def X86pcmpgtd : SDNode<"X86ISD::PCMPGTD", SDTIntBinOp>;
def X86pcmpgtq : SDNode<"X86ISD::PCMPGTQ", SDTIntBinOp>;
-def SDTX86CmpPTest : SDTypeProfile<0, 2, [SDTCisVT<0, v4f32>,
- SDTCisVT<1, v4f32>]>;
+def SDTX86CmpPTest : SDTypeProfile<1, 2, [SDTCisVT<0, i32>,
+ SDTCisVT<1, v4f32>,
+ SDTCisVT<2, v4f32>]>;
def X86ptest : SDNode<"X86ISD::PTEST", SDTX86CmpPTest>;
//===----------------------------------------------------------------------===//
return cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;
-def alignedloadfsf32 : PatFrag<(ops node:$ptr),
+def alignedloadfsf32 : PatFrag<(ops node:$ptr),
(f32 (alignedload node:$ptr))>;
-def alignedloadfsf64 : PatFrag<(ops node:$ptr),
+def alignedloadfsf64 : PatFrag<(ops node:$ptr),
(f64 (alignedload node:$ptr))>;
-def alignedloadv4f32 : PatFrag<(ops node:$ptr),
+def alignedloadv4f32 : PatFrag<(ops node:$ptr),
(v4f32 (alignedload node:$ptr))>;
-def alignedloadv2f64 : PatFrag<(ops node:$ptr),
+def alignedloadv2f64 : PatFrag<(ops node:$ptr),
(v2f64 (alignedload node:$ptr))>;
-def alignedloadv4i32 : PatFrag<(ops node:$ptr),
+def alignedloadv4i32 : PatFrag<(ops node:$ptr),
(v4i32 (alignedload node:$ptr))>;
-def alignedloadv2i64 : PatFrag<(ops node:$ptr),
+def alignedloadv2i64 : PatFrag<(ops node:$ptr),
(v2i64 (alignedload node:$ptr))>;
// Like 'load', but uses special alignment checks suitable for use in
EFLAGS)))]>;
}
+//===----------------------------------------------------------------------===//
+// SSE 1 & 2 Instructions Classes
+//===----------------------------------------------------------------------===//
+
+/// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
+multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ RegisterClass RC, X86MemOperand x86memop> {
+ let isCommutable = 1 in {
+ def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
+ OpcodeStr, [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
+ }
+ def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
+ OpcodeStr, [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
+}
+
+/// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
+multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
+ string asm, string SSEVer, string FPSizeStr,
+ Operand memopr, ComplexPattern mem_cpat> {
+ def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
+ asm, [(set RC:$dst, (
+ !nameconcat<Intrinsic>("int_x86_sse",
+ !strconcat(SSEVer, !strconcat("_",
+ !strconcat(OpcodeStr, FPSizeStr))))
+ RC:$src1, RC:$src2))]>;
+ def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
+ asm, [(set RC:$dst, (
+ !nameconcat<Intrinsic>("int_x86_sse",
+ !strconcat(SSEVer, !strconcat("_",
+ !strconcat(OpcodeStr, FPSizeStr))))
+ RC:$src1, mem_cpat:$src2))]>;
+}
+
+/// sse12_fp_packed - SSE 1 & 2 packed instructions class
+multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ RegisterClass RC, ValueType vt,
+ X86MemOperand x86memop, PatFrag mem_frag,
+ Domain d, bit MayLoad = 0> {
+ let isCommutable = 1 in
+ def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
+ OpcodeStr, [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))],d>;
+ let mayLoad = MayLoad in
+ def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
+ OpcodeStr, [(set RC:$dst, (OpNode RC:$src1,
+ (mem_frag addr:$src2)))],d>;
+}
+
+/// 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> {
+ let isCommutable = 1 in
+ def rr : PI<opc, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2), OpcodeStr, pat_rr, d>;
+ def rm : PI<opc, MRMSrcMem, (outs RC:$dst),
+ (ins RC:$src1, x86memop:$src2), OpcodeStr, pat_rm, 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> {
+ def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
+ asm, [(set RC:$dst, (
+ !nameconcat<Intrinsic>("int_x86_sse",
+ !strconcat(SSEVer, !strconcat("_",
+ !strconcat(OpcodeStr, FPSizeStr))))
+ RC:$src1, RC:$src2))], d>;
+ def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
+ asm, [(set RC:$dst, (
+ !nameconcat<Intrinsic>("int_x86_sse",
+ !strconcat(SSEVer, !strconcat("_",
+ !strconcat(OpcodeStr, FPSizeStr))))
+ RC:$src1, (mem_frag addr:$src2)))], d>;
+}
+
//===----------------------------------------------------------------------===//
// SSE1 Instructions
//===----------------------------------------------------------------------===//
-// Move Instructions
-let neverHasSideEffects = 1 in
-def MOVSSrr : SSI<0x10, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- "movss\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+// Conversion Instructions
+
+// Match intrinsics which expect XMM operand(s).
+def CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
+ "cvtss2si{l}\t{$src, $dst|$dst, $src}", []>;
+def CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
+ "cvtss2si{l}\t{$src, $dst|$dst, $src}", []>;
+
+def CVTDQ2PSrr : PSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtdq2ps\t{$src, $dst|$dst, $src}", []>;
+def CVTDQ2PSrm : PSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvtdq2ps\t{$src, $dst|$dst, $src}", []>;
+
+// Aliases for intrinsics
+def Int_CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
+ "cvttss2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst,
+ (int_x86_sse_cvttss2si VR128:$src))]>;
+def Int_CVTTSS2SIrm : SSI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
+ "cvttss2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst,
+ (int_x86_sse_cvttss2si(load addr:$src)))]>;
+
+let Constraints = "$src1 = $dst" in {
+ def Int_CVTSI2SSrr : SSI<0x2A, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, GR32:$src2),
+ "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
+ GR32:$src2))]>;
+ def Int_CVTSI2SSrm : SSI<0x2A, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i32mem:$src2),
+ "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
+ (loadi32 addr:$src2)))]>;
+}
+
+// Compare Instructions
+let Defs = [EFLAGS] in {
+def COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
+ "comiss\t{$src2, $src1|$src1, $src2}", []>;
+def COMISSrm: PSI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
+ "comiss\t{$src2, $src1|$src1, $src2}", []>;
+} // Defs = [EFLAGS]
+
+//===----------------------------------------------------------------------===//
+// SSE 1 & 2 - Move Instructions
+//===----------------------------------------------------------------------===//
+
+// Move Instructions. Register-to-register movss/movsd is not used for FR32/64
+// register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
+// is used instead. Register-to-register movss/movsd is not modeled as an
+// INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
+// in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
+let Constraints = "$src1 = $dst" in {
+def MOVSSrr : SSI<0x10, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, FR32:$src2),
+ "movss\t{$src2, $dst|$dst, $src2}",
+ [(set (v4f32 VR128:$dst),
+ (movl VR128:$src1, (scalar_to_vector FR32:$src2)))]>;
+def MOVSDrr : SDI<0x10, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, FR64:$src2),
+ "movsd\t{$src2, $dst|$dst, $src2}",
+ [(set (v2f64 VR128:$dst),
+ (movl VR128:$src1, (scalar_to_vector FR64:$src2)))]>;
+}
+
+// Loading from memory automatically zeroing upper bits.
+let canFoldAsLoad = 1, isReMaterializable = 1 in {
def MOVSSrm : SSI<0x10, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
"movss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (loadf32 addr:$src))]>;
+let AddedComplexity = 20 in
+def MOVSDrm : SDI<0x10, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
+ "movsd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (loadf64 addr:$src))]>;
+}
+
+let AddedComplexity = 15 in {
+// Extract the low 32-bit value from one vector and insert it into another.
+def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
+ (MOVSSrr (v4f32 VR128:$src1),
+ (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
+// Extract the low 64-bit value from one vector and insert it into another.
+def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
+ (MOVSDrr (v2f64 VR128:$src1),
+ (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
+}
+
+// Implicitly promote a 32-bit scalar to a vector.
+def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
+// Implicitly promote a 64-bit scalar to a vector.
+def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
+ (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
+
+let AddedComplexity = 20 in {
+// MOVSSrm zeros the high parts of the register; represent this
+// with SUBREG_TO_REG.
+def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
+// MOVSDrm zeros the high parts of the register; represent this
+// with SUBREG_TO_REG.
+def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+def : Pat<(v2f64 (X86vzload addr:$src)),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
+}
+
+// Store scalar value to memory.
def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
"movss\t{$src, $dst|$dst, $src}",
[(store FR32:$src, addr:$dst)]>;
+def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
+ "movsd\t{$src, $dst|$dst, $src}",
+ [(store FR64:$src, addr:$dst)]>;
+
+// Extract and store.
+def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
+ addr:$dst),
+ (MOVSSmr addr:$dst,
+ (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
+def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
+ addr:$dst),
+ (MOVSDmr addr:$dst,
+ (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
+
+//===----------------------------------------------------------------------===//
+// SSE 1 & 2 - Conversion Instructions
+//===----------------------------------------------------------------------===//
// Conversion instructions
def CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
def CVTTSS2SIrm : SSI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
"cvttss2si\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
+def CVTTSD2SIrr : SDI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (fp_to_sint FR64:$src))]>;
+def CVTTSD2SIrm : SDI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f64mem:$src),
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
+
def CVTSI2SSrr : SSI<0x2A, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
"cvtsi2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (sint_to_fp GR32:$src))]>;
def CVTSI2SSrm : SSI<0x2A, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
"cvtsi2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
+def CVTSI2SDrr : SDI<0x2A, MRMSrcReg, (outs FR64:$dst), (ins GR32:$src),
+ "cvtsi2sd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (sint_to_fp GR32:$src))]>;
+def CVTSI2SDrm : SDI<0x2A, MRMSrcMem, (outs FR64:$dst), (ins i32mem:$src),
+ "cvtsi2sd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
// Match intrinsics which expect XMM operand(s).
