itins,
+ IsCommutable, 0>, VEX_4V, VEX_L;
}
-let Constraints = "$src1 = $dst" in {
-defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1>;
-defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1>;
-defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 1>;
-defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
- i128mem, SSE_BIT_ITINS_P, 0>;
-} // Constraints = "$src1 = $dst"
+// These are ordered here for pattern ordering requirements with the fp versions
-let Predicates = [HasAVX2] in {
-defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
- i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
- i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
- i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
-defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
- i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
-}
+defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
+defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
+defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
+defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 0>;
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Logical Instructions
SDNode OpNode, OpndItins itins> {
defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
- TB, VEX_4V;
+ PS, VEX_4V;
defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
- TB, OpSize, VEX_4V;
+ PD, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
- TB;
+ PS;
defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
- TB, OpSize;
+ PD;
}
}
// Alias bitwise logical operations using SSE logical ops on packed FP values.
-defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
- SSE_BIT_ITINS_P>;
-defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
- SSE_BIT_ITINS_P>;
-defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
- SSE_BIT_ITINS_P>;
-
-let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
- defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
+let isCodeGenOnly = 1 in {
+ defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
SSE_BIT_ITINS_P>;
+ defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
+ SSE_BIT_ITINS_P>;
+ defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
+ SSE_BIT_ITINS_P>;
+
+ let isCommutable = 0 in
+ defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", X86fandn,
+ SSE_BIT_ITINS_P>;
+}
/// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
///
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
SDNode OpNode> {
+ defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
+ !strconcat(OpcodeStr, "ps"), f256mem,
+ [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
+ (loadv4i64 addr:$src2)))], 0>, PS, VEX_4V, VEX_L;
+
+ defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
+ !strconcat(OpcodeStr, "pd"), f256mem,
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+ (bc_v4i64 (v4f64 VR256:$src2))))],
+ [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
+ (loadv4i64 addr:$src2)))], 0>,
+ PD, VEX_4V, VEX_L;
+
// In AVX no need to add a pattern for 128-bit logical rr ps, because they
// are all promoted to v2i64, and the patterns are covered by the int
// version. This is needed in SSE only, because v2i64 isn't supported on
defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f128mem, [],
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
+ (loadv2i64 addr:$src2)))], 0>, PS, VEX_4V;
defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f128mem,
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
(bc_v2i64 (v2f64 VR128:$src2))))],
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))], 0>,
- TB, OpSize, VEX_4V;
+ (loadv2i64 addr:$src2)))], 0>,
+ PD, VEX_4V;
+
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f128mem,
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>, TB;
+ (memopv2i64 addr:$src2)))]>, PS;
defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f128mem,
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
(bc_v2i64 (v2f64 VR128:$src2))))],
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>, TB, OpSize;
+ (memopv2i64 addr:$src2)))]>, PD;
}
}
-/// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
-///
-multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
- SDNode OpNode> {
- defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
- !strconcat(OpcodeStr, "ps"), f256mem,
- [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
- [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
- (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
-
- defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
- !strconcat(OpcodeStr, "pd"), f256mem,
- [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
- (bc_v4i64 (v4f64 VR256:$src2))))],
- [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
- (memopv4i64 addr:$src2)))], 0>,
- TB, OpSize, VEX_4V;
-}
-
-// AVX 256-bit packed logical ops forms
-defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
-defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
-defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
-defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
-
defm AND : sse12_fp_packed_logical<0x54, "and", and>;
defm OR : sse12_fp_packed_logical<0x56, "or", or>;
defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
let isCommutable = 0 in
defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
+// AVX1 requires type coercions in order to fold loads directly into logical
+// operations.
+let Predicates = [HasAVX1Only] in {
+ def : Pat<(bc_v8f32 (and VR256:$src1, (loadv4i64 addr:$src2))),
+ (VANDPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (or VR256:$src1, (loadv4i64 addr:$src2))),
+ (VORPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (xor VR256:$src1, (loadv4i64 addr:$src2))),
+ (VXORPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (X86andnp VR256:$src1, (loadv4i64 addr:$src2))),
+ (VANDNPSYrm VR256:$src1, addr:$src2)>;
+}
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Arithmetic Instructions
//===----------------------------------------------------------------------===//
/// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
/// classes below
-multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
- SizeItins itins,
- bit Is2Addr = 1> {
- defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
- OpNode, FR32, f32mem,
- itins.s, Is2Addr>, XS;
- defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
- OpNode, FR64, f64mem,
- itins.d, Is2Addr>, XD;
-}
-
-multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
- SizeItins itins,
- bit Is2Addr = 1> {
- defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
- v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
- TB;
- defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
- v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
- TB, OpSize;
-}
-
-multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
- SDNode OpNode,
- SizeItins itins> {
- defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
- v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
- TB;
- defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
- v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
- TB, OpSize;
-}
+multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
+ SDNode OpNode, SizeItins itins> {
+ defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
+ VR128, v4f32, f128mem, loadv4f32,
+ SSEPackedSingle, itins.s, 0>, PS, VEX_4V;
+ defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
+ VR128, v2f64, f128mem, loadv2f64,
+ SSEPackedDouble, itins.d, 0>, PD, VEX_4V;
-multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
- SizeItins itins,
- bit Is2Addr = 1> {
- defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
- !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
- itins.s, Is2Addr>, XS;
- defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
- !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
- itins.d, Is2Addr>, XD;
+ defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
+ OpNode, VR256, v8f32, f256mem, loadv8f32,
+ SSEPackedSingle, itins.s, 0>, PS, VEX_4V, VEX_L;
+ defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
+ OpNode, VR256, v4f64, f256mem, loadv4f64,
+ SSEPackedDouble, itins.d, 0>, PD, VEX_4V, VEX_L;
+
+ let Constraints = "$src1 = $dst" in {
+ defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
+ v4f32, f128mem, memopv4f32, SSEPackedSingle,
+ itins.s>, PS;
+ defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
+ v2f64, f128mem, memopv2f64, SSEPackedDouble,
+ itins.d>, PD;
+ }
}
-multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
- SizeItins itins,
- bit Is2Addr = 1> {
- defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
- !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
- SSEPackedSingle, itins.s, Is2Addr>,
- TB;
+multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ SizeItins itins> {
+ defm V#NAME#SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
+ OpNode, FR32, f32mem, itins.s, 0>, XS, VEX_4V, VEX_LIG;
+ defm V#NAME#SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
+ OpNode, FR64, f64mem, itins.d, 0>, XD, VEX_4V, VEX_LIG;
- defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
- !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
- SSEPackedDouble, itins.d, Is2Addr>,
- TB, OpSize;
+ let Constraints = "$src1 = $dst" in {
+ defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
+ OpNode, FR32, f32mem, itins.s>, XS;
+ defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
+ OpNode, FR64, f64mem, itins.d>, XD;
+ }
}
-multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
- SizeItins itins> {
- defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
- !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
- SSEPackedSingle, itins.s, 0>, TB;
+multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
+ SizeItins itins> {
+ defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
+ itins.s, 0>, XS, VEX_4V, VEX_LIG;
+ defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
+ itins.d, 0>, XD, VEX_4V, VEX_LIG;
- defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
- !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
- SSEPackedDouble, itins.d, 0>, TB, OpSize;
+ let Constraints = "$src1 = $dst" in {
+ defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
+ itins.s>, XS;
+ defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
+ itins.d>, XD;
+ }
}
// Binary Arithmetic instructions
-defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
-defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
- VEX_4V;
-defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
-defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
- VEX_4V;
-
+defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
+defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
+ basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
let isCommutable = 0 in {
- defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
- VEX_4V;
- defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
- VEX_4V;
- defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
- VEX_4V;
- defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
- basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
- VEX_4V;
+ defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
+ defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
+ defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
+ defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
+ basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
}
-let Constraints = "$src1 = $dst" in {
- defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
- defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
- basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
+let isCodeGenOnly = 1 in {
+ defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
+ defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>,
+ basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
+}
- let isCommutable = 0 in {
- defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
- defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
- basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
- defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
- defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
- basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
- }
+// Patterns used to select SSE scalar fp arithmetic instructions from
+// a scalar fp operation followed by a blend.
+//
+// These patterns know, for example, how to select an ADDSS from a
+// float add plus vector insert.
+//
+// The effect is that the backend no longer emits unnecessary vector
+// insert instructions immediately after SSE scalar fp instructions
+// like addss or mulss.
+//
+// For example, given the following code:
+// __m128 foo(__m128 A, __m128 B) {
+// A[0] += B[0];
+// return A;
+// }
+//
+// previously we generated:
+// addss %xmm0, %xmm1
+// movss %xmm1, %xmm0
+//
+// we now generate:
+// addss %xmm1, %xmm0
+
+let Predicates = [UseSSE1] in {
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))))),
+ (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))))),
+ (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))))),
+ (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))))),
+ (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
}
-let isCodeGenOnly = 1 in {
- defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>, VEX_4V;
- defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
- VEX_4V, VEX_LIG;
- defm VMINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P, 0>,
- basic_sse12_fp_binop_p_y<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>, VEX_4V;
- let Constraints = "$src1 = $dst" in {
- defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
- defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>,
- basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
- }
+let Predicates = [UseSSE2] in {
+ // SSE2 patterns to select scalar double-precision fp arithmetic instructions
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+}
+
+let Predicates = [UseSSE41] in {
+ // If the subtarget has SSE4.1 but not AVX, the vector insert instruction is
+ // lowered into a X86insertps or a X86Blendi rather than a X86Movss. When
+ // selecting SSE scalar single-precision fp arithmetic instructions, make
+ // sure that we correctly match them.
+
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+}
+
+let Predicates = [HasAVX] in {
+ // The following patterns select AVX Scalar single/double precision fp
+ // arithmetic instructions.
+
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))))),
+ (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ (fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (iPTR 0))),
+ (VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+}
+
+// Patterns used to select SSE scalar fp arithmetic instructions from
+// a vector packed single/double fp operation followed by a vector insert.
+//
+// The effect is that the backend converts the packed fp instruction
+// followed by a vector insert into a single SSE scalar fp instruction.
+//
+// For example, given the following code:
+// __m128 foo(__m128 A, __m128 B) {
+// __m128 C = A + B;
+// return (__m128) {c[0], a[1], a[2], a[3]};
+// }
+//
+// previously we generated:
+// addps %xmm0, %xmm1
+// movss %xmm1, %xmm0
+//
+// we now generate:
+// addss %xmm1, %xmm0
+
+let Predicates = [UseSSE1] in {
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (ADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (SUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (MULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (DIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+}
+
+let Predicates = [UseSSE2] in {
+ // SSE2 patterns to select scalar double-precision fp arithmetic instructions
+ // from a packed double-precision fp instruction plus movsd.
+
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+}
+
+let Predicates = [UseSSE41] in {
+ // With SSE4.1 we may see these operations using X86Blendi rather than
+ // X86Movs{s,d}.
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (ADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (SUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (MULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (DIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+}
+
+let Predicates = [HasAVX] in {
+ // The following patterns select AVX Scalar single/double precision fp
+ // arithmetic instructions from a packed single precision fp instruction
+ // plus movss/movsd.
+
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (VADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (VSUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (VMULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
+ (VDIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
+ (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ // Also handle X86Blendi-based patterns.
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VSUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VMULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VDIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
}
/// Unop Arithmetic
///
/// And, we have a special variant form for a full-vector intrinsic form.
-def SSE_SQRTP : OpndItins<
- IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
+let Sched = WriteFSqrt in {
+def SSE_SQRTPS : OpndItins<
+ IIC_SSE_SQRTPS_RR, IIC_SSE_SQRTPS_RM
+>;
+
+def SSE_SQRTSS : OpndItins<
+ IIC_SSE_SQRTSS_RR, IIC_SSE_SQRTSS_RM
+>;
+
+def SSE_SQRTPD : OpndItins<
+ IIC_SSE_SQRTPD_RR, IIC_SSE_SQRTPD_RM
+>;
+
+def SSE_SQRTSD : OpndItins<
+ IIC_SSE_SQRTSD_RR, IIC_SSE_SQRTSD_RM
>;
+}
-def SSE_SQRTS : OpndItins<
- IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
+let Sched = WriteFRsqrt in {
+def SSE_RSQRTPS : OpndItins<
+ IIC_SSE_RSQRTPS_RR, IIC_SSE_RSQRTPS_RM
>;
+def SSE_RSQRTSS : OpndItins<
+ IIC_SSE_RSQRTSS_RR, IIC_SSE_RSQRTSS_RM
+>;
+}
+
+let Sched = WriteFRcp in {
def SSE_RCPP : OpndItins<
IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
>;
def SSE_RCPS : OpndItins<
IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
>;
+}
/// sse1_fp_unop_s - SSE1 unops in scalar form.
multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+ (ins FR32:$src1, FR32:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
+ let mayLoad = 1 in {
+ def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+ (ins FR32:$src1,f32mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ let isCodeGenOnly = 1 in
+ def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ }
+}
+
def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode FR32:$src))]>;
+ [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
// For scalar unary operations, fold a load into the operation
// only in OptForSize mode. It eliminates an instruction, but it also
// eliminates a whole-register clobber (the load), so it introduces a
def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
[(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
- Requires<[UseSSE1, OptForSize]>;
+ Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
+let isCodeGenOnly = 1 in {
def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
+ [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
+ [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
+ Sched<[itins.Sched.Folded]>;
+}
}
-/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
- def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
+multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+ (ins FR32:$src1, FR32:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
let mayLoad = 1 in {
- def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+ (ins FR32:$src1,f32mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ let isCodeGenOnly = 1 in
+ def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ }
+}
+
+ def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
+ // For scalar unary operations, fold a load into the operation
+ // only in OptForSize mode. It eliminates an instruction, but it also
+ // eliminates a whole-register clobber (the load), so it introduces a
+ // partial register update condition.
+ def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
+ Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
+ let isCodeGenOnly = 1, Constraints = "$src1 = $dst" in {
+ def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rr>, Sched<[itins.Sched]>;
+ let mayLoad = 1, hasSideEffects = 0 in
+ def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
/// sse1_fp_unop_p - SSE1 unops in packed form.
multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
OpndItins itins> {
- def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
- def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
+let Predicates = [HasAVX] in {
+ def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
+ itins.rr>, VEX, Sched<[itins.Sched]>;
+ def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (OpNode (loadv4f32 addr:$src)))],
+ itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+ def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
+ itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+ def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (OpNode (loadv8f32 addr:$src)))],
+ itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
}
-/// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
-multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
- OpndItins itins> {
- def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
- itins.rr>;
- def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
- itins.rm>;
+ [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>,
+ Sched<[itins.Sched]>;
+ def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>,
+ Sched<[itins.Sched.Folded]>;
}
/// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
- Intrinsic V4F32Int, OpndItins itins> {
+ Intrinsic V4F32Int, Intrinsic V8F32Int,
+ OpndItins itins> {
+let isCodeGenOnly = 1 in {
+let Predicates = [HasAVX] in {
+ def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (V4F32Int VR128:$src))],
+ itins.rr>, VEX, Sched<[itins.Sched]>;
+ def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (V4F32Int (loadv4f32 addr:$src)))],
+ itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+ def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (V8F32Int VR256:$src))],
+ itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+ def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
+ (ins f256mem:$src),
+ !strconcat("v", OpcodeStr,
+ "ps\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (V8F32Int (loadv8f32 addr:$src)))],
+ itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
+}
+
def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int VR128:$src))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
- itins.rm>;
-}
-
-/// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
-multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
- Intrinsic V4F32Int, OpndItins itins> {
- def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (V4F32Int VR256:$src))],
- itins.rr>;
- def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded]>;
+} // isCodeGenOnly = 1
}
/// sse2_fp_unop_s - SSE2 unops in scalar form.
multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
+ (ins FR64:$src1, FR64:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
+ let mayLoad = 1 in {
+ def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
+ (ins FR64:$src1,f64mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ let isCodeGenOnly = 1 in
+ def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, sdmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+ }
+}
+
def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
- [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
+ [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
// See the comments in sse1_fp_unop_s for why this is OptForSize.
def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
[(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
- Requires<[UseSSE2, OptForSize]>;
+ Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>;
+let isCodeGenOnly = 1 in {
def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
+ [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
+ [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>,
+ Sched<[itins.Sched.Folded]>;
}
-
-/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
-let hasSideEffects = 0 in
-multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
- def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- let mayLoad = 1 in {
- def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, sdmem:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- }
}
/// sse2_fp_unop_p - SSE2 unops in vector forms.
multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
SDNode OpNode, OpndItins itins> {
- def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
- def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
+let Predicates = [HasAVX] in {
+ def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ !strconcat("v", OpcodeStr,
+ "pd\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
+ itins.rr>, VEX, Sched<[itins.Sched]>;
+ def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ !strconcat("v", OpcodeStr,
+ "pd\t{$src, $dst|$dst, $src}"),
+ [(set VR128:$dst, (OpNode (loadv2f64 addr:$src)))],
+ itins.rm>, VEX, Sched<[itins.Sched.Folded]>;
+ def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ !strconcat("v", OpcodeStr,
+ "pd\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
+ itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>;
+ def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ !strconcat("v", OpcodeStr,
+ "pd\t{$src, $dst|$dst, $src}"),
+ [(set VR256:$dst, (OpNode (loadv4f64 addr:$src)))],
+ itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>;
}
-/// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
-multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
- OpndItins itins> {
- def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
+ def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
- itins.rr>;
- def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
+ [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>,
+ Sched<[itins.Sched]>;
+ def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
!strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
- itins.rm>;
-}
-
-/// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
-multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
- Intrinsic V2F64Int, OpndItins itins> {
- def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (V2F64Int VR128:$src))],
- itins.rr>;
- def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
- itins.rm>;
-}
-
-/// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
-multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
- Intrinsic V2F64Int, OpndItins itins> {
- def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (V2F64Int VR256:$src))],
- itins.rr>;
- def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
- itins.rm>;
+ [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>,
+ Sched<[itins.Sched.Folded]>;
}
-let Predicates = [HasAVX] in {
- // Square root.
- defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
- sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
-
- defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
- sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
- sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
- sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
- sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
- SSE_SQRTP>,
- sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
- SSE_SQRTP>,
- sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
- SSE_SQRTP>,
- sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
- SSE_SQRTP>,
- VEX;
-
- // Reciprocal approximations. Note that these typically require refinement
- // in order to obtain suitable precision.
- defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
- defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
- sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
- sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
- SSE_SQRTP>,
- sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
- SSE_SQRTP>, VEX;
-
- defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
- defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
- sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
- sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
- SSE_RCPP>,
- sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
- SSE_RCPP>, VEX;
-}
-
-def : Pat<(f32 (fsqrt FR32:$src)),
- (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
-def : Pat<(f32 (fsqrt (load addr:$src))),
- (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
-def : Pat<(f64 (fsqrt FR64:$src)),
- (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
-def : Pat<(f64 (fsqrt (load addr:$src))),
- (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
-
-def : Pat<(f32 (X86frsqrt FR32:$src)),
- (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
-def : Pat<(f32 (X86frsqrt (load addr:$src))),
- (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
-
-def : Pat<(f32 (X86frcp FR32:$src)),
- (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
-def : Pat<(f32 (X86frcp (load addr:$src))),
- (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[HasAVX, OptForSize]>;
+// Square root.
+defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
+ SSE_SQRTSS>,
+ sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
+ sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
+ SSE_SQRTSD>,
+ sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
-let Predicates = [HasAVX] in {
+// Reciprocal approximations. Note that these typically require refinement
+// in order to obtain suitable precision.
+defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_RSQRTSS>,
+ sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_RSQRTPS>,
+ sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
+ int_x86_avx_rsqrt_ps_256, SSE_RSQRTPS>;
+defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
+ sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
+ sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
+ int_x86_avx_rcp_ps_256, SSE_RCPP>;
+
+let Predicates = [UseAVX] in {
+ def : Pat<(f32 (fsqrt FR32:$src)),
+ (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+ def : Pat<(f32 (fsqrt (load addr:$src))),
+ (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+ def : Pat<(f64 (fsqrt FR64:$src)),
+ (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
+ def : Pat<(f64 (fsqrt (load addr:$src))),
+ (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+ def : Pat<(f32 (X86frsqrt FR32:$src)),
+ (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+ def : Pat<(f32 (X86frsqrt (load addr:$src))),
+ (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+
+ def : Pat<(f32 (X86frcp FR32:$src)),
+ (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
+ def : Pat<(f32 (X86frcp (load addr:$src))),
+ (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[HasAVX, OptForSize]>;
+}
+let Predicates = [UseAVX] in {
def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
(COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
(COPY_TO_REGCLASS VR128:$src, FR32)),
VR128)>;
def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
(VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
+}
+let Predicates = [HasAVX] in {
def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
(COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
(COPY_TO_REGCLASS VR128:$src, FR32)),
(VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
}
-// Square root.
-defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
- SSE_SQRTS>,
- sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
- sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
- sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
- SSE_SQRTS>,
- sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
- sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
-
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
-defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
- SSE_SQRTS>,
- sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
- sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
- SSE_SQRTS>;
-defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
- SSE_RCPS>,
- sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
- sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
+let Predicates = [UseSSE1] in {
+ def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
+ (RSQRTSSr_Int VR128:$src, VR128:$src)>;
+ def : Pat<(int_x86_sse_rcp_ss VR128:$src),
+ (RCPSSr_Int VR128:$src, VR128:$src)>;
+}
// There is no f64 version of the reciprocal approximation instructions.
//===----------------------------------------------------------------------===//
let AddedComplexity = 400 in { // Prefer non-temporal versions
- def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f32 VR128:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
- def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v2f64 VR128:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
-
- let ExeDomain = SSEPackedInt in
- def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
- (ins f128mem:$dst, VR128:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v2i64 VR128:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
-
- def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
- (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
-
- def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntps\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v8f32 VR256:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
- def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntpd\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4f64 VR256:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
- let ExeDomain = SSEPackedInt in
- def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
- (ins f256mem:$dst, VR256:$src),
- "movntdq\t{$src, $dst|$dst, $src}",
- [(alignednontemporalstore (v4i64 VR256:$src),
- addr:$dst)],
- IIC_SSE_MOVNT>, VEX;
+let SchedRW = [WriteStore] in {
+let Predicates = [HasAVX, NoVLX] in {
+def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movntps\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f32 VR128:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX;
+def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movntpd\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v2f64 VR128:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX;
+
+let ExeDomain = SSEPackedInt in
+def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
+ (ins f128mem:$dst, VR128:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v2i64 VR128:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX;
+
+def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntps\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v8f32 VR256:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX, VEX_L;
+def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntpd\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f64 VR256:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX, VEX_L;
+let ExeDomain = SSEPackedInt in
+def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
+ (ins f256mem:$dst, VR256:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4i64 VR256:$src),
+ addr:$dst)],
+ IIC_SSE_MOVNT>, VEX, VEX_L;
}
-let AddedComplexity = 400 in { // Prefer non-temporal versions
def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntps\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
[(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
IIC_SSE_MOVNT>;
-def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
- (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
-
// There is no AVX form for instructions below this point
def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
"movnti{l}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i32 GR32:$src), addr:$dst)],
IIC_SSE_MOVNT>,
- TB, Requires<[HasSSE2]>;
+ PS, Requires<[HasSSE2]>;
def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
"movnti{q}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i64 GR64:$src), addr:$dst)],
IIC_SSE_MOVNT>,
- TB, Requires<[HasSSE2]>;
+ PS, Requires<[HasSSE2]>;
+} // SchedRW = [WriteStore]
+
+let Predicates = [HasAVX, NoVLX] in {
+ def : Pat<(alignednontemporalstore (v4i32 VR128:$src), addr:$dst),
+ (VMOVNTPSmr addr:$dst, VR128:$src)>;
}
+def : Pat<(alignednontemporalstore (v4i32 VR128:$src), addr:$dst),
+ (MOVNTPSmr addr:$dst, VR128:$src)>;
+
+} // AddedComplexity
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Prefetch and memory fence
//===----------------------------------------------------------------------===//
// Prefetch intrinsic.
-let Predicates = [HasSSE1] in {
+let Predicates = [HasSSE1], SchedRW = [WriteLoad] in {
def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
"prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
IIC_SSE_PREFETCH>, TB;
IIC_SSE_PREFETCH>, TB;
}
+// FIXME: How should flush instruction be modeled?
+let SchedRW = [WriteLoad] in {
// Flush cache
def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
"clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
+}
+let SchedRW = [WriteNop] in {
// Pause. This "instruction" is encoded as "rep; nop", so even though it
// was introduced with SSE2, it's backward compatible.
-def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
+def PAUSE : I<0x90, RawFrm, (outs), (ins),
+ "pause", [(int_x86_sse2_pause)], IIC_SSE_PAUSE>,
+ OBXS, Requires<[HasSSE2]>;
+}
+let SchedRW = [WriteFence] in {
// Load, store, and memory fence
def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
"sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
"mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
TB, Requires<[HasSSE2]>;
+} // SchedRW
def : Pat<(X86SFence), (SFENCE)>;
def : Pat<(X86LFence), (LFENCE)>;
def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
"ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
- IIC_SSE_LDMXCSR>, VEX;
+ IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>;
def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
"stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
- IIC_SSE_STMXCSR>, VEX;
+ IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
"ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
- IIC_SSE_LDMXCSR>;
+ IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
"stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
- IIC_SSE_STMXCSR>;
+ IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
//===---------------------------------------------------------------------===//
// SSE2 - Move Aligned/Unaligned Packed Integer Instructions
let ExeDomain = SSEPackedInt in { // SSE integer instructions
-let neverHasSideEffects = 1 in {
+let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
VEX;
def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
- VEX;
-}
+ VEX, VEX_L;
def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
VEX;
def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
"movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
- VEX;
+ VEX, VEX_L;
+}
// For Disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
+ SchedRW = [WriteMove] in {
def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [],
IIC_SSE_MOVA_P_RR>,
VEX;
def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
"movdqa\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVA_P_RR>,
- VEX;
+ IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movdqu\t{$src, $dst|$dst, $src}", [],
IIC_SSE_MOVU_P_RR>,
VEX;
def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
"movdqu\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVU_P_RR>,
- VEX;
+ IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
}
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+ neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
VEX;
def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
- VEX;
+ VEX, VEX_L;
let Predicates = [HasAVX] in {
def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
XS, VEX;
def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
- XS, VEX;
+ XS, VEX, VEX_L;
}
}
-let mayStore = 1 in {
+let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
(ins i128mem:$dst, VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
(ins i256mem:$dst, VR256:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
- VEX;
+ VEX, VEX_L;
let Predicates = [HasAVX] in {
def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
XS, VEX;
def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
"vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
- XS, VEX;
+ XS, VEX, VEX_L;
}
}
+let SchedRW = [WriteMove] in {
let neverHasSideEffects = 1 in
def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
[], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
// For Disassembler
-let isCodeGenOnly = 1 in {
+let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in {
def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [],
IIC_SSE_MOVA_P_RR>;
"movdqu\t{$src, $dst|$dst, $src}",
[], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
}
+} // SchedRW
-let canFoldAsLoad = 1, mayLoad = 1 in {
+let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
+ neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}",
[/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
XS, Requires<[UseSSE2]>;
}
-let mayStore = 1 in {
+let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}",
[/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
XS, Requires<[UseSSE2]>;
}
-// Intrinsic forms of MOVDQU load and store
-def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "vmovdqu\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
- IIC_SSE_MOVU_P_MR>,
- XS, VEX, Requires<[HasAVX]>;
-
-def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
- "movdqu\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
- IIC_SSE_MOVU_P_MR>,
- XS, Requires<[UseSSE2]>;
-
} // ExeDomain = SSEPackedInt
let Predicates = [HasAVX] in {
+ def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+ (VMOVDQUmr addr:$dst, VR128:$src)>;
def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
(VMOVDQUYmr addr:$dst, VR256:$src)>;
}
+let Predicates = [UseSSE2] in
+def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
+ (MOVDQUmr addr:$dst, VR128:$src)>;
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Arithmetic Instructions
//===---------------------------------------------------------------------===//
+let Sched = WriteVecIMul in
def SSE_PMADD : OpndItins<
IIC_SSE_PMADD, IIC_SSE_PMADD
>;
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
+ [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
+ Intrinsic IntId256, OpndItins itins,
+ bit IsCommutable = 0> {
+let Predicates = [HasAVX] in
+ defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
+ VR128, loadv2i64, i128mem, itins,
+ IsCommutable, 0>, VEX_4V;
+
+let Constraints = "$src1 = $dst" in
+ defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
+ i128mem, itins, IsCommutable, 1>;
+
+let Predicates = [HasAVX2] in
+ defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
+ VR256, loadv4i64, i256mem, itins,
+ IsCommutable, 0>, VEX_4V, VEX_L;
}
multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
- itins.rr>;
+ itins.rr>, Sched<[WriteVecShift]>;
def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (DstVT (OpNode RC:$src1,
- (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
+ (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
+ Sched<[WriteVecShiftLd, ReadAfterLd]>;
def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
- (ins RC:$src1, i32i8imm:$src2),
+ (ins RC:$src1, i8imm:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
+ [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i8 imm:$src2))))], itins.ri>,
+ Sched<[WriteVecShift]>;
}
-/// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
+/// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType DstVT, ValueType SrcVT, RegisterClass RC,
PatFrag memop_frag, X86MemOperand x86memop,
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
+ [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
+ Sched<[itins.Sched]>;
def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
- (bitconvert (memop_frag addr:$src2)))))]>;
+ (bitconvert (memop_frag addr:$src2)))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
} // ExeDomain = SSEPackedInt
-// 128-bit Integer Arithmetic
-
-let Predicates = [HasAVX] in {
-defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
- VEX_4V;
-defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
- i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
- i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
-defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
- memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
- VEX_4V;
+defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 1>;
+defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 1>;
+defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
+ SSE_INTALU_ITINS_P, 1>;
+defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
+ SSE_INTALUQ_ITINS_P, 1>;
+defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
+ SSE_INTMUL_ITINS_P, 1>;
+defm PMULHUW : PDI_binop_all<0xE4, "pmulhuw", mulhu, v8i16, v16i16,
+ SSE_INTMUL_ITINS_P, 1>;
+defm PMULHW : PDI_binop_all<0xE5, "pmulhw", mulhs, v8i16, v16i16,
+ SSE_INTMUL_ITINS_P, 1>;
+defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 0>;
+defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 0>;
+defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
+ SSE_INTALU_ITINS_P, 0>;
+defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
+ SSE_INTALUQ_ITINS_P, 0>;
+defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 0>;
+defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 0>;
+defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 1>;
+defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 1>;
+defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 1>;
+defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 1>;
// Intrinsic forms
-defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
- VR128, memopv2i64, i128mem,
- SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
- VR128, memopv2i64, i128mem,
- SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
- VR128, memopv2i64, i128mem,
- SSE_PMADD, 1, 0>, VEX_4V;
-defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-}
+defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
+ int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
+defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
+ int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
+defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
+ int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
+defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
+ int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
+defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
+ int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
+defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
+ int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
+defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
+ int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
+defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
+ int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
+defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
+ int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
+defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
+ int_x86_avx2_psad_bw, SSE_PMADD, 1>;
-let Predicates = [HasAVX2] in {
-defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
- i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
- i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
- i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
- i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
+let Predicates = [HasAVX] in
+defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
+ loadv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
+ VEX_4V;
+let Predicates = [HasAVX2] in
defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
- VR256, memopv4i64, i256mem,
- SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-
-// Intrinsic forms
-defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
- VR256, memopv4i64, i256mem,
- SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
- VR256, memopv4i64, i256mem,
- SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
-defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
- VR256, memopv4i64, i256mem,
- SSE_PMADD, 1, 0>, VEX_4V;
-defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
-}
-
-let Constraints = "$src1 = $dst" in {
-defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P, 1>;
-defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
- i128mem, SSE_INTALUQ_ITINS_P, 1>;
-defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTMUL_ITINS_P, 1>;
-defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
- i128mem, SSE_INTALU_ITINS_P>;
-defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
- i128mem, SSE_INTALUQ_ITINS_P>;
+ VR256, loadv4i64, i256mem,
+ SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+let Constraints = "$src1 = $dst" in
defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
-// Intrinsic forms
-defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
- VR128, memopv2i64, i128mem,
- SSE_INTMUL_ITINS_P, 1>;
-defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
- VR128, memopv2i64, i128mem,
- SSE_INTMUL_ITINS_P, 1>;
-defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
- VR128, memopv2i64, i128mem,
- SSE_PMADD, 1>;
-defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
-
-} // Constraints = "$src1 = $dst"
-
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Logical Instructions
//===---------------------------------------------------------------------===//
VR128, v4i32, v4i32, bc_v4i32,
SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
// 128-bit logical shifts.
def VPSLLDQri : PDIi8<0x73, MRM7r,
(outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
let Predicates = [HasAVX2] in {
defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
VR256, v16i16, v8i16, bc_v8i16,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
VR256, v8i32, v4i32, bc_v4i32,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
VR256, v4i64, v2i64, bc_v2i64,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
VR256, v16i16, v8i16, bc_v8i16,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
VR256, v8i32, v4i32, bc_v4i32,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
VR256, v4i64, v2i64, bc_v2i64,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
VR256, v16i16, v8i16, bc_v8i16,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
VR256, v8i32, v4i32, bc_v4i32,
- SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
+ SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
// 256-bit logical shifts.
def VPSLLDQYri : PDIi8<0x73, MRM7r,
(outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
"vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR256:$dst,
(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
- VEX_4V;
+ VEX_4V, VEX_L;
def VPSRLDQYri : PDIi8<0x73, MRM3r,
(outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
"vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR256:$dst,
(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
- VEX_4V;
+ VEX_4V, VEX_L;
// PSRADQYri doesn't exist in SSE[1-3].
}
} // Predicates = [HasAVX2]
VR128, v4i32, v4i32, bc_v4i32,
SSE_INTSHIFT_ITINS_P>;
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in {
// 128-bit logical shifts.
def PSLLDQri : PDIi8<0x73, MRM7r,
(outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
"pslldq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
+ (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))],
+ IIC_SSE_INTSHDQ_P_RI>;
def PSRLDQri : PDIi8<0x73, MRM3r,
(outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
"psrldq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
+ (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))],
+ IIC_SSE_INTSHDQ_P_RI>;
// PSRADQri doesn't exist in SSE[1-3].
}
} // Constraints = "$src1 = $dst"
// SSE2 - Packed Integer Comparison Instructions
//===---------------------------------------------------------------------===//
+defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 1>;
+defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 1>;
+defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
+ SSE_INTALU_ITINS_P, 1>;
+defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
+ SSE_INTALU_ITINS_P, 0>;
+defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
+ SSE_INTALU_ITINS_P, 0>;
+defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
+ SSE_INTALU_ITINS_P, 0>;
+
+//===---------------------------------------------------------------------===//
+// SSE2 - Packed Integer Shuffle Instructions
+//===---------------------------------------------------------------------===//
+
+let ExeDomain = SSEPackedInt in {
+multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
+ SDNode OpNode> {
let Predicates = [HasAVX] in {
- defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
- defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
- defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
+ def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, i8imm:$src2),
+ !strconcat("v", OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR128:$dst,
+ (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
+ IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
+ def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
+ (ins i128mem:$src1, i8imm:$src2),
+ !strconcat("v", OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR128:$dst,
+ (vt128 (OpNode (bitconvert (loadv2i64 addr:$src1)),
+ (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX,
+ Sched<[WriteShuffleLd]>;
}
let Predicates = [HasAVX2] in {
- defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
- defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
- defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
- defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
+ def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
+ (ins VR256:$src1, i8imm:$src2),
+ !strconcat("v", OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
+ IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
+ def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
+ (ins i256mem:$src1, i8imm:$src2),
+ !strconcat("v", OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (vt256 (OpNode (bitconvert (loadv4i64 addr:$src1)),
+ (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>, VEX, VEX_L,
+ Sched<[WriteShuffleLd]>;
}
-let Constraints = "$src1 = $dst" in {
- defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
- defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
- defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 1>;
- defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
- defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
- defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-} // Constraints = "$src1 = $dst"
+let Predicates = [UseSSE2] in {
+ def ri : Ii8<0x70, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR128:$dst,
+ (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
+ IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
+ def mi : Ii8<0x70, MRMSrcMem,
+ (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR128:$dst,
+ (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
+ (i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+}
+} // ExeDomain = SSEPackedInt
-//===---------------------------------------------------------------------===//
-// SSE2 - Packed Integer Pack Instructions
-//===---------------------------------------------------------------------===//
+defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, PD;
+defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
+defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
let Predicates = [HasAVX] in {
-defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
+ def : Pat<(v4f32 (X86PShufd (loadv4f32 addr:$src1), (i8 imm:$imm))),
+ (VPSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (VPSHUFDri VR128:$src1, imm:$imm)>;
}
-let Predicates = [HasAVX2] in {
-defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
-defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
- VR256, memopv4i64, i256mem,
- SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
+let Predicates = [UseSSE2] in {
+ def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
+ (PSHUFDmi addr:$src1, imm:$imm)>;
+ def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
+ (PSHUFDri VR128:$src1, imm:$imm)>;
}
-let Constraints = "$src1 = $dst" in {
-defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
- VR128, memopv2i64, i128mem,
- SSE_INTALU_ITINS_P>;
-} // Constraints = "$src1 = $dst"
-
//===---------------------------------------------------------------------===//
-// SSE2 - Packed Integer Shuffle Instructions
+// Packed Integer Pack Instructions (SSE & AVX)
//===---------------------------------------------------------------------===//
let ExeDomain = SSEPackedInt in {
-multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
-def ri : Ii8<0x70, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
- IIC_SSE_PSHUF>;
-def mi : Ii8<0x70, MRMSrcMem,
- (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128:$dst,
- (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
- (i8 imm:$src2))))],
- IIC_SSE_PSHUF>;
-}
-
-multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
-def Yri : Ii8<0x70, MRMSrcReg,
- (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
-def Ymi : Ii8<0x70, MRMSrcMem,
- (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
- !strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst,
- (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
- (i8 imm:$src2))))]>;
+multiclass sse2_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag,
+ bit Is2Addr = 1> {
+ def rr : PDI<opc, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def rm : PDI<opc, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode VR128:$src1,
+ (bc_frag (memopv2i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
}
-} // ExeDomain = SSEPackedInt
-let Predicates = [HasAVX] in {
- let AddedComplexity = 5 in
- defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
+multiclass sse2_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag> {
+ def Yrr : PDI<opc, MRMSrcReg,
+ (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def Yrm : PDI<opc, MRMSrcMem,
+ (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode VR256:$src1,
+ (bc_frag (memopv4i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+
+multiclass sse4_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag,
+ bit Is2Addr = 1> {
+ def rr : SS48I<opc, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def rm : SS48I<opc, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode VR128:$src1,
+ (bc_frag (memopv2i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
- // SSE2 with ImmT == Imm8 and XS prefix.
- defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
+multiclass sse4_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag> {
+ def Yrr : SS48I<opc, MRMSrcReg,
+ (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def Yrm : SS48I<opc, MRMSrcMem,
+ (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode VR256:$src1,
+ (bc_frag (memopv4i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
- // SSE2 with ImmT == Imm8 and XD prefix.
- defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
+let Predicates = [HasAVX] in {
+ defm VPACKSSWB : sse2_pack<0x63, "vpacksswb", v16i8, v8i16, X86Packss,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPACKSSDW : sse2_pack<0x6B, "vpackssdw", v8i16, v4i32, X86Packss,
+ bc_v4i32, 0>, VEX_4V;
- def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
- (VPSHUFDmi addr:$src1, imm:$imm)>;
- def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (VPSHUFDri VR128:$src1, imm:$imm)>;
+ defm VPACKUSWB : sse2_pack<0x67, "vpackuswb", v16i8, v8i16, X86Packus,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPACKUSDW : sse4_pack<0x2B, "vpackusdw", v8i16, v4i32, X86Packus,
+ bc_v4i32, 0>, VEX_4V;
}
let Predicates = [HasAVX2] in {
- defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
- defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
- defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
-}
+ defm VPACKSSWB : sse2_pack_y<0x63, "vpacksswb", v32i8, v16i16, X86Packss,
+ bc_v16i16>, VEX_4V, VEX_L;
+ defm VPACKSSDW : sse2_pack_y<0x6B, "vpackssdw", v16i16, v8i32, X86Packss,
+ bc_v8i32>, VEX_4V, VEX_L;
-let Predicates = [UseSSE2] in {
- let AddedComplexity = 5 in
- defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
+ defm VPACKUSWB : sse2_pack_y<0x67, "vpackuswb", v32i8, v16i16, X86Packus,
+ bc_v16i16>, VEX_4V, VEX_L;
+ defm VPACKUSDW : sse4_pack_y<0x2B, "vpackusdw", v16i16, v8i32, X86Packus,
+ bc_v8i32>, VEX_4V, VEX_L;
+}
- // SSE2 with ImmT == Imm8 and XS prefix.
- defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
+let Constraints = "$src1 = $dst" in {
+ defm PACKSSWB : sse2_pack<0x63, "packsswb", v16i8, v8i16, X86Packss,
+ bc_v8i16>;
+ defm PACKSSDW : sse2_pack<0x6B, "packssdw", v8i16, v4i32, X86Packss,
+ bc_v4i32>;
- // SSE2 with ImmT == Imm8 and XD prefix.
- defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
+ defm PACKUSWB : sse2_pack<0x67, "packuswb", v16i8, v8i16, X86Packus,
+ bc_v8i16>;
- def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
- (PSHUFDmi addr:$src1, imm:$imm)>;
- def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
- (PSHUFDri VR128:$src1, imm:$imm)>;
+ let Predicates = [HasSSE41] in
+ defm PACKUSDW : sse4_pack<0x2B, "packusdw", v8i16, v4i32, X86Packus,
+ bc_v4i32>;
}
+} // ExeDomain = SSEPackedInt
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Unpack Instructions
!strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
- IIC_SSE_UNPCK>;
+ IIC_SSE_UNPCK>, Sched<[WriteShuffle]>;
def rm : PDI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
[(set VR128:$dst, (OpNode VR128:$src1,
(bc_frag (memopv2i64
addr:$src2))))],
- IIC_SSE_UNPCK>;
+ IIC_SSE_UNPCK>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
}
multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
def Yrr : PDI<opc, MRMSrcReg,
(outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
+ [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>,
+ Sched<[WriteShuffle]>;
def Yrm : PDI<opc, MRMSrcMem,
(outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst, (OpNode VR256:$src1,
- (bc_frag (memopv4i64 addr:$src2))))]>;
+ (bc_frag (memopv4i64 addr:$src2))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let Predicates = [HasAVX2] in {
defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
- bc_v32i8>, VEX_4V;
+ bc_v32i8>, VEX_4V, VEX_L;
defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
- bc_v16i16>, VEX_4V;
+ bc_v16i16>, VEX_4V, VEX_L;
defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
- bc_v8i32>, VEX_4V;
+ bc_v8i32>, VEX_4V, VEX_L;
defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
- bc_v4i64>, VEX_4V;
+ bc_v4i64>, VEX_4V, VEX_L;
defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
- bc_v32i8>, VEX_4V;
+ bc_v32i8>, VEX_4V, VEX_L;
defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
- bc_v16i16>, VEX_4V;
+ bc_v16i16>, VEX_4V, VEX_L;
defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
- bc_v8i32>, VEX_4V;
+ bc_v8i32>, VEX_4V, VEX_L;
defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
- bc_v4i64>, VEX_4V;
+ bc_v4i64>, VEX_4V, VEX_L;
}
let Constraints = "$src1 = $dst" in {
multiclass sse2_pinsrw<bit Is2Addr = 1> {
def rri : Ii8<0xC4, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1,
- GR32:$src2, i32i8imm:$src3),
+ GR32orGR64:$src2, i32i8imm:$src3),
!if(Is2Addr,
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
"vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst,
- (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
+ (X86pinsrw VR128:$src1, GR32orGR64:$src2, imm:$src3))],
+ IIC_SSE_PINSRW>, Sched<[WriteShuffle]>;
def rmi : Ii8<0xC4, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1,
i16mem:$src2, i32i8imm:$src3),
"vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set VR128:$dst,
(X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
- imm:$src3))], IIC_SSE_PINSRW>;
+ imm:$src3))], IIC_SSE_PINSRW>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
}
// Extract
let Predicates = [HasAVX] in
def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
- (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
+ (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
"vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
- imm:$src2))]>, TB, OpSize, VEX;
+ [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
+ imm:$src2))]>, PD, VEX,
+ Sched<[WriteShuffle]>;
def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
- (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
+ (outs GR32orGR64:$dst), (ins VR128:$src1, i32i8imm:$src2),
"pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
- imm:$src2))], IIC_SSE_PEXTRW>;
+ [(set GR32orGR64:$dst, (X86pextrw (v8i16 VR128:$src1),
+ imm:$src2))], IIC_SSE_PEXTRW>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
// Insert
-let Predicates = [HasAVX] in {
- defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
- def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
- "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, TB, OpSize, VEX_4V;
-}
+let Predicates = [HasAVX] in
+defm VPINSRW : sse2_pinsrw<0>, PD, VEX_4V;
-let Constraints = "$src1 = $dst" in
- defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
+let Predicates = [UseSSE2], Constraints = "$src1 = $dst" in
+defm PINSRW : sse2_pinsrw, PD;
} // ExeDomain = SSEPackedInt
// SSE2 - Packed Mask Creation
//===---------------------------------------------------------------------===//
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in {
-def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
+def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
+ (ins VR128:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
+ [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
IIC_SSE_MOVMSK>, VEX;
-def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
- "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
let Predicates = [HasAVX2] in {
-def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
+def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst),
+ (ins VR256:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
-def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
- "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
+ [(set GR32orGR64:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>,
+ VEX, VEX_L;
}
-def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
+def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32orGR64:$dst), (ins VR128:$src),
"pmovmskb\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
+ [(set GR32orGR64:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
IIC_SSE_MOVMSK>;
} // ExeDomain = SSEPackedInt
// SSE2 - Conditional Store
//===---------------------------------------------------------------------===//
-let ExeDomain = SSEPackedInt in {
+let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in {
-let Uses = [EDI] in
+let Uses = [EDI], Predicates = [HasAVX,Not64BitMode] in
def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
(ins VR128:$src, VR128:$mask),
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
IIC_SSE_MASKMOV>, VEX;
-let Uses = [RDI] in
+let Uses = [RDI], Predicates = [HasAVX,In64BitMode] in
def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
(ins VR128:$src, VR128:$mask),
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
IIC_SSE_MASKMOV>, VEX;
-let Uses = [EDI] in
+let Uses = [EDI], Predicates = [UseSSE2,Not64BitMode] in
def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
IIC_SSE_MASKMOV>;
-let Uses = [RDI] in
+let Uses = [RDI], Predicates = [UseSSE2,In64BitMode] in
def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
//===---------------------------------------------------------------------===//
// Move Int Doubleword to Packed Double Int
//
-def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
+def VMOVDI2PDIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
- VEX;
-def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
+ VEX, Sched<[WriteMove]>;
+def VMOVDI2PDIrm : VS2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v4i32 (scalar_to_vector (loadi32 addr:$src))))],
IIC_SSE_MOVDQ>,
- VEX;
-def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}",
+ VEX, Sched<[WriteLoad]>;
+def VMOV64toPQIrr : VRS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+ "movq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (scalar_to_vector GR64:$src)))],
- IIC_SSE_MOVDQ>, VEX;
-def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}",
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
+let isCodeGenOnly = 1 in
+def VMOV64toSDrr : VRS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
+ "movq\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (bitconvert GR64:$src))],
- IIC_SSE_MOVDQ>, VEX;
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
-def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
+def MOVDI2PDIrr : S2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
- (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
-def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
+ (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
+ Sched<[WriteMove]>;
+def MOVDI2PDIrm : S2I<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v4i32 (scalar_to_vector (loadi32 addr:$src))))],
- IIC_SSE_MOVDQ>;
-def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+ IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
+def MOV64toPQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
"mov{d|q}\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (scalar_to_vector GR64:$src)))],
- IIC_SSE_MOVDQ>;
-def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
+ IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
+let isCodeGenOnly = 1 in
+def MOV64toSDrr : RS2I<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
"mov{d|q}\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (bitconvert GR64:$src))],
- IIC_SSE_MOVDQ>;
+ IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
//===---------------------------------------------------------------------===//
// Move Int Doubleword to Single Scalar
//
-def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (bitconvert GR32:$src))],
- IIC_SSE_MOVDQ>, VEX;
-
-def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
- IIC_SSE_MOVDQ>,
- VEX;
-def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (bitconvert GR32:$src))],
- IIC_SSE_MOVDQ>;
+let isCodeGenOnly = 1 in {
+ def VMOVDI2SSrr : VS2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (bitconvert GR32:$src))],
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
+
+ def VMOVDI2SSrm : VS2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
+ IIC_SSE_MOVDQ>,
+ VEX, Sched<[WriteLoad]>;
+ def MOVDI2SSrr : S2I<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (bitconvert GR32:$src))],
+ IIC_SSE_MOVDQ>, Sched<[WriteMove]>;
-def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
- IIC_SSE_MOVDQ>;
+ def MOVDI2SSrm : S2I<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
+ IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
+}
//===---------------------------------------------------------------------===//
// Move Packed Doubleword Int to Packed Double Int
//
-def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
+def VMOVPDI2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
- (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
-def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
+ (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX,
+ Sched<[WriteMove]>;
+def VMOVPDI2DImr : VS2I<0x7E, MRMDestMem, (outs),
(ins i32mem:$dst, VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
- VEX;
-def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
+ VEX, Sched<[WriteStore]>;
+def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
- (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
-def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
+ (iPTR 0)))], IIC_SSE_MOVD_ToGP>,
+ Sched<[WriteMove]>;
+def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)],
- IIC_SSE_MOVDQ>;
+ IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
+
+def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
+ (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
+
+def : Pat<(v4i64 (X86Vinsert (bc_v4i64 (v8i32 immAllZerosV)), GR64:$src2, (iPTR 0))),
+ (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
+
+def : Pat<(v8i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))),
+ (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
+
+def : Pat<(v4i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
+ (SUBREG_TO_REG (i32 0), (VMOV64toPQIrr GR64:$src2), sub_xmm)>;
//===---------------------------------------------------------------------===//
// Move Packed Doubleword Int first element to Doubleword Int
//
-def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
- "vmov{d|q}\t{$src, $dst|$dst, $src}",
+let SchedRW = [WriteMove] in {
+def VMOVPQIto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
+ "movq\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
(iPTR 0)))],
IIC_SSE_MOVD_ToGP>,
- TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
+ VEX;
-def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
+def MOVPQIto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
"mov{d|q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
(iPTR 0)))],
IIC_SSE_MOVD_ToGP>;
+} //SchedRW
//===---------------------------------------------------------------------===//
// Bitcast FR64 <-> GR64
//
-let Predicates = [HasAVX] in
-def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
- "vmovq\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
- VEX;
-def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
+let isCodeGenOnly = 1 in {
+ let Predicates = [UseAVX] in
+ def VMOV64toSDrm : VS2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
+ VEX, Sched<[WriteLoad]>;
+ def VMOVSDto64rr : VRS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(set GR64:$dst, (bitconvert FR64:$src))],
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>;
+ def VMOVSDto64mr : VRS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
+
+ def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
+ "movq\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
+ IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
+ def MOVSDto64rr : RS2I<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
"mov{d|q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (bitconvert FR64:$src))],
- IIC_SSE_MOVDQ>, VEX;
-def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
+ IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
+ def MOVSDto64mr : RS2I<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (i64 (bitconvert FR64:$src)), addr:$dst)],
- IIC_SSE_MOVDQ>, VEX;
-
-def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
- "movq\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
- IIC_SSE_MOVDQ>;
-def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}",
- [(set GR64:$dst, (bitconvert FR64:$src))],
- IIC_SSE_MOVD_ToGP>;
-def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
- "movq\t{$src, $dst|$dst, $src}",
- [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
- IIC_SSE_MOVDQ>;
+ IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
+}
//===---------------------------------------------------------------------===//
// Move Scalar Single to Double Int
//
-def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (bitconvert FR32:$src))],
- IIC_SSE_MOVD_ToGP>, VEX;
-def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
- IIC_SSE_MOVDQ>, VEX;
-def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (bitconvert FR32:$src))],
- IIC_SSE_MOVD_ToGP>;
-def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
- IIC_SSE_MOVDQ>;
+let isCodeGenOnly = 1 in {
+ def VMOVSS2DIrr : VS2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (bitconvert FR32:$src))],
+ IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>;
+ def VMOVSS2DImr : VS2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
+ IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>;
+ def MOVSS2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (bitconvert FR32:$src))],
+ IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>;
+ def MOVSS2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
+ "movd\t{$src, $dst|$dst, $src}",
+ [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
+ IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
+}
//===---------------------------------------------------------------------===//
// Patterns and instructions to describe movd/movq to XMM register zero-extends
//
+let isCodeGenOnly = 1, SchedRW = [WriteMove] in {
let AddedComplexity = 15 in {
-def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector GR32:$src)))))],
- IIC_SSE_MOVDQ>, VEX;
-def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
- "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
+def VMOVZQI2PQIrr : VS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+ "movq\t{$src, $dst|$dst, $src}", // X86-64 only
[(set VR128:$dst, (v2i64 (X86vzmovl
(v2i64 (scalar_to_vector GR64:$src)))))],
IIC_SSE_MOVDQ>,
VEX, VEX_W;
-}
-let AddedComplexity = 15 in {
-def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector GR32:$src)))))],
- IIC_SSE_MOVDQ>;
-def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
+def MOVZQI2PQIrr : RS2I<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
"mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
[(set VR128:$dst, (v2i64 (X86vzmovl
(v2i64 (scalar_to_vector GR64:$src)))))],
IIC_SSE_MOVDQ>;
}
+} // isCodeGenOnly, SchedRW
-let AddedComplexity = 20 in {
-def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
- (loadi32 addr:$src))))))],
- IIC_SSE_MOVDQ>, VEX;
-def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
- "movd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
- (loadi32 addr:$src))))))],
- IIC_SSE_MOVDQ>;
-}
+let Predicates = [UseAVX] in {
+ let AddedComplexity = 15 in
+ def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
+ (VMOVDI2PDIrr GR32:$src)>;
-let Predicates = [HasAVX] in {
// AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
let AddedComplexity = 20 in {
+ def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
+ (VMOVDI2PDIrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
- (VMOVZDI2PDIrm addr:$src)>;
+ (VMOVDI2PDIrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
- (VMOVZDI2PDIrm addr:$src)>;
+ (VMOVDI2PDIrm addr:$src)>;
}
// Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
(v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
+ (SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src), sub_xmm)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
(SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
}
-let Predicates = [UseSSE2], AddedComplexity = 20 in {
- def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
- (MOVZDI2PDIrm addr:$src)>;
- def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
- (MOVZDI2PDIrm addr:$src)>;
+let Predicates = [UseSSE2] in {
+ let AddedComplexity = 15 in
+ def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
+ (MOVDI2PDIrr GR32:$src)>;
+
+ let AddedComplexity = 20 in {
+ def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
+ (MOVDI2PDIrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
+ (MOVDI2PDIrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
+ (MOVDI2PDIrm addr:$src)>;
+ }
}
// These are the correct encodings of the instructions so that we know how to
// compatibility with Darwin's buggy assembler.
def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
(MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
-def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
- (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
(MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
-def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
- (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
-def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
- (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
-def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
- (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
+// Allow "vmovd" but print "vmovq" since we don't need compatibility for AVX.
+def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
+ (VMOV64toPQIrr VR128:$dst, GR64:$src), 0>;
+def : InstAlias<"vmovd\t{$src, $dst|$dst, $src}",
+ (VMOVPQIto64rr GR64:$dst, VR128:$src), 0>;
//===---------------------------------------------------------------------===//
// SSE2 - Move Quadword
//===---------------------------------------------------------------------===//
// Move Quadword Int to Packed Quadword Int
//
+
+let SchedRW = [WriteLoad] in {
def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
- VEX, Requires<[HasAVX]>;
+ VEX, Requires<[UseAVX]>;
def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (scalar_to_vector (loadi64 addr:$src))))],
IIC_SSE_MOVDQ>, XS,
Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
+} // SchedRW
//===---------------------------------------------------------------------===//
// Move Packed Quadword Int to Quadword Int
//
-def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
+let SchedRW = [WriteStore] in {
+def VMOVPQI2QImr : VS2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (i64 (vector_extract (v2i64 VR128:$src),
(iPTR 0))), addr:$dst)],
IIC_SSE_MOVDQ>, VEX;
-def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
+def MOVPQI2QImr : S2I<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
"movq\t{$src, $dst|$dst, $src}",
[(store (i64 (vector_extract (v2i64 VR128:$src),
(iPTR 0))), addr:$dst)],
IIC_SSE_MOVDQ>;
+} // SchedRW
+
+// For disassembler only
+let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
+ SchedRW = [WriteVecLogic] in {
+def VMOVPQI2QIrr : VS2I<0xD6, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
+ "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, VEX;
+def MOVPQI2QIrr : S2I<0xD6, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
+ "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>;
+}
//===---------------------------------------------------------------------===//
// Store / copy lower 64-bits of a XMM register.
