//====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
-//
+//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
-//
+//
//===----------------------------------------------------------------------===//
//
// This file describes the X86 SSE instruction set, defining the instructions,
def X86fsrl : SDNode<"X86ISD::FSRL", SDTX86FPShiftOp>;
def X86comi : SDNode<"X86ISD::COMI", SDTX86CmpTest>;
def X86ucomi : SDNode<"X86ISD::UCOMI", SDTX86CmpTest>;
-def X86pshufb : SDNode<"X86ISD::PSHUFB",
+def X86pshufb : SDNode<"X86ISD::PSHUFB",
SDTypeProfile<1, 2, [SDTCisVT<0, v16i8>, SDTCisSameAs<0,1>,
SDTCisSameAs<0,2>]>>;
def X86pextrb : SDNode<"X86ISD::PEXTRB",
SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<2>]>>;
def X86pextrw : SDNode<"X86ISD::PEXTRW",
SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisPtrTy<2>]>>;
-def X86pinsrb : SDNode<"X86ISD::PINSRB",
+def X86pinsrb : SDNode<"X86ISD::PINSRB",
SDTypeProfile<1, 3, [SDTCisVT<0, v16i8>, SDTCisSameAs<0,1>,
SDTCisVT<2, i32>, SDTCisPtrTy<3>]>>;
-def X86pinsrw : SDNode<"X86ISD::PINSRW",
+def X86pinsrw : SDNode<"X86ISD::PINSRW",
SDTypeProfile<1, 3, [SDTCisVT<0, v8i16>, SDTCisSameAs<0,1>,
SDTCisVT<2, i32>, SDTCisPtrTy<3>]>>;
-def X86insrtps : SDNode<"X86ISD::INSERTPS",
+def X86insrtps : SDNode<"X86ISD::INSERTPS",
SDTypeProfile<1, 3, [SDTCisVT<0, v4f32>, SDTCisSameAs<0,1>,
- SDTCisVT<2, f32>, SDTCisPtrTy<3>]>>;
+ SDTCisVT<2, v4f32>, SDTCisPtrTy<3>]>>;
def X86vzmovl : SDNode<"X86ISD::VZEXT_MOVL",
SDTypeProfile<1, 1, [SDTCisSameAs<0,1>]>>;
def X86vzload : SDNode<"X86ISD::VZEXT_LOAD", SDTLoad,
def X86pcmpgtd : SDNode<"X86ISD::PCMPGTD", SDTIntBinOp>;
def X86pcmpgtq : SDNode<"X86ISD::PCMPGTQ", SDTIntBinOp>;
+def X86aesimc : SDNode<"X86ISD::AESIMC", SDTIntBinOp>;
+def X86aesenc : SDNode<"X86ISD::AESENC", SDTIntBinOp>;
+def X86aesenclast : SDNode<"X86ISD::AESENCLAST", SDTIntBinOp>;
+def X86aesdec : SDNode<"X86ISD::AESDEC", SDTIntBinOp>;
+def X86aesdeclast : SDNode<"X86ISD::AESDECLAST", SDTIntBinOp>;
+
+def SDTX86CmpPTest : SDTypeProfile<0, 2, [SDTCisVT<0, v4f32>,
+ SDTCisVT<1, v4f32>]>;
+def X86ptest : SDNode<"X86ISD::PTEST", SDTX86CmpPTest>;
+
//===----------------------------------------------------------------------===//
// SSE Complex Patterns
//===----------------------------------------------------------------------===//
def ssmem : Operand<v4f32> {
let PrintMethod = "printf32mem";
- let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm, i8imm);
+ let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc_nosp, i32imm, i8imm);
+ let ParserMatchClass = X86MemAsmOperand;
}
def sdmem : Operand<v2f64> {
let PrintMethod = "printf64mem";
- let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm, i8imm);
+ let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc_nosp, i32imm, i8imm);
+ let ParserMatchClass = X86MemAsmOperand;
}
//===----------------------------------------------------------------------===//
return cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;
-def alignedloadfsf32 : PatFrag<(ops node:$ptr), (f32 (alignedload node:$ptr))>;
-def alignedloadfsf64 : PatFrag<(ops node:$ptr), (f64 (alignedload node:$ptr))>;
-def alignedloadv4f32 : PatFrag<(ops node:$ptr), (v4f32 (alignedload node:$ptr))>;
-def alignedloadv2f64 : PatFrag<(ops node:$ptr), (v2f64 (alignedload node:$ptr))>;
-def alignedloadv4i32 : PatFrag<(ops node:$ptr), (v4i32 (alignedload node:$ptr))>;
-def alignedloadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (alignedload node:$ptr))>;
+def alignedloadfsf32 : PatFrag<(ops node:$ptr),
+ (f32 (alignedload node:$ptr))>;
+def alignedloadfsf64 : PatFrag<(ops node:$ptr),
+ (f64 (alignedload node:$ptr))>;
+def alignedloadv4f32 : PatFrag<(ops node:$ptr),
+ (v4f32 (alignedload node:$ptr))>;
+def alignedloadv2f64 : PatFrag<(ops node:$ptr),
+ (v2f64 (alignedload node:$ptr))>;
+def alignedloadv4i32 : PatFrag<(ops node:$ptr),
+ (v4i32 (alignedload node:$ptr))>;
+def alignedloadv2i64 : PatFrag<(ops node:$ptr),
+ (v2i64 (alignedload node:$ptr))>;
// Like 'load', but uses special alignment checks suitable for use in
// memory operands in most SSE instructions, which are required to
-// be naturally aligned on some targets but not on others.
-// FIXME: Actually implement support for targets that don't require the
-// alignment. This probably wants a subtarget predicate.
+// be naturally aligned on some targets but not on others. If the subtarget
+// allows unaligned accesses, match any load, though this may require
+// setting a feature bit in the processor (on startup, for example).
+// Opteron 10h and later implement such a feature.
def memop : PatFrag<(ops node:$ptr), (load node:$ptr), [{
- return cast<LoadSDNode>(N)->getAlignment() >= 16;
+ return Subtarget->hasVectorUAMem()
+ || cast<LoadSDNode>(N)->getAlignment() >= 16;
}]>;
def memopfsf32 : PatFrag<(ops node:$ptr), (f32 (memop node:$ptr))>;
def memopv8i16 : PatFrag<(ops node:$ptr), (v8i16 (memop64 node:$ptr))>;
def memopv2i32 : PatFrag<(ops node:$ptr), (v2i32 (memop64 node:$ptr))>;
+// MOVNT Support
+// Like 'store', but requires the non-temporal bit to be set
+def nontemporalstore : PatFrag<(ops node:$val, node:$ptr),
+ (st node:$val, node:$ptr), [{
+ if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
+ return ST->isNonTemporal();
+ return false;
+}]>;
+
+def alignednontemporalstore : PatFrag<(ops node:$val, node:$ptr),
+ (st node:$val, node:$ptr), [{
+ if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
+ return ST->isNonTemporal() && !ST->isTruncatingStore() &&
+ ST->getAddressingMode() == ISD::UNINDEXED &&
+ ST->getAlignment() >= 16;
+ return false;
+}]>;
+
+def unalignednontemporalstore : PatFrag<(ops node:$val, node:$ptr),
+ (st node:$val, node:$ptr), [{
+ if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
+ return ST->isNonTemporal() &&
+ ST->getAlignment() < 16;
+ return false;
+}]>;
+
def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
return N->isExactlyValue(+0.0);
}]>;
-def PSxLDQ_imm : SDNodeXForm<imm, [{
+// BYTE_imm - Transform bit immediates into byte immediates.