-def CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
- "cvtss2si{l}\t{$src, $dst|$dst, $src}", []>;
-def CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
- "cvtss2si{l}\t{$src, $dst|$dst, $src}", []>;
-
def Int_CVTSS2SIrr : SSI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
"cvtss2si\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (int_x86_sse_cvtss2si VR128:$src))]>;
"cvtss2si\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (int_x86_sse_cvtss2si
(load addr:$src)))]>;
+def Int_CVTSD2SIrr : SDI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
+ "cvtsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (int_x86_sse2_cvtsd2si VR128:$src))]>;
+def Int_CVTSD2SIrm : SDI<0x2D, MRMSrcMem, (outs GR32:$dst), (ins f128mem:$src),
+ "cvtsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (int_x86_sse2_cvtsd2si
+ (load addr:$src)))]>;
-// Match intrinisics which expect MM and XMM operand(s).
+// Match intrinsics which expect MM and XMM operand(s).
def Int_CVTPS2PIrr : PSI<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"cvtps2pi\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (int_x86_sse_cvtps2pi VR128:$src))]>;
"cvtps2pi\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (int_x86_sse_cvtps2pi
(load addr:$src)))]>;
+def Int_CVTPD2PIrr : PDI<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
+ "cvtpd2pi\t{$src, $dst|$dst, $src}",
+ [(set VR64:$dst, (int_x86_sse_cvtpd2pi VR128:$src))]>;
+def Int_CVTPD2PIrm : PDI<0x2D, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
+ "cvtpd2pi\t{$src, $dst|$dst, $src}",
+ [(set VR64:$dst, (int_x86_sse_cvtpd2pi
+ (memop addr:$src)))]>;
+
+// Match intrinsics which expect MM and XMM operand(s).
def Int_CVTTPS2PIrr: PSI<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
"cvttps2pi\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (int_x86_sse_cvttps2pi VR128:$src))]>;
"cvttps2pi\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (int_x86_sse_cvttps2pi
(load addr:$src)))]>;
+def Int_CVTTPD2PIrr: PDI<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
+ "cvttpd2pi\t{$src, $dst|$dst, $src}",
+ [(set VR64:$dst, (int_x86_sse_cvttpd2pi VR128:$src))]>;
+def Int_CVTTPD2PIrm: PDI<0x2C, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
+ "cvttpd2pi\t{$src, $dst|$dst, $src}",
+ [(set VR64:$dst, (int_x86_sse_cvttpd2pi
+ (memop addr:$src)))]>;
+
let Constraints = "$src1 = $dst" in {
def Int_CVTPI2PSrr : PSI<0x2A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR64:$src2),
(load addr:$src2)))]>;
}
-// Aliases for intrinsics
-def Int_CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
- "cvttss2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst,
- (int_x86_sse_cvttss2si VR128:$src))]>;
-def Int_CVTTSS2SIrm : SSI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f32mem:$src),
- "cvttss2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst,
- (int_x86_sse_cvttss2si(load addr:$src)))]>;
-
-let Constraints = "$src1 = $dst" in {
- def Int_CVTSI2SSrr : SSI<0x2A, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, GR32:$src2),
- "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
- GR32:$src2))]>;
- def Int_CVTSI2SSrm : SSI<0x2A, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i32mem:$src2),
- "cvtsi2ss\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
- (loadi32 addr:$src2)))]>;
-}
+//===----------------------------------------------------------------------===//
+// SSE 1 & 2 - Compare Instructions
+//===----------------------------------------------------------------------===//
// Comparison instructions
let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
def CMPSSrr : SSIi8<0xC2, MRMSrcReg,
(outs FR32:$dst), (ins FR32:$src1, FR32:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
-let mayLoad = 1 in
+ let mayLoad = 1 in
def CMPSSrm : SSIi8<0xC2, MRMSrcMem,
(outs FR32:$dst), (ins FR32:$src1, f32mem:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
+
+ def CMPSDrr : SDIi8<0xC2, MRMSrcReg,
+ (outs FR64:$dst), (ins FR64:$src1, FR64:$src, SSECC:$cc),
+ "cmp${cc}sd\t{$src, $dst|$dst, $src}", []>;
+ let mayLoad = 1 in
+ def CMPSDrm : SDIi8<0xC2, MRMSrcMem,
+ (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, SSECC:$cc),
+ "cmp${cc}sd\t{$src, $dst|$dst, $src}", []>;
+
+// Accept explicit immediate argument form instead of comparison code.
+let isAsmParserOnly = 1 in {
+ def CMPSSrr_alt : SSIi8<0xC2, MRMSrcReg,
+ (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
+ "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+ let mayLoad = 1 in
+ def CMPSSrm_alt : SSIi8<0xC2, MRMSrcMem,
+ (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
+ "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+
+ def CMPSDrr_alt : SDIi8<0xC2, MRMSrcReg,
+ (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
+ "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+ let mayLoad = 1 in
+ def CMPSDrm_alt : SDIi8<0xC2, MRMSrcMem,
+ (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
+ "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+}
}
let Defs = [EFLAGS] in {
def UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins FR32:$src1, FR32:$src2),
"ucomiss\t{$src2, $src1|$src1, $src2}",
- [(X86cmp FR32:$src1, FR32:$src2), (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86cmp FR32:$src1, FR32:$src2))]>;
def UCOMISSrm: PSI<0x2E, MRMSrcMem, (outs), (ins FR32:$src1, f32mem:$src2),
"ucomiss\t{$src2, $src1|$src1, $src2}",
- [(X86cmp FR32:$src1, (loadf32 addr:$src2)),
- (implicit EFLAGS)]>;
-
-def COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "comiss\t{$src2, $src1|$src1, $src2}", []>;
-def COMISSrm: PSI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
- "comiss\t{$src2, $src1|$src1, $src2}", []>;
-
+ [(set EFLAGS, (X86cmp FR32:$src1, (loadf32 addr:$src2)))]>;
+def UCOMISDrr: PDI<0x2E, MRMSrcReg, (outs), (ins FR64:$src1, FR64:$src2),
+ "ucomisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86cmp FR64:$src1, FR64:$src2))]>;
+def UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs), (ins FR64:$src1, f64mem:$src2),
+ "ucomisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86cmp FR64:$src1, (loadf64 addr:$src2)))]>;
} // Defs = [EFLAGS]
// Aliases to match intrinsics which expect XMM operand(s).
let Constraints = "$src1 = $dst" in {
def Int_CMPSSrr : SSIi8<0xC2, MRMSrcReg,
- (outs VR128:$dst),
+ (outs VR128:$dst),
(ins VR128:$src1, VR128:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_cmp_ss
+ [(set VR128:$dst, (int_x86_sse_cmp_ss
VR128:$src1,
VR128:$src, imm:$cc))]>;
def Int_CMPSSrm : SSIi8<0xC2, MRMSrcMem,
- (outs VR128:$dst),
+ (outs VR128:$dst),
(ins VR128:$src1, f32mem:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
(load addr:$src), imm:$cc))]>;
+
+ def Int_CMPSDrr : SDIi8<0xC2, MRMSrcReg,
+ (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src, SSECC:$cc),
+ "cmp${cc}sd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cmp_sd VR128:$src1,
+ VR128:$src, imm:$cc))]>;
+ def Int_CMPSDrm : SDIi8<0xC2, MRMSrcMem,
+ (outs VR128:$dst),
+ (ins VR128:$src1, f64mem:$src, SSECC:$cc),
+ "cmp${cc}sd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cmp_sd VR128:$src1,
+ (load addr:$src), imm:$cc))]>;
}
let Defs = [EFLAGS] in {
def Int_UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"ucomiss\t{$src2, $src1|$src1, $src2}",
- [(X86ucomi (v4f32 VR128:$src1), VR128:$src2),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86ucomi (v4f32 VR128:$src1),
+ VR128:$src2))]>;
def Int_UCOMISSrm: PSI<0x2E, MRMSrcMem, (outs),(ins VR128:$src1, f128mem:$src2),
"ucomiss\t{$src2, $src1|$src1, $src2}",
- [(X86ucomi (v4f32 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86ucomi (v4f32 VR128:$src1),
+ (load addr:$src2)))]>;
+def Int_UCOMISDrr: PDI<0x2E, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
+ "ucomisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ucomi (v2f64 VR128:$src1),
+ VR128:$src2))]>;
+def Int_UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs),(ins VR128:$src1, f128mem:$src2),
+ "ucomisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86ucomi (v2f64 VR128:$src1),
+ (load addr:$src2)))]>;
def Int_COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"comiss\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v4f32 VR128:$src1), VR128:$src2),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86comi (v4f32 VR128:$src1),
+ VR128:$src2))]>;
def Int_COMISSrm: PSI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"comiss\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v4f32 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86comi (v4f32 VR128:$src1),
+ (load addr:$src2)))]>;
+def Int_COMISDrr: PDI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
+ "comisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86comi (v2f64 VR128:$src1),
+ VR128:$src2))]>;
+def Int_COMISDrm: PDI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
+ "comisd\t{$src2, $src1|$src1, $src2}",
+ [(set EFLAGS, (X86comi (v2f64 VR128:$src1),
+ (load addr:$src2)))]>;
} // Defs = [EFLAGS]
-// Aliases of packed SSE1 instructions for scalar use. These all have names
-// that start with 'Fs'.
+// Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
+// names that start with 'Fs'.
// Alias instructions that map fld0 to pxor for sse.
let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
- canFoldAsLoad = 1 in
+ canFoldAsLoad = 1 in {
// FIXME: Set encoding to pseudo!
def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
[(set FR32:$dst, fp32imm0)]>,
Requires<[HasSSE1]>, TB, OpSize;
+def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
+ [(set FR64:$dst, fpimm0)]>,
+ Requires<[HasSSE2]>, TB, OpSize;
+}
-// Alias instruction to do FR32 reg-to-reg copy using movaps. Upper bits are
-// disregarded.
-let neverHasSideEffects = 1 in
+// Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
+// bits are disregarded.
+let neverHasSideEffects = 1 in {
def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
"movaps\t{$src, $dst|$dst, $src}", []>;
+def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
+ "movapd\t{$src, $dst|$dst, $src}", []>;
+}
-// Alias instruction to load FR32 from f128mem using movaps. Upper bits are
-// disregarded.
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+// Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
+// bits are disregarded.
+let canFoldAsLoad = 1, isReMaterializable = 1 in {
def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
+def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
+ "movapd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
+}
-// Alias bitwise logical operations using SSE logical ops on packed FP values.