//
-def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
- "movq\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
-def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
- "movq\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
- IIC_SSE_MOVDQ>;
+let Predicates = [UseAVX] in
+def : Pat<(int_x86_sse2_storel_dq addr:$dst, VR128:$src),
+ (VMOVPQI2QImr addr:$dst, VR128:$src)>;
+let Predicates = [UseSSE2] in
+def : Pat<(int_x86_sse2_storel_dq addr:$dst, VR128:$src),
+ (MOVPQI2QImr addr:$dst, VR128:$src)>;
-let AddedComplexity = 20 in
+let isCodeGenOnly = 1, AddedComplexity = 20 in {
def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (X86vzmovl (v2i64 (scalar_to_vector
(loadi64 addr:$src))))))],
IIC_SSE_MOVDQ>,
- XS, VEX, Requires<[HasAVX]>;
+ XS, VEX, Requires<[UseAVX]>, Sched<[WriteLoad]>;
-let AddedComplexity = 20 in
def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (X86vzmovl (v2i64 (scalar_to_vector
(loadi64 addr:$src))))))],
IIC_SSE_MOVDQ>,
- XS, Requires<[UseSSE2]>;
+ XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>;
+}
-let Predicates = [HasAVX], AddedComplexity = 20 in {
- def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
- (VMOVZQI2PQIrm addr:$src)>;
+let Predicates = [UseAVX], AddedComplexity = 20 in {
def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
(VMOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzload addr:$src)),
}
let Predicates = [UseSSE2], AddedComplexity = 20 in {
- def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
- (MOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
(MOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
// Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
// IA32 document. movq xmm1, xmm2 does clear the high bits.
//
+let SchedRW = [WriteVecLogic] in {
let AddedComplexity = 15 in
def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
IIC_SSE_MOVQ_RR>,
- XS, VEX, Requires<[HasAVX]>;
+ XS, VEX, Requires<[UseAVX]>;
let AddedComplexity = 15 in
def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
IIC_SSE_MOVQ_RR>,
XS, Requires<[UseSSE2]>;
+} // SchedRW
+let isCodeGenOnly = 1, SchedRW = [WriteVecLogicLd] in {
let AddedComplexity = 20 in
def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2i64 (X86vzmovl
(loadv2i64 addr:$src))))],
IIC_SSE_MOVDQ>,
- XS, VEX, Requires<[HasAVX]>;
+ XS, VEX, Requires<[UseAVX]>;
let AddedComplexity = 20 in {
def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movq\t{$src, $dst|$dst, $src}",
IIC_SSE_MOVDQ>,
XS, Requires<[UseSSE2]>;
}
+} // isCodeGenOnly, SchedRW
let AddedComplexity = 20 in {
- let Predicates = [HasAVX] in {
- def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
- (VMOVZPQILo2PQIrm addr:$src)>;
+ let Predicates = [UseAVX] in {
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(VMOVZPQILo2PQIrr VR128:$src)>;
}
let Predicates = [UseSSE2] in {
- def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
- (MOVZPQILo2PQIrm addr:$src)>;
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(MOVZPQILo2PQIrr VR128:$src)>;
}
}
-// Instructions to match in the assembler
-def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
- "movq\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVDQ>, VEX, VEX_W;
-def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
- "movq\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVDQ>, VEX, VEX_W;
-// Recognize "movd" with GR64 destination, but encode as a "movq"
-def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
- "movd\t{$src, $dst|$dst, $src}", [],
- IIC_SSE_MOVDQ>, VEX, VEX_W;
-
-// Instructions for the disassembler
-// xr = XMM register
-// xm = mem64
-
-let Predicates = [HasAVX] in
-def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
-def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
-
//===---------------------------------------------------------------------===//
// SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
//===---------------------------------------------------------------------===//
def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set RC:$dst, (vt (OpNode RC:$src)))],
- IIC_SSE_MOV_LH>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set RC:$dst, (OpNode (mem_frag addr:$src)))],
- IIC_SSE_MOV_LH>;
+ IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
}
let Predicates = [HasAVX] in {
defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
- v4f32, VR128, memopv4f32, f128mem>, VEX;
+ v4f32, VR128, loadv4f32, f128mem>, VEX;
defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
- v4f32, VR128, memopv4f32, f128mem>, VEX;
+ v4f32, VR128, loadv4f32, f128mem>, VEX;
defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
- v8f32, VR256, memopv8f32, f256mem>, VEX;
+ v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
- v8f32, VR256, memopv8f32, f256mem>, VEX;
+ v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
}
defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
memopv4f32, f128mem>;
let Predicates = [HasAVX] in {
def : Pat<(v4i32 (X86Movshdup VR128:$src)),
(VMOVSHDUPrr VR128:$src)>;
- def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
+ def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
(VMOVSHDUPrm addr:$src)>;
def : Pat<(v4i32 (X86Movsldup VR128:$src)),
(VMOVSLDUPrr VR128:$src)>;
- def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
+ def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (loadv2i64 addr:$src)))),
(VMOVSLDUPrm addr:$src)>;
def : Pat<(v8i32 (X86Movshdup VR256:$src)),
(VMOVSHDUPYrr VR256:$src)>;
- def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
+ def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (loadv4i64 addr:$src)))),
(VMOVSHDUPYrm addr:$src)>;
def : Pat<(v8i32 (X86Movsldup VR256:$src)),
(VMOVSLDUPYrr VR256:$src)>;
- def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
+ def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (loadv4i64 addr:$src)))),
(VMOVSLDUPYrm addr:$src)>;
}
let neverHasSideEffects = 1 in
def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [], IIC_SSE_MOV_LH>;
+ [], IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
(v2f64 (X86Movddup
(scalar_to_vector (loadf64 addr:$src)))))],
- IIC_SSE_MOV_LH>;
+ IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
}
// FIXME: Merge with above classe when there're patterns for the ymm version
multiclass sse3_replicate_dfp_y<string OpcodeStr> {
def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
+ [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
+ Sched<[WriteFShuffle]>;
def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
(v4f64 (X86Movddup
- (scalar_to_vector (loadf64 addr:$src)))))]>;
+ (scalar_to_vector (loadf64 addr:$src)))))]>,
+ Sched<[WriteLoad]>;
}
let Predicates = [HasAVX] in {
defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
- defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
+ defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
}
defm MOVDDUP : sse3_replicate_dfp<"movddup">;
let Predicates = [HasAVX] in {
- def : Pat<(X86Movddup (memopv2f64 addr:$src)),
+ def : Pat<(X86Movddup (loadv2f64 addr:$src)),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
- def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
+ def : Pat<(X86Movddup (bc_v2f64 (loadv4f32 addr:$src))),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
- def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
+ def : Pat<(X86Movddup (bc_v2f64 (loadv2i64 addr:$src))),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
def : Pat<(X86Movddup (bc_v2f64
(v2i64 (scalar_to_vector (loadi64 addr:$src))))),
(VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
// 256-bit version
- def : Pat<(X86Movddup (memopv4f64 addr:$src)),
+ def : Pat<(X86Movddup (loadv4f64 addr:$src)),
(VMOVDDUPYrm addr:$src)>;
- def : Pat<(X86Movddup (memopv4i64 addr:$src)),
+ def : Pat<(X86Movddup (loadv4i64 addr:$src)),
(VMOVDDUPYrm addr:$src)>;
def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
(VMOVDDUPYrm addr:$src)>;
(VMOVDDUPYrr VR256:$src)>;
}
+let Predicates = [UseAVX, OptForSize] in {
+ def : Pat<(v2f64 (X86VBroadcast (loadf64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>;
+ def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>;
+}
+
let Predicates = [UseSSE3] in {
def : Pat<(X86Movddup (memopv2f64 addr:$src)),
(MOVDDUPrm addr:$src)>;
// SSE3 - Move Unaligned Integer
//===---------------------------------------------------------------------===//
+let SchedRW = [WriteLoad] in {
let Predicates = [HasAVX] in {
def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vlddqu\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
"vlddqu\t{$src, $dst|$dst, $src}",
- [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
+ [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
+ VEX, VEX_L;
}
def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"lddqu\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
IIC_SSE_LDDQU>;
+}
//===---------------------------------------------------------------------===//
// SSE3 - Arithmetic
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
+ [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : I<0xD0, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
+ [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let ExeDomain = SSEPackedSingle in {
defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
- f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
+ f128mem, SSE_ALU_F32P, 0>, XD, VEX_4V;
defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
- f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
+ f256mem, SSE_ALU_F32P, 0>, XD, VEX_4V, VEX_L;
}
let ExeDomain = SSEPackedDouble in {
defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
- f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
+ f128mem, SSE_ALU_F64P, 0>, PD, VEX_4V;
defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
- f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
+ f256mem, SSE_ALU_F64P, 0>, PD, VEX_4V, VEX_L;
}
}
let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
let ExeDomain = SSEPackedSingle in
defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
- f128mem, SSE_ALU_F32P>, TB, XD;
+ f128mem, SSE_ALU_F32P>, XD;
let ExeDomain = SSEPackedDouble in
defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
- f128mem, SSE_ALU_F64P>, TB, OpSize;
+ f128mem, SSE_ALU_F64P>, PD;
+}
+
+// Patterns used to select 'addsub' instructions.
+let Predicates = [HasAVX] in {
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 VR128:$rhs))),
+ (VADDSUBPSrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 (memop addr:$rhs)))),
+ (VADDSUBPSrm VR128:$lhs, f128mem:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 VR128:$rhs))),
+ (VADDSUBPDrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 (memop addr:$rhs)))),
+ (VADDSUBPDrm VR128:$lhs, f128mem:$rhs)>;
+
+ def : Pat<(v8f32 (X86Addsub (v8f32 VR256:$lhs), (v8f32 VR256:$rhs))),
+ (VADDSUBPSYrr VR256:$lhs, VR256:$rhs)>;
+ def : Pat<(v8f32 (X86Addsub (v8f32 VR256:$lhs), (v8f32 (memop addr:$rhs)))),
+ (VADDSUBPSYrm VR256:$lhs, f256mem:$rhs)>;
+ def : Pat<(v4f64 (X86Addsub (v4f64 VR256:$lhs), (v4f64 VR256:$rhs))),
+ (VADDSUBPDYrr VR256:$lhs, VR256:$rhs)>;
+ def : Pat<(v4f64 (X86Addsub (v4f64 VR256:$lhs), (v4f64 (memop addr:$rhs)))),
+ (VADDSUBPDYrm VR256:$lhs, f256mem:$rhs)>;
+}
+
+let Predicates = [UseSSE3] in {
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 VR128:$rhs))),
+ (ADDSUBPSrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 (memop addr:$rhs)))),
+ (ADDSUBPSrm VR128:$lhs, f128mem:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 VR128:$rhs))),
+ (ADDSUBPDrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 (memop addr:$rhs)))),
+ (ADDSUBPDrm VR128:$lhs, f128mem:$rhs)>;
}
//===---------------------------------------------------------------------===//
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
+ [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
+ Sched<[WriteFAdd]>;
def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
- IIC_SSE_HADDSUB_RM>;
+ IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
}
multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
+ [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>,
+ Sched<[WriteFAdd]>;
def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
- IIC_SSE_HADDSUB_RM>;
+ IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
X86fhsub, 0>, VEX_4V;
defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
- X86fhadd, 0>, VEX_4V;
+ X86fhadd, 0>, VEX_4V, VEX_L;
defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
- X86fhsub, 0>, VEX_4V;
+ X86fhsub, 0>, VEX_4V, VEX_L;
}
let ExeDomain = SSEPackedDouble in {
defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
X86fhsub, 0>, VEX_4V;
defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
- X86fhadd, 0>, VEX_4V;
+ X86fhadd, 0>, VEX_4V, VEX_L;
defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
- X86fhsub, 0>, VEX_4V;
+ X86fhsub, 0>, VEX_4V, VEX_L;
}
}
(ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
- OpSize;
+ Sched<[WriteVecALU]>;
def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src),
[(set VR128:$dst,
(IntId128
(bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
- OpSize;
+ Sched<[WriteVecALULd]>;
}
/// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
(ins VR256:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (IntId256 VR256:$src))]>,
- OpSize;
+ Sched<[WriteVecALU]>;
def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
(ins i256mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
(IntId256
- (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
+ (bitconvert (memopv4i64 addr:$src))))]>,
+ Sched<[WriteVecALULd]>;
}
+// Helper fragments to match sext vXi1 to vXiY.
+def v16i1sextv16i8 : PatLeaf<(v16i8 (X86pcmpgt (bc_v16i8 (v4i32 immAllZerosV)),
+ VR128:$src))>;
+def v8i1sextv8i16 : PatLeaf<(v8i16 (X86vsrai VR128:$src, (i8 15)))>;
+def v4i1sextv4i32 : PatLeaf<(v4i32 (X86vsrai VR128:$src, (i8 31)))>;
+def v32i1sextv32i8 : PatLeaf<(v32i8 (X86pcmpgt (bc_v32i8 (v8i32 immAllZerosV)),
+ VR256:$src))>;
+def v16i1sextv16i16: PatLeaf<(v16i16 (X86vsrai VR256:$src, (i8 15)))>;
+def v8i1sextv8i32 : PatLeaf<(v8i32 (X86vsrai VR256:$src, (i8 31)))>;
+
let Predicates = [HasAVX] in {
defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
int_x86_ssse3_pabs_b_128>, VEX;
int_x86_ssse3_pabs_w_128>, VEX;
defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
int_x86_ssse3_pabs_d_128>, VEX;
+
+ def : Pat<(xor
+ (bc_v2i64 (v16i1sextv16i8)),
+ (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
+ (VPABSBrr128 VR128:$src)>;
+ def : Pat<(xor
+ (bc_v2i64 (v8i1sextv8i16)),
+ (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
+ (VPABSWrr128 VR128:$src)>;
+ def : Pat<(xor
+ (bc_v2i64 (v4i1sextv4i32)),
+ (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
+ (VPABSDrr128 VR128:$src)>;
}
let Predicates = [HasAVX2] in {
defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
- int_x86_avx2_pabs_b>, VEX;
+ int_x86_avx2_pabs_b>, VEX, VEX_L;
defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
- int_x86_avx2_pabs_w>, VEX;
+ int_x86_avx2_pabs_w>, VEX, VEX_L;
defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
- int_x86_avx2_pabs_d>, VEX;
+ int_x86_avx2_pabs_d>, VEX, VEX_L;
+
+ def : Pat<(xor
+ (bc_v4i64 (v32i1sextv32i8)),
+ (bc_v4i64 (add (v32i8 VR256:$src), (v32i1sextv32i8)))),
+ (VPABSBrr256 VR256:$src)>;
+ def : Pat<(xor
+ (bc_v4i64 (v16i1sextv16i16)),
+ (bc_v4i64 (add (v16i16 VR256:$src), (v16i1sextv16i16)))),
+ (VPABSWrr256 VR256:$src)>;
+ def : Pat<(xor
+ (bc_v4i64 (v8i1sextv8i32)),
+ (bc_v4i64 (add (v8i32 VR256:$src), (v8i1sextv8i32)))),
+ (VPABSDrr256 VR256:$src)>;
}
defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
int_x86_ssse3_pabs_d_128>;
+let Predicates = [HasSSSE3] in {
+ def : Pat<(xor
+ (bc_v2i64 (v16i1sextv16i8)),
+ (bc_v2i64 (add (v16i8 VR128:$src), (v16i1sextv16i8)))),
+ (PABSBrr128 VR128:$src)>;
+ def : Pat<(xor
+ (bc_v2i64 (v8i1sextv8i16)),
+ (bc_v2i64 (add (v8i16 VR128:$src), (v8i1sextv8i16)))),
+ (PABSWrr128 VR128:$src)>;
+ def : Pat<(xor
+ (bc_v2i64 (v4i1sextv4i32)),
+ (bc_v2i64 (add (v4i32 VR128:$src), (v4i1sextv4i32)))),
+ (PABSDrr128 VR128:$src)>;
+}
+
//===---------------------------------------------------------------------===//
// SSSE3 - Packed Binary Operator Instructions
//===---------------------------------------------------------------------===//
+let Sched = WriteVecALU in {
def SSE_PHADDSUBD : OpndItins<
IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
>;
def SSE_PHADDSUBW : OpndItins<
IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
>;
+}
+let Sched = WriteShuffle in
def SSE_PSHUFB : OpndItins<
IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
>;
+let Sched = WriteVecALU in
def SSE_PSIGN : OpndItins<
IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
>;
+let Sched = WriteVecIMul in
def SSE_PMULHRSW : OpndItins<
IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
>;
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
- OpSize;
+ Sched<[itins.Sched]>;
def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst,
(OpVT (OpNode RC:$src1,
- (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
+ (bitconvert (memop_frag addr:$src2)))))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
/// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
- OpSize;
+ Sched<[itins.Sched]>;
def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
(IntId128 VR128:$src1,
- (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
+ (bitconvert (memopv2i64 addr:$src2))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId256> {
+ Intrinsic IntId256,
+ X86FoldableSchedWrite Sched> {
let isCommutable = 1 in
def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
- OpSize;
+ Sched<[Sched]>;
def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (IntId256 VR256:$src1,
- (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
+ (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
let ImmT = NoImm, Predicates = [HasAVX] in {
let isCommutable = 0 in {
defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PHADDSUBW, 0>, VEX_4V;
defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PHADDSUBD, 0>, VEX_4V;
defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PHADDSUBW, 0>, VEX_4V;
defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PHADDSUBD, 0>, VEX_4V;
defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PSIGN, 0>, VEX_4V;
defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PSIGN, 0>, VEX_4V;
defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PSIGN, 0>, VEX_4V;
defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
- memopv2i64, i128mem,
+ loadv2i64, i128mem,
SSE_PSHUFB, 0>, VEX_4V;
defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
int_x86_ssse3_phadd_sw_128,
let ImmT = NoImm, Predicates = [HasAVX2] in {
let isCommutable = 0 in {
defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
- memopv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V;
+ loadv4i64, i256mem,
+ SSE_PSHUFB, 0>, VEX_4V, VEX_L;
defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
- int_x86_avx2_phadd_sw>, VEX_4V;
+ int_x86_avx2_phadd_sw,
+ WriteVecALU>, VEX_4V, VEX_L;
defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
- int_x86_avx2_phsub_sw>, VEX_4V;
+ int_x86_avx2_phsub_sw,
+ WriteVecALU>, VEX_4V, VEX_L;
defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
- int_x86_avx2_pmadd_ub_sw>, VEX_4V;
+ int_x86_avx2_pmadd_ub_sw,
+ WriteVecIMul>, VEX_4V, VEX_L;
}
defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
- int_x86_avx2_pmul_hr_sw>, VEX_4V;
+ int_x86_avx2_pmul_hr_sw,
+ WriteVecIMul>, VEX_4V, VEX_L;
}
// None of these have i8 immediate fields.