+def BYTE_imm : SDNodeXForm<imm, [{
// Transformation function: imm >> 3
return getI32Imm(N->getZExtValue() >> 3);
}]>;
return getI8Imm(X86::getShuffleSHUFImmediate(N));
}]>;
-// SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to
+// SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to
// PSHUFHW imm.
def SHUFFLE_get_pshufhw_imm : SDNodeXForm<vector_shuffle, [{
return getI8Imm(X86::getShufflePSHUFHWImmediate(N));
}]>;
-// SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to
+// SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to
// PSHUFLW imm.
def SHUFFLE_get_pshuflw_imm : SDNodeXForm<vector_shuffle, [{
return getI8Imm(X86::getShufflePSHUFLWImmediate(N));
}]>;
+// SHUFFLE_get_palign_imm xform function: convert vector_shuffle mask to
+// a PALIGNR imm.
+def SHUFFLE_get_palign_imm : SDNodeXForm<vector_shuffle, [{
+ return getI8Imm(X86::getShufflePALIGNRImmediate(N));
+}]>;
+
def splat_lo : PatFrag<(ops node:$lhs, node:$rhs),
(vector_shuffle node:$lhs, node:$rhs), [{
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(N);
return X86::isMOVHLPS_v_undef_Mask(cast<ShuffleVectorSDNode>(N));
}]>;
-def movhp : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
- return X86::isMOVHPMask(cast<ShuffleVectorSDNode>(N));
+def movlhps : PatFrag<(ops node:$lhs, node:$rhs),
+ (vector_shuffle node:$lhs, node:$rhs), [{
+ return X86::isMOVLHPSMask(cast<ShuffleVectorSDNode>(N));
}]>;
def movlp : PatFrag<(ops node:$lhs, node:$rhs),
return X86::isPSHUFLWMask(cast<ShuffleVectorSDNode>(N));
}], SHUFFLE_get_pshuflw_imm>;
+def palign : PatFrag<(ops node:$lhs, node:$rhs),
+ (vector_shuffle node:$lhs, node:$rhs), [{
+ return X86::isPALIGNRMask(cast<ShuffleVectorSDNode>(N));
+}], SHUFFLE_get_palign_imm>;
+
//===----------------------------------------------------------------------===//
// SSE scalar FP Instructions
//===----------------------------------------------------------------------===//
-// CMOV* - Used to implement the SSE SELECT DAG operation. Expanded by the
-// scheduler into a branch sequence.
-// These are expanded by the scheduler.
-let Uses = [EFLAGS], usesCustomDAGSchedInserter = 1 in {
+// CMOV* - Used to implement the SSE SELECT DAG operation. Expanded after
+// instruction selection into a branch sequence.
+let Uses = [EFLAGS], usesCustomInserter = 1 in {
def CMOV_FR32 : I<0, Pseudo,
(outs FR32:$dst), (ins FR32:$t, FR32:$f, i8imm:$cond),
"#CMOV_FR32 PSEUDO!",
// SSE1 Instructions
//===----------------------------------------------------------------------===//
-// Move Instructions
-let neverHasSideEffects = 1 in
-def MOVSSrr : SSI<0x10, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- "movss\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+// Move Instructions. Register-to-register movss is not used for FR32
+// register copies because it's a partial register update; FsMOVAPSrr is
+// used instead. Register-to-register movss is not modeled as an INSERT_SUBREG
+// because INSERT_SUBREG requires that the insert be implementable in terms of
+// a copy, and just mentioned, we don't use movss for copies.
+let Constraints = "$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)))]>;
+
+// Extract the low 32-bit value from one vector and insert it into another.
+let AddedComplexity = 15 in
+def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
+ (MOVSSrr (v4f32 VR128:$src1),
+ (EXTRACT_SUBREG (v4f32 VR128:$src2), x86_subreg_ss))>;
+
+// Implicitly promote a 32-bit scalar to a vector.
+def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
+ (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, x86_subreg_ss)>;
+
+// 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:$src))]>;
+
+// MOVSSrm zeros the high parts of the register; represent this
+// with SUBREG_TO_REG.
+let AddedComplexity = 20 in {
+def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), x86_subreg_ss)>;
+def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), x86_subreg_ss)>;
+def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
+ (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), x86_subreg_ss)>;
+}
+
+// Store scalar value to memory.
def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
"movss\t{$src, $dst|$dst, $src}",
[(store FR32:$src, addr:$dst)]>;
+// Extract and store.
+def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
+ addr:$dst),
+ (MOVSSmr addr:$dst,
+ (EXTRACT_SUBREG (v4f32 VR128:$src), x86_subreg_ss))>;
+
// Conversion instructions
def CVTTSS2SIrr : SSI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
"cvttss2si\t{$src, $dst|$dst, $src}",
[(set FR32:$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:$src))]>;
[(set VR64:$dst, (int_x86_sse_cvtps2pi VR128:$src))]>;
def Int_CVTPS2PIrm : PSI<0x2D, MRMSrcMem, (outs VR64:$dst), (ins f64mem:$src),
"cvtps2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvtps2pi
+ [(set VR64:$dst, (int_x86_sse_cvtps2pi
(load addr:$src)))]>;
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:$src))]>;
def Int_CVTTPS2PIrm: PSI<0x2C, MRMSrcMem, (outs VR64:$dst), (ins f64mem:$src),
"cvttps2pi\t{$src, $dst|$dst, $src}",
- [(set VR64:$dst, (int_x86_sse_cvttps2pi
+ [(set VR64:$dst, (int_x86_sse_cvttps2pi
(load addr:$src)))]>;
let Constraints = "$src1 = $dst" in {
- def Int_CVTPI2PSrr : PSI<0x2A, MRMSrcReg,
+ def Int_CVTPI2PSrr : PSI<0x2A, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR64:$src2),
"cvtpi2ps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst, (int_x86_sse_cvtpi2ps VR128:$src1,
VR64:$src2))]>;
- def Int_CVTPI2PSrm : PSI<0x2A, MRMSrcMem,
+ def Int_CVTPI2PSrm : PSI<0x2A, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i64mem:$src2),
"cvtpi2ps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst, (int_x86_sse_cvtpi2ps VR128:$src1,
+ [(set VR128:$dst, (int_x86_sse_cvtpi2ps VR128:$src1,
(load addr:$src2)))]>;
}
// Comparison instructions
let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
- def CMPSSrr : SSIi8<0xC2, MRMSrcReg,
+ def CMPSSrr : SSIi8<0xC2, MRMSrcReg,
(outs FR32:$dst), (ins FR32:$src1, FR32:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
- def CMPSSrm : SSIi8<0xC2, MRMSrcMem,
+ def CMPSSrm : SSIi8<0xC2, MRMSrcMem,
(outs FR32:$dst), (ins FR32:$src1, f32mem:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}", []>;
}
let Defs = [EFLAGS] in {
def UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins FR32:$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:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86cmp FR32:$src1, (loadf32 addr:$src2)))]>;
+
+def COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$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, $src2}", []>;
+
} // Defs = [EFLAGS]
// Aliases to match intrinsics which expect XMM operand(s).