-let Constraints = "$src1 = $dst" in {
-let isCommutable = 1 in {
- def FsANDPSrr : PSI<0x54, MRMSrcReg, (outs FR32:$dst),
- (ins FR32:$src1, FR32:$src2),
- "andps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86fand FR32:$src1, FR32:$src2))]>;
- def FsORPSrr : PSI<0x56, MRMSrcReg, (outs FR32:$dst),
- (ins FR32:$src1, FR32:$src2),
- "orps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86for FR32:$src1, FR32:$src2))]>;
- def FsXORPSrr : PSI<0x57, MRMSrcReg, (outs FR32:$dst),
- (ins FR32:$src1, FR32:$src2),
- "xorps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86fxor FR32:$src1, FR32:$src2))]>;
-}
-
-def FsANDPSrm : PSI<0x54, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1, f128mem:$src2),
- "andps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86fand FR32:$src1,
- (memopfsf32 addr:$src2)))]>;
-def FsORPSrm : PSI<0x56, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1, f128mem:$src2),
- "orps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86for FR32:$src1,
- (memopfsf32 addr:$src2)))]>;
-def FsXORPSrm : PSI<0x57, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1, f128mem:$src2),
- "xorps\t{$src2, $dst|$dst, $src2}",
- [(set FR32:$dst, (X86fxor FR32:$src1,
- (memopfsf32 addr:$src2)))]>;
+/// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
+///
+multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
+ SDNode OpNode, bit MayLoad = 0> {
+ let isAsmParserOnly = 1 in {
+ defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode, FR32,
+ f32, f128mem, memopfsf32, SSEPackedSingle, MayLoad>, VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode, FR64,
+ f64, f128mem, memopfsf64, SSEPackedDouble, MayLoad>, OpSize,
+ VEX_4V;
+ }
-let neverHasSideEffects = 1 in {
-def FsANDNPSrr : PSI<0x55, MRMSrcReg,
- (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
- "andnps\t{$src2, $dst|$dst, $src2}", []>;
-let mayLoad = 1 in
-def FsANDNPSrm : PSI<0x55, MRMSrcMem,
- (outs FR32:$dst), (ins FR32:$src1, f128mem:$src2),
- "andnps\t{$src2, $dst|$dst, $src2}", []>;
-}
+ let Constraints = "$src1 = $dst" in {
+ defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $dst|$dst, $src2}"), OpNode, FR32, f32,
+ f128mem, memopfsf32, SSEPackedSingle, MayLoad>, TB;
+
+ defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $dst|$dst, $src2}"), OpNode, FR64, f64,
+ f128mem, memopfsf64, SSEPackedDouble, MayLoad>, TB, OpSize;
+ }
}
-/// basic_sse1_fp_binop_rm - SSE1 binops come in both scalar and vector forms.
+// Alias bitwise logical operations using SSE logical ops on packed FP values.
+defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
+defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
+defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
+
+let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
+ defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef, 1>;
+
+/// basic_sse12_fp_binop_rm - SSE 1 & 2 binops come in both scalar and
+/// vector forms.
///
/// In addition, we also have a special variant of the scalar form here to
/// represent the associated intrinsic operation. This form is unlike the
/// These three forms can each be reg+reg or reg+mem, so there are a total of
/// six "instructions".
///
-let Constraints = "$src1 = $dst" in {
-multiclass basic_sse1_fp_binop_rm<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F32Int,
- bit Commutable = 0> {
- // Scalar operation, reg+reg.
- def SSrr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
- let isCommutable = Commutable;
+multiclass basic_sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
+ SDNode OpNode> {
+
+ let isAsmParserOnly = 1 in {
+ defm V#NAME#SS : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ OpNode, FR32, f32mem>, XS, VEX_4V;
+
+ defm V#NAME#SD : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ OpNode, FR64, f64mem>, XD, VEX_4V;
+
+ defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode,
+ VR128, v4f32, f128mem, memopv4f32, SSEPackedSingle>,
+ VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode,
+ VR128, v2f64, f128mem, memopv2f64, SSEPackedDouble>,
+ OpSize, VEX_4V;
+
+ defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "", "_ss", ssmem, sse_load_f32>, XS, VEX_4V;
+
+ defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "2", "_sd", sdmem, sse_load_f64>, XD, VEX_4V;
}
- // Scalar operation, reg+mem.
- def SSrm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1, f32mem:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set FR32:$dst, (OpNode FR32:$src1, (load addr:$src2)))]>;
+ let Constraints = "$src1 = $dst" in {
+ defm SS : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ OpNode, FR32, f32mem>, XS;
- // Vector operation, reg+reg.
- def PSrr : PSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (v4f32 (OpNode VR128:$src1, VR128:$src2)))]> {
- let isCommutable = Commutable;
- }
+ defm SD : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
+ OpNode, FR64, f64mem>, XD;
- // Vector operation, reg+mem.
- def PSrm : PSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode VR128:$src1, (memopv4f32 addr:$src2)))]>;
+ defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $dst|$dst, $src2}"), OpNode, VR128, v4f32,
+ f128mem, memopv4f32, SSEPackedSingle>, TB;
- // Intrinsic operation, reg+reg.
- def SSrr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]>;
+ defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $dst|$dst, $src2}"), OpNode, VR128, v2f64,
+ f128mem, memopv2f64, SSEPackedDouble>, TB, OpSize;
- // Intrinsic operation, reg+mem.
- def SSrm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F32Int VR128:$src1,
- sse_load_f32:$src2))]>;
-}
+ defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ "", "_ss", ssmem, sse_load_f32>, XS;
+
+ defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
+ "2", "_sd", sdmem, sse_load_f64>, XD;
+ }
}
// Arithmetic instructions
-defm ADD : basic_sse1_fp_binop_rm<0x58, "add", fadd, int_x86_sse_add_ss, 1>;
-defm MUL : basic_sse1_fp_binop_rm<0x59, "mul", fmul, int_x86_sse_mul_ss, 1>;
-defm SUB : basic_sse1_fp_binop_rm<0x5C, "sub", fsub, int_x86_sse_sub_ss>;
-defm DIV : basic_sse1_fp_binop_rm<0x5E, "div", fdiv, int_x86_sse_div_ss>;
+defm ADD : basic_sse12_fp_binop_rm<0x58, "add", fadd>;
+defm MUL : basic_sse12_fp_binop_rm<0x59, "mul", fmul>;
+
+let isCommutable = 0 in {
+ defm SUB : basic_sse12_fp_binop_rm<0x5C, "sub", fsub>;
+ defm DIV : basic_sse12_fp_binop_rm<0x5E, "div", fdiv>;
+}
-/// sse1_fp_binop_rm - Other SSE1 binops
+/// sse12_fp_binop_rm - Other SSE 1 & 2 binops
///
-/// This multiclass is like basic_sse1_fp_binop_rm, with the addition of
+/// This multiclass is like basic_sse12_fp_binop_rm, with the addition of
/// instructions for a full-vector intrinsic form. Operations that map
/// onto C operators don't use this form since they just use the plain
/// vector form instead of having a separate vector intrinsic form.
///
/// This provides a total of eight "instructions".
///
-let Constraints = "$src1 = $dst" in {
-multiclass sse1_fp_binop_rm<bits<8> opc, string OpcodeStr,
- SDNode OpNode,
- Intrinsic F32Int,
- Intrinsic V4F32Int,
- bit Commutable = 0> {
-
- // Scalar operation, reg+reg.
- def SSrr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
- let isCommutable = Commutable;
- }
-
- // Scalar operation, reg+mem.
- def SSrm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1, f32mem:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set FR32:$dst, (OpNode FR32:$src1, (load addr:$src2)))]>;
-
- // Vector operation, reg+reg.
- def PSrr : PSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (v4f32 (OpNode VR128:$src1, VR128:$src2)))]> {
- let isCommutable = Commutable;
- }
-
- // Vector operation, reg+mem.
- def PSrm : PSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode VR128:$src1, (memopv4f32 addr:$src2)))]>;
-
- // Intrinsic operation, reg+reg.
- def SSrr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]> {
- let isCommutable = Commutable;
+multiclass sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
+ SDNode OpNode> {
+
+ let isAsmParserOnly = 1 in {
+ // Scalar operation, reg+reg.
+ defm V#NAME#SS : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ OpNode, FR32, f32mem>, XS, VEX_4V;
+
+ defm V#NAME#SD : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ OpNode, FR64, f64mem>, XD, VEX_4V;
+
+ defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode,
+ VR128, v4f32, f128mem, memopv4f32, SSEPackedSingle>,
+ VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), OpNode,
+ VR128, v2f64, f128mem, memopv2f64, SSEPackedDouble>,
+ OpSize, VEX_4V;
+
+ defm V#NAME#SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "", "_ss", ssmem, sse_load_f32>, XS, VEX_4V;
+
+ defm V#NAME#SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "2", "_sd", sdmem, sse_load_f64>, XD, VEX_4V;
+
+ defm V#NAME#PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "", "_ps", f128mem, memopv4f32, SSEPackedSingle>, VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ "2", "_pd", f128mem, memopv2f64, SSEPackedDouble>, OpSize,
+ VEX_4V;
}
- // Intrinsic operation, reg+mem.
- def SSrm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F32Int VR128:$src1,
- sse_load_f32:$src2))]>;
-
- // Vector intrinsic operation, reg+reg.
- def PSrr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (V4F32Int VR128:$src1, VR128:$src2))]> {
- let isCommutable = Commutable;
+ let Constraints = "$src1 = $dst" in {
+ // Scalar operation, reg+reg.
+ defm SS : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ OpNode, FR32, f32mem>, XS;
+ defm SD : sse12_fp_scalar<opc,
+ !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
+ OpNode, FR64, f64mem>, XD;
+ defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "ps\t{$src2, $dst|$dst, $src2}"), OpNode, VR128, v4f32,
+ f128mem, memopv4f32, SSEPackedSingle>, TB;
+
+ defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr,
+ "pd\t{$src2, $dst|$dst, $src2}"), OpNode, VR128, v2f64,
+ f128mem, memopv2f64, SSEPackedDouble>, TB, OpSize;
+
+ defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ "", "_ss", ssmem, sse_load_f32>, XS;
+
+ defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
+ "2", "_sd", sdmem, sse_load_f64>, XD;
+
+ defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
+ "", "_ps", f128mem, memopv4f32, SSEPackedSingle>, TB;
+
+ defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
+ !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
+ "2", "_pd", f128mem, memopv2f64, SSEPackedDouble>, TB, OpSize;
}
-
- // Vector intrinsic operation, reg+mem.