// SSSE3 - Packed Align Instruction Patterns
//===---------------------------------------------------------------------===//
-multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
let neverHasSideEffects = 1 in {
def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [], IIC_SSE_PALIGNR>, OpSize;
+ [], IIC_SSE_PALIGNRR>, Sched<[WriteShuffle]>;
let mayLoad = 1 in
def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [], IIC_SSE_PALIGNR>, OpSize;
+ [], IIC_SSE_PALIGNRM>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
}
-multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
let neverHasSideEffects = 1 in {
def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
- []>, OpSize;
+ []>, Sched<[WriteShuffle]>;
let mayLoad = 1 in
def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2, i8imm:$src3),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
- []>, OpSize;
+ []>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
}
let Predicates = [HasAVX] in
- defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
+ defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
let Predicates = [HasAVX2] in
- defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
+ defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
- defm PALIGN : ssse3_palign<"palignr">;
+ defm PALIGN : ssse3_palignr<"palignr">;
let Predicates = [HasAVX2] in {
-def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
}
let Predicates = [HasAVX] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
let Predicates = [UseSSSE3] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
// SSSE3 - Thread synchronization
//===---------------------------------------------------------------------===//
+let SchedRW = [WriteSystem] in {
let usesCustomInserter = 1 in {
def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
[(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
[(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
TB, Requires<[HasSSE3]>;
+} // SchedRW
-def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
-def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
+def : InstAlias<"mwait\t{%eax, %ecx|ecx, eax}", (MWAITrr)>, Requires<[Not64BitMode]>;
+def : InstAlias<"mwait\t{%rax, %rcx|rcx, rax}", (MWAITrr)>, Requires<[In64BitMode]>;
-def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
- Requires<[In32BitMode]>;
-def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
+def : InstAlias<"monitor\t{%eax, %ecx, %edx|edx, ecx, eax}", (MONITORrrr)>,
+ Requires<[Not64BitMode]>;
+def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
Requires<[In64BitMode]>;
//===----------------------------------------------------------------------===//
// SSE4.1 - Packed Move with Sign/Zero Extend
//===----------------------------------------------------------------------===//
-multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
+multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
+ OpndItins itins = DEFAULT_ITINS> {
def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR128:$dst, (IntId VR128:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
- OpSize;
+ (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
+ itins.rm>, Sched<[itins.Sched.Folded]>;
}
multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
+ Intrinsic IntId, X86FoldableSchedWrite Sched> {
def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>;
def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
+ [(set VR256:$dst, (IntId (load addr:$src)))]>,
+ Sched<[Sched.Folded]>;
}
let Predicates = [HasAVX] in {
-defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
- VEX;
-defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
- VEX;
-defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
- VEX;
-defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
- VEX;
-defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
- VEX;
-defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
- VEX;
+defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
+ int_x86_sse41_pmovsxbw,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
+ int_x86_sse41_pmovsxwd,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
+ int_x86_sse41_pmovsxdq,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
+ int_x86_sse41_pmovzxbw,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
+ int_x86_sse41_pmovzxwd,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
+ int_x86_sse41_pmovzxdq,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
}
let Predicates = [HasAVX2] in {
defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
- int_x86_avx2_pmovsxbw>, VEX;
+ int_x86_avx2_pmovsxbw,
+ WriteShuffle>, VEX, VEX_L;
defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
- int_x86_avx2_pmovsxwd>, VEX;
+ int_x86_avx2_pmovsxwd,
+ WriteShuffle>, VEX, VEX_L;
defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
- int_x86_avx2_pmovsxdq>, VEX;
+ int_x86_avx2_pmovsxdq,
+ WriteShuffle>, VEX, VEX_L;
defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
- int_x86_avx2_pmovzxbw>, VEX;
+ int_x86_avx2_pmovzxbw,
+ WriteShuffle>, VEX, VEX_L;
defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
- int_x86_avx2_pmovzxwd>, VEX;
+ int_x86_avx2_pmovzxwd,
+ WriteShuffle>, VEX, VEX_L;
defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
- int_x86_avx2_pmovzxdq>, VEX;
-}
-
-defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
-defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
-defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
-defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
-defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
-defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
+ int_x86_avx2_pmovzxdq,
+ WriteShuffle>, VEX, VEX_L;
+}
+
+defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq,
+ SSE_INTALU_ITINS_SHUFF_P>;
let Predicates = [HasAVX] in {
// Common patterns involving scalar load.
(VPMOVSXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
(VPMOVSXBWrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
+ (VPMOVSXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
(VPMOVSXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
(VPMOVSXWDrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
+ (VPMOVSXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
(VPMOVSXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
(VPMOVSXDQrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
+ (VPMOVSXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
(VPMOVZXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
(VPMOVZXBWrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
+ (VPMOVZXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
(VPMOVZXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
(VPMOVZXWDrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
+ (VPMOVZXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
(VPMOVZXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
(VPMOVZXDQrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
+ (VPMOVZXDQrm addr:$src)>;
}
let Predicates = [UseSSE41] in {
(PMOVSXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
(PMOVSXBWrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
+ (PMOVSXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
(PMOVSXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
(PMOVSXWDrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
+ (PMOVSXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
(PMOVSXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
(PMOVSXDQrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
+ (PMOVSXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
(PMOVZXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
(PMOVZXBWrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
+ (PMOVZXBWrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
(PMOVZXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
(PMOVZXWDrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
+ (PMOVZXWDrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
(PMOVZXDQrm addr:$src)>;
def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
(PMOVZXDQrm addr:$src)>;
+ def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
+ (PMOVZXDQrm addr:$src)>;
}
-let Predicates = [HasAVX2] in {
- let AddedComplexity = 15 in {
- def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
- (VPMOVZXDQYrr VR128:$src)>;
- def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
- (VPMOVZXWDYrr VR128:$src)>;
- }
-
- def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
- def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
-}
-
-let Predicates = [HasAVX] in {
- def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
- def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
-}
-
-let Predicates = [UseSSE41] in {
- def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
- def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
-}
-
-
-multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
+multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
+ OpndItins itins = DEFAULT_ITINS> {
def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR128:$dst, (IntId VR128:$src))], itins.rr>,
+ Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
- OpSize;
+ (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
+ itins.rm>, Sched<[itins.Sched.Folded]>;
}
multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
+ Intrinsic IntId, X86FoldableSchedWrite Sched> {
def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>;
def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
(IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
- OpSize;
+ Sched<[Sched.Folded]>;
}
let Predicates = [HasAVX] in {
-defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
- VEX;
-defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
- VEX;
-defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
- VEX;
-defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
- VEX;
+defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
+defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq,
+ DEFAULT_ITINS_SHUFFLESCHED>, VEX;
}
let Predicates = [HasAVX2] in {
defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
- int_x86_avx2_pmovsxbd>, VEX;
+ int_x86_avx2_pmovsxbd, WriteShuffle>,
+ VEX, VEX_L;
defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
- int_x86_avx2_pmovsxwq>, VEX;
+ int_x86_avx2_pmovsxwq, WriteShuffle>,
+ VEX, VEX_L;
defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
- int_x86_avx2_pmovzxbd>, VEX;
+ int_x86_avx2_pmovzxbd, WriteShuffle>,
+ VEX, VEX_L;
defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
- int_x86_avx2_pmovzxwq>, VEX;
+ int_x86_avx2_pmovzxwq, WriteShuffle>,
+ VEX, VEX_L;
}
-defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
-defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
-defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
-defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
+defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
+ SSE_INTALU_ITINS_SHUFF_P>;
+defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
+ SSE_INTALU_ITINS_SHUFF_P>;
let Predicates = [HasAVX] in {
// Common patterns involving scalar load
(PMOVZXWQrm addr:$src)>;
}
-multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
+multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
+ X86FoldableSchedWrite Sched> {
def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR128:$dst, (IntId VR128:$src))]>, Sched<[Sched]>;
// Expecting a i16 load any extended to i32 value.
def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (IntId (bitconvert
(v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
- OpSize;
+ Sched<[Sched.Folded]>;
}
multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
+ Intrinsic IntId, X86FoldableSchedWrite Sched> {
def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
+ [(set VR256:$dst, (IntId VR128:$src))]>, Sched<[Sched]>;
// Expecting a i16 load any extended to i32 value.
def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (IntId (bitconvert
(v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
- OpSize;
+ Sched<[Sched.Folded]>;
+}
+
+let Predicates = [HasAVX] in {
+defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq,
+ WriteShuffle>, VEX;
+defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq,
+ WriteShuffle>, VEX;
}
+let Predicates = [HasAVX2] in {
+defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq", int_x86_avx2_pmovsxbq,
+ WriteShuffle>, VEX, VEX_L;
+defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq", int_x86_avx2_pmovzxbq,
+ WriteShuffle>, VEX, VEX_L;
+}
+defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
+ WriteShuffle>;
+defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
+ WriteShuffle>;
+
+let Predicates = [HasAVX2] in {
+ def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
+ def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
+
+ def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
+
+ def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
+ (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
+ (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
+ (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (X86vsext (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
+ (VPMOVSXWDYrm addr:$src)>;
+ def : Pat<(v4i64 (X86vsext (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
+ (VPMOVSXDQYrm addr:$src)>;
+
+ def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXBDYrm addr:$src)>;
+ def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXBDYrm addr:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXWQYrm addr:$src)>;
+ def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXWQYrm addr:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXBQYrm addr:$src)>;
+}
+
+let Predicates = [HasAVX] in {
+ // Common patterns involving scalar load
+ def : Pat<(int_x86_sse41_pmovsxbq
+ (bitconvert (v4i32 (X86vzmovl
+ (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXBQrm addr:$src)>;
+
+ def : Pat<(int_x86_sse41_pmovzxbq
+ (bitconvert (v4i32 (X86vzmovl
+ (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVZXBQrm addr:$src)>;
+}
+
+let Predicates = [UseSSE41] in {
+ def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
+
+ def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
+
+ // Common patterns involving scalar load
+ def : Pat<(int_x86_sse41_pmovsxbq
+ (bitconvert (v4i32 (X86vzmovl
+ (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVSXBQrm addr:$src)>;
+
+ def : Pat<(int_x86_sse41_pmovzxbq
+ (bitconvert (v4i32 (X86vzmovl
+ (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVSXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVSXWQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (extloadi32i16 addr:$src))))))),
+ (PMOVSXBQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXDQrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVSXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVSXBWrm addr:$src)>;
+}
+
+let Predicates = [HasAVX2] in {
+ def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
+ def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
+
+ def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
+
+ def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
+
+ def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
+ (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
+ (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
+ (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
+ (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+}
+
+let Predicates = [HasAVX] in {
+ def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
+
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXBWrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVZXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
+ (VPMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVZXWQrm addr:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
+ (VPMOVZXDQrm addr:$src)>;
-let Predicates = [HasAVX] in {
-defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
- VEX;
-defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
- VEX;
-}
-let Predicates = [HasAVX2] in {
-defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
- int_x86_avx2_pmovsxbq>, VEX;
-defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
- int_x86_avx2_pmovzxbq>, VEX;
-}
-defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
-defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
+ def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
-let Predicates = [HasAVX] in {
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVSXBQrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVZXBQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXDQrm addr:$src)>;
+ def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXDQrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
+ (scalar_to_vector (loadi64 addr:$src))))))),
+ (VPMOVSXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
+ (scalar_to_vector (loadf64 addr:$src))))))),
+ (VPMOVSXBWrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
+ (scalar_to_vector (loadi32 addr:$src))))))),
+ (VPMOVSXWQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
+ (scalar_to_vector (extloadi32i16 addr:$src))))))),
+ (VPMOVSXBQrm addr:$src)>;
}
let Predicates = [UseSSE41] in {
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVSXBQrm addr:$src)>;
+ def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
+
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXBWrm addr:$src)>;
+ def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXBWrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVZXBDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
(PMOVZXBQrm addr:$src)>;
+
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXWDrm addr:$src)>;
+ def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXWDrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
+ (PMOVZXWQrm addr:$src)>;
+
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
+ (PMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
+ (PMOVZXDQrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
+ (PMOVZXDQrm addr:$src)>;
}
//===----------------------------------------------------------------------===//
/// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
- def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
+ def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
(ins VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
- OpSize;
- let neverHasSideEffects = 1, mayStore = 1 in
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set GR32orGR64:$dst, (X86pextrb (v16i8 VR128:$src1),
+ imm:$src2))]>,
+ Sched<[WriteShuffle]>;
+ let neverHasSideEffects = 1, mayStore = 1,
+ SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, OpSize;
-// FIXME:
-// There's an AssertZext in the way of writing the store pattern
-// (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(store (i8 (trunc (assertzext (X86pextrb (v16i8 VR128:$src1),
+ imm:$src2)))), addr:$dst)]>;
}
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX] in
defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
- def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
- (ins VR128:$src1, i32i8imm:$src2),
- "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
-}
defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
/// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
- let neverHasSideEffects = 1, mayStore = 1 in
+ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
+ def rr_REV : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
+ (ins VR128:$src1, i32i8imm:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, Sched<[WriteShuffle]>;
+
+ let neverHasSideEffects = 1, mayStore = 1,
+ SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, OpSize;
-// FIXME:
-// There's an AssertZext in the way of writing the store pattern
-// (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
+ [(store (i16 (trunc (assertzext (X86pextrw (v8i16 VR128:$src1),
+ imm:$src2)))), addr:$dst)]>;
}
let Predicates = [HasAVX] in
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32:$dst,
- (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
+ (extractelt (v4i32 VR128:$src1), imm:$src2))]>,
+ Sched<[WriteShuffle]>;
+ let SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(store (extractelt (v4i32 VR128:$src1), imm:$src2),
- addr:$dst)]>, OpSize;
+ addr:$dst)]>;
}
let Predicates = [HasAVX] in
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR64:$dst,
- (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
+ (extractelt (v2i64 VR128:$src1), imm:$src2))]>,
+ Sched<[WriteShuffle]>, REX_W;
+ let SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(store (extractelt (v2i64 VR128:$src1), imm:$src2),
- addr:$dst)]>, OpSize, REX_W;
+ addr:$dst)]>, REX_W;
}
let Predicates = [HasAVX] in
/// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
/// destination
-multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
- def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
+multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr,
+ OpndItins itins = DEFAULT_ITINS> {
+ def rr : SS4AIi8<opc, MRMDestReg, (outs GR32orGR64:$dst),
(ins VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set GR32:$dst,
- (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
- OpSize;
+ [(set GR32orGR64:$dst,
+ (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
+ itins.rr>, Sched<[WriteFBlend]>;
+ let SchedRW = [WriteFBlendLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
- addr:$dst)]>, OpSize;
+ addr:$dst)], itins.rm>;
}
let ExeDomain = SSEPackedSingle in {
- let Predicates = [HasAVX] in {
+ let Predicates = [UseAVX] in
defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
- def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
- (ins VR128:$src1, i32i8imm:$src2),
- "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
- []>, OpSize, VEX;
- }
- defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
+ defm EXTRACTPS : SS41I_extractf32<0x17, "extractps", SSE_EXTRACT_ITINS>;
}
// Also match an EXTRACTPS store when the store is done as f32 instead of i32.
multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
+ (ins VR128:$src1, GR32orGR64:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
+ (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>,
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
- imm:$src3))]>, OpSize;
+ imm:$src3))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
- OpSize;
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
- imm:$src3)))]>, OpSize;
+ imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
- OpSize;
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
- imm:$src3)))]>, OpSize;
+ imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
// are optimized inserts that won't zero arbitrary elements in the destination
// vector. The next one matches the intrinsic and could zero arbitrary elements
// in the target vector.
-multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
+multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
+ OpndItins itins = DEFAULT_ITINS> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
- OpSize;
+ (X86insertps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
+ Sched<[WriteFShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
+ (ins VR128:$src1, f32mem:$src2, i8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86insrtps VR128:$src1,
+ (X86insertps VR128:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
- imm:$src3))]>, OpSize;
+ imm:$src3))], itins.rm>,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
let ExeDomain = SSEPackedSingle in {
- let Predicates = [HasAVX] in
+ let Predicates = [UseAVX] in
defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
- defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
+ defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
+}
+
+let Predicates = [UseSSE41] in {
+ // If we're inserting an element from a load or a null pshuf of a load,
+ // fold the load into the insertps instruction.
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1), (X86PShufd (v4f32
+ (scalar_to_vector (loadf32 addr:$src2))), (i8 0)),
+ imm:$src3)),
+ (INSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1), (X86PShufd
+ (loadv4f32 addr:$src2), (i8 0)), imm:$src3)),
+ (INSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+}
+
+let Predicates = [UseAVX] in {
+ // If we're inserting an element from a vbroadcast of a load, fold the
+ // load into the X86insertps instruction.
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1),
+ (X86VBroadcast (loadf32 addr:$src2)), imm:$src3)),
+ (VINSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1),
+ (X86VBroadcast (loadv4f32 addr:$src2)), imm:$src3)),
+ (VINSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
}
//===----------------------------------------------------------------------===//
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
- OpSize;
+ [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAdd]>;
// Vector intrinsic operation, mem
def PSm : SS4AIi8<opcps, MRMSrcMem,
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
- (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
- OpSize;
+ (V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
+ IIC_SSE_ROUNDPS_MEM>, Sched<[WriteFAddLd]>;
} // ExeDomain = SSEPackedSingle
let ExeDomain = SSEPackedDouble in {
(outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
- OpSize;
+ [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAdd]>;
// Vector intrinsic operation, mem
def PDm : SS4AIi8<opcpd, MRMSrcMem,
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
- (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
- OpSize;
+ (V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAddLd]>;
} // ExeDomain = SSEPackedDouble
}
Intrinsic F64Int, bit Is2Addr = 1> {
let ExeDomain = GenericDomain in {
// Operation, reg.
+ let hasSideEffects = 0 in
def SSr : SS4AIi8<opcss, MRMSrcReg,
(outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
!if(Is2Addr,
"ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>, OpSize;
+ []>, Sched<[WriteFAdd]>;
// Intrinsic operation, reg.
+ let isCodeGenOnly = 1 in
def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
- OpSize;
+ Sched<[WriteFAdd]>;
// Intrinsic operation, mem.
def SSm : SS4AIi8<opcss, MRMSrcMem,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
- OpSize;
+ Sched<[WriteFAddLd, ReadAfterLd]>;
// Operation, reg.
+ let hasSideEffects = 0 in
def SDr : SS4AIi8<opcsd, MRMSrcReg,
(outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
!if(Is2Addr,
"sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>, OpSize;
+ []>, Sched<[WriteFAdd]>;
// Intrinsic operation, reg.