let Constraints = "$src1 = $dst" in {
- def Int_CMPSSrr : SSIi8<0xC2, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
+ def Int_CMPSSrr : SSIi8<0xC2, MRMSrcReg,
+ (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src, SSECC:$cc),
"cmp${cc}ss\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
- VR128:$src, imm:$cc))]>;
- def Int_CMPSSrm : SSIi8<0xC2, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f32mem:$src, SSECC:$cc),
+ [(set VR128:$dst, (int_x86_sse_cmp_ss
+ VR128:$src1,
+ VR128:$src, imm:$cc))]>;
+ def Int_CMPSSrm : SSIi8<0xC2, MRMSrcMem,
+ (outs VR128:$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:$cc))]>;
let Defs = [EFLAGS] in {
def Int_UCOMISSrr: PSI<0x2E, MRMSrcReg, (outs), (ins VR128:$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:$src2)))]>;
def Int_COMISSrr: PSI<0x2F, MRMSrcReg, (outs), (ins VR128:$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:$src2)))]>;
} // Defs = [EFLAGS]
-// Aliases of packed SSE1 instructions for scalar use. These all have names that
-// start with 'Fs'.
+// Aliases of packed SSE1 instructions for scalar use. These all have names
+// that start with 'Fs'.
// Alias instructions that map fld0 to pxor for sse.
-let isReMaterializable = 1, isAsCheapAsAMove = 1 in
-def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins),
- "pxor\t$dst, $dst", [(set FR32:$dst, fp32imm0)]>,
- Requires<[HasSSE1]>, TB, OpSize;
+let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
+ 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;
// Alias instruction to do FR32 reg-to-reg copy using movaps. Upper bits are
// disregarded.
-let neverHasSideEffects = 1 in
+let neverHasSideEffects = 1 in
def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
"movaps\t{$src, $dst|$dst, $src}", []>;
// Alias instruction to load FR32 from f128mem using movaps. Upper bits are
// disregarded.
-let canFoldAsLoad = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
(ins FR32:$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:$src2),
(ins FR32:$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:$src2),
// SSE packed FP Instructions
// Move Instructions
-let neverHasSideEffects = 1 in
+let neverHasSideEffects = 1 in
def MOVAPSrr : PSI<0x28, MRMSrcReg, (outs VR128:$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, $src}", []>;
-let canFoldAsLoad = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVUPSrm : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movups\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (loadv4f32 addr:$src))]>;
[(store (v4f32 VR128:$src), addr:$dst)]>;
// Intrinsic forms of MOVUPS load and store
-let canFoldAsLoad = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"movups\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
def MOVLPSrm : PSI<0x12, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
"movlps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
+ [(set VR128:$dst,
(movlp VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))]>;
def MOVHPSrm : PSI<0x16, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
"movhps\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (movhp VR128:$src1,
+ [(set VR128:$dst,
+ (movlhps VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))]>;
} // AddedComplexity
} // Constraints = "$src1 = $dst"
+def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
+ (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
+
def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
"movlps\t{$src, $dst|$dst, $src}",
[(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
(ins VR128:$src1, VR128:$src2),
"movlhps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
- (v4f32 (movhp VR128:$src1, VR128:$src2)))]>;
+ (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
let AddedComplexity = 20 in {
def : Pat<(v4f32 (movddup VR128:$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:$src))>;
}
}
// Scalar operation, mem.
- def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
+ def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode (load addr:$src)))]>;
-
+ [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
+ Requires<[HasSSE1, OptForSize]>;
+
// Vector operation, reg.
def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
}
let Constraints = "$src1 = $dst" in {
- def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
+ def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
(outs VR128:$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:$src1,
VR128:$src, imm:$cc))]>;
- def CMPPSrmi : PSIi8<0xC2, MRMSrcMem,
+ def CMPPSrmi : PSIi8<0xC2, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f128mem:$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:$cc)>;
+ (CMPPSrri (v4f32 VR128:$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:$cc)>;
// Shuffle and unpack instructions
let Constraints = "$src1 = $dst" in {
let isConvertibleToThreeAddress = 1 in // Convert to pshufd
- def SHUFPSrri : PSIi8<0xC6, MRMSrcReg,
+ def SHUFPSrri : PSIi8<0xC6, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1,
VR128:$src2, i8imm:$src3),
"shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst,
(v4f32 (shufp:$src3 VR128:$src1, VR128:$src2)))]>;
- def SHUFPSrmi : PSIi8<0xC6, MRMSrcMem,
+ def SHUFPSrmi : PSIi8<0xC6, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1,
f128mem:$src2, i8imm:$src3),
"shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
VR128:$src1, (memopv4f32 addr:$src2))))]>;
let AddedComplexity = 10 in {
- def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
+ def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"unpckhps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (unpckh VR128:$src1, VR128:$src2)))]>;
- def UNPCKHPSrm : PSI<0x15, MRMSrcMem,
+ def UNPCKHPSrm : PSI<0x15, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
"unpckhps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (unpckh VR128:$src1,
(memopv4f32 addr:$src2))))]>;
- def UNPCKLPSrr : PSI<0x14, MRMSrcReg,
+ def UNPCKLPSrr : PSI<0x14, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"unpcklps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (unpckl VR128:$src1, VR128:$src2)))]>;
- def UNPCKLPSrm : PSI<0x14, MRMSrcMem,
+ def UNPCKLPSrm : PSI<0x14, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
"unpcklps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
"prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
// Non-temporal stores
-def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
+def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"movntps\t{$src, $dst|$dst, $src}",
[(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
+let AddedComplexity = 400 in { // Prefer non-temporal versions
+def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
+ "movntps\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
+
+def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(alignednontemporalstore (v2f64 VR128:$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:$dst)]>,
+ TB, Requires<[HasSSE2]>;
+
+def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
+ "movnti\t{$src, $dst|$dst, $src}",
+ [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
+ TB, Requires<[HasSSE2]>;
+}
+
// Load, store, and memory fence
def SFENCE : PSI<0xAE, MRM7r, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>;
// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
-let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1 in
-def V_SET0 : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins),
- "xorps\t$dst, $dst",
+// FIXME: Change encoding to pseudo!