- def PSrm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (V4F32Int VR128:$src1, (memopv4f32 addr:$src2)))]>;
-}
}
-defm MAX : sse1_fp_binop_rm<0x5F, "max", X86fmax,
- int_x86_sse_max_ss, int_x86_sse_max_ps>;
-defm MIN : sse1_fp_binop_rm<0x5D, "min", X86fmin,
- int_x86_sse_min_ss, int_x86_sse_min_ps>;
+let isCommutable = 0 in {
+ defm MAX : sse12_fp_binop_rm<0x5F, "max", X86fmax>;
+ defm MIN : sse12_fp_binop_rm<0x5D, "min", X86fmin>;
+}
//===----------------------------------------------------------------------===//
// SSE packed FP Instructions
let neverHasSideEffects = 1 in
def MOVAPSrr : PSI<0x28, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movaps\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVAPSrm : PSI<0x28, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (alignedloadv4f32 addr:$src))]>;
let neverHasSideEffects = 1 in
def MOVUPSrr : PSI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movups\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVUPSrm : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movups\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (loadv4f32 addr:$src))]>;
[(store (v4f32 VR128:$src), addr:$dst)]>;
// Intrinsic forms of MOVUPS load and store
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movups\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
- (MOVHPSrm VR128:$src1, addr:$src2)>;
+ (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
let AddedComplexity = 20 in {
def : Pat<(v4f32 (movddup VR128:$src, (undef))),
- (MOVLHPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
def : Pat<(v2i64 (movddup VR128:$src, (undef))),
- (MOVLHPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
}
defm RCP : sse1_fp_unop_rm<0x53, "rcp", X86frcp,
int_x86_sse_rcp_ss, int_x86_sse_rcp_ps>;
-// Logical
-let Constraints = "$src1 = $dst" in {
- let isCommutable = 1 in {
- def ANDPSrr : PSI<0x54, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "andps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (v2i64
- (and VR128:$src1, VR128:$src2)))]>;
- def ORPSrr : PSI<0x56, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "orps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (v2i64
- (or VR128:$src1, VR128:$src2)))]>;
- def XORPSrr : PSI<0x57, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "xorps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (v2i64
- (xor VR128:$src1, VR128:$src2)))]>;
+/// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
+///
+multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
+ SDNode OpNode, int HasPat = 0,
+ list<list<dag>> Pattern = []> {
+ let isAsmParserOnly = 1 in {
+ defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
+ !strconcat(OpcodeStr, "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ f128mem,
+ !if(HasPat, Pattern[0], // rr
+ [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
+ VR128:$src2)))]),
+ !if(HasPat, Pattern[2], // rm
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
+ (memopv2i64 addr:$src2)))])>,
+ VEX_4V;
+
+ defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
+ !strconcat(OpcodeStr, "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ f128mem,
+ !if(HasPat, Pattern[1], // rr
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (bc_v2i64 (v2f64
+ VR128:$src2))))]),
+ !if(HasPat, Pattern[3], // rm
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (memopv2i64 addr:$src2)))])>,
+ OpSize, VEX_4V;
+ }
+ let Constraints = "$src1 = $dst" in {
+ defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
+ !strconcat(OpcodeStr, "ps\t{$src2, $dst|$dst, $src2}"), f128mem,
+ !if(HasPat, Pattern[0], // rr
+ [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
+ VR128:$src2)))]),
+ !if(HasPat, Pattern[2], // rm
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
+ (memopv2i64 addr:$src2)))])>, TB;
+
+ defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
+ !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"), f128mem,
+ !if(HasPat, Pattern[1], // rr
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (bc_v2i64 (v2f64
+ VR128:$src2))))]),
+ !if(HasPat, Pattern[3], // rm
+ [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
+ (memopv2i64 addr:$src2)))])>,
+ TB, OpSize;
}
-
- def ANDPSrm : PSI<0x54, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "andps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (and (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def ORPSrm : PSI<0x56, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "orps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (or (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def XORPSrm : PSI<0x57, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "xorps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (xor (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def ANDNPSrr : PSI<0x55, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "andnps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2i64 (and (xor VR128:$src1,
- (bc_v2i64 (v4i32 immAllOnesV))),
- VR128:$src2)))]>;
- def ANDNPSrm : PSI<0x55, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1,f128mem:$src2),
- "andnps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
- (bc_v2i64 (v4i32 immAllOnesV))),
- (memopv2i64 addr:$src2))))]>;
}
+// Logical
+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", undef /* dummy */, 1, [
+ // single r+r
+ [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
+ (bc_v2i64 (v4i32 immAllOnesV))),
+ VR128:$src2)))],
+ // double r+r
+ [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
+ (bc_v2i64 (v2f64 VR128:$src2))))],
+ // single r+m
+ [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
+ (bc_v2i64 (v4i32 immAllOnesV))),
+ (memopv2i64 addr:$src2))))],
+ // double r+m
+ [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
+ (memopv2i64 addr:$src2)))]]>;
+
let Constraints = "$src1 = $dst" in {
def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
"cmp${cc}ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
(memop addr:$src), imm:$cc))]>;
-}
-def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
- (CMPPSrri VR128:$src1, VR128:$src2, imm:$cc)>;
+ def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
+ "cmp${cc}pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
+ VR128:$src, imm:$cc))]>;
+ def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, SSECC:$cc),
+ "cmp${cc}pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
+ (memop addr:$src), imm:$cc))]>;
+
+ // Accept explicit immediate argument form instead of comparison code.
+let isAsmParserOnly = 1 in {
+ def CMPPSrri_alt : PSIi8<0xC2, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src, i8imm:$src2),
+ "cmpps\t{$src2, $src, $dst|$dst, $src, $src}", []>;
+ def CMPPSrmi_alt : PSIi8<0xC2, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, i8imm:$src2),
+ "cmpps\t{$src2, $src, $dst|$dst, $src, $src}", []>;
+ def CMPPDrri_alt : PDIi8<0xC2, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src, i8imm:$src2),
+ "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+ def CMPPDrmi_alt : PDIi8<0xC2, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, i8imm:$src2),
+ "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}", []>;
+}
+}
+def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
+ (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
- (CMPPSrmi VR128:$src1, addr:$src2, imm:$cc)>;
+ (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
+def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
+ (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
+def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
+ (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
// Shuffle and unpack instructions
let Constraints = "$src1 = $dst" in {
[(set VR128:$dst,
(v4f32 (shufp:$src3
VR128:$src1, (memopv4f32 addr:$src2))))]>;
+ def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ [(set VR128:$dst,
+ (v2f64 (shufp:$src3 VR128:$src1, VR128:$src2)))]>;
+ def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1,
+ f128mem:$src2, i8imm:$src3),
+ "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
+ [(set VR128:$dst,
+ (v2f64 (shufp:$src3
+ VR128:$src1, (memopv2f64 addr:$src2))))]>;
let AddedComplexity = 10 in {
def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
"unpcklps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(unpckl VR128:$src1, (memopv4f32 addr:$src2)))]>;
+ def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ "unpckhpd\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst,
+ (v2f64 (unpckh VR128:$src1, VR128:$src2)))]>;
+ def UNPCKHPDrm : PDI<0x15, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
+ "unpckhpd\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst,
+ (v2f64 (unpckh VR128:$src1,
+ (memopv2f64 addr:$src2))))]>;
+
+ def UNPCKLPDrr : PDI<0x14, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ "unpcklpd\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst,
+ (v2f64 (unpckl VR128:$src1, VR128:$src2)))]>;
+ def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
+ "unpcklpd\t{$src2, $dst|$dst, $src2}",
+ [(set VR128:$dst,
+ (unpckl VR128:$src1, (memopv2f64 addr:$src2)))]>;
} // AddedComplexity
} // Constraints = "$src1 = $dst"
"movntdq\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
-def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
- (MOVNTDQ_64mr VR128:$src, addr:$dst)>;
-
def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
"movnti\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
}
// Load, store, and memory fence
-def SFENCE : PSI<0xAE, MRM7r, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>;
+def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
+ TB, Requires<[HasSSE1]>;
// MXCSR register
def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
// load of an all-zeros value if folding it would be beneficial.
// FIXME: Change encoding to pseudo!
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isCodeGenOnly = 1 in
-def V_SET0 : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
+ isCodeGenOnly = 1 in {
+def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4f32 immAllZerosV))]>;
+def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v2f64 immAllZerosV))]>;
+let ExeDomain = SSEPackedInt in
+def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllZerosV))]>;
-
-let Predicates = [HasSSE1] in {
- def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
- def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
- def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
- def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
- def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
}
-// FR32 to 128-bit vector conversion.
-let isAsCheapAsAMove = 1 in
-def MOVSS2PSrr : SSI<0x10, MRMSrcReg, (outs VR128:$dst), (ins FR32:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v4f32 (scalar_to_vector FR32:$src)))]>;
-def MOVSS2PSrm : SSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f32mem:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v4f32 (scalar_to_vector (loadf32 addr:$src))))]>;
-
-// FIXME: may not be able to eliminate this movss with coalescing the src and
-// dest register classes are different. We really want to write this pattern
-// like this:
-// def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
-// (f32 FR32:$src)>;
-let isAsCheapAsAMove = 1 in
-def MOVPS2SSrr : SSI<0x10, MRMSrcReg, (outs FR32:$dst), (ins VR128:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (vector_extract (v4f32 VR128:$src),
- (iPTR 0)))]>;
-def MOVPS2SSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(store (f32 (vector_extract (v4f32 VR128:$src),
- (iPTR 0))), addr:$dst)]>;
-
-
-// Move to lower bits of a VR128, leaving upper bits alone.
-// Three operand (but two address) aliases.
-let Constraints = "$src1 = $dst" in {
-let neverHasSideEffects = 1 in
- def MOVLSS2PSrr : SSI<0x10, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, FR32:$src2),
- "movss\t{$src2, $dst|$dst, $src2}", []>;
-
- let AddedComplexity = 15 in
- def MOVLPSrr : SSI<0x10, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "movss\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v4f32 (movl VR128:$src1, VR128:$src2)))]>;
-}
-
-// Move to lower bits of a VR128 and zeroing upper bits.
-// Loading from memory automatically zeroing upper bits.