+ let isCodeGenOnly = 1 in
def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
- OpSize;
+ Sched<[WriteFAdd]>;
// Intrinsic operation, mem.
def SDm : SS4AIi8<opcsd, MRMSrcMem,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
- OpSize;
+ Sched<[WriteFAddLd, ReadAfterLd]>;
} // ExeDomain = GenericDomain
}
let Predicates = [HasAVX] in {
// Intrinsic form
defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
- memopv4f32, memopv2f64,
+ loadv4f32, loadv2f64,
int_x86_sse41_round_ps,
int_x86_sse41_round_pd>, VEX;
defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
- memopv8f32, memopv4f64,
+ loadv8f32, loadv4f64,
int_x86_avx_round_ps_256,
- int_x86_avx_round_pd_256>, VEX;
+ int_x86_avx_round_pd_256>, VEX, VEX_L;
defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
int_x86_sse41_round_ss,
int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
def : Pat<(v4f32 (ffloor VR128:$src)),
(VROUNDPSr VR128:$src, (i32 0x1))>;
+ def : Pat<(v4f32 (fnearbyint VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0xC))>;
+ def : Pat<(v4f32 (fceil VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x2))>;
+ def : Pat<(v4f32 (frint VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x4))>;
+ def : Pat<(v4f32 (ftrunc VR128:$src)),
+ (VROUNDPSr VR128:$src, (i32 0x3))>;
+
def : Pat<(v2f64 (ffloor VR128:$src)),
(VROUNDPDr VR128:$src, (i32 0x1))>;
+ def : Pat<(v2f64 (fnearbyint VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0xC))>;
+ def : Pat<(v2f64 (fceil VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x2))>;
+ def : Pat<(v2f64 (frint VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x4))>;
+ def : Pat<(v2f64 (ftrunc VR128:$src)),
+ (VROUNDPDr VR128:$src, (i32 0x3))>;
+
def : Pat<(v8f32 (ffloor VR256:$src)),
(VROUNDYPSr VR256:$src, (i32 0x1))>;
+ def : Pat<(v8f32 (fnearbyint VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0xC))>;
+ def : Pat<(v8f32 (fceil VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x2))>;
+ def : Pat<(v8f32 (frint VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x4))>;
+ def : Pat<(v8f32 (ftrunc VR256:$src)),
+ (VROUNDYPSr VR256:$src, (i32 0x3))>;
+
def : Pat<(v4f64 (ffloor VR256:$src)),
(VROUNDYPDr VR256:$src, (i32 0x1))>;
+ def : Pat<(v4f64 (fnearbyint VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0xC))>;
+ def : Pat<(v4f64 (fceil VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x2))>;
+ def : Pat<(v4f64 (frint VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x4))>;
+ def : Pat<(v4f64 (ftrunc VR256:$src)),
+ (VROUNDYPDr VR256:$src, (i32 0x3))>;
}
defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
def : Pat<(v4f32 (ffloor VR128:$src)),
(ROUNDPSr VR128:$src, (i32 0x1))>;
+ def : Pat<(v4f32 (fnearbyint VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0xC))>;
+ def : Pat<(v4f32 (fceil VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x2))>;
+ def : Pat<(v4f32 (frint VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x4))>;
+ def : Pat<(v4f32 (ftrunc VR128:$src)),
+ (ROUNDPSr VR128:$src, (i32 0x3))>;
+
def : Pat<(v2f64 (ffloor VR128:$src)),
(ROUNDPDr VR128:$src, (i32 0x1))>;
+ def : Pat<(v2f64 (fnearbyint VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0xC))>;
+ def : Pat<(v2f64 (fceil VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x2))>;
+ def : Pat<(v2f64 (frint VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x4))>;
+ def : Pat<(v2f64 (ftrunc VR128:$src)),
+ (ROUNDPDr VR128:$src, (i32 0x3))>;
}
//===----------------------------------------------------------------------===//
def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
- OpSize, VEX;
+ Sched<[WriteVecLogic]>, VEX;
def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
- OpSize, VEX;
+ [(set EFLAGS,(X86ptest VR128:$src1, (loadv2i64 addr:$src2)))]>,
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX;
def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
- OpSize, VEX;
+ Sched<[WriteVecLogic]>, VEX, VEX_L;
def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
- OpSize, VEX;
+ [(set EFLAGS,(X86ptest VR256:$src1, (loadv4i64 addr:$src2)))]>,
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX, VEX_L;
}
let Defs = [EFLAGS] in {
def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"ptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
- OpSize;
+ Sched<[WriteVecLogic]>;
def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"ptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
- OpSize;
+ Sched<[WriteVecLogicLd, ReadAfterLd]>;
}
// The bit test instructions below are AVX only
X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
- [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
+ [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>,
+ Sched<[WriteVecLogic]>, VEX;
def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
[(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
- OpSize, VEX;
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX;
}
let Defs = [EFLAGS], Predicates = [HasAVX] in {
let ExeDomain = SSEPackedSingle in {
-defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
-defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
+defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, loadv4f32, v4f32>;
+defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, loadv8f32, v8f32>,
+ VEX_L;
}
let ExeDomain = SSEPackedDouble in {
-defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
-defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
+defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, loadv2f64, v2f64>;
+defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, loadv4f64, v4f64>,
+ VEX_L;
}
}
let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
"popcnt{w}\t{$src, $dst|$dst, $src}",
- [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
- OpSize, XS;
+ [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>,
+ OpSize16, XS;
def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
"popcnt{w}\t{$src, $dst|$dst, $src}",
[(set GR16:$dst, (ctpop (loadi16 addr:$src))),
- (implicit EFLAGS)]>, OpSize, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, OpSize16, XS;
def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
"popcnt{l}\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
- XS;
+ [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>,
+ OpSize32, XS;
+
def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
"popcnt{l}\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (ctpop (loadi32 addr:$src))),
- (implicit EFLAGS)]>, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, OpSize32, XS;
def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
"popcnt{q}\t{$src, $dst|$dst, $src}",
- [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
- XS;
+ [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>, XS;
def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
"popcnt{q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (ctpop (loadi64 addr:$src))),
- (implicit EFLAGS)]>, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, XS;
}
// SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
- Intrinsic IntId128> {
+ Intrinsic IntId128,
+ X86FoldableSchedWrite Sched> {
def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
+ [(set VR128:$dst, (IntId128 VR128:$src))]>,
+ Sched<[Sched]>;
def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (IntId128
- (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
+ (IntId128 (bitconvert (memopv2i64 addr:$src))))]>,
+ Sched<[Sched.Folded]>;
}
+// PHMIN has the same profile as PSAD, thus we use the same scheduling
+// model, although the naming is misleading.
let Predicates = [HasAVX] in
defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
- int_x86_sse41_phminposuw>, VEX;
+ int_x86_sse41_phminposuw,
+ WriteVecIMul>, VEX;
defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
- int_x86_sse41_phminposuw>;
+ int_x86_sse41_phminposuw,
+ WriteVecIMul>;
/// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
- Intrinsic IntId128, bit Is2Addr = 1> {
+ Intrinsic IntId128, bit Is2Addr = 1,
+ OpndItins itins = DEFAULT_ITINS> {
let isCommutable = 1 in
def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
+ [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
+ itins.rr>, Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
- (IntId128 VR128:$src1,
- (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
+ (IntId128 VR128:$src1, (bitconvert (memopv2i64 addr:$src2))))],
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
/// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId256> {
+ Intrinsic IntId256,
+ X86FoldableSchedWrite Sched> {
let isCommutable = 1 in
def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
+ [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
+ Sched<[Sched]>;
def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (IntId256 VR256:$src1,
- (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
-}
-
-let Predicates = [HasAVX] in {
- let isCommutable = 0 in
- defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
- 0>, VEX_4V;
- defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
- 0>, VEX_4V;
- defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
- 0>, VEX_4V;
- defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
- 0>, VEX_4V;
- defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
- 0>, VEX_4V;
- defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
- 0>, VEX_4V;
- defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
- 0>, VEX_4V;
- defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
- 0>, VEX_4V;
- defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
- 0>, VEX_4V;
- defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
- 0>, VEX_4V;
-}
-
-let Predicates = [HasAVX2] in {
- let isCommutable = 0 in
- defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
- int_x86_avx2_packusdw>, VEX_4V;
- defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
- int_x86_avx2_pmins_b>, VEX_4V;
- defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
- int_x86_avx2_pmins_d>, VEX_4V;
- defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
- int_x86_avx2_pminu_d>, VEX_4V;
- defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
- int_x86_avx2_pminu_w>, VEX_4V;
- defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
- int_x86_avx2_pmaxs_b>, VEX_4V;
- defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
- int_x86_avx2_pmaxs_d>, VEX_4V;
- defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
- int_x86_avx2_pmaxu_d>, VEX_4V;
- defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
- int_x86_avx2_pmaxu_w>, VEX_4V;
- defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
- int_x86_avx2_pmul_dq>, VEX_4V;
+ (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
-let Constraints = "$src1 = $dst" in {
- let isCommutable = 0 in
- defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
- defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
- defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
- defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
- defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
- defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
- defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
- defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
- defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
- defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
-}
/// SS48I_binop_rm - Simple SSE41 binary operator.
multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
- X86MemOperand x86memop, bit Is2Addr = 1> {
+ X86MemOperand x86memop, bit Is2Addr = 1,
+ OpndItins itins = SSE_INTALU_ITINS_P> {
let isCommutable = 1 in
def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
+ [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
+ Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst,
- (OpVT (OpNode RC:$src1,
- (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
+ (OpVT (OpNode RC:$src1, (bitconvert (memop_frag addr:$src2)))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+/// SS48I_binop_rm2 - Simple SSE41 binary operator with different src and dst
+/// types.
+multiclass SS48I_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ ValueType DstVT, ValueType SrcVT, RegisterClass RC,
+ PatFrag memop_frag, X86MemOperand x86memop,
+ OpndItins itins,
+ bit IsCommutable = 0, bit Is2Addr = 1> {
+ let isCommutable = IsCommutable in
+ def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
+ Sched<[itins.Sched]>;
+ def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
+ (ins RC:$src1, x86memop:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
+ (bitconvert (memop_frag addr:$src2)))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+let Predicates = [HasAVX] in {
+ let isCommutable = 0 in
+ defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMULDQ : SS48I_binop_rm2<0x28, "vpmuldq", X86pmuldq, v2i64, v4i32,
+ VR128, loadv2i64, i128mem,
+ SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
+}
+
+let Predicates = [HasAVX2] in {
+ let isCommutable = 0 in
+ defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMULDQY : SS48I_binop_rm2<0x28, "vpmuldq", X86pmuldq, v4i64, v8i32,
+ VR256, loadv4i64, i256mem,
+ SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
+}
+
+let Constraints = "$src1 = $dst" in {
+ let isCommutable = 0 in
+ defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
+ memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
+ defm PMULDQ : SS48I_binop_rm2<0x28, "pmuldq", X86pmuldq, v2i64, v4i32,
+ VR128, memopv2i64, i128mem,
+ SSE_INTMUL_ITINS_P, 1>;
}
let Predicates = [HasAVX] in {
defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
- memopv2i64, i128mem, 0>, VEX_4V;
+ memopv2i64, i128mem, 0, SSE_PMULLD_ITINS>,
+ VEX_4V;
defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
- memopv2i64, i128mem, 0>, VEX_4V;
+ memopv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
}
let Predicates = [HasAVX2] in {
defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
- memopv4i64, i256mem, 0>, VEX_4V;
+ memopv4i64, i256mem, 0, SSE_PMULLD_ITINS>,
+ VEX_4V, VEX_L;
defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
- memopv4i64, i256mem, 0>, VEX_4V;
+ memopv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
}
let Constraints = "$src1 = $dst" in {
defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
- memopv2i64, i128mem>;
+ memopv2i64, i128mem, 1, SSE_PMULLD_ITINS>;
defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
- memopv2i64, i128mem>;
+ memopv2i64, i128mem, 1, SSE_INTALUQ_ITINS_P>;
}
/// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
- X86MemOperand x86memop, bit Is2Addr = 1> {
+ X86MemOperand x86memop, bit Is2Addr = 1,
+ OpndItins itins = DEFAULT_ITINS> {
let isCommutable = 1 in
def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, u32u8imm:$src3),
+ (ins RC:$src1, RC:$src2, i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
- OpSize;
+ [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
+ Sched<[itins.Sched]>;
def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, i8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set RC:$dst,
(IntId RC:$src1,
- (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
- OpSize;
+ (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let isCommutable = 0 in {
let ExeDomain = SSEPackedSingle in {
defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
- VR128, memopv4f32, f128mem, 0>, VEX_4V;
+ VR128, loadv4f32, f128mem, 0,
+ DEFAULT_ITINS_FBLENDSCHED>, VEX_4V;
defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
- int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
+ int_x86_avx_blend_ps_256, VR256, loadv8f32,
+ f256mem, 0, DEFAULT_ITINS_FBLENDSCHED>,
+ VEX_4V, VEX_L;
}
let ExeDomain = SSEPackedDouble in {
defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
- VR128, memopv2f64, f128mem, 0>, VEX_4V;
+ VR128, loadv2f64, f128mem, 0,
+ DEFAULT_ITINS_FBLENDSCHED>, VEX_4V;
defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
- int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
+ int_x86_avx_blend_pd_256,VR256, loadv4f64,
+ f256mem, 0, DEFAULT_ITINS_FBLENDSCHED>,
+ VEX_4V, VEX_L;
}
defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
- VR128, memopv2i64, i128mem, 0>, VEX_4V;
+ VR128, loadv2i64, i128mem, 0,
+ DEFAULT_ITINS_BLENDSCHED>, VEX_4V;
defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
- VR128, memopv2i64, i128mem, 0>, VEX_4V;
+ VR128, loadv2i64, i128mem, 0,
+ DEFAULT_ITINS_MPSADSCHED>, VEX_4V;
}
let ExeDomain = SSEPackedSingle in
defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
- VR128, memopv4f32, f128mem, 0>, VEX_4V;
+ VR128, loadv4f32, f128mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V;
let ExeDomain = SSEPackedDouble in
defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
- VR128, memopv2f64, f128mem, 0>, VEX_4V;
+ VR128, loadv2f64, f128mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V;
let ExeDomain = SSEPackedSingle in
defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
- VR256, memopv8f32, i256mem, 0>, VEX_4V;
+ VR256, loadv8f32, i256mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX2] in {
let isCommutable = 0 in {
defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
- VR256, memopv4i64, i256mem, 0>, VEX_4V;
+ VR256, loadv4i64, i256mem, 0,
+ DEFAULT_ITINS_BLENDSCHED>, VEX_4V, VEX_L;
defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
- VR256, memopv4i64, i256mem, 0>, VEX_4V;
+ VR256, loadv4i64, i256mem, 0,
+ DEFAULT_ITINS_MPSADSCHED>, VEX_4V, VEX_L;
}
}
let isCommutable = 0 in {
let ExeDomain = SSEPackedSingle in
defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
- VR128, memopv4f32, f128mem>;
+ VR128, memopv4f32, f128mem,
+ 1, SSE_INTALU_ITINS_FBLEND_P>;
let ExeDomain = SSEPackedDouble in
defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
- VR128, memopv2f64, f128mem>;
+ VR128, memopv2f64, f128mem,
+ 1, SSE_INTALU_ITINS_FBLEND_P>;
defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
- VR128, memopv2i64, i128mem>;
+ VR128, memopv2i64, i128mem,
+ 1, SSE_INTALU_ITINS_BLEND_P>;
defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
- VR128, memopv2i64, i128mem>;
+ VR128, memopv2i64, i128mem,
+ 1, SSE_MPSADBW_ITINS>;
}
let ExeDomain = SSEPackedSingle in
defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
- VR128, memopv4f32, f128mem>;
+ VR128, memopv4f32, f128mem, 1,
+ SSE_DPPS_ITINS>;
let ExeDomain = SSEPackedDouble in
defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
- VR128, memopv2f64, f128mem>;
+ VR128, memopv2f64, f128mem, 1,
+ SSE_DPPD_ITINS>;
}
/// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
RegisterClass RC, X86MemOperand x86memop,
- PatFrag mem_frag, Intrinsic IntId> {
+ PatFrag mem_frag, Intrinsic IntId,
+ X86FoldableSchedWrite Sched> {
def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
- IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
+ NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM,
+ Sched<[Sched]>;
def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, RC:$src3),
[(set RC:$dst,
(IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
RC:$src3))],
- IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
+ NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let ExeDomain = SSEPackedDouble in {
defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
- memopv2f64, int_x86_sse41_blendvpd>;
+ loadv2f64, int_x86_sse41_blendvpd,
+ WriteFVarBlend>;
defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
- memopv4f64, int_x86_avx_blendv_pd_256>;
+ loadv4f64, int_x86_avx_blendv_pd_256,
+ WriteFVarBlend>, VEX_L;
} // ExeDomain = SSEPackedDouble
let ExeDomain = SSEPackedSingle in {
defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
- memopv4f32, int_x86_sse41_blendvps>;
+ loadv4f32, int_x86_sse41_blendvps,
+ WriteFVarBlend>;
defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
- memopv8f32, int_x86_avx_blendv_ps_256>;
+ loadv8f32, int_x86_avx_blendv_ps_256,
+ WriteFVarBlend>, VEX_L;
} // ExeDomain = SSEPackedSingle
defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
- memopv2i64, int_x86_sse41_pblendvb>;
+ loadv2i64, int_x86_sse41_pblendvb,
+ WriteVarBlend>;
}
let Predicates = [HasAVX2] in {
defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
- memopv4i64, int_x86_avx2_pblendvb>;
+ loadv4i64, int_x86_avx2_pblendvb,
+ WriteVarBlend>, VEX_L;
}
let Predicates = [HasAVX] in {
(v4f64 VR256:$src2))),
(VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
- def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
+ def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
(imm:$mask))),
- (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
- def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
+ (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+ def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
(imm:$mask))),
- (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+ (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
- def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+ def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
(imm:$mask))),
- (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+ (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
(imm:$mask))),
- (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+ (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
(imm:$mask))),
- (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+ (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
}
let Predicates = [HasAVX2] in {
def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
(v32i8 VR256:$src2))),
(VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
- def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
+ def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
(imm:$mask))),
- (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
+ (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+}
+
+// Patterns
+let Predicates = [UseAVX] in {
+ let AddedComplexity = 15 in {
+ // Move scalar to XMM zero-extended, zeroing a VR128 then do a
+ // MOVS{S,D} to the lower bits.
+ def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
+ (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
+ def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
+ (VBLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
+ (VBLENDPSrri (v4i32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
+ (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
+
+ // Move low f32 and clear high bits.
+ def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
+ (VBLENDPSYrri (v8f32 (AVX_SET0)), VR256:$src, (i8 1))>;
+ def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
+ (VBLENDPSYrri (v8i32 (AVX_SET0)), VR256:$src, (i8 1))>;
+ }
+
+ def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
+ (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
+ (SUBREG_TO_REG (i32 0),
+ (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
+ sub_xmm)>;
+ def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
+ (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
+ (SUBREG_TO_REG (i64 0),
+ (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
+ sub_xmm)>;
+
+ // Move low f64 and clear high bits.