+let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
+ isCodeGenOnly = 1 in
+def V_SET0 : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
[(set VR128:$dst, (v4i32 immAllZerosV))]>;
-let Predicates = [HasSSE1] in {
- def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
- def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
- def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
- def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
- def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
-}
+def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
+def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
+def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
+def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
+def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
-// FR32 to 128-bit vector conversion.
-let isAsCheapAsAMove = 1 in
-def MOVSS2PSrr : SSI<0x10, MRMSrcReg, (outs VR128:$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, $src2}", []>;
+def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
+ (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), x86_subreg_ss))>;
- let AddedComplexity = 15 in
- def MOVLPSrr : SSI<0x10, MRMSrcReg,
- (outs VR128:$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:$src)>;
-
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// SSE2 Instructions
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
+
+// Move Instructions. Register-to-register movsd is not used for FR64
+// register copies because it's a partial register update; FsMOVAPDrr is
+// used instead. Register-to-register movsd is not modeled as an INSERT_SUBREG
+// because INSERT_SUBREG requires that the insert be implementable in terms of
+// a copy, and just mentioned, we don't use movsd for copies.
+let Constraints = "$src1 = $dst" in
+def MOVSDrr : SDI<0x10, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, FR64:$src2),
+ "movsd\t{$src2, $dst|$dst, $src2}",
+ [(set (v2f64 VR128:$dst),
+ (movl VR128:$src1, (scalar_to_vector FR64:$src2)))]>;
+
+// Extract the low 64-bit value from one vector and insert it into another.
+let AddedComplexity = 15 in
+def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
+ (MOVSDrr (v2f64 VR128:$src1),
+ (EXTRACT_SUBREG (v2f64 VR128:$src2), x86_subreg_sd))>;
-// Move Instructions
-let neverHasSideEffects = 1 in
-def MOVSDrr : SDI<0x10, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}", []>;
-let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+// Implicitly promote a 64-bit scalar to a vector.
+def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
+ (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, x86_subreg_sd)>;
+
+// Loading from memory automatically zeroing upper bits.
+let canFoldAsLoad = 1, isReMaterializable = 1, AddedComplexity = 20 in
def MOVSDrm : SDI<0x10, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
"movsd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (loadf64 addr:$src))]>;
+
+// MOVSDrm zeros the high parts of the register; represent this
+// with SUBREG_TO_REG.
+let AddedComplexity = 20 in {
+def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), x86_subreg_sd)>;
+def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), x86_subreg_sd)>;
+def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), x86_subreg_sd)>;
+def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), x86_subreg_sd)>;
+def : Pat<(v2f64 (X86vzload addr:$src)),
+ (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), x86_subreg_sd)>;
+}
+
+// Store scalar value to memory.
def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
"movsd\t{$src, $dst|$dst, $src}",
[(store FR64:$src, addr:$dst)]>;
+// Extract and store.
+def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
+ addr:$dst),
+ (MOVSDmr addr:$dst,
+ (EXTRACT_SUBREG (v2f64 VR128:$src), x86_subreg_sd))>;
+
// Conversion instructions
def CVTTSD2SIrr : SDI<0x2C, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
"cvttsd2si\t{$src, $dst|$dst, $src}",
def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
"cvtsd2ss\t{$src, $dst|$dst, $src}",
[(set FR32:$dst, (fround FR64:$src))]>;
-def CVTSD2SSrm : SDI<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
+def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
"cvtsd2ss\t{$src, $dst|$dst, $src}",
- [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
+ [(set FR32:$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:$src))]>;
"cvtsi2sd\t{$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, $src}",
def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
"cvtss2sd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
- Requires<[HasSSE2]>;
+ Requires<[HasSSE2, OptForSize]>;
+
+def : Pat<(extloadf32 addr:$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:$src),
[(set VR64:$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
+ [(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
+ [(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
+ [(set VR128:$dst, (int_x86_sse_cvtpi2pd
(load addr:$src)))]>;
// Aliases for intrinsics
// Comparison instructions
let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
- def CMPSDrr : SDIi8<0xC2, MRMSrcReg,
+ def CMPSDrr : SDIi8<0xC2, MRMSrcReg,
(outs FR64:$dst), (ins FR64:$src1, FR64:$src, SSECC:$cc),
"cmp${cc}sd\t{$src, $dst|$dst, $src}", []>;
let mayLoad = 1 in
- def CMPSDrm : SDIi8<0xC2, MRMSrcMem,
+ def CMPSDrm : SDIi8<0xC2, MRMSrcMem,
(outs FR64:$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:$src2), (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86cmp FR64:$src1, FR64:$src2))]>;
def UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs), (ins FR64:$src1, f64mem:$src2),
"ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86cmp FR64:$src1, (loadf64 addr:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86cmp FR64:$src1, (loadf64 addr:$src2)))]>;
} // Defs = [EFLAGS]
// Aliases to match intrinsics which expect XMM operand(s).
let Constraints = "$src1 = $dst" in {
- def Int_CMPSDrr : SDIi8<0xC2, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, VR128:$src, SSECC:$cc),
+ def Int_CMPSDrr : SDIi8<0xC2, MRMSrcReg,
+ (outs VR128:$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:$cc),
+ def Int_CMPSDrm : SDIi8<0xC2, MRMSrcMem,
+ (outs VR128:$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:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86ucomi (v2f64 VR128:$src1),
+ VR128:$src2))]>;
def Int_UCOMISDrm: PDI<0x2E, MRMSrcMem, (outs),(ins VR128:$src1, f128mem:$src2),
"ucomisd\t{$src2, $src1|$src1, $src2}",
- [(X86ucomi (v2f64 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86ucomi (v2f64 VR128:$src1),
+ (load addr:$src2)))]>;
def Int_COMISDrr: PDI<0x2F, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"comisd\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v2f64 VR128:$src1), (v2f64 VR128:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86comi (v2f64 VR128:$src1),
+ VR128:$src2))]>;
def Int_COMISDrm: PDI<0x2F, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"comisd\t{$src2, $src1|$src1, $src2}",
- [(X86comi (v2f64 VR128:$src1), (load addr:$src2)),
- (implicit EFLAGS)]>;
+ [(set EFLAGS, (X86comi (v2f64 VR128:$src1),
+ (load addr:$src2)))]>;
} // Defs = [EFLAGS]
-// Aliases of packed SSE2 instructions for scalar use. These all have names that
-// start with 'Fs'.