-let AddedComplexity = 20 in
-def MOVZSS2PSrm : SSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f32mem:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (v4f32 (X86vzmovl (v4f32 (scalar_to_vector
- (loadf32 addr:$src))))))]>;
-
-def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
- (MOVZSS2PSrm addr:$src)>;
-
-//===---------------------------------------------------------------------===//
-// SSE2 Instructions
-//===---------------------------------------------------------------------===//
-
-// Move Instructions
-let neverHasSideEffects = 1 in
-def MOVSDrr : SDI<0x10, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
-def MOVSDrm : SDI<0x10, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (loadf64 addr:$src))]>;
-def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store FR64:$src, addr:$dst)]>;
-
-// Conversion instructions
-def CVTTSD2SIrr : SDI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
- "cvttsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (fp_to_sint FR64:$src))]>;
-def CVTTSD2SIrm : SDI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f64mem:$src),
- "cvttsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
-def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
- "cvtsd2ss\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (fround FR64:$src))]>;
-def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
- "cvtsd2ss\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
- Requires<[HasSSE2, OptForSize]>;
-def CVTSI2SDrr : SDI<0x2A, MRMSrcReg, (outs FR64:$dst), (ins GR32:$src),
- "cvtsi2sd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (sint_to_fp GR32:$src))]>;
-def CVTSI2SDrm : SDI<0x2A, MRMSrcMem, (outs FR64:$dst), (ins i32mem:$src),
- "cvtsi2sd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
-
-def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
-def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
-def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
-def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
-def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
-def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
-def CVTDQ2PSrr : PSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- "cvtdq2ps\t{$src, $dst|$dst, $src}", []>;
-def CVTDQ2PSrm : PSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- "cvtdq2ps\t{$src, $dst|$dst, $src}", []>;
-def COMISDrr: PDI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "comisd\t{$src2, $src1|$src1, $src2}", []>;
-def COMISDrm: PDI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
- "comisd\t{$src2, $src1|$src1, $src2}", []>;
-
-// SSE2 instructions with XS prefix
-def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
- "cvtss2sd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (fextend FR32:$src))]>, XS,
- Requires<[HasSSE2]>;
-def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
- "cvtss2sd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
- Requires<[HasSSE2, OptForSize]>;
-
-def : Pat<(extloadf32 addr:$src),
- (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
-
-// Match intrinsics which expect XMM operand(s).
-def Int_CVTSD2SIrr : SDI<0x2D, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
- "cvtsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_sse2_cvtsd2si VR128:$src))]>;
-def Int_CVTSD2SIrm : SDI<0x2D, MRMSrcMem, (outs GR32:$dst), (ins f128mem:$src),
- "cvtsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_sse2_cvtsd2si
- (load addr:$src)))]>;
-
-// Match intrinisics which expect MM and XMM operand(s).
-def Int_CVTPD2PIrr : PDI<0x2D, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
- "cvtpd2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvtpd2pi VR128:$src))]>;
-def Int_CVTPD2PIrm : PDI<0x2D, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
- "cvtpd2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvtpd2pi
- (memop addr:$src)))]>;
-def Int_CVTTPD2PIrr: PDI<0x2C, MRMSrcReg, (outs VR64:$dst), (ins VR128:$src),
- "cvttpd2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvttpd2pi VR128:$src))]>;
-def Int_CVTTPD2PIrm: PDI<0x2C, MRMSrcMem, (outs VR64:$dst), (ins f128mem:$src),
- "cvttpd2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvttpd2pi
- (memop addr:$src)))]>;
-def Int_CVTPI2PDrr : PDI<0x2A, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
- "cvtpi2pd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_cvtpi2pd VR64:$src))]>;
-def Int_CVTPI2PDrm : PDI<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
- "cvtpi2pd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_cvtpi2pd
- (load addr:$src)))]>;
-
-// Aliases for intrinsics
-def Int_CVTTSD2SIrr : SDI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
- "cvttsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst,
- (int_x86_sse2_cvttsd2si VR128:$src))]>;
-def Int_CVTTSD2SIrm : SDI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f128mem:$src),
- "cvttsd2si\t{$src, $dst|$dst, $src}",
- [(set GR32:$dst, (int_x86_sse2_cvttsd2si
- (load addr:$src)))]>;
-
-// Comparison instructions
-let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
- def CMPSDrr : SDIi8<0xC2, MRMSrcReg,
- (outs FR64:$dst), (ins FR64:$src1, FR64:$src, SSECC:$cc),
- "cmp${cc}sd\t{$src, $dst|$dst, $src}", []>;
-let mayLoad = 1 in
- def CMPSDrm : SDIi8<0xC2, MRMSrcMem,
- (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, SSECC:$cc),
- "cmp${cc}sd\t{$src, $dst|$dst, $src}", []>;
-}
-
-let Defs = [EFLAGS] in {
-def UCOMISDrr: PDI<0x2E, MRMSrcReg, (outs), (ins FR64:$src1, FR64:$src2),
- "ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86cmp FR64:$src1, FR64:$src2), (implicit EFLAGS)]>;
-def UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs), (ins FR64:$src1, f64mem:$src2),
- "ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86cmp FR64:$src1, (loadf64 addr:$src2)),
- (implicit EFLAGS)]>;
-} // Defs = [EFLAGS]
-
-// Aliases to match intrinsics which expect XMM operand(s).
-let Constraints = "$src1 = $dst" in {
- def Int_CMPSDrr : SDIi8<0xC2, MRMSrcReg,
- (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src, SSECC:$cc),
- "cmp${cc}sd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cmp_sd VR128:$src1,
- VR128:$src, imm:$cc))]>;
- def Int_CMPSDrm : SDIi8<0xC2, MRMSrcMem,
- (outs VR128:$dst),
- (ins VR128:$src1, f64mem:$src, SSECC:$cc),
- "cmp${cc}sd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cmp_sd VR128:$src1,
- (load addr:$src), imm:$cc))]>;
-}
-
-let Defs = [EFLAGS] in {
-def Int_UCOMISDrr: PDI<0x2E, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86ucomi (v2f64 VR128:$src1), (v2f64 VR128:$src2)),
- (implicit EFLAGS)]>;
-def Int_UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs),(ins VR128:$src1, f128mem:$src2),
- "ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86ucomi (v2f64 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
-
-def Int_COMISDrr: PDI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "comisd\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v2f64 VR128:$src1), (v2f64 VR128:$src2)),
- (implicit EFLAGS)]>;
-def Int_COMISDrm: PDI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
- "comisd\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v2f64 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
-} // Defs = [EFLAGS]
-
-// Aliases of packed SSE2 instructions for scalar use. These all have names
-// that start with 'Fs'.
-
-// Alias instructions that map fld0 to pxor for sse.
-let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
- canFoldAsLoad = 1 in
-def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
- [(set FR64:$dst, fpimm0)]>,
- Requires<[HasSSE2]>, TB, OpSize;
-
-// Alias instruction to do FR64 reg-to-reg copy using movapd. Upper bits are
-// disregarded.
-let neverHasSideEffects = 1 in
-def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
- "movapd\t{$src, $dst|$dst, $src}", []>;
-
-// Alias instruction to load FR64 from f128mem using movapd. Upper bits are
-// disregarded.
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
-def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
- "movapd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
-
-// Alias bitwise logical operations using SSE logical ops on packed FP values.
-let Constraints = "$src1 = $dst" in {
-let isCommutable = 1 in {
- def FsANDPDrr : PDI<0x54, MRMSrcReg, (outs FR64:$dst),
- (ins FR64:$src1, FR64:$src2),
- "andpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86fand FR64:$src1, FR64:$src2))]>;
- def FsORPDrr : PDI<0x56, MRMSrcReg, (outs FR64:$dst),
- (ins FR64:$src1, FR64:$src2),
- "orpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86for FR64:$src1, FR64:$src2))]>;
- def FsXORPDrr : PDI<0x57, MRMSrcReg, (outs FR64:$dst),
- (ins FR64:$src1, FR64:$src2),
- "xorpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86fxor FR64:$src1, FR64:$src2))]>;
-}
-
-def FsANDPDrm : PDI<0x54, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f128mem:$src2),
- "andpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86fand FR64:$src1,
- (memopfsf64 addr:$src2)))]>;
-def FsORPDrm : PDI<0x56, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f128mem:$src2),
- "orpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86for FR64:$src1,
- (memopfsf64 addr:$src2)))]>;
-def FsXORPDrm : PDI<0x57, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f128mem:$src2),
- "xorpd\t{$src2, $dst|$dst, $src2}",
- [(set FR64:$dst, (X86fxor FR64:$src1,
- (memopfsf64 addr:$src2)))]>;
-
-let neverHasSideEffects = 1 in {
-def FsANDNPDrr : PDI<0x55, MRMSrcReg,
- (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
- "andnpd\t{$src2, $dst|$dst, $src2}", []>;
-let mayLoad = 1 in
-def FsANDNPDrm : PDI<0x55, MRMSrcMem,
- (outs FR64:$dst), (ins FR64:$src1, f128mem:$src2),
- "andnpd\t{$src2, $dst|$dst, $src2}", []>;
-}
-}
-
-/// basic_sse2_fp_binop_rm - SSE2 binops come in both scalar and vector forms.
-///
-/// In addition, we also have a special variant of the scalar form here to
-/// represent the associated intrinsic operation. This form is unlike the
-/// plain scalar form, in that it takes an entire vector (instead of a scalar)
-/// and leaves the top elements unmodified (therefore these cannot be commuted).
-///
-/// These three forms can each be reg+reg or reg+mem, so there are a total of
-/// six "instructions".
-///
-let Constraints = "$src1 = $dst" in {
-multiclass basic_sse2_fp_binop_rm<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F64Int,
- bit Commutable = 0> {
- // Scalar operation, reg+reg.
- def SDrr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set FR64:$dst, (OpNode FR64:$src1, FR64:$src2))]> {
- let isCommutable = Commutable;
- }
-
- // Scalar operation, reg+mem.
- def SDrm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f64mem:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set FR64:$dst, (OpNode FR64:$src1, (load addr:$src2)))]>;
-
- // Vector operation, reg+reg.
- def PDrr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (v2f64 (OpNode VR128:$src1, VR128:$src2)))]> {
- let isCommutable = Commutable;
- }
-
- // Vector operation, reg+mem.