+ def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
+ (VBLENDPDYrri (v4f64 (AVX_SET0)), VR256:$src, (i8 1))>;
+
+ def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
+ (VBLENDPDYrri (v4i64 (AVX_SET0)), VR256:$src, (i8 1))>;
+}
+
+let Predicates = [UseSSE41] in {
+ // With SSE41 we can use blends for these patterns.
+ def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
+ (BLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
+ (BLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
+ (BLENDPDrri (v2f64 (V_SET0)), VR128:$src, (i8 1))>;
}
+
/// SS41I_ternary_int - SSE 4.1 ternary operator
let Uses = [XMM0], Constraints = "$src1 = $dst" in {
multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
- X86MemOperand x86memop, Intrinsic IntId> {
+ X86MemOperand x86memop, Intrinsic IntId,
+ OpndItins itins = DEFAULT_ITINS> {
def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
- OpSize;
+ [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
+ itins.rr>, Sched<[itins.Sched]>;
def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, x86memop:$src2),
"\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst,
(IntId VR128:$src1,
- (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
+ (bitconvert (mem_frag addr:$src2)), XMM0))],
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
let ExeDomain = SSEPackedDouble in
defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
- int_x86_sse41_blendvpd>;
+ int_x86_sse41_blendvpd,
+ DEFAULT_ITINS_FBLENDSCHED>;
let ExeDomain = SSEPackedSingle in
defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
- int_x86_sse41_blendvps>;
+ int_x86_sse41_blendvps,
+ DEFAULT_ITINS_FBLENDSCHED>;
defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
- int_x86_sse41_pblendvb>;
+ int_x86_sse41_pblendvb,
+ DEFAULT_ITINS_VARBLENDSCHED>;
// Aliases with the implicit xmm0 argument
-def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
-def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
-def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
-def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
-def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
-def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
+def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
(PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
let Predicates = [UseSSE41] in {
(v2f64 VR128:$src2))),
(BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
- def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
+ def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
(imm:$mask))),
- (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
+ (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
(imm:$mask))),
- (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
- def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
+ (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
(imm:$mask))),
- (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
+ (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
}
+let SchedRW = [WriteLoad] in {
let Predicates = [HasAVX] in
def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
- OpSize, VEX;
+ VEX;
let Predicates = [HasAVX2] in
def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
[(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
- OpSize, VEX;
+ VEX, VEX_L;
def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movntdqa\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
- OpSize;
+ [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>;
+} // SchedRW
//===----------------------------------------------------------------------===//
// SSE4.2 - Compare Instructions
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
- OpSize;
+ [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst,
- (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
+ (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>;
}
let Predicates = [HasAVX] in
defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
- memopv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0>, VEX_4V;
let Predicates = [HasAVX2] in
defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
- memopv4i64, i256mem, 0>, VEX_4V;
+ loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in
defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
imm:$src3))]>;
def MEM : PseudoI<(outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
- VR128:$src1, (load addr:$src2), imm:$src3))]>;
+ [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
+ (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
let Defs = [EFLAGS], usesCustomInserter = 1 in {
defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
}
-let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
- def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+multiclass pcmpistrm_SS42AI<string asm> {
+ def rr : SS42AI<0x62, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, Sched<[WritePCmpIStrM]>;
let mayLoad = 1 in
- def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
+ def rm :SS42AI<0x62, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, Sched<[WritePCmpIStrMLd, ReadAfterLd]>;
}
let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
- def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
- let mayLoad = 1 in
- def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+ let Predicates = [HasAVX] in
+ defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
+ defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
}
// Packed Compare Explicit Length Strings, Return Mask
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
def MEM : PseudoI<(outs VR128:$dst),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
- VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
+ [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
+ (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
}
let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
}
-let Predicates = [HasAVX],
- Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+multiclass SS42AI_pcmpestrm<string asm> {
+ def rr : SS42AI<0x60, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, Sched<[WritePCmpEStrM]>;
let mayLoad = 1 in
- def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
+ def rm : SS42AI<0x60, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, Sched<[WritePCmpEStrMLd, ReadAfterLd]>;
}
let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
- let mayLoad = 1 in
- def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
+ let Predicates = [HasAVX] in
+ defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
+ defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
}
// Packed Compare Implicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
- multiclass SS42AI_pcmpistri<string asm> {
- def rr : SS42AI<0x63, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>, OpSize;
- let mayLoad = 1 in
- def rm : SS42AI<0x63, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
- !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>, OpSize;
- }
+multiclass pseudo_pcmpistri<string asm> {
+ def REG : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
+ def MEM : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
+ (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
-let Predicates = [HasAVX] in
-defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
-defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
+let Defs = [EFLAGS], usesCustomInserter = 1 in {
+ defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
+ defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpistri<string asm> {
+ def rr : SS42AI<0x63, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, Sched<[WritePCmpIStrI]>;
+ let mayLoad = 1 in
+ def rm : SS42AI<0x63, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
+ []>, Sched<[WritePCmpIStrILd, ReadAfterLd]>;
+}
+
+let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
+ let Predicates = [HasAVX] in
+ defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
+ defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
+}
// Packed Compare Explicit Length Strings, Return Index
-let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
- multiclass SS42AI_pcmpestri<string asm> {
- def rr : SS42AI<0x61, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
- !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>, OpSize;
- let mayLoad = 1 in
- def rm : SS42AI<0x61, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
- !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>, OpSize;
- }
+multiclass pseudo_pcmpestri<string asm> {
+ def REG : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
+ def MEM : PseudoI<(outs GR32:$dst),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ [(set GR32:$dst, EFLAGS,
+ (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
+ imm:$src5))]>;
}
-let Predicates = [HasAVX] in
-defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
-defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
+let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
+ defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
+ defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
+}
+
+multiclass SS42AI_pcmpestri<string asm> {
+ def rr : SS42AI<0x61, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, Sched<[WritePCmpEStrI]>;
+ let mayLoad = 1 in
+ def rm : SS42AI<0x61, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
+ []>, Sched<[WritePCmpEStrILd, ReadAfterLd]>;
+}
+
+let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+ let Predicates = [HasAVX] in
+ defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
+ defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
+}
//===----------------------------------------------------------------------===//
// SSE4.2 - CRC Instructions
// crc intrinsic instruction
// This set of instructions are only rm, the only difference is the size
// of r and m.
+class SS42I_crc32r<bits<8> opc, string asm, RegisterClass RCOut,
+ RegisterClass RCIn, SDPatternOperator Int> :
+ SS42FI<opc, MRMSrcReg, (outs RCOut:$dst), (ins RCOut:$src1, RCIn:$src2),
+ !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
+ [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>,
+ Sched<[WriteFAdd]>;
+
+class SS42I_crc32m<bits<8> opc, string asm, RegisterClass RCOut,
+ X86MemOperand x86memop, SDPatternOperator Int> :
+ SS42FI<opc, MRMSrcMem, (outs RCOut:$dst), (ins RCOut:$src1, x86memop:$src2),
+ !strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
+ [(set RCOut:$dst, (Int RCOut:$src1, (load addr:$src2)))],
+ IIC_CRC32_MEM>, Sched<[WriteFAddLd, ReadAfterLd]>;
+
let Constraints = "$src1 = $dst" in {
- def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
- (ins GR32:$src1, i8mem:$src2),
- "crc32{b} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_8 GR32:$src1,
- (load addr:$src2)))]>;
- def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
- (ins GR32:$src1, GR8:$src2),
- "crc32{b} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
- def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
- (ins GR32:$src1, i16mem:$src2),
- "crc32{w} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_16 GR32:$src1,
- (load addr:$src2)))]>,
- OpSize;
- def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
- (ins GR32:$src1, GR16:$src2),
- "crc32{w} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
- OpSize;
- def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
- (ins GR32:$src1, i32mem:$src2),
- "crc32{l} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_32 GR32:$src1,
- (load addr:$src2)))]>;
- def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
- (ins GR32:$src1, GR32:$src2),
- "crc32{l} \t{$src2, $src1|$src1, $src2}",
- [(set GR32:$dst,
- (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
- def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
- (ins GR64:$src1, i8mem:$src2),
- "crc32{b} \t{$src2, $src1|$src1, $src2}",
- [(set GR64:$dst,
- (int_x86_sse42_crc32_64_8 GR64:$src1,
- (load addr:$src2)))]>,
- REX_W;
- def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
- (ins GR64:$src1, GR8:$src2),
- "crc32{b} \t{$src2, $src1|$src1, $src2}",
- [(set GR64:$dst,
- (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
- REX_W;
- def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
- (ins GR64:$src1, i64mem:$src2),
- "crc32{q} \t{$src2, $src1|$src1, $src2}",
- [(set GR64:$dst,
- (int_x86_sse42_crc32_64_64 GR64:$src1,
- (load addr:$src2)))]>,
- REX_W;
- def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
- (ins GR64:$src1, GR64:$src2),
- "crc32{q} \t{$src2, $src1|$src1, $src2}",
- [(set GR64:$dst,
- (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
- REX_W;
+ def CRC32r32m8 : SS42I_crc32m<0xF0, "crc32{b}", GR32, i8mem,
+ int_x86_sse42_crc32_32_8>;
+ def CRC32r32r8 : SS42I_crc32r<0xF0, "crc32{b}", GR32, GR8,
+ int_x86_sse42_crc32_32_8>;
+ def CRC32r32m16 : SS42I_crc32m<0xF1, "crc32{w}", GR32, i16mem,
+ int_x86_sse42_crc32_32_16>, OpSize16;
+ def CRC32r32r16 : SS42I_crc32r<0xF1, "crc32{w}", GR32, GR16,
+ int_x86_sse42_crc32_32_16>, OpSize16;
+ def CRC32r32m32 : SS42I_crc32m<0xF1, "crc32{l}", GR32, i32mem,
+ int_x86_sse42_crc32_32_32>, OpSize32;
+ def CRC32r32r32 : SS42I_crc32r<0xF1, "crc32{l}", GR32, GR32,
+ int_x86_sse42_crc32_32_32>, OpSize32;
+ def CRC32r64m64 : SS42I_crc32m<0xF1, "crc32{q}", GR64, i64mem,
+ int_x86_sse42_crc32_64_64>, REX_W;
+ def CRC32r64r64 : SS42I_crc32r<0xF1, "crc32{q}", GR64, GR64,
+ int_x86_sse42_crc32_64_64>, REX_W;
+ let hasSideEffects = 0 in {
+ let mayLoad = 1 in
+ def CRC32r64m8 : SS42I_crc32m<0xF0, "crc32{b}", GR64, i8mem,
+ null_frag>, REX_W;
+ def CRC32r64r8 : SS42I_crc32r<0xF0, "crc32{b}", GR64, GR8,
+ null_frag>, REX_W;
+ }
+}
+
+//===----------------------------------------------------------------------===//
+// SHA-NI Instructions
+//===----------------------------------------------------------------------===//
+
+multiclass SHAI_binop<bits<8> Opc, string OpcodeStr, Intrinsic IntId,
+ bit UsesXMM0 = 0> {
+ def rr : I<Opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [!if(UsesXMM0,
+ (set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0)),
+ (set VR128:$dst, (IntId VR128:$src1, VR128:$src2)))]>, T8;
+
+ def rm : I<Opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [!if(UsesXMM0,
+ (set VR128:$dst, (IntId VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)), XMM0)),
+ (set VR128:$dst, (IntId VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)))))]>, T8;
+}
+
+let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
+ def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ [(set VR128:$dst,
+ (int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
+ (i8 imm:$src3)))]>, TA;
+ def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ [(set VR128:$dst,
+ (int_x86_sha1rnds4 VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2)),
+ (i8 imm:$src3)))]>, TA;
+
+ defm SHA1NEXTE : SHAI_binop<0xC8, "sha1nexte", int_x86_sha1nexte>;
+ defm SHA1MSG1 : SHAI_binop<0xC9, "sha1msg1", int_x86_sha1msg1>;
+ defm SHA1MSG2 : SHAI_binop<0xCA, "sha1msg2", int_x86_sha1msg2>;
+
+ let Uses=[XMM0] in
+ defm SHA256RNDS2 : SHAI_binop<0xCB, "sha256rnds2", int_x86_sha256rnds2, 1>;
+
+ defm SHA256MSG1 : SHAI_binop<0xCC, "sha256msg1", int_x86_sha256msg1>;
+ defm SHA256MSG2 : SHAI_binop<0xCD, "sha256msg2", int_x86_sha256msg2>;
}
+// Aliases with explicit %xmm0
+def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
+ (SHA256RNDS2rr VR128:$dst, VR128:$src2)>;
+def : InstAlias<"sha256rnds2\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
+ (SHA256RNDS2rm VR128:$dst, i128mem:$src2)>;
+
//===----------------------------------------------------------------------===//
// AES-NI Instructions
//===----------------------------------------------------------------------===//
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
- OpSize;
+ Sched<[WriteAESDecEnc]>;
def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
- (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
+ (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>,
+ Sched<[WriteAESDecEncLd, ReadAfterLd]>;
}
// Perform One Round of an AES Encryption/Decryption Flow
(ins VR128:$src1),
"vaesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst,
- (int_x86_aesni_aesimc VR128:$src1))]>,
- OpSize, VEX;
+ (int_x86_aesni_aesimc VR128:$src1))]>, Sched<[WriteAESIMC]>,
+ VEX;
def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1),
"vaesimc\t{$src1, $dst|$dst, $src1}",
- [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
- OpSize, VEX;
+ [(set VR128:$dst, (int_x86_aesni_aesimc (loadv2i64 addr:$src1)))]>,
+ Sched<[WriteAESIMCLd]>, VEX;
}
def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1),
"aesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst,
- (int_x86_aesni_aesimc VR128:$src1))]>,
- OpSize;
+ (int_x86_aesni_aesimc VR128:$src1))]>, Sched<[WriteAESIMC]>;
def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1),
"aesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
- OpSize;
+ Sched<[WriteAESIMCLd]>;
// AES Round Key Generation Assist
let Predicates = [HasAVX, HasAES] in {
"vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
- OpSize, VEX;
+ Sched<[WriteAESKeyGen]>, VEX;
def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1, i8imm:$src2),
"vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
- (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
- OpSize, VEX;
+ (int_x86_aesni_aeskeygenassist (loadv2i64 addr:$src1), imm:$src2))]>,
+ Sched<[WriteAESKeyGenLd]>, VEX;
}
def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, i8imm:$src2),
"aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
- OpSize;
+ Sched<[WriteAESKeyGen]>;
def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1, i8imm:$src2),
"aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
- OpSize;
+ Sched<[WriteAESKeyGenLd]>;
//===----------------------------------------------------------------------===//
// PCLMUL Instructions
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128:$dst,
- (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
+ (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>,
+ Sched<[WriteCLMul]>;
def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
- (memopv2i64 addr:$src2), imm:$src3))]>;
+ (loadv2i64 addr:$src2), imm:$src3))]>,
+ Sched<[WriteCLMulLd, ReadAfterLd]>;
// Carry-less Multiplication instructions
let Constraints = "$src1 = $dst" in {
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst,
- (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
+ (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
+ IIC_SSE_PCLMULQDQ_RR>, Sched<[WriteCLMul]>;
def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
- (memopv2i64 addr:$src2), imm:$src3))]>;
+ (memopv2i64 addr:$src2), imm:$src3))],
+ IIC_SSE_PCLMULQDQ_RM>,
+ Sched<[WriteCLMulLd, ReadAfterLd]>;
} // Constraints = "$src1 = $dst"
multiclass pclmul_alias<string asm, int immop> {
def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
- (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
+ (PCLMULQDQrr VR128:$dst, VR128:$src, immop), 0>;
def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
- (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
+ (PCLMULQDQrm VR128:$dst, i128mem:$src, immop), 0>;
def : InstAlias<!strconcat("vpclmul", asm,
"dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
- (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
+ (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop),
+ 0>;
def : InstAlias<!strconcat("vpclmul", asm,
"dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
- (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
+ (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop),
+ 0>;
}
defm : pclmul_alias<"hqhq", 0x11>;
defm : pclmul_alias<"hqlq", 0x01>;
let Predicates = [HasSSE4A] in {
let Constraints = "$src = $dst" in {
-def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
+def EXTRQI : Ii8<0x78, MRMXr, (outs VR128:$dst),
(ins VR128:$src, i8imm:$len, i8imm:$idx),
"extrq\t{$idx, $len, $src|$src, $len, $idx}",
[(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
- imm:$idx))]>, TB, OpSize;
+ imm:$idx))]>, PD;
def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src, VR128:$mask),
"extrq\t{$mask, $src|$src, $mask}",
[(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
- VR128:$mask))]>, TB, OpSize;
+ VR128:$mask))]>, PD;
def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
// destination operand
//
class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
- X86MemOperand x86memop, Intrinsic Int> :
+ X86MemOperand x86memop, Intrinsic Int, SchedWrite Sched> :
AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (Int addr:$src))]>, VEX;
+ [(set RC:$dst, (Int addr:$src))]>, Sched<[Sched]>, VEX;
+
+class avx_broadcast_no_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
+ X86MemOperand x86memop, ValueType VT,
+ PatFrag ld_frag, SchedWrite Sched> :
+ AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (VT (X86VBroadcast (ld_frag addr:$src))))]>,
+ Sched<[Sched]>, VEX {
+ let mayLoad = 1;
+}
// AVX2 adds register forms
class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
- Intrinsic Int> :
+ Intrinsic Int, SchedWrite Sched> :
AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (Int VR128:$src))]>, VEX;
+ [(set RC:$dst, (Int VR128:$src))]>, Sched<[Sched]>, VEX;
let ExeDomain = SSEPackedSingle in {
- def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
- int_x86_avx_vbroadcast_ss>;
- def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
- int_x86_avx_vbroadcast_ss_256>;
+ def VBROADCASTSSrm : avx_broadcast_no_int<0x18, "vbroadcastss", VR128,
+ f32mem, v4f32, loadf32, WriteLoad>;
+ def VBROADCASTSSYrm : avx_broadcast_no_int<0x18, "vbroadcastss", VR256,
+ f32mem, v8f32, loadf32,
+ WriteFShuffleLd>, VEX_L;
}
let ExeDomain = SSEPackedDouble in
-def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
- int_x86_avx_vbroadcast_sd_256>;
+def VBROADCASTSDYrm : avx_broadcast_no_int<0x19, "vbroadcastsd", VR256, f64mem,
+ v4f64, loadf64, WriteFShuffleLd>, VEX_L;
def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
- int_x86_avx_vbroadcastf128_pd_256>;
+ int_x86_avx_vbroadcastf128_pd_256,
+ WriteFShuffleLd>, VEX_L;
let ExeDomain = SSEPackedSingle in {
def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
- int_x86_avx2_vbroadcast_ss_ps>;
+ int_x86_avx2_vbroadcast_ss_ps,
+ WriteFShuffle>;
def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
- int_x86_avx2_vbroadcast_ss_ps_256>;
+ int_x86_avx2_vbroadcast_ss_ps_256,
+ WriteFShuffle256>, VEX_L;
}
let ExeDomain = SSEPackedDouble in
def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
- int_x86_avx2_vbroadcast_sd_pd_256>;
+ int_x86_avx2_vbroadcast_sd_pd_256,
+ WriteFShuffle256>, VEX_L;
let Predicates = [HasAVX2] in
def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
- int_x86_avx2_vbroadcasti128>;
+ int_x86_avx2_vbroadcasti128, WriteLoad>,
+ VEX_L;
let Predicates = [HasAVX] in
def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR128:$src2, i8imm:$src3),
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V;
+ []>, Sched<[WriteFShuffle]>, VEX_4V, VEX_L;
let mayLoad = 1 in
def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, f128mem:$src2, i8imm:$src3),
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V;
+ []>, Sched<[WriteFShuffleLd, ReadAfterLd]>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX] in {
-def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
+def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
+def : Pat<(vinsert128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
}
let Predicates = [HasAVX1Only] in {
-def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