+// Aliases of packed SSE2 instructions for scalar use. These all have names
+// that start with 'Fs'.
// Alias instructions that map fld0 to pxor for sse.
-let isReMaterializable = 1, isAsCheapAsAMove = 1 in
-def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins),
- "pxor\t$dst, $dst", [(set FR64:$dst, fpimm0)]>,
+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
// Alias instruction to load FR64 from f128mem using movapd. Upper bits are
// disregarded.
-let canFoldAsLoad = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
"movapd\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
(ins FR64:$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:$src2),
(ins FR64:$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:$src2),
defm MIN : sse2_fp_binop_rm<0x5D, "min", X86fmin,
int_x86_sse2_min_sd, int_x86_sse2_min_pd>;
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// SSE packed FP Instructions
// Move Instructions
let neverHasSideEffects = 1 in
def MOVAPDrr : PDI<0x28, MRMSrcReg, (outs VR128:$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:$src))]>;
def MOVLPDrm : PDI<0x12, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
"movlpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
+ [(set VR128:$dst,
(v2f64 (movlp VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))]>;
def MOVHPDrm : PDI<0x16, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
"movhpd\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (v2f64 (movhp VR128:$src1,
+ [(set VR128:$dst,
+ (v2f64 (movlhps VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))]>;
} // AddedComplexity
} // Constraints = "$src1 = $dst"
[(set VR128:$dst, (int_x86_sse2_cvtps2dq
(memop addr:$src)))]>;
// SSE2 packed instructions with XS prefix
+def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
+def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
+
def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))]>,
+ [(set VR128:$dst,
+ (int_x86_sse2_cvttps2dq VR128:$src))]>,
XS, Requires<[HasSSE2]>;
def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttps2dq\t{$src, $dst|$dst, $src}",
(memop addr:$src)))]>;
// SSE2 instructions without OpSize prefix
+def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
+def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
+ "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
+
def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
(load addr:$src)))]>,
TB, Requires<[HasSSE2]>;
+def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
+def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
+ "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
+
+
def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
[(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
VR128:$src2))]>;
def Int_CVTSD2SSrm: SDI<0x5A, MRMSrcMem,
- (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
+ (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
"cvtsd2ss\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
(load addr:$src2)))]>;
def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
[(set FR64:$dst, (OpNode (load addr:$src)))]>;
-
+
// Vector operation, reg.
def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
}
let Constraints = "$src1 = $dst" in {
- def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
+ def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
(outs VR128:$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:$cc))]>;
- def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
+ def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
(outs VR128:$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:$src1,
// Shuffle and unpack instructions
let Constraints = "$src1 = $dst" in {
- def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
+ def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
(outs VR128:$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:$src2)))]>;
- def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
+ def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1,
f128mem:$src2, i8imm:$src3),
"shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
VR128:$src1, (memopv2f64 addr:$src2))))]>;
let AddedComplexity = 10 in {
- def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
+ def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
(outs VR128:$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,
+ 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,
+ 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,
+ def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
"unpcklpd\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
} // Constraints = "$src1 = $dst"
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// SSE integer instructions
// Move Instructions
multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
bit Commutable = 0> {
- def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ 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:$dst), (ins VR128:$src1, i128mem:$src2),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1,
- (bitconvert (memopv2i64 addr:$src2))))]>;
+ (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), (ins VR128:$src1, VR128:$src2),
+ 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))]>;
- def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1,
- (bitconvert (memopv2i64 addr:$src2))))]>;
- def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
+ (bitconvert (memopv2i64 addr:$src2))))]>;
+ 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:$dst), (ins VR128:$src1, VR128:$src2),
+ def rr : PDI<opc, MRMSrcReg, (outs VR128:$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:$dst), (ins VR128:$src1, i128mem:$src2),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (OpVT (OpNode VR128:$src1,
- (bitconvert (memopv2i64 addr:$src2)))))]>;
+ (bitconvert (memopv2i64 addr:$src2)))))]>;
}
/// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
///
multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
bit Commutable = 0> {
- def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]> {
let isCommutable = Commutable;
}
- def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
- [(set VR128:$dst, (OpNode VR128:$src1,(memopv2i64 addr:$src2)))]>;
+ [(set VR128:$dst, (OpNode VR128:$src1,
+ (memopv2i64 addr:$src2)))]>;
}
} // Constraints = "$src1 = $dst"
let Predicates = [HasSSE2] in {
def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
- (v2i64 (PSLLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
+ (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
- (v2i64 (PSRLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
+ (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
(v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
(v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
- (v2f64 (PSRLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
+ (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
// Shift up / down and insert zero's.
def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
- (v2i64 (PSLLDQri VR128:$src, (PSxLDQ_imm imm:$amt)))>;
+ (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
- (v2i64 (PSRLDQri VR128:$src, (PSxLDQ_imm imm:$amt)))>;
+ (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
}
// Logical
defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
// Shuffle and unpack instructions
+let AddedComplexity = 5 in {
def PSHUFDri : PDIi8<0x70, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
"pshufd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
(outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
"pshufd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (v4i32 (pshufd:$src2
- (bc_v4i32(memopv2i64 addr:$src1)),
- (undef))))]>;
+ (bc_v4i32 (memopv2i64 addr:$src1)),
+ (undef))))]>;
+}
// SSE2 with ImmT == Imm8 and XS prefix.
def PSHUFHWri : Ii8<0x70, MRMSrcReg,
(outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
"pshufhw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (v8i16 (pshufhw:$src2
- (bc_v8i16 (memopv2i64 addr:$src1)),
- (undef))))]>,
+ (bc_v8i16 (memopv2i64 addr:$src1)),
+ (undef))))]>,
XS, Requires<[HasSSE2]>;
// SSE2 with ImmT == Imm8 and XD prefix.