- def PDrm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode VR128:$src1, (memopv2f64 addr:$src2)))]>;
-
- // Intrinsic operation, reg+reg.
- def SDrr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2))]>;
-
- // Intrinsic operation, reg+mem.
- def SDrm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, sdmem:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F64Int VR128:$src1,
- sse_load_f64:$src2))]>;
-}
-}
-
-// Arithmetic instructions
-defm ADD : basic_sse2_fp_binop_rm<0x58, "add", fadd, int_x86_sse2_add_sd, 1>;
-defm MUL : basic_sse2_fp_binop_rm<0x59, "mul", fmul, int_x86_sse2_mul_sd, 1>;
-defm SUB : basic_sse2_fp_binop_rm<0x5C, "sub", fsub, int_x86_sse2_sub_sd>;
-defm DIV : basic_sse2_fp_binop_rm<0x5E, "div", fdiv, int_x86_sse2_div_sd>;
-
-/// sse2_fp_binop_rm - Other SSE2 binops
-///
-/// This multiclass is like basic_sse2_fp_binop_rm, with the addition of
-/// instructions for a full-vector intrinsic form. Operations that map
-/// onto C operators don't use this form since they just use the plain
-/// vector form instead of having a separate vector intrinsic form.
-///
-/// This provides a total of eight "instructions".
-///
-let Constraints = "$src1 = $dst" in {
-multiclass sse2_fp_binop_rm<bits<8> opc, string OpcodeStr,
- SDNode OpNode,
- Intrinsic F64Int,
- Intrinsic V2F64Int,
- bit Commutable = 0> {
-
- // Scalar operation, reg+reg.
- def SDrr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set FR64:$dst, (OpNode FR64:$src1, FR64:$src2))]> {
- let isCommutable = Commutable;
- }
-
- // Scalar operation, reg+mem.
- def SDrm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
- (ins FR64:$src1, f64mem:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set FR64:$dst, (OpNode FR64:$src1, (load addr:$src2)))]>;
-
- // Vector operation, reg+reg.
- def PDrr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (v2f64 (OpNode VR128:$src1, VR128:$src2)))]> {
- let isCommutable = Commutable;
- }
+def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
+def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
+def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
- // Vector operation, reg+mem.
- def PDrm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode VR128:$src1, (memopv2f64 addr:$src2)))]>;
+def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
+ (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
- // Intrinsic operation, reg+reg.
- def SDrr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2))]> {
- let isCommutable = Commutable;
- }
+//===---------------------------------------------------------------------===//
+// SSE2 Instructions
+//===---------------------------------------------------------------------===//
- // Intrinsic operation, reg+mem.
- def SDrm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, sdmem:$src2),
- !strconcat(OpcodeStr, "sd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (F64Int VR128:$src1,
- sse_load_f64:$src2))]>;
+// Conversion instructions
+def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
+ "cvtsd2ss\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (fround FR64:$src))]>;
+def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
+ "cvtsd2ss\t{$src, $dst|$dst, $src}",
+ [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
+ Requires<[HasSSE2, OptForSize]>;
- // Vector intrinsic operation, reg+reg.
- def PDrr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (V2F64Int VR128:$src1, VR128:$src2))]> {
- let isCommutable = Commutable;
- }
+def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
+def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
+def COMISDrr: PDI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
+ "comisd\t{$src2, $src1|$src1, $src2}", []>;
+def COMISDrm: PDI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
+ "comisd\t{$src2, $src1|$src1, $src2}", []>;
- // Vector intrinsic operation, reg+mem.
- def PDrm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f128mem:$src2),
- !strconcat(OpcodeStr, "pd\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (V2F64Int VR128:$src1,
- (memopv2f64 addr:$src2)))]>;
-}
-}
+// SSE2 instructions with XS prefix
+def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
+ "cvtss2sd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (fextend FR32:$src))]>, XS,
+ Requires<[HasSSE2]>;
+def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
+ "cvtss2sd\t{$src, $dst|$dst, $src}",
+ [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
+ Requires<[HasSSE2, OptForSize]>;
+
+def : Pat<(extloadf32 addr:$src),
+ (CVTSS2SDrr (MOVSSrm addr:$src))>,
+ Requires<[HasSSE2, OptForSpeed]>;
+
+// Match intrinsics which expect MM and XMM operand(s).
+def Int_CVTPI2PDrr : PDI<0x2A, MRMSrcReg, (outs VR128:$dst), (ins VR64:$src),
+ "cvtpi2pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_cvtpi2pd VR64:$src))]>;
+def Int_CVTPI2PDrm : PDI<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
+ "cvtpi2pd\t{$src, $dst|$dst, $src}",
+ [(set VR128:$dst, (int_x86_sse_cvtpi2pd
+ (load addr:$src)))]>;
-defm MAX : sse2_fp_binop_rm<0x5F, "max", X86fmax,
- int_x86_sse2_max_sd, int_x86_sse2_max_pd>;
-defm MIN : sse2_fp_binop_rm<0x5D, "min", X86fmin,
- int_x86_sse2_min_sd, int_x86_sse2_min_pd>;
+// Aliases for intrinsics
+def Int_CVTTSD2SIrr : SDI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst,
+ (int_x86_sse2_cvttsd2si VR128:$src))]>;
+def Int_CVTTSD2SIrm : SDI<0x2C, MRMSrcMem, (outs GR32:$dst), (ins f128mem:$src),
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
+ [(set GR32:$dst, (int_x86_sse2_cvttsd2si
+ (load addr:$src)))]>;
//===---------------------------------------------------------------------===//
// SSE packed FP Instructions
let neverHasSideEffects = 1 in
def MOVAPDrr : PDI<0x28, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movapd\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVAPDrm : PDI<0x28, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movapd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (alignedloadv2f64 addr:$src))]>;
def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
+ [(set VR128:$dst,
(int_x86_sse2_cvttps2dq VR128:$src))]>,
XS, Requires<[HasSSE2]>;
def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
// There is no f64 version of the reciprocal approximation instructions.
-// Logical
-let Constraints = "$src1 = $dst" in {
- let isCommutable = 1 in {
- def ANDPDrr : PDI<0x54, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "andpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (and (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64 VR128:$src2))))]>;
- def ORPDrr : PDI<0x56, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "orpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (or (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64 VR128:$src2))))]>;
- def XORPDrr : PDI<0x57, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "xorpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (xor (bc_v2i64 (v2f64 VR128:$src1)),
- (bc_v2i64 (v2f64 VR128:$src2))))]>;
- }
-
- def ANDPDrm : PDI<0x54, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "andpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (and (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def ORPDrm : PDI<0x56, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "orpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (or (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def XORPDrm : PDI<0x57, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "xorpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (xor (bc_v2i64 (v2f64 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>;
- def ANDNPDrr : PDI<0x55, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "andnpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
- (bc_v2i64 (v2f64 VR128:$src2))))]>;
- def ANDNPDrm : PDI<0x55, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1,f128mem:$src2),
- "andnpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
- (memopv2i64 addr:$src2)))]>;
-}
-
-let Constraints = "$src1 = $dst" in {
- def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
- "cmp${cc}pd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
- VR128:$src, imm:$cc))]>;
- def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src, SSECC:$cc),
- "cmp${cc}pd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
- (memop addr:$src), imm:$cc))]>;
-}
-def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
- (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
-def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
- (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
-
-// Shuffle and unpack instructions
-let Constraints = "$src1 = $dst" in {
- def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i8imm:$src3),
- "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR128:$dst,
- (v2f64 (shufp:$src3 VR128:$src1, VR128:$src2)))]>;
- def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1,
- f128mem:$src2, i8imm:$src3),
- "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR128:$dst,
- (v2f64 (shufp:$src3
- VR128:$src1, (memopv2f64 addr:$src2))))]>;
-
- let AddedComplexity = 10 in {
- def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "unpckhpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2f64 (unpckh VR128:$src1, VR128:$src2)))]>;
- def UNPCKHPDrm : PDI<0x15, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "unpckhpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2f64 (unpckh VR128:$src1,
- (memopv2f64 addr:$src2))))]>;
-
- def UNPCKLPDrr : PDI<0x14, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "unpcklpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2f64 (unpckl VR128:$src1, VR128:$src2)))]>;
- def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
- "unpcklpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckl VR128:$src1, (memopv2f64 addr:$src2)))]>;
- } // AddedComplexity
-} // Constraints = "$src1 = $dst"
-
-
//===---------------------------------------------------------------------===//
// SSE integer instructions
+let ExeDomain = SSEPackedInt in {
// Move Instructions
let neverHasSideEffects = 1 in
multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
bit Commutable = 0> {
- def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
+ def rr : PDI<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))]> {
let isCommutable = Commutable;
}
- def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1,
- (bitconvert (memopv2i64
+ (bitconvert (memopv2i64
addr:$src2))))]>;
}
multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
string OpcodeStr,
Intrinsic IntId, Intrinsic IntId2> {
- def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
+ def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1,
(bitconvert (memopv2i64 addr:$src2))))]>;
- def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
+ def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
(ins VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
/// PDI_binop_rm - Simple SSE2 binary operator.
multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType OpVT, bit Commutable = 0> {
- def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
+ def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]> {
let isCommutable = Commutable;
}
- def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (OpVT (OpNode VR128:$src1,
}
} // Constraints = "$src1 = $dst"
+} // ExeDomain = SSEPackedInt
// 128-bit Integer Arithmetic
int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
// 128-bit logical shifts.