+def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
(iPTR imm)),
(VINSERTF128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
+def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
- (bc_v4i32 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
+ (bc_v4i32 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
- (bc_v16i8 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
+ (bc_v16i8 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
- (bc_v8i16 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
+ (bc_v8i16 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTF128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
}
//===----------------------------------------------------------------------===//
def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
(ins VR256:$src1, i8imm:$src2),
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- []>, VEX;
+ []>, Sched<[WriteFShuffle]>, VEX, VEX_L;
let mayStore = 1 in
def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
(ins f128mem:$dst, VR256:$src1, i8imm:$src2),
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- []>, VEX;
+ []>, Sched<[WriteStore]>, VEX, VEX_L;
}
// AVX1 patterns
let Predicates = [HasAVX] in {
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v4f32 (VEXTRACTF128rr
(v8f32 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v2f64 (VEXTRACTF128rr
(v4f64 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
-def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
- (iPTR imm))), addr:$dst),
+def : Pat<(store (v4f32 (vextract128_extract:$ext (v8f32 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
- (iPTR imm))), addr:$dst),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(store (v2f64 (vextract128_extract:$ext (v4f64 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
}
let Predicates = [HasAVX1Only] in {
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v2i64 (VEXTRACTF128rr
(v4i64 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v4i32 (VEXTRACTF128rr
(v8i32 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v8i16 (VEXTRACTF128rr
(v16i16 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v16i8 (VEXTRACTF128rr
(v32i8 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
-def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
+def : Pat<(alignedstore (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
(iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
(iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
(iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(alignedstore (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
(iPTR imm))), addr:$dst),
(VEXTRACTF128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
}
//===----------------------------------------------------------------------===//
(ins VR256:$src1, f256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
- VEX_4V;
+ VEX_4V, VEX_L;
def mr : AVX8I<opc_mr, MRMDestMem, (outs),
(ins f128mem:$dst, VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
(ins f256mem:$dst, VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
+ [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
}
let ExeDomain = SSEPackedSingle in
def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
+ [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V,
+ Sched<[WriteFShuffle]>;
def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop_i:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (IntVar RC:$src1,
- (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
+ (bitconvert (i_frag addr:$src2))))]>, VEX_4V,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
+ [(set RC:$dst, (vt (X86VPermilpi RC:$src1, (i8 imm:$src2))))]>, VEX,
+ Sched<[WriteFShuffle]>;
def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
(ins x86memop_f:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
- (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
+ (vt (X86VPermilpi (memop addr:$src1), (i8 imm:$src2))))]>, VEX,
+ Sched<[WriteFShuffleLd]>;
}
let ExeDomain = SSEPackedSingle in {
defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
- memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
+ loadv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
- memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
+ loadv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
}
let ExeDomain = SSEPackedDouble in {
defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
- memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
+ loadv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
- memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
+ loadv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
}
let Predicates = [HasAVX] in {
-def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v8f32 (X86VPermilpv VR256:$src1, (v8i32 VR256:$src2))),
+ (VPERMILPSYrr VR256:$src1, VR256:$src2)>;
+def : Pat<(v8f32 (X86VPermilpv VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
+ (VPERMILPSYrm VR256:$src1, addr:$src2)>;
+def : Pat<(v4f64 (X86VPermilpv VR256:$src1, (v4i64 VR256:$src2))),
+ (VPERMILPDYrr VR256:$src1, VR256:$src2)>;
+def : Pat<(v4f64 (X86VPermilpv VR256:$src1, (loadv4i64 addr:$src2))),
+ (VPERMILPDYrm VR256:$src1, addr:$src2)>;
+
+def : Pat<(v8i32 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPSYri VR256:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPDYri VR256:$src1, imm:$imm)>;
-def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
+def : Pat<(v8i32 (X86VPermilpi (bc_v8i32 (loadv4i64 addr:$src1)),
(i8 imm:$imm))),
(VPERMILPSYmi addr:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi (loadv4i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDYmi addr:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86VPermilpv VR128:$src1, (v4i32 VR128:$src2))),
+ (VPERMILPSrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v4f32 (X86VPermilpv VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)))),
+ (VPERMILPSrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2f64 (X86VPermilpv VR128:$src1, (v2i64 VR128:$src2))),
+ (VPERMILPDrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2f64 (X86VPermilpv VR128:$src1, (loadv2i64 addr:$src2))),
+ (VPERMILPDrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v2i64 (X86VPermilpi VR128:$src1, (i8 imm:$imm))),
(VPERMILPDri VR128:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v2i64 (X86VPermilpi (loadv2i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDmi addr:$src1, imm:$imm)>;
}
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
- (i8 imm:$src3))))]>, VEX_4V;
+ (i8 imm:$src3))))]>, VEX_4V, VEX_L,
+ Sched<[WriteFShuffle]>;
def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, f256mem:$src2, i8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
- (i8 imm:$src3)))]>, VEX_4V;
+ [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv8f32 addr:$src2),
+ (i8 imm:$src3)))]>, VEX_4V, VEX_L,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
- (memopv4f64 addr:$src2), (i8 imm:$imm))),
+ (loadv4f64 addr:$src2), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
}
(VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
- (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v8i32 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
- (memopv4i64 addr:$src2), (i8 imm:$imm))),
+ (loadv4i64 addr:$src2), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
- (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v32i8 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
- (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
}
YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
// Zero All YMM registers
def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
- [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
+ [(int_x86_avx_vzeroall)]>, PS, VEX, VEX_L, Requires<[HasAVX]>;
// Zero Upper bits of YMM registers
def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
- [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
+ [(int_x86_avx_vzeroupper)]>, PS, VEX, Requires<[HasAVX]>;
}
//===----------------------------------------------------------------------===//
def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
"vcvtph2ps\t{$src, $dst|$dst, $src}",
[(set RC:$dst, (Int VR128:$src))]>,
- T8, OpSize, VEX;
+ T8PD, VEX, Sched<[WriteCvtF2F]>;
let neverHasSideEffects = 1, mayLoad = 1 in
def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
+ "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8PD, VEX,
+ Sched<[WriteCvtF2FLd]>;
}
multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
(ins RC:$src1, i32i8imm:$src2),
"vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
- TA, OpSize, VEX;
- let neverHasSideEffects = 1, mayStore = 1 in
+ TAPD, VEX, Sched<[WriteCvtF2F]>;
+ let neverHasSideEffects = 1, mayStore = 1,
+ SchedRW = [WriteCvtF2FLd, WriteRMW] in
def mr : Ii8<0x1D, MRMDestMem, (outs),
(ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
"vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
- TA, OpSize, VEX;
+ TAPD, VEX;
}
-let Predicates = [HasAVX, HasF16C] in {
+let Predicates = [HasF16C] in {
defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
- defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
+ defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
- defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
+ defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
+
+ // Pattern match vcvtph2ps of a scalar i64 load.
+ def : Pat<(int_x86_vcvtph2ps_128 (vzmovl_v2i64 addr:$src)),
+ (VCVTPH2PSrm addr:$src)>;
+ def : Pat<(int_x86_vcvtph2ps_128 (vzload_v2i64 addr:$src)),
+ (VCVTPH2PSrm addr:$src)>;
+}
+
+// Patterns for matching conversions from float to half-float and vice versa.
+let Predicates = [HasF16C] in {
+ def : Pat<(fp_to_f16 FR32:$src),
+ (i16 (EXTRACT_SUBREG (VMOVPDI2DIrr (VCVTPS2PHrr
+ (COPY_TO_REGCLASS FR32:$src, VR128), 0)), sub_16bit))>;
+
+ def : Pat<(f16_to_fp GR16:$src),
+ (f32 (COPY_TO_REGCLASS (VCVTPH2PSrr
+ (COPY_TO_REGCLASS (MOVSX32rr16 GR16:$src), VR128)), FR32)) >;
+
+ def : Pat<(f16_to_fp (i16 (fp_to_f16 FR32:$src))),
+ (f32 (COPY_TO_REGCLASS (VCVTPH2PSrr
+ (VCVTPS2PHrr (COPY_TO_REGCLASS FR32:$src, VR128), 0)), FR32)) >;
}
//===----------------------------------------------------------------------===//
X86MemOperand x86memop> {
let isCommutable = 1 in
def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, u32u8imm:$src3),
+ (ins RC:$src1, RC:$src2, i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
- VEX_4V;
+ Sched<[WriteBlend]>, VEX_4V;
def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst,
(IntId RC:$src1,
(bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
- VEX_4V;
+ Sched<[WriteBlendLd, ReadAfterLd]>, VEX_4V;
}
let isCommutable = 0 in {
defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
- VR128, memopv2i64, i128mem>;
+ VR128, loadv2i64, i128mem>;
defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
- VR256, memopv4i64, i256mem>;
+ VR256, loadv4i64, i256mem>, VEX_L;
}
+def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
+ imm:$mask)),
+ (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
+def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
+ imm:$mask)),
+ (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
+
//===----------------------------------------------------------------------===//
// VPBROADCAST - Load from memory and broadcast to all elements of the
// destination operand
Intrinsic Int128, Intrinsic Int256> {
def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
+ [(set VR128:$dst, (Int128 VR128:$src))]>,
+ Sched<[WriteShuffle]>, VEX;
def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
+ (Int128 (scalar_to_vector (ld_frag addr:$src))))]>,
+ Sched<[WriteLoad]>, VEX;
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
+ [(set VR256:$dst, (Int256 VR128:$src))]>,
+ Sched<[WriteShuffle256]>, VEX, VEX_L;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
- (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
+ (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
+ Sched<[WriteLoad]>, VEX, VEX_L;
}
defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
(VBROADCASTSDYrr VR128:$src)>;
+ // Provide aliases for broadcast from the same regitser class that
+ // automatically does the extract.
+ def : Pat<(v32i8 (X86VBroadcast (v32i8 VR256:$src))),
+ (VPBROADCASTBYrr (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v16i16 (X86VBroadcast (v16i16 VR256:$src))),
+ (VPBROADCASTWYrr (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v8i32 (X86VBroadcast (v8i32 VR256:$src))),
+ (VPBROADCASTDYrr (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v4i64 (X86VBroadcast (v4i64 VR256:$src))),
+ (VPBROADCASTQYrr (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v8f32 (X86VBroadcast (v8f32 VR256:$src))),
+ (VBROADCASTSSYrr (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v4f64 (X86VBroadcast (v4f64 VR256:$src))),
+ (VBROADCASTSDYrr (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src),
+ sub_xmm)))>;
+
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
let AddedComplexity = 20 in {
(VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
(VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
+
+ def : Pat<(v16i8 (X86VBroadcast GR8:$src)),
+ (VPBROADCASTBrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+ VR128))>;
+ def : Pat<(v32i8 (X86VBroadcast GR8:$src)),
+ (VPBROADCASTBYrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+ VR128))>;
+
+ def : Pat<(v8i16 (X86VBroadcast GR16:$src)),
+ (VPBROADCASTWrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+ VR128))>;
+ def : Pat<(v16i16 (X86VBroadcast GR16:$src)),
+ (VPBROADCASTWYrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+ VR128))>;
+
+ // The patterns for VPBROADCASTD are not needed because they would match
+ // the exact same thing as VBROADCASTSS patterns.
+
+ def : Pat<(v2i64 (X86VBroadcast GR64:$src)),
+ (VPBROADCASTQrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
+ // The v4i64 pattern is not needed because VBROADCASTSDYrr already match.
}
}
}
let Predicates = [HasAVX] in {
-def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
- (VBROADCASTSSYrm addr:$src)>;
-def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
- (VBROADCASTSDYrm addr:$src)>;
-def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
- (VBROADCASTSSrm addr:$src)>;
-
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
let AddedComplexity = 20 in {
(VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
(VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
}
+
+ def : Pat<(v2f64 (X86VBroadcast f64:$src)),
+ (VMOVDDUPrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
}
//===----------------------------------------------------------------------===//
//
multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
- ValueType OpVT> {
+ ValueType OpVT, X86FoldableSchedWrite Sched> {
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
+ (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
+ Sched<[Sched]>, VEX_4V, VEX_L;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr,
[(set VR256:$dst,
(OpVT (X86VPermv VR256:$src1,
(bitconvert (mem_frag addr:$src2)))))]>,
- VEX_4V;
+ Sched<[Sched.Folded, ReadAfterLd]>, VEX_4V, VEX_L;
}
-defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
+defm VPERMD : avx2_perm<0x36, "vpermd", loadv4i64, v8i32, WriteShuffle256>;
let ExeDomain = SSEPackedSingle in
-defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
+defm VPERMPS : avx2_perm<0x16, "vpermps", loadv8f32, v8f32, WriteFShuffle256>;
multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
- ValueType OpVT> {
+ ValueType OpVT, X86FoldableSchedWrite Sched> {
def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
+ (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
+ Sched<[Sched]>, VEX, VEX_L;
def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
(ins i256mem:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (X86VPermi (mem_frag addr:$src1),
- (i8 imm:$src2))))]>, VEX;
+ (i8 imm:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>, VEX, VEX_L;
}
-defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
+defm VPERMQ : avx2_perm_imm<0x00, "vpermq", loadv4i64, v4i64,
+ WriteShuffle256>, VEX_W;
let ExeDomain = SSEPackedDouble in
-defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
+defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", loadv4f64, v4f64,
+ WriteFShuffle256>, VEX_W;
//===----------------------------------------------------------------------===//
// VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
- (i8 imm:$src3))))]>, VEX_4V;
+ (i8 imm:$src3))))]>, Sched<[WriteShuffle256]>,
+ VEX_4V, VEX_L;
def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, f256mem:$src2, i8imm:$src3),
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
- (i8 imm:$src3)))]>, VEX_4V;
+ [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv4i64 addr:$src2),
+ (i8 imm:$src3)))]>,
+ Sched<[WriteShuffle256Ld, ReadAfterLd]>, VEX_4V, VEX_L;
let Predicates = [HasAVX2] in {
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
-def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
+def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (loadv4i64 addr:$src2)),
(i8 imm:$imm))),
(VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
- (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
+ (bc_v16i16 (loadv4i64 addr:$src2)), (i8 imm:$imm))),
(VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
-def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
+def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)),
(i8 imm:$imm))),
(VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
}
def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR128:$src2, i8imm:$src3),
"vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V;
+ []>, Sched<[WriteShuffle256]>, VEX_4V, VEX_L;
let mayLoad = 1 in
def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i128mem:$src2, i8imm:$src3),
"vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V;
+ []>, Sched<[WriteShuffle256Ld, ReadAfterLd]>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX2] in {
-def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
+def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
(iPTR imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
(iPTR imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
(iPTR imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
(iPTR imm)),
(VINSERTI128rr VR256:$src1, VR128:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
+def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
(iPTR imm)),
(VINSERTI128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
- (bc_v4i32 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v8i32 VR256:$src1),
+ (bc_v4i32 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTI128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
- (bc_v16i8 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v32i8 VR256:$src1),
+ (bc_v16i8 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTI128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
-def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
- (bc_v8i16 (memopv2i64 addr:$src2)),
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
+def : Pat<(vinsert128_insert:$ins (v16i16 VR256:$src1),
+ (bc_v8i16 (loadv2i64 addr:$src2)),
(iPTR imm)),
(VINSERTI128rm VR256:$src1, addr:$src2,
- (INSERT_get_vinsertf128_imm VR256:$ins))>;
+ (INSERT_get_vinsert128_imm VR256:$ins))>;
}
//===----------------------------------------------------------------------===//
"vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
- VEX;
+ Sched<[WriteShuffle256]>, VEX, VEX_L;
let neverHasSideEffects = 1, mayStore = 1 in
def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
(ins i128mem:$dst, VR256:$src1, i8imm:$src2),
- "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
+ "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
+ Sched<[WriteStore]>, VEX, VEX_L;
let Predicates = [HasAVX2] in {
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v2i64 (VEXTRACTI128rr
(v4i64 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v4i32 (VEXTRACTI128rr
(v8i32 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v8i16 (VEXTRACTI128rr
(v16i16 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
-def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
+def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
(v16i8 (VEXTRACTI128rr
(v32i8 VR256:$src1),
- (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext)))>;
-def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
- (iPTR imm))), addr:$dst),
+def : Pat<(store (v2i64 (vextract128_extract:$ext (v4i64 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTI128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
- (iPTR imm))), addr:$dst),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(store (v4i32 (vextract128_extract:$ext (v8i32 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTI128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
- (iPTR imm))), addr:$dst),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(store (v8i16 (vextract128_extract:$ext (v16i16 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTI128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
-def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
- (iPTR imm))), addr:$dst),
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
+def : Pat<(store (v16i8 (vextract128_extract:$ext (v32i8 VR256:$src1),
+ (iPTR imm))), addr:$dst),
(VEXTRACTI128mr addr:$dst, VR256:$src1,
- (EXTRACT_get_vextractf128_imm VR128:$ext))>;
+ (EXTRACT_get_vextract128_imm VR128:$ext))>;
}
//===----------------------------------------------------------------------===//
def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
+ [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
+ VEX_4V, VEX_L;
def mr : AVX28I<0x8e, MRMDestMem, (outs),
(ins i128mem:$dst, VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
(ins i256mem:$dst, VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
+ [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
}
defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
- VEX_4V;
+ VEX_4V, Sched<[WriteVarVecShift]>;
def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1,
- (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
- VEX_4V;
+ (vt128 (bitconvert (loadv2i64 addr:$src2))))))]>,
+ VEX_4V, Sched<[WriteVarVecShiftLd, ReadAfterLd]>;
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
- VEX_4V;
+ VEX_4V, VEX_L, Sched<[WriteVarVecShift]>;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1,
- (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
- VEX_4V;
+ (vt256 (bitconvert (loadv4i64 addr:$src2))))))]>,
+ VEX_4V, VEX_L, Sched<[WriteVarVecShiftLd, ReadAfterLd]>;
}
defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
[]>, VEX_4VOp3, VEX_L;
}
-let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
- defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
- defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
- defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
- defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
+let mayLoad = 1, Constraints
+ = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
+ in {
defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;
+
+ let ExeDomain = SSEPackedDouble in {
+ defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
+ defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
+ }
+
+ let ExeDomain = SSEPackedSingle in {
+ defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
+ defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
+ }
}