let Constraints = "$src1 = $dst" in {
- def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
+ def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
(outs VR128:$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,
+ 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,
+ 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,
+ 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,
+ 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,
+ 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))))]>;
- def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
+ def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpcklqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
- def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
+ def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"punpcklqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v2i64 (unpckl VR128:$src1,
(memopv2i64 addr:$src2))))]>;
-
- def PUNPCKHBWrr : PDI<0x68, MRMSrcReg,
+
+ 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,
+ def PUNPCKHBWrm : PDI<0x68, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"punpckhbw\t{$src2, $dst|$dst, $src2}",
- [(set VR128:$dst,
- (unpckh VR128:$src1,
+ [(set VR128:$dst,
+ (unpckh VR128:$src1,
(bc_v16i8 (memopv2i64 addr:$src2))))]>;
- def PUNPCKHWDrr : PDI<0x69, MRMSrcReg,
+ 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,
+ 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,
+ 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,
+ 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))))]>;
- def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
+ def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
"punpckhqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
- def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
+ def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
"punpckhqdq\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(outs VR128:$dst), (ins VR128:$src1,
i16mem:$src2, i32i8imm:$src3),
"pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR128:$dst,
+ [(set VR128:$dst,
(X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
imm:$src3))]>;
}
[(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
// Non-temporal stores
-def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
+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:$src)]>;
+def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
+ "movntdq\t{$src, $dst|$dst, $src}",
+ [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
+def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
+ "movnti\t{$src, $dst|$dst, $src}",
+ [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
+ TB, Requires<[HasSSE2]>;
+
+let AddedComplexity = 400 in { // Prefer non-temporal versions
+def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntpd\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
+ [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
+
def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntdq\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
-def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
- "movnti\t{$src, $dst|$dst, $src}",
- [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
- TB, Requires<[HasSSE2]>;
+ [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
+}
// Flush cache
def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
TB, Requires<[HasSSE2]>;
// Load, store, and memory fence
-def LFENCE : I<0xAE, MRM5r, (outs), (ins),
+def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
"lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
-def MFENCE : I<0xAE, MRM6r, (outs), (ins),
+def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
"mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
//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 in
- def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins),
- "pcmpeqd\t$dst, $dst",
+let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
+ isCodeGenOnly = 1 in
+ // FIXME: Change encoding to pseudo.
+ def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$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:$src), x86_subreg_sd))>;
+
def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$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:$src),
(MOVZPQILo2PQIrm addr:$src)>;
}
-//===----------------------------------------------------------------------===//
+// Instructions for the disassembler
+// xr = XMM register
+// xm = mem64
+
+def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
+ "movq\t{$src, $dst|$dst, $src}", []>, XS;
+
+//===---------------------------------------------------------------------===//
// SSE3 Instructions
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// Move Instructions
def MOVSHDUPrr : S3SI<0x16, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
}
// Thread synchronization
-def MONITOR : I<0x01, MRM1r, (outs), (ins), "monitor",
+def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
[(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
-def MWAIT : I<0x01, MRM1r, (outs), (ins), "mwait",
+def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
[(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
// vector_shuffle v1, <undef> <1, 1, 3, 3>
def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
(MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// SSSE3 Instructions
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
/// SS3I_unop_rm_int_8 - Simple SSSE3 unary operator whose type is v*i8.
multiclass SS3I_unop_rm_int_8<bits<8> opc, string OpcodeStr,
let Constraints = "$src1 = $dst" in {
def PALIGNR64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
- (ins VR64:$src1, VR64:$src2, i16imm:$src3),
+ (ins VR64:$src1, VR64:$src2, i8imm:$src3),
"palignr\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR64:$dst,
- (int_x86_ssse3_palign_r
- VR64:$src1, VR64:$src2,
- imm:$src3))]>;
+ []>;
def PALIGNR64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
- (ins VR64:$src1, i64mem:$src2, i16imm:$src3),
+ (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
"palignr\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR64:$dst,
- (int_x86_ssse3_palign_r
- VR64:$src1,
- (bitconvert (memopv2i32 addr:$src2)),
- imm:$src3))]>;
+ []>;
def PALIGNR128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i32imm:$src3),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
"palignr\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR128:$dst,
- (int_x86_ssse3_palign_r_128
- VR128:$src1, VR128:$src2,
- imm:$src3))]>, OpSize;
+ []>, OpSize;
def PALIGNR128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i32imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
"palignr\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- [(set VR128:$dst,
- (int_x86_ssse3_palign_r_128
- VR128:$src1,
- (bitconvert (memopv4i32 addr:$src2)),
- imm:$src3))]>, OpSize;
+ []>, OpSize;
+}
+
+// palignr patterns.
+def : Pat<(int_x86_ssse3_palign_r VR64:$src1, VR64:$src2, (i8 imm:$src3)),
+ (PALIGNR64rr VR64:$src1, VR64:$src2, (BYTE_imm imm:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(int_x86_ssse3_palign_r VR64:$src1,
+ (memop64 addr:$src2),
+ (i8 imm:$src3)),
+ (PALIGNR64rm VR64:$src1, addr:$src2, (BYTE_imm imm:$src3))>,
+ Requires<[HasSSSE3]>;
+
+def : Pat<(int_x86_ssse3_palign_r_128 VR128:$src1, VR128:$src2, (i8 imm:$src3)),
+ (PALIGNR128rr VR128:$src1, VR128:$src2, (BYTE_imm imm:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(int_x86_ssse3_palign_r_128 VR128:$src1,
+ (memopv2i64 addr:$src2),
+ (i8 imm:$src3)),
+ (PALIGNR128rm VR128:$src1, addr:$src2, (BYTE_imm imm:$src3))>,
+ Requires<[HasSSSE3]>;
+
+let AddedComplexity = 5 in {
+def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
+ (PALIGNR128rr VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_palign_imm VR128:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
+ (PALIGNR128rr VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_palign_imm VR128:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
+ (PALIGNR128rr VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_palign_imm VR128:$src3))>,
+ Requires<[HasSSSE3]>;
+def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
+ (PALIGNR128rr VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_palign_imm VR128:$src3))>,
+ Requires<[HasSSSE3]>;
}
def : Pat<(X86pshufb VR128:$src, VR128:$mask),
def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
(PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
// Non-Instruction Patterns
-//===----------------------------------------------------------------------===//
+//===---------------------------------------------------------------------===//
-// extload f32 -> f64. This matches load+fextend because we have a hack in
-// the isel (PreprocessForFPConvert) that can introduce loads after dag combine.
+// extload f32 -> f64. This matches load+fextend because we have a hack in
+// the isel (PreprocessForFPConvert) that can introduce loads after dag
+// combine.
// Since these loads aren't folded into the fextend, we have to match it
// explicitly here.
let Predicates = [HasSSE2] in
let AddedComplexity = 15 in {
// Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
- (MOVLSD2PDrr (V_SET0), FR64:$src)>, Requires<[HasSSE2]>;
+ (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
- (MOVLSS2PSrr (V_SET0), FR32:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
- (MOVLPSrr (V_SET0), VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4f32 (V_SET0)),
+ (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), x86_subreg_ss)))>;
def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
- (MOVLPSrr (V_SET0), VR128:$src)>, Requires<[HasSSE1]>;
+ (MOVSSrr (v4i32 (V_SET0)),
+ (EXTRACT_SUBREG (v4i32 VR128:$src), x86_subreg_ss))>;
}
// Splat v2f64 / v2i64
// Special unary SHUFPSrri case.
def : Pat<(v4f32 (pshufd:$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:$src2))>,
Requires<[HasSSE2]>;
// Special unary SHUFPDrri case.
def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFPDrri VR128:$src1, VR128:$src1,
(SHUFFLE_get_shuf_imm VR128:$src3))>,
Requires<[HasSSE2]>;
// Special unary SHUFPDrri case.
def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
- (SHUFPDrri VR128:$src1, VR128:$src1,
+ (SHUFPDrri VR128:$src1, VR128:$src1,
(SHUFFLE_get_shuf_imm VR128:$src3))>,
Requires<[HasSSE2]>;
// Unary v4f32 shuffle with PSHUF* in order to fold a load.