-let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
+let Constraints = "$src1 = $dst", neverHasSideEffects = 1,
+ ExeDomain = SSEPackedInt in {
def PSLLDQri : PDIi8<0x73, MRM7r,
(outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
"pslldq\t{$src2, $dst|$dst, $src2}", []>;
defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or , 1>;
defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
-let Constraints = "$src1 = $dst" in {
+let Constraints = "$src1 = $dst", ExeDomain = SSEPackedInt in {
def PANDNrr : PDI<0xDF, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"pandn\t{$src2, $dst|$dst, $src2}",
defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
+let ExeDomain = SSEPackedInt in {
+
// Shuffle and unpack instructions
let AddedComplexity = 5 in {
def PSHUFDri : PDIi8<0x70, MRMSrcReg,
(undef))))]>,
XD, Requires<[HasSSE2]>;
+// Unpack instructions
+multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
+ PatFrag unp_frag, PatFrag bc_frag> {
+ def rr : PDI<opc, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
+ def rm : PDI<opc, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst, (unp_frag VR128:$src1,
+ (bc_frag (memopv2i64
+ addr:$src2))))]>;
+}
let Constraints = "$src1 = $dst" in {
- def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpcklbw\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v16i8 (unpckl VR128:$src1, VR128:$src2)))]>;
- def PUNPCKLBWrm : PDI<0x60, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpcklbw\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckl VR128:$src1,
- (bc_v16i8 (memopv2i64 addr:$src2))))]>;
- def PUNPCKLWDrr : PDI<0x61, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpcklwd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v8i16 (unpckl VR128:$src1, VR128:$src2)))]>;
- def PUNPCKLWDrm : PDI<0x61, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpcklwd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckl VR128:$src1,
- (bc_v8i16 (memopv2i64 addr:$src2))))]>;
- def PUNPCKLDQrr : PDI<0x62, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpckldq\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v4i32 (unpckl VR128:$src1, VR128:$src2)))]>;
- def PUNPCKLDQrm : PDI<0x62, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpckldq\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckl VR128:$src1,
- (bc_v4i32 (memopv2i64 addr:$src2))))]>;
+ defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
+ defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
+ defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
+
+ /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
+ /// knew to collapse (bitconvert VT to VT) into its operand.
def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpcklqdq\t{$src2, $dst|$dst, $src2}",
(v2i64 (unpckl VR128:$src1,
(memopv2i64 addr:$src2))))]>;
- def PUNPCKHBWrr : PDI<0x68, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpckhbw\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v16i8 (unpckh VR128:$src1, VR128:$src2)))]>;
- def PUNPCKHBWrm : PDI<0x68, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpckhbw\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckh VR128:$src1,
- (bc_v16i8 (memopv2i64 addr:$src2))))]>;
- def PUNPCKHWDrr : PDI<0x69, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpckhwd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v8i16 (unpckh VR128:$src1, VR128:$src2)))]>;
- def PUNPCKHWDrm : PDI<0x69, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpckhwd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckh VR128:$src1,
- (bc_v8i16 (memopv2i64 addr:$src2))))]>;
- def PUNPCKHDQrr : PDI<0x6A, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "punpckhdq\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v4i32 (unpckh VR128:$src1, VR128:$src2)))]>;
- def PUNPCKHDQrm : PDI<0x6A, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
- "punpckhdq\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckh VR128:$src1,
- (bc_v4i32 (memopv2i64 addr:$src2))))]>;
+ defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
+ defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
+ defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
+
+ /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
+ /// knew to collapse (bitconvert VT to VT) into its operand.
def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpckhqdq\t{$src2, $dst|$dst, $src2}",
"maskmovdqu\t{$mask, $src|$src, $mask}",
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
+} // ExeDomain = SSEPackedInt
+
// Non-temporal stores
def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"movntpd\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
+let ExeDomain = SSEPackedInt in
def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntdq\t{$src, $dst|$dst, $src}",
[(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
"movntpd\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
+let ExeDomain = SSEPackedInt in
def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntdq\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
-
-def : Pat<(alignednontemporalstore (v4i32 VR128:$src), addr:$dst),
- (MOVNTDQmr VR128:$src, addr:$dst)>;
}
// Flush cache
def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
"mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
+// Pause. This "instruction" is encoded as "rep; nop", so even though it
+// was introduced with SSE2, it's backward compatible.
+def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
+
//TODO: custom lower this so as to never even generate the noop
-def : Pat<(membarrier (i8 imm:$ll), (i8 imm:$ls), (i8 imm:$sl), (i8 imm:$ss),
+def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
(i8 0)), (NOOP)>;
def : Pat<(membarrier (i8 0), (i8 0), (i8 0), (i8 1), (i8 1)), (SFENCE)>;
def : Pat<(membarrier (i8 1), (i8 0), (i8 0), (i8 0), (i8 1)), (LFENCE)>;
-def : Pat<(membarrier (i8 imm:$ll), (i8 imm:$ls), (i8 imm:$sl), (i8 imm:$ss),
+def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
(i8 1)), (MFENCE)>;
// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-ones value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isCodeGenOnly = 1 in
+ isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
// FIXME: Change encoding to pseudo.
def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllOnesV))]>;
-// FR64 to 128-bit vector conversion.
-let isAsCheapAsAMove = 1 in
-def MOVSD2PDrr : SDI<0x10, MRMSrcReg, (outs VR128:$dst), (ins FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v2f64 (scalar_to_vector FR64:$src)))]>;
-def MOVSD2PDrm : SDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v2f64 (scalar_to_vector (loadf64 addr:$src))))]>;
-
def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
[(store (i64 (vector_extract (v2i64 VR128:$src),
(iPTR 0))), addr:$dst)]>;
-// FIXME: may not be able to eliminate this movss with coalescing the src and
-// dest register classes are different. We really want to write this pattern
-// like this:
-// def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
-// (f32 FR32:$src)>;
-let isAsCheapAsAMove = 1 in
-def MOVPD2SDrr : SDI<0x10, MRMSrcReg, (outs FR64:$dst), (ins VR128:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(set FR64:$dst, (vector_extract (v2f64 VR128:$src),
- (iPTR 0)))]>;
-def MOVPD2SDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store (f64 (vector_extract (v2f64 VR128:$src),
- (iPTR 0))), addr:$dst)]>;
+def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
+ (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
+
def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
-
-// Move to lower bits of a VR128, leaving upper bits alone.
-// Three operand (but two address) aliases.
-let Constraints = "$src1 = $dst" in {
- let neverHasSideEffects = 1 in
- def MOVLSD2PDrr : SDI<0x10, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, FR64:$src2),
- "movsd\t{$src2, $dst|$dst, $src2}", []>;
-
- let AddedComplexity = 15 in
- def MOVLPDrr : SDI<0x10, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
- "movsd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2f64 (movl VR128:$src1, VR128:$src2)))]>;
-}
-
// Store / copy lower 64-bits of a XMM register.
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:$src)]>;
-// Move to lower bits of a VR128 and zeroing upper bits.
-// Loading from memory automatically zeroing upper bits.
-let AddedComplexity = 20 in {
-def MOVZSD2PDrm : SDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst,
- (v2f64 (X86vzmovl (v2f64 (scalar_to_vector
- (loadf64 addr:$src))))))]>;
-
-def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
- (MOVZSD2PDrm addr:$src)>;
-def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
- (MOVZSD2PDrm addr:$src)>;
-def : Pat<(v2f64 (X86vzload addr:$src)), (MOVZSD2PDrm addr:$src)>;
-}
-
// movd / movq to XMM register zero-extends
let AddedComplexity = 15 in {
def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
// SSE3 Instructions
//===---------------------------------------------------------------------===//
+// Conversion Instructions
+def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
+def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
+def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
+def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
+
// Move Instructions
def MOVSHDUPrr : S3SI<0x16, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movshdup\t{$src, $dst|$dst, $src}",
}
}
+let ImmT = NoImm in { // None of these have i8 immediate fields.
defm PHADDW : SS3I_binop_rm_int_16<0x01, "phaddw",
int_x86_ssse3_phadd_w,
int_x86_ssse3_phadd_w_128>;
defm PMULHRSW : SS3I_binop_rm_int_16<0x0B, "pmulhrsw",
int_x86_ssse3_pmul_hr_sw,
int_x86_ssse3_pmul_hr_sw_128, 1>;
+
defm PSHUFB : SS3I_binop_rm_int_8 <0x00, "pshufb",
int_x86_ssse3_pshuf_b,
int_x86_ssse3_pshuf_b_128>;
defm PSIGND : SS3I_binop_rm_int_32<0x0A, "psignd",
int_x86_ssse3_psign_d,
int_x86_ssse3_psign_d_128>;
+}
+// palignr patterns.
let Constraints = "$src1 = $dst" in {
def PALIGNR64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
(ins VR64:$src1, VR64:$src2, i8imm:$src3),
[]>, OpSize;
}
-// palignr patterns.
-def : Pat<(int_x86_ssse3_palign_r VR64:$src1, VR64:$src2, (i8 imm:$src3)),
- (PALIGNR64rr VR64:$src1, VR64:$src2, (BYTE_imm imm:$src3))>,
+let AddedComplexity = 5 in {
+
+def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
+ (PALIGNR64rr VR64:$src2, VR64:$src1,
+ (SHUFFLE_get_palign_imm VR64:$src3))>,
Requires<[HasSSSE3]>;
-def : Pat<(int_x86_ssse3_palign_r VR64:$src1,
- (memop64 addr:$src2),
- (i8 imm:$src3)),
- (PALIGNR64rm VR64:$src1, addr:$src2, (BYTE_imm imm:$src3))>,
+def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
+ (PALIGNR64rr VR64:$src2, VR64:$src1,
+ (SHUFFLE_get_palign_imm VR64:$src3))>,
Requires<[HasSSSE3]>;
-
-def : Pat<(int_x86_ssse3_palign_r_128 VR128:$src1, VR128:$src2, (i8 imm:$src3)),
- (PALIGNR128rr VR128:$src1, VR128:$src2, (BYTE_imm imm:$src3))>,
+def : Pat<(v2f32 (palign:$src3 VR64:$src1, VR64:$src2)),
+ (PALIGNR64rr VR64:$src2, VR64:$src1,
+ (SHUFFLE_get_palign_imm VR64:$src3))>,
Requires<[HasSSSE3]>;
-def : Pat<(int_x86_ssse3_palign_r_128 VR128:$src1,
- (memopv2i64 addr:$src2),
- (i8 imm:$src3)),
- (PALIGNR128rm VR128:$src1, addr:$src2, (BYTE_imm imm:$src3))>,
+def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
+ (PALIGNR64rr VR64:$src2, VR64:$src1,
+ (SHUFFLE_get_palign_imm VR64:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
+ (PALIGNR64rr VR64:$src2, VR64:$src1,
+ (SHUFFLE_get_palign_imm VR64:$src3))>,
Requires<[HasSSSE3]>;
-let AddedComplexity = 5 in {
def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
(PALIGNR128rr VR128:$src2, VR128:$src1,
(SHUFFLE_get_palign_imm VR128:$src3))>,
let AddedComplexity = 15 in {
// Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
- (MOVLSD2PDrr (V_SET0), FR64:$src)>, Requires<[HasSSE2]>;
+ (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
- (MOVLSS2PSrr (V_SET0), FR32:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
- (MOVLPSrr (V_SET0), VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4f32 (V_SET0PS)),
+ (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
- (MOVLPSrr (V_SET0), VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4i32 (V_SET0PI)),
+ (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
}
// Splat v2f64 / v2i64
// Special unary SHUFPSrri case.
