diff options
Diffstat (limited to 'contrib/llvm/lib/Target/X86/X86InstrSSE.td')
| -rw-r--r-- | contrib/llvm/lib/Target/X86/X86InstrSSE.td | 2554 |
1 files changed, 1312 insertions, 1242 deletions
diff --git a/contrib/llvm/lib/Target/X86/X86InstrSSE.td b/contrib/llvm/lib/Target/X86/X86InstrSSE.td index 6f48d7e..3842387 100644 --- a/contrib/llvm/lib/Target/X86/X86InstrSSE.td +++ b/contrib/llvm/lib/Target/X86/X86InstrSSE.td @@ -16,6 +16,8 @@ class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> { InstrItinClass rr = arg_rr; InstrItinClass rm = arg_rm; + // InstrSchedModel info. + X86FoldableSchedWrite Sched = WriteFAdd; } class SizeItins<OpndItins arg_s, OpndItins arg_d> { @@ -33,6 +35,7 @@ class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm, // scalar +let Sched = WriteFAdd in { def SSE_ALU_F32S : OpndItins< IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM >; @@ -40,11 +43,13 @@ def SSE_ALU_F32S : OpndItins< def SSE_ALU_F64S : OpndItins< IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM >; +} def SSE_ALU_ITINS_S : SizeItins< SSE_ALU_F32S, SSE_ALU_F64S >; +let Sched = WriteFMul in { def SSE_MUL_F32S : OpndItins< IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM >; @@ -52,11 +57,13 @@ def SSE_MUL_F32S : OpndItins< def SSE_MUL_F64S : OpndItins< IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM >; +} def SSE_MUL_ITINS_S : SizeItins< SSE_MUL_F32S, SSE_MUL_F64S >; +let Sched = WriteFDiv in { def SSE_DIV_F32S : OpndItins< IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM >; @@ -64,12 +71,14 @@ def SSE_DIV_F32S : OpndItins< def SSE_DIV_F64S : OpndItins< IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM >; +} def SSE_DIV_ITINS_S : SizeItins< SSE_DIV_F32S, SSE_DIV_F64S >; // parallel +let Sched = WriteFAdd in { def SSE_ALU_F32P : OpndItins< IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM >; @@ -77,11 +86,13 @@ def SSE_ALU_F32P : OpndItins< def SSE_ALU_F64P : OpndItins< IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM >; +} def SSE_ALU_ITINS_P : SizeItins< SSE_ALU_F32P, SSE_ALU_F64P >; +let Sched = WriteFMul in { def SSE_MUL_F32P : OpndItins< IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM >; @@ -89,11 +100,13 @@ def SSE_MUL_F32P : OpndItins< def SSE_MUL_F64P : OpndItins< IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM >; +} def SSE_MUL_ITINS_P : SizeItins< SSE_MUL_F32P, SSE_MUL_F64P >; +let Sched = WriteFDiv in { def SSE_DIV_F32P : OpndItins< IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM >; @@ -101,6 +114,7 @@ def SSE_DIV_F32P : OpndItins< def SSE_DIV_F64P : OpndItins< IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM >; +} def SSE_DIV_ITINS_P : SizeItins< SSE_DIV_F32P, SSE_DIV_F64P @@ -110,6 +124,7 @@ def SSE_BIT_ITINS_P : OpndItins< IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM >; +let Sched = WriteVecALU in { def SSE_INTALU_ITINS_P : OpndItins< IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM >; @@ -117,7 +132,9 @@ def SSE_INTALU_ITINS_P : OpndItins< def SSE_INTALUQ_ITINS_P : OpndItins< IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM >; +} +let Sched = WriteVecIMul in def SSE_INTMUL_ITINS_P : OpndItins< IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM >; @@ -148,13 +165,15 @@ multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>; + [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; } def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>; + [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class @@ -169,14 +188,16 @@ multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC, !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (!cast<Intrinsic>( !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, RC:$src2))], itins.rr>; + RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2), !if(Is2Addr, !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, mem_cpat:$src2))], itins.rm>; + RC:$src1, mem_cpat:$src2))], itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// sse12_fp_packed - SSE 1 & 2 packed instructions class @@ -189,54 +210,36 @@ multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>; + [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>, + Sched<[itins.Sched]>; let mayLoad = 1 in def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], - itins.rm, d>; + itins.rm, d>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d, string OpcodeStr, X86MemOperand x86memop, list<dag> pat_rr, list<dag> pat_rm, - bit Is2Addr = 1, - bit rr_hasSideEffects = 0> { - let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in + bit Is2Addr = 1> { + let isCommutable = 1, hasSideEffects = 0 in def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - pat_rr, IIC_DEFAULT, d>; + pat_rr, NoItinerary, d>, + Sched<[WriteVecLogic]>; def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - pat_rm, IIC_DEFAULT, d>; -} - -/// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class -multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC, - string asm, string SSEVer, string FPSizeStr, - X86MemOperand x86memop, PatFrag mem_frag, - Domain d, OpndItins itins, bit Is2Addr = 1> { - def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2), - !if(Is2Addr, - !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), - !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (!cast<Intrinsic>( - !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, RC:$src2))], IIC_DEFAULT, d>; - def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2), - !if(Is2Addr, - !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), - !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (!cast<Intrinsic>( - !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr)) - RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>; + pat_rm, NoItinerary, d>, + Sched<[WriteVecLogicLd, ReadAfterLd]>; } //===----------------------------------------------------------------------===// @@ -367,7 +370,7 @@ let Predicates = [HasAVX] in { // Alias instructions that map fld0 to xorps for sse or vxorps for avx. // This is expanded by ExpandPostRAPseudos. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "", [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>; def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "", @@ -384,7 +387,7 @@ let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, // We set canFoldAsLoad because this can be converted to a constant-pool // load of an all-zeros value if folding it would be beneficial. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", [(set VR128:$dst, (v4f32 immAllZerosV))]>; } @@ -401,7 +404,7 @@ def : Pat<(v16i8 immAllZerosV), (V_SET0)>; // at the rename stage without using any execution unit, so SET0PSY // and SET0PDY can be used for vector int instructions without penalty let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1, Predicates = [HasAVX] in { + isPseudo = 1, Predicates = [HasAVX], SchedRW = [WriteZero] in { def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "", [(set VR256:$dst, (v8f32 immAllZerosV))]>; } @@ -439,7 +442,7 @@ def : Pat<(bc_v4i64 (v8f32 immAllZerosV)), // We set canFoldAsLoad because this can be converted to a constant-pool // load of an all-ones value if folding it would be beneficial. let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, - isPseudo = 1 in { + isPseudo = 1, SchedRW = [WriteZero] in { def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "", [(set VR128:$dst, (v4i32 immAllOnesV))]>; let Predicates = [HasAVX2] in @@ -458,93 +461,70 @@ let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, // in terms of a copy, and just mentioned, we don't use movss/movsd for copies. //===----------------------------------------------------------------------===// -class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> : - SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm, - [(set VR128:$dst, (vt (OpNode VR128:$src1, - (scalar_to_vector RC:$src2))))], - IIC_SSE_MOV_S_RR>; +multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, + X86MemOperand x86memop, string base_opc, + string asm_opr> { + def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, RC:$src2), + !strconcat(base_opc, asm_opr), + [(set VR128:$dst, (vt (OpNode VR128:$src1, + (scalar_to_vector RC:$src2))))], + IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>; -// Loading from memory automatically zeroing upper bits. -class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop, - PatFrag mem_pat, string OpcodeStr> : - SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), - !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set RC:$dst, (mem_pat addr:$src))], - IIC_SSE_MOV_S_RM>; - -// AVX -def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32, - "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V, - VEX_LIG; -def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64, - "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V, - VEX_LIG; - -// For the disassembler -let isCodeGenOnly = 1 in { - def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), - (ins VR128:$src1, FR32:$src2), - "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_MOV_S_RR>, - XS, VEX_4V, VEX_LIG; - def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), - (ins VR128:$src1, FR64:$src2), - "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_MOV_S_RR>, - XD, VEX_4V, VEX_LIG; + // For the disassembler + let isCodeGenOnly = 1, hasSideEffects = 0 in + def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), + (ins VR128:$src1, RC:$src2), + !strconcat(base_opc, asm_opr), + [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>; } -let canFoldAsLoad = 1, isReMaterializable = 1 in { - def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX, - VEX_LIG; - let AddedComplexity = 20 in - def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX, - VEX_LIG; -} +multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt, + X86MemOperand x86memop, string OpcodeStr> { + // AVX + defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, + VEX_4V, VEX_LIG; -def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src), - "movss\t{$src, $dst|$dst, $src}", - [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, - XS, VEX, VEX_LIG; -def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src), - "movsd\t{$src, $dst|$dst, $src}", - [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, - XD, VEX, VEX_LIG; + def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, + VEX, VEX_LIG, Sched<[WriteStore]>; + // SSE1 & 2 + let Constraints = "$src1 = $dst" in { + defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr, + "\t{$src2, $dst|$dst, $src2}">; + } -// SSE1 & 2 -let Constraints = "$src1 = $dst" in { - def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32, - "movss\t{$src2, $dst|$dst, $src2}">, XS; - def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64, - "movsd\t{$src2, $dst|$dst, $src2}">, XD; + def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, + Sched<[WriteStore]>; +} - // For the disassembler - let isCodeGenOnly = 1 in { - def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), - (ins VR128:$src1, FR32:$src2), - "movss\t{$src2, $dst|$dst, $src2}", [], - IIC_SSE_MOV_S_RR>, XS; - def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst), - (ins VR128:$src1, FR64:$src2), - "movsd\t{$src2, $dst|$dst, $src2}", [], - IIC_SSE_MOV_S_RR>, XD; - } +// Loading from memory automatically zeroing upper bits. +multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop, + PatFrag mem_pat, string OpcodeStr> { + def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (mem_pat addr:$src))], + IIC_SSE_MOV_S_RM>, VEX, VEX_LIG, Sched<[WriteLoad]>; + def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (mem_pat addr:$src