// Special binary v4i32 shuffle cases with SHUFPS.
def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
- (SHUFPSrri VR128:$src1, VR128:$src2,
+ (SHUFPSrri VR128:$src1, VR128:$src2,
(SHUFFLE_get_shuf_imm VR128:$src3))>,
Requires<[HasSSE2]>;
def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
- (SHUFPSrmi VR128:$src1, addr:$src2,
+ (SHUFPSrmi VR128:$src1, addr:$src2,
(SHUFFLE_get_shuf_imm VR128:$src3))>,
Requires<[HasSSE2]>;
// Special binary v2i64 shuffle cases using SHUFPDrri.
def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
- (SHUFPDrri VR128:$src1, VR128:$src2,
+ (SHUFPDrri VR128:$src1, VR128:$src2,
(SHUFFLE_get_shuf_imm VR128:$src3))>,
Requires<[HasSSE2]>;
}
let AddedComplexity = 10 in {
def : Pat<(v4f32 (unpckl_undef VR128:$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:$src)>;
}
let AddedComplexity = 20 in {
// vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
-def : Pat<(v4i32 (movhp VR128:$src1, VR128:$src2)),
+def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
(MOVLHPSrr VR128:$src1, VR128:$src2)>;
// vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
let AddedComplexity = 20 in {
// vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
-// vector_shuffle v1, (load v2) <0, 1, 4, 5> using MOVHPS
def : Pat<(v4f32 (movlp VR128:$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:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v4f32 (movhp VR128:$src1, (load addr:$src2))),
- (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
-def : Pat<(v2f64 (movhp VR128:$src1, (load addr:$src2))),
- (MOVHPDrm VR128:$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:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(v4i32 (movhp VR128:$src1, (load addr:$src2))),
- (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
-def : Pat<(v2i64 (movhp VR128:$src1, (load addr:$src2))),
- (MOVHPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
}
// (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
-// (store (vector_shuffle (load addr), v2, <0, 1, 4, 5>), addr) using MOVHPS
def : Pat<(store (v4f32 (movlp (load addr:$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:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(store (v4f32 (movhp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVHPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>;
-def : Pat<(store (v2f64 (movhp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVHPDmr addr:$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:$src2)>, Requires<[HasSSE2]>;
-def : Pat<(store (v4i32 (movhp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
- addr:$src1),
- (MOVHPSmr addr:$src1, VR128:$src2)>, Requires<[HasSSE1]>;
-def : Pat<(store (v2i64 (movhp (load addr:$src1), VR128:$src2)), addr:$src1),
- (MOVHPDmr addr:$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:$src2), x86_subreg_ss))>;
def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr (v2i64 VR128:$src1),
+ (EXTRACT_SUBREG (v2i64 VR128:$src2), x86_subreg_sd))>;
-// vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
+// vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, x86_subreg_sd))>,
+ Requires<[HasSSE2]>;
def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
- (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
+ (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, x86_subreg_sd))>,
+ Requires<[HasSSE2]>;
}
+// vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
+// fall back to this for SSE1)
+def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
+ (SHUFPSrri VR128:$src2, VR128:$src1,
+ (SHUFFLE_get_shuf_imm VR128:$src3))>;
+
// Set lowest element and zero upper elements.
let AddedComplexity = 15 in
def : Pat<(v2f64 (movl immAllZerosV_bc, VR128:$src)),
// 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
//===----------------------------------------------------------------------===//
Intrinsic V2F64Int> {
// Intrinsic operation, reg.
// Vector intrinsic operation, reg
- def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
+ def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
(outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
OpSize;
// Vector intrinsic operation, mem
- def PSm_Int : SS4AIi8<opcps, MRMSrcMem,
+ def PSm_Int : Ii8<opcps, MRMSrcMem,
(outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(V4F32Int (memopv4f32 addr:$src1),imm:$src2))]>,
- OpSize;
+ TA, OpSize,
+ Requires<[HasSSE41]>;
// Vector intrinsic operation, reg
def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
Intrinsic F64Int> {
// Intrinsic operation, reg.
def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
- (outs VR128:$dst),
+ (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
OpSize;
// Intrinsic operation, mem.
- def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
- (outs VR128:$dst),
+ def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
+ (outs VR128:$dst),
(ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
OpSize;
// Intrinsic operation, reg.
def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
- (outs VR128:$dst),
+ (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
OpSize;
// Intrinsic operation, mem.
def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
- (outs VR128:$dst),
+ (outs VR128:$dst),
(ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
OpSize;
}
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst,
- (OpNode VR128:$src1, (memop addr:$src2)))]>, OpSize;
+ (OpVT (OpNode VR128:$src1, (memop addr:$src2))))]>, OpSize;
def rm_int : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
Intrinsic IntId128, bit Commutable = 0> {
def rri : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(IntId128 VR128:$src1, VR128:$src2, imm:$src3))]>,
OpSize {
let isCommutable = Commutable;
multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
[(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
OpSize;
multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
(ins VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
OpSize;
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[]>, OpSize;
// FIXME:
multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[]>, OpSize;
// FIXME:
multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
(ins VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32:$dst,
(extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(store (extractelt (v4i32 VR128:$src1), imm:$src2),
addr:$dst)]>, OpSize;
multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
(ins VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32:$dst,
(extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
OpSize;
- def mr : SS4AIi8<opc, MRMDestMem, (outs),
+ def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
addr:$dst)]>, OpSize;
multiclass SS41I_insert8<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
imm:$src3))]>, OpSize;
}
multiclass SS41I_insert32<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
- !strconcat(OpcodeStr,
+ !strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
+ [(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
imm:$src3)))]>, OpSize;
}
defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
+// insertps has a few different modes, there's the first two here below which
+// are optimized inserts that won't zero arbitrary elements in the destination
+// vector. The next one matches the intrinsic and could zero arbitrary elements
+// in the target vector.
let Constraints = "$src1 = $dst" in {
multiclass SS41I_insertf32<bits<8> opc, string OpcodeStr> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, FR32:$src2, i32i8imm:$src3),
- !strconcat(OpcodeStr,
+ (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
+ !strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
- (X86insrtps VR128:$src1, FR32:$src2, imm:$src3))]>, OpSize;
+ [(set VR128:$dst,
+ (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
+ OpSize;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
- [(set VR128:$dst,
- (X86insrtps VR128:$src1, (loadf32 addr:$src2),
+ [(set VR128:$dst,
+ (X86insrtps VR128:$src1,
+ (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
imm:$src3))]>, OpSize;
}
}
defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
+def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
+ (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>;
+
+// ptest instruction we'll lower to this in X86ISelLowering primarily from
+// the intel intrinsic that corresponds to this.