def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
(SHUFPSrri VR128:$src1, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>,
- Requires<[HasSSE1]>;
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
let AddedComplexity = 5 in
def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
(PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
}
let AddedComplexity = 10 in {
def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
- (UNPCKLPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
+ (UNPCKLPSrr VR128:$src, VR128:$src)>;
def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLBWrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKLBWrr VR128:$src, VR128:$src)>;
def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLWDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKLWDrr VR128:$src, VR128:$src)>;
def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
- (PUNPCKLDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKLDQrr VR128:$src, VR128:$src)>;
}
// vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
}
let AddedComplexity = 10 in {
def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
- (UNPCKHPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
+ (UNPCKHPSrr VR128:$src, VR128:$src)>;
def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHBWrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKHBWrr VR128:$src, VR128:$src)>;
def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHWDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKHWDrr VR128:$src, VR128:$src)>;
def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
- (PUNPCKHDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
+ (PUNPCKHDQrr VR128:$src, VR128:$src)>;
}
let AddedComplexity = 20 in {
let AddedComplexity = 20 in {
// vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
- (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
+ (MOVLPSrm VR128:$src1, addr:$src2)>;
def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
- (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
- (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPSrm VR128:$src1, addr:$src2)>;
def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
- (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
}
// (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVLPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>;
+ (MOVLPSmr addr:$src1, VR128:$src2)>;
def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVLPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPDmr addr:$src1, VR128:$src2)>;
def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
addr:$src1),
- (MOVLPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>;
+ (MOVLPSmr addr:$src1, VR128:$src2)>;
def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVLPDmr addr:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPDmr addr:$src1, VR128:$src2)>;
let AddedComplexity = 15 in {
// Setting the lowest element in the vector.
def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
- (MOVLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSSrr (v4i32 VR128:$src1),
+ (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr (v2i64 VR128:$src1),
+ (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
-// vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
+// vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
+ Requires<[HasSSE2]>;
def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
+ Requires<[HasSSE2]>;
}
// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
// fall back to this for SSE1)
def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
(SHUFPSrri VR128:$src2, VR128:$src1,
- (SHUFFLE_get_shuf_imm VR128:$src3))>, Requires<[HasSSE1]>;
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
// Set lowest element and zero upper elements.
-let AddedComplexity = 15 in
-def : Pat<(v2f64 (movl immAllZerosV_bc, VR128:$src)),
- (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
// Use movaps / movups for SSE integer load / store (one byte shorter).
def : Pat<(alignedloadv4i32 addr:$src),
- (MOVAPSrm addr:$src)>, Requires<[HasSSE1]>;
+ (MOVAPSrm addr:$src)>;
def : Pat<(loadv4i32 addr:$src),
- (MOVUPSrm addr:$src)>, Requires<[HasSSE1]>;
+ (MOVUPSrm addr:$src)>;
def : Pat<(alignedloadv2i64 addr:$src),
- (MOVAPSrm addr:$src)>, Requires<[HasSSE2]>;
+ (MOVAPSrm addr:$src)>;
def : Pat<(loadv2i64 addr:$src),
- (MOVUPSrm addr:$src)>, Requires<[HasSSE2]>;
+ (MOVUPSrm addr:$src)>;
def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVAPSmr addr:$dst, VR128:$src)>;
def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVAPSmr addr:$dst, VR128:$src)>;
def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVAPSmr addr:$dst, VR128:$src)>;
def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
- (MOVAPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVAPSmr addr:$dst, VR128:$src)>;
def : Pat<(store (v2i64 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVUPSmr addr:$dst, VR128:$src)>;
def : Pat<(store (v4i32 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVUPSmr addr:$dst, VR128:$src)>;
def : Pat<(store (v8i16 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVUPSmr addr:$dst, VR128:$src)>;
def : Pat<(store (v16i8 VR128:$src), addr:$dst),
- (MOVUPSmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
+ (MOVUPSmr addr:$dst, VR128:$src)>;
//===----------------------------------------------------------------------===//
// SSE4.1 Instructions
OpSize;
}
}
-defm PMULLD : SS41I_binop_patint<0x40, "pmulld", v4i32, mul,
- int_x86_sse41_pmulld, 1>;
+
+/// SS48I_binop_rm - Simple SSE41 binary operator.
+let Constraints = "$src1 = $dst" in {
+multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ ValueType OpVT, bit Commutable = 0> {
+ def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
+ OpSize {
+ let isCommutable = Commutable;
+ }
+ def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst, (OpNode VR128:$src1,
+ (bc_v4i32 (memopv2i64 addr:$src2))))]>,
+ OpSize;
+}
+}
+
+defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, 1>;
/// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
let Constraints = "$src1 = $dst" in {
defm DPPD : SS41I_binop_rmi_int<0x41, "dppd",
int_x86_sse41_dppd, 1>;
defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw",
- int_x86_sse41_mpsadbw, 1>;
+ int_x86_sse41_mpsadbw, 0>;
/// SS41I_ternary_int - SSE 4.1 ternary operator
let Defs = [EFLAGS] in {
def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"ptest \t{$src2, $src1|$src1, $src2}",
- [(X86ptest VR128:$src1, VR128:$src2),
- (implicit EFLAGS)]>, OpSize;
+ [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
+ OpSize;
def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
"ptest \t{$src2, $src1|$src1, $src2}",
- [(X86ptest VR128:$src1, (load addr:$src2)),
- (implicit EFLAGS)]>, OpSize;
+ [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
+ OpSize;
}
def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
let Constraints = "$src1 = $dst" in {
def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i8mem:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
(int_x86_sse42_crc32_8 GR32:$src1,
- (load addr:$src2)))]>, OpSize;
+ (load addr:$src2)))]>;
def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR8:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>,
- OpSize;
+ (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i16mem:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{w} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
(int_x86_sse42_crc32_16 GR32:$src1,
(load addr:$src2)))]>,
OpSize;
def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR16:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{w} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
(int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
OpSize;
def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
(ins GR32:$src1, i32mem:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{l} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
(int_x86_sse42_crc32_32 GR32:$src1,
- (load addr:$src2)))]>, OpSize;
+ (load addr:$src2)))]>;
def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
(ins GR32:$src1, GR32:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{l} \t{$src2, $src1|$src1, $src2}",
[(set GR32:$dst,
- (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>,
- OpSize;
- def CRC64m64 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
+ (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
+ def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
+ (ins GR64:$src1, i8mem:$src2),
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
+ [(set GR64:$dst,
+ (int_x86_sse42_crc64_8 GR64:$src1,
+ (load addr:$src2)))]>,
+ REX_W;
+ def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
+ (ins GR64:$src1, GR8:$src2),
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
+ [(set GR64:$dst,
+ (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
+ REX_W;
+ def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
(ins GR64:$src1, i64mem:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{q} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc32_64 GR64:$src1,
+ (int_x86_sse42_crc64_64 GR64:$src1,
(load addr:$src2)))]>,
- OpSize, REX_W;
- def CRC64r64 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
+ REX_W;
+ def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
(ins GR64:$src1, GR64:$src2),
- "crc32 \t{$src2, $src1|$src1, $src2}",
+ "crc32{q} \t{$src2, $src1|$src1, $src2}",
[(set GR64:$dst,
- (int_x86_sse42_crc32_64 GR64:$src1, GR64:$src2))]>,
- OpSize, REX_W;
+ (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
+ REX_W;
}
// String/text processing instructions.
def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src3, i8imm:$src5),
"#PCMPESTRM128rr PSEUDO!",
- [(set VR128:$dst,
- (int_x86_sse42_pcmpestrm128
+ [(set VR128:$dst,
+ (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>, OpSize;
def PCMPESTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"#PCMPESTRM128rm PSEUDO!",
- [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
- VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
+ [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
+ VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
OpSize;
}
let Defs = [ECX, EFLAGS] in {
multiclass SS42AI_pcmpistri<Intrinsic IntId128> {
- def rr : SS42AI<0x63, MRMSrcReg, (outs),
+ def rr : SS42AI<0x63, MRMSrcReg, (outs),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
"pcmpistri\t{$src3, $src2, $src1|$src1, $src2, $src3}",
[(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
def rm : SS42AI<0x61, MRMSrcMem, (outs),
(ins VR128:$src1, i128mem:$src3, i8imm:$src5),
"pcmpestri\t{$src5, $src3, $src1|$src1, $src3, $src5}",
- [(set ECX,
+ [(set ECX,
(IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
(implicit EFLAGS)]>, OpSize;
}
defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
+
+//===----------------------------------------------------------------------===//
+// AES-NI Instructions
+//===----------------------------------------------------------------------===//
+
+let Constraints = "$src1 = $dst" in {
+ multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
+ Intrinsic IntId128, bit Commutable = 0> {
+ def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
+ OpSize {
+ let isCommutable = Commutable;
+ }
+ def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ [(set VR128:$dst,
+ (IntId128 VR128:$src1,
+ (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
+ }
+}
+
+defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
+ int_x86_aesni_aesenc>;
+defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
+ int_x86_aesni_aesenclast>;
+defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
+ int_x86_aesni_aesdec>;
+defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
+ int_x86_aesni_aesdeclast>;
+
+def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
+ (AESENCrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
+ (AESENCrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
+ (AESENCLASTrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
+ (AESENCLASTrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
+ (AESDECrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
+ (AESDECrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
+ (AESDECLASTrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
+ (AESDECLASTrm VR128:$src1, addr:$src2)>;
+
+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:$src1))]>,
+ OpSize;
+
+def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
+ (ins i128mem:$src1),
+ "aesimc\t{$src1, $dst|$dst, $src1}",
+ [(set VR128:$dst,
+ (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
+ OpSize;
+
+def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$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;
+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 (bitconvert (memopv2i64 addr:$src1)),
+ imm:$src2))]>,
+ OpSize;
projects / oota-llvm.git / blobdiff