Index: lib/Target/X86/X86InstrAVX512.td =================================================================== --- lib/Target/X86/X86InstrAVX512.td +++ lib/Target/X86/X86InstrAVX512.td @@ -2723,22 +2723,22 @@ } multiclass avx512_store opc, string OpcodeStr, X86VectorVTInfo _, - PatFrag st_frag, PatFrag mstore> { + PatFrag st_frag, PatFrag mstore, string Name> { let hasSideEffects = 0 in { def rr_REV : AVX512PI, EVEX; + [], _.ExeDomain>, EVEX, FoldGenData; def rrk_REV : AVX512PI, EVEX, EVEX_K; + [], _.ExeDomain>, EVEX, EVEX_K, FoldGenData; def rrkz_REV : AVX512PI, EVEX, EVEX_KZ; + [], _.ExeDomain>, EVEX, EVEX_KZ, FoldGenData; } def mr : AVX512PI opc, string OpcodeStr, - AVX512VLVectorVTInfo _, Predicate prd> { + AVX512VLVectorVTInfo _, Predicate prd, + string Name> { let Predicates = [prd] in defm Z : avx512_store, EVEX_V512; + masked_store_unaligned, Name#Z>, EVEX_V512; let Predicates = [prd, HasVLX] in { defm Z256 : avx512_store, EVEX_V256; + masked_store_unaligned, Name#Z256>, EVEX_V256; defm Z128 : avx512_store, EVEX_V128; + masked_store_unaligned, Name#Z128>, EVEX_V128; } } multiclass avx512_alignedstore_vl opc, string OpcodeStr, - AVX512VLVectorVTInfo _, Predicate prd> { + AVX512VLVectorVTInfo _, Predicate prd, + string Name> { let Predicates = [prd] in defm Z : avx512_store, EVEX_V512; + masked_store_aligned512, Name#Z>, EVEX_V512; let Predicates = [prd, HasVLX] in { defm Z256 : avx512_store, EVEX_V256; + masked_store_aligned256, Name#Z256>, EVEX_V256; defm Z128 : avx512_store, EVEX_V128; + masked_store_aligned128, Name#Z128>, EVEX_V128; } } defm VMOVAPS : avx512_alignedload_vl<0x28, "vmovaps", avx512vl_f32_info, HasAVX512>, avx512_alignedstore_vl<0x29, "vmovaps", avx512vl_f32_info, - HasAVX512>, PS, EVEX_CD8<32, CD8VF>; + HasAVX512, "VMOVAPS">, + PS, EVEX_CD8<32, CD8VF>; defm VMOVAPD : avx512_alignedload_vl<0x28, "vmovapd", avx512vl_f64_info, HasAVX512>, avx512_alignedstore_vl<0x29, "vmovapd", avx512vl_f64_info, - HasAVX512>, PD, VEX_W, EVEX_CD8<64, CD8VF>; + HasAVX512, "VMOVAPD">, + PD, VEX_W, EVEX_CD8<64, CD8VF>; defm VMOVUPS : avx512_load_vl<0x10, "vmovups", avx512vl_f32_info, HasAVX512, null_frag>, - avx512_store_vl<0x11, "vmovups", avx512vl_f32_info, HasAVX512>, + avx512_store_vl<0x11, "vmovups", avx512vl_f32_info, HasAVX512, + "VMOVUPS">, PS, EVEX_CD8<32, CD8VF>; defm VMOVUPD : avx512_load_vl<0x10, "vmovupd", avx512vl_f64_info, HasAVX512, null_frag>, - avx512_store_vl<0x11, "vmovupd", avx512vl_f64_info, HasAVX512>, + avx512_store_vl<0x11, "vmovupd", avx512vl_f64_info, HasAVX512, + "VMOVUPD">, PD, VEX_W, EVEX_CD8<64, CD8VF>; defm VMOVDQA32 : avx512_alignedload_vl<0x6F, "vmovdqa32", avx512vl_i32_info, HasAVX512>, avx512_alignedstore_vl<0x7F, "vmovdqa32", avx512vl_i32_info, - HasAVX512>, PD, EVEX_CD8<32, CD8VF>; + HasAVX512, "VMOVDQA32">, + PD, EVEX_CD8<32, CD8VF>; defm VMOVDQA64 : avx512_alignedload_vl<0x6F, "vmovdqa64", avx512vl_i64_info, HasAVX512>, avx512_alignedstore_vl<0x7F, "vmovdqa64", avx512vl_i64_info, - HasAVX512>, PD, VEX_W, EVEX_CD8<64, CD8VF>; + HasAVX512, "VMOVDQA64">, + PD, VEX_W, EVEX_CD8<64, CD8VF>; defm VMOVDQU8 : avx512_load_vl<0x6F, "vmovdqu8", avx512vl_i8_info, HasBWI>, - avx512_store_vl<0x7F, "vmovdqu8", avx512vl_i8_info, - HasBWI>, XD, EVEX_CD8<8, CD8VF>; + avx512_store_vl<0x7F, "vmovdqu8", avx512vl_i8_info, + HasBWI, "VMOVDQU8">, + XD, EVEX_CD8<8, CD8VF>; defm VMOVDQU16 : avx512_load_vl<0x6F, "vmovdqu16", avx512vl_i16_info, HasBWI>, avx512_store_vl<0x7F, "vmovdqu16", avx512vl_i16_info, - HasBWI>, XD, VEX_W, EVEX_CD8<16, CD8VF>; + HasBWI, "VMOVDQU16">, + XD, VEX_W, EVEX_CD8<16, CD8VF>; defm VMOVDQU32 : avx512_load_vl<0x6F, "vmovdqu32", avx512vl_i32_info, HasAVX512, null_frag>, avx512_store_vl<0x7F, "vmovdqu32", avx512vl_i32_info, - HasAVX512>, XS, EVEX_CD8<32, CD8VF>; + HasAVX512, "VMOVDQU32">, + XS, EVEX_CD8<32, CD8VF>; defm VMOVDQU64 : avx512_load_vl<0x6F, "vmovdqu64", avx512vl_i64_info, HasAVX512, null_frag>, avx512_store_vl<0x7F, "vmovdqu64", avx512vl_i64_info, - HasAVX512>, XS, VEX_W, EVEX_CD8<64, CD8VF>; + HasAVX512, "VMOVDQU64">, + XS, VEX_W, EVEX_CD8<64, CD8VF>; // Special instructions to help with spilling when we don't have VLX. We need // to load or store from a ZMM register instead. These are converted in @@ -3384,17 +3396,52 @@ (VMOVSSZmrk addr:$dst, (i1 (COPY_TO_REGCLASS (i32 (INSERT_SUBREG (IMPLICIT_DEF), GR8:$mask, sub_8bit)), VK1WM)), (COPY_TO_REGCLASS VR128X:$src, FR32X))>; -let hasSideEffects = 0 in -defm VMOVSSZrr_REV : AVX512_maskable_in_asm<0x11, MRMDestReg, f32x_info, - (outs VR128X:$dst), (ins VR128X:$src1, FR32X:$src2), - "vmovss.s", "$src2, $src1", "$src1, $src2", []>, - XS, EVEX_4V, VEX_LIG; - -let hasSideEffects = 0 in -defm VMOVSDZrr_REV : AVX512_maskable_in_asm<0x11, MRMDestReg, f64x_info, - (outs VR128X:$dst), (ins VR128X:$src1, FR64X:$src2), - "vmovsd.s", "$src2, $src1", "$src1, $src2", []>, - XD, EVEX_4V, VEX_LIG, VEX_W; +let hasSideEffects = 0 in { + def VMOVSSZrr_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins VR128X:$src1, FR32X:$src2), + "vmovss.s\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [], NoItinerary>, XS, EVEX_4V, VEX_LIG, + FoldGenData<"VMOVSSZrr">; + +let Constraints = "$src0 = $dst" in + def VMOVSSZrrk_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins f32x_info.RC:$src0, f32x_info.KRCWM:$mask, + VR128X:$src1, FR32X:$src2), + "vmovss.s\t{$src2, $src1, $dst {${mask}}|"# + "$dst {${mask}}, $src1, $src2}", + [], NoItinerary>, EVEX_K, XS, EVEX_4V, VEX_LIG, + FoldGenData<"VMOVSSZrrk">; + + def VMOVSSZrrkz_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins f32x_info.KRCWM:$mask, VR128X:$src1, FR32X:$src2), + "vmovss.s\t{$src2, $src1, $dst {${mask}} {z}|"# + "$dst {${mask}} {z}, $src1, $src2}", + [], NoItinerary>, EVEX_KZ, XS, EVEX_4V, VEX_LIG, + FoldGenData<"VMOVSSZrrkz">; + + def VMOVSDZrr_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins VR128X:$src1, FR64X:$src2), + "vmovsd.s\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [], NoItinerary>, XD, EVEX_4V, VEX_LIG, VEX_W, + FoldGenData<"VMOVSDZrr">; + +let Constraints = "$src0 = $dst" in + def VMOVSDZrrk_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins f64x_info.RC:$src0, f64x_info.KRCWM:$mask, + VR128X:$src1, FR64X:$src2), + "vmovsd.s\t{$src2, $src1, $dst {${mask}}|"# + "$dst {${mask}}, $src1, $src2}", + [], NoItinerary>, EVEX_K, XD, EVEX_4V, VEX_LIG, + VEX_W, FoldGenData<"VMOVSDZrrk">; + + def VMOVSDZrrkz_REV: AVX512<0x11, MRMDestReg, (outs VR128X:$dst), + (ins f64x_info.KRCWM:$mask, VR128X:$src1, + FR64X:$src2), + "vmovsd.s\t{$src2, $src1, $dst {${mask}} {z}|"# + "$dst {${mask}} {z}, $src1, $src2}", + [], NoItinerary>, EVEX_KZ, XD, EVEX_4V, VEX_LIG, + VEX_W, FoldGenData<"VMOVSDZrrkz">; +} let Predicates = [HasAVX512] in { let AddedComplexity = 15 in { @@ -8786,7 +8833,7 @@ def rr_REV : AVX512Ii8<0x15, MRMDestReg, (outs GR32orGR64:$dst), (ins _.RC:$src1, u8imm:$src2), OpcodeStr#".s\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, - EVEX, TAPD; + EVEX, TAPD, FoldGenData; defm NAME : avx512_extract_elt_bw_m<0x15, OpcodeStr, X86pextrw, _>, TAPD; } Index: lib/Target/X86/X86InstrArithmetic.td =================================================================== --- lib/Target/X86/X86InstrArithmetic.td +++ lib/Target/X86/X86InstrArithmetic.td @@ -964,10 +964,10 @@ } // isConvertibleToThreeAddress } // isCommutable - def NAME#8rr_REV : BinOpRR_Rev; - def NAME#16rr_REV : BinOpRR_Rev; - def NAME#32rr_REV : BinOpRR_Rev; - def NAME#64rr_REV : BinOpRR_Rev; + def NAME#8rr_REV : BinOpRR_Rev, FoldGenData; + def NAME#16rr_REV : BinOpRR_Rev, FoldGenData; + def NAME#32rr_REV : BinOpRR_Rev, FoldGenData; + def NAME#64rr_REV : BinOpRR_Rev, FoldGenData; def NAME#8rm : BinOpRM_RF; def NAME#16rm : BinOpRM_RF; @@ -1047,10 +1047,10 @@ } // isConvertibleToThreeAddress } // isCommutable - def NAME#8rr_REV : BinOpRR_RFF_Rev; - def NAME#16rr_REV : BinOpRR_RFF_Rev; - def NAME#32rr_REV : BinOpRR_RFF_Rev; - def NAME#64rr_REV : BinOpRR_RFF_Rev; + def NAME#8rr_REV : BinOpRR_RFF_Rev, FoldGenData; + def NAME#16rr_REV : BinOpRR_RFF_Rev, FoldGenData; + def NAME#32rr_REV : BinOpRR_RFF_Rev, FoldGenData; + def NAME#64rr_REV : BinOpRR_RFF_Rev, FoldGenData; def NAME#8rm : BinOpRM_RFF; def NAME#16rm : BinOpRM_RFF; @@ -1127,10 +1127,10 @@ } } // isCommutable - def NAME#8rr_REV : BinOpRR_F_Rev; - def NAME#16rr_REV : BinOpRR_F_Rev; - def NAME#32rr_REV : BinOpRR_F_Rev; - def NAME#64rr_REV : BinOpRR_F_Rev; + def NAME#8rr_REV : BinOpRR_F_Rev, FoldGenData; + def NAME#16rr_REV : BinOpRR_F_Rev, FoldGenData; + def NAME#32rr_REV : BinOpRR_F_Rev, FoldGenData; + def NAME#64rr_REV : BinOpRR_F_Rev, FoldGenData; def NAME#8rm : BinOpRM_F; def NAME#16rm : BinOpRM_F; Index: lib/Target/X86/X86InstrFMA.td =================================================================== --- lib/Target/X86/X86InstrFMA.td +++ lib/Target/X86/X86InstrFMA.td @@ -297,7 +297,7 @@ (ins RC:$src1, RC:$src2, RC:$src3), !strconcat(OpcodeStr, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>, - VEX_LIG; + VEX_LIG, FoldGenData; } multiclass fma4s_int opc, string OpcodeStr, Operand memop, @@ -321,6 +321,12 @@ "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), [(set VR128:$dst, (Int VR128:$src1, mem_cpat:$src2, VR128:$src3))]>, VEX_LIG; +let hasSideEffects = 0 in + def rr_Int_REV : FMA4, VEX_LIG, FoldGenData; } // isCodeGenOnly = 1 } @@ -372,12 +378,13 @@ def rr_REV : FMA4; + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), []>, + FoldGenData; def Yrr_REV : FMA4, - VEX_L; + VEX_L, FoldGenData; } // isCodeGenOnly = 1 } Index: lib/Target/X86/X86InstrFormats.td =================================================================== --- lib/Target/X86/X86InstrFormats.td +++ lib/Target/X86/X86InstrFormats.td @@ -225,6 +225,12 @@ class XOP { Encoding OpEnc = EncXOP; } class XOP_4V : XOP { bit hasVEX_4V = 1; } +// Specify the alternative register form instruction to replace the current +// instruction in case it was picked during generation of memory folding tables +class FoldGenData { + string FoldGenRegForm = _RegisterForm; +} + class X86Inst opcod, Format f, ImmType i, dag outs, dag ins, string AsmStr, InstrItinClass itin, @@ -304,6 +310,10 @@ CD8_EltSize, !srl(VectSize, CD8_Form{1-0}))), 0); + // Used in the memory folding generation (TableGen backend) to point to an alternative + // instruction to replace the current one in case it got picked during generation. + string FoldGenRegForm = ?; + // TSFlags layout should be kept in sync with X86BaseInfo.h. let TSFlags{6-0} = FormBits; let TSFlags{8-7} = OpSizeBits; Index: lib/Target/X86/X86InstrInfo.td =================================================================== --- lib/Target/X86/X86InstrInfo.td +++ lib/Target/X86/X86InstrInfo.td @@ -1420,11 +1420,14 @@ // Longer forms that use a ModR/M byte. Needed for disassembler let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { def MOV8ri_alt : Ii8 <0xC6, MRM0r, (outs GR8 :$dst), (ins i8imm :$src), - "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; + "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, + FoldGenData<"MOV8ri">; def MOV16ri_alt : Ii16<0xC7, MRM0r, (outs GR16:$dst), (ins i16imm:$src), - "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize16; + "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize16, + FoldGenData<"MOV16ri">; def MOV32ri_alt : Ii32<0xC7, MRM0r, (outs GR32:$dst), (ins i32imm:$src), - "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize32; + "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize32, + FoldGenData<"MOV32ri">; } } // SchedRW @@ -1547,13 +1550,17 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def MOV8rr_REV : I<0x8A, MRMSrcReg, (outs GR8:$dst), (ins GR8:$src), - "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; + "mov{b}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, + FoldGenData<"MOV8rr">; def MOV16rr_REV : I<0x8B, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src), - "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize16; + "mov{w}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize16, + FoldGenData<"MOV16rr">; def MOV32rr_REV : I<0x8B, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src), - "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize32; + "mov{l}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, OpSize32, + FoldGenData<"MOV32rr">; def MOV64rr_REV : RI<0x8B, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src), - "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>; + "mov{q}\t{$src, $dst|$dst, $src}", [], IIC_MOV>, + FoldGenData<"MOV64rr">; } let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in { Index: lib/Target/X86/X86InstrMMX.td =================================================================== --- lib/Target/X86/X86InstrMMX.td +++ lib/Target/X86/X86InstrMMX.td @@ -248,7 +248,8 @@ "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (MMX_X86movd2w (x86mmx VR64:$src)))], - IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>; + IIC_MMX_MOV_REG_MM>, Sched<[WriteMove]>, + FoldGenData<"MMX_MOVD64rr">; let isBitcast = 1 in def MMX_MOVD64to64rr : MMXRI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR64:$src), @@ -277,7 +278,7 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { def MMX_MOVQ64rr_REV : MMXI<0x7F, MRMDestReg, (outs VR64:$dst), (ins VR64:$src), "movq\t{$src, $dst|$dst, $src}", [], - IIC_MMX_MOVQ_RR>; + IIC_MMX_MOVQ_RR>, FoldGenData<"MMX_MOVQ64rr">; } } // SchedRW Index: lib/Target/X86/X86InstrSSE.td =================================================================== --- lib/Target/X86/X86InstrSSE.td +++ lib/Target/X86/X86InstrSSE.td @@ -503,7 +503,8 @@ multiclass sse12_move_rr { + string asm_opr, Domain d = GenericDomain, + string Name> { let isCommutable = 1 in def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), @@ -517,15 +518,17 @@ def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), !strconcat(base_opc, asm_opr), - [], IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>; + [], IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>, + FoldGenData; } multiclass sse12_move { + Domain d = GenericDomain, string Name> { // AVX defm V#NAME : sse12_move_rr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}", d, + "V"#Name>, VEX_4V, VEX_LIG, VEX_WIG; def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), @@ -535,7 +538,7 @@ // SSE1 & 2 let Constraints = "$src1 = $dst" in { defm NAME : sse12_move_rr; + "\t{$src2, $dst|$dst, $src2}", d, Name>; } def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), @@ -559,9 +562,9 @@ } defm MOVSS : sse12_move, XS; + SSEPackedSingle, "MOVSS">, XS; defm MOVSD : sse12_move, XD; + SSEPackedDouble, "MOVSD">, XD; let canFoldAsLoad = 1, isReMaterializable = 1 in { defm MOVSS : sse12_move_rm, VEX, VEX_WIG; + IIC_SSE_MOVA_P_RR>, VEX, VEX_WIG, + FoldGenData<"VMOVAPSrr">; def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movapd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>, VEX, VEX_WIG; + IIC_SSE_MOVA_P_RR>, VEX, VEX_WIG, + FoldGenData<"VMOVAPDrr">; def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movups\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>, VEX, VEX_WIG; + IIC_SSE_MOVU_P_RR>, VEX, VEX_WIG, + FoldGenData<"VMOVUPSrr">; def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movupd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>, VEX, VEX_WIG; + IIC_SSE_MOVU_P_RR>, VEX, VEX_WIG, + FoldGenData<"VMOVUPDrr">; def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movaps\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVAPSYrr">; def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movapd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVAPDYrr">; def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movups\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVUPSYrr">; def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movupd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVUPDYrr">; } // Aliases to help the assembler pick two byte VEX encodings by swapping the @@ -934,16 +945,16 @@ SchedRW = [WriteFShuffle] in { def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>; + IIC_SSE_MOVA_P_RR>, FoldGenData<"MOVAPSrr">; def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movapd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>; + IIC_SSE_MOVA_P_RR>, FoldGenData<"MOVAPDrr">; def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movups\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>; + IIC_SSE_MOVU_P_RR>, FoldGenData<"MOVUPSrr">; def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movupd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>; + IIC_SSE_MOVU_P_RR>, FoldGenData<"MOVUPDrr">; } let Predicates = [HasAVX, NoVLX] in { @@ -3748,17 +3759,19 @@ def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, - VEX, VEX_WIG; + VEX, VEX_WIG, FoldGenData<"VMOVDQArr">; def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movdqa\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVA_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVDQAYrr">; def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>, - VEX, VEX_WIG; + VEX, VEX_WIG, FoldGenData<"VMOVDQUrr">; def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), "movdqu\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG; + IIC_SSE_MOVU_P_RR>, VEX, VEX_L, VEX_WIG, + FoldGenData<"VMOVDQUYrr">; } let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1, @@ -3816,11 +3829,12 @@ let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in { def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], - IIC_SSE_MOVA_P_RR>; + IIC_SSE_MOVA_P_RR>, FoldGenData<"MOVDQArr">; def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqu\t{$src, $dst|$dst, $src}", - [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>; + [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>, + FoldGenData<"MOVDQUrr">; } } // SchedRW @@ -5911,7 +5925,7 @@ (ins VR128:$src1, u8imm:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - []>, Sched<[WriteShuffle]>; + []>, Sched<[WriteShuffle]>, FoldGenData; let hasSideEffects = 0, mayStore = 1, SchedRW = [WriteShuffleLd, WriteRMW] in Index: lib/Target/X86/X86InstrXOP.td =================================================================== --- lib/Target/X86/X86InstrXOP.td +++ lib/Target/X86/X86InstrXOP.td @@ -111,7 +111,7 @@ (ins VR128:$src1, VR128:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, - XOP_4V, VEX_W, Sched<[WriteVarVecShift]>; + XOP_4V, VEX_W, Sched<[WriteVarVecShift]>, FoldGenData; } let ExeDomain = SSEPackedInt in { @@ -282,7 +282,7 @@ (ins VR128:$src1, VR128:$src2, VR128:$src3), !strconcat(OpcodeStr, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, XOP_4V, VEX_W; + []>, XOP_4V, VEX_W, FoldGenData; } let ExeDomain = SSEPackedInt in { @@ -318,7 +318,7 @@ (ins RC:$src1, RC:$src2, RC:$src3), !strconcat(OpcodeStr, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, XOP_4V, VEX_W; + []>, XOP_4V, VEX_W, FoldGenData; } let ExeDomain = SSEPackedInt in { @@ -357,7 +357,7 @@ (ins RC:$src1, RC:$src2, RC:$src3, u8imm:$src4), !strconcat(OpcodeStr, "\t{$src4, $src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3, $src4}"), - []>, VEX_W; + []>, VEX_W, FoldGenData; } let ExeDomain = SSEPackedDouble in {