let Defs = [EFLAGS] in {
def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
- "ptest \t{$src2, $src1|$src1, $src2}", []>, OpSize;
+ "ptest \t{$src2, $src1|$src1, $src2}",
+ [(X86ptest VR128:$src1, VR128:$src2),
+ (implicit EFLAGS)]>, OpSize;
def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
- "ptest \t{$src2, $src1|$src1, $src2}", []>, OpSize;
+ "ptest \t{$src2, $src1|$src1, $src2}",
+ [(X86ptest VR128:$src1, (load addr:$src2)),
+ (implicit EFLAGS)]>, OpSize;
}
def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movntdqa\t{$src, $dst|$dst, $src}",
- [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>;
+ [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
+ OpSize;
+
+
+//===----------------------------------------------------------------------===//
+// SSE4.2 Instructions
+//===----------------------------------------------------------------------===//
/// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
let Constraints = "$src1 = $dst" in {
(PCMPGTQrr VR128:$src1, VR128:$src2)>;
def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
(PCMPGTQrm VR128:$src1, addr:$src2)>;
+
+defm AESIMC : SS42I_binop_rm_int<0xDB, "aesimc",
+ int_x86_sse42_aesimc>;
+defm AESENC : SS42I_binop_rm_int<0xDC, "aesenc",
+ int_x86_sse42_aesenc>;
+defm AESENCLAST : SS42I_binop_rm_int<0xDD, "aesenclast",
+ int_x86_sse42_aesenclast>;
+defm AESDEC : SS42I_binop_rm_int<0xDE, "aesdec",
+ int_x86_sse42_aesdec>;
+defm AESDECLAST : SS42I_binop_rm_int<0xDF, "aesdeclast",
+ int_x86_sse42_aesdeclast>;
+
+def : Pat<(v2i64 (X86aesimc VR128:$src1, VR128:$src2)),
+ (AESIMCrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (X86aesimc VR128:$src1, (memop addr:$src2))),
+ (AESIMCrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (X86aesenc VR128:$src1, VR128:$src2)),
+ (AESENCrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (X86aesenc VR128:$src1, (memop addr:$src2))),
+ (AESENCrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (X86aesenclast VR128:$src1, VR128:$src2)),
+ (AESENCLASTrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (X86aesenclast VR128:$src1, (memop addr:$src2))),
+ (AESENCLASTrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (X86aesdec VR128:$src1, VR128:$src2)),
+ (AESDECrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (X86aesdec VR128:$src1, (memop addr:$src2))),
+ (AESDECrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2i64 (X86aesdeclast VR128:$src1, VR128:$src2)),
+ (AESDECLASTrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2i64 (X86aesdeclast VR128:$src1, (memop addr:$src2))),
+ (AESDECLASTrm VR128:$src1, addr:$src2)>;
+
+def AESKEYGENASSIST128rr : SS42AI<0xDF, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "aeskeygenassist\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+def AESKEYGENASSIST128rm : SS42AI<0xDF, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "aeskeygenassist\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+
+// crc intrinsic instruction
+// This set of instructions are only rm, the only difference is the size
+// of r and m.
+let Constraints = "$src1 = $dst" in {
+ def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
+ (ins GR32:$src1, i8mem:$src2),
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_8 GR32:$src1,
+ (load addr:$src2)))]>;
+ def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
+ (ins GR32:$src1, GR8:$src2),
+ "crc32{b} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
+ def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
+ (ins GR32:$src1, i16mem:$src2),
+ "crc32{w} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_16 GR32:$src1,
+ (load addr:$src2)))]>,
+ OpSize;
+ def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
+ (ins GR32:$src1, GR16:$src2),
+ "crc32{w} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
+ OpSize;
+ def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
+ (ins GR32:$src1, i32mem:$src2),
+ "crc32{l} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_32 GR32:$src1,
+ (load addr:$src2)))]>;
+ def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
+ (ins GR32:$src1, GR32:$src2),
+ "crc32{l} \t{$src2, $src1|$src1, $src2}",
+ [(set GR32:$dst,
+ (int_x86_sse42_crc32_32 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:$src2),
+ "crc32{q} \t{$src2, $src1|$src1, $src2}",
+ [(set GR64:$dst,
+ (int_x86_sse42_crc64_64 GR64:$src1,
+ (load addr:$src2)))]>,
+ REX_W;
+ def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
+ (ins GR64:$src1, GR64:$src2),
+ "crc32{q} \t{$src2, $src1|$src1, $src2}",
+ [(set GR64:$dst,
+ (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
+ REX_W;
+}
+
+// String/text processing instructions.
+let Defs = [EFLAGS], usesCustomInserter = 1 in {
+def PCMPISTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "#PCMPISTRM128rr PSEUDO!",
+ [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
+ imm:$src3))]>, OpSize;
+def PCMPISTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "#PCMPISTRM128rm PSEUDO!",
+ [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, (load addr:$src2),
+ imm:$src3))]>, OpSize;
+}
+
+let Defs = [XMM0, EFLAGS] in {
+def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
+}
+
+let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
+def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ "#PCMPESTRM128rr PSEUDO!",
+ [(set VR128:$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:$src5))]>,
+ OpSize;
+}
+
+let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
+def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
+def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
+}
+
+let Defs = [ECX, EFLAGS] in {
+ multiclass SS42AI_pcmpistri<Intrinsic IntId128> {
+ def rr : SS42AI<0x63, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ "pcmpistri\t{$src3, $src2, $src1|$src1, $src2, $src3}",
+ [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
+ (implicit EFLAGS)]>, OpSize;
+ def rm : SS42AI<0x63, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ "pcmpistri\t{$src3, $src2, $src1|$src1, $src2, $src3}",
+ [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
+ (implicit EFLAGS)]>, OpSize;
+ }
+}
+
+defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
+defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
+defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
+defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
+defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
+defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
+
+let Defs = [ECX, EFLAGS] in {
+let Uses = [EAX, EDX] in {
+ multiclass SS42AI_pcmpestri<Intrinsic IntId128> {
+ def rr : SS42AI<0x61, MRMSrcReg, (outs),
+ (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ "pcmpestri\t{$src5, $src3, $src1|$src1, $src3, $src5}",
+ [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
+ (implicit EFLAGS)]>, OpSize;
+ def rm : SS42AI<0x61, MRMSrcMem, (outs),
+ (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ "pcmpestri\t{$src5, $src3, $src1|$src1, $src3, $src5}",
+ [(set ECX,
+ (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
+ (implicit EFLAGS)]>, OpSize;
+ }
+}
+}
+
+defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
+defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
+defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
+defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
+defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
+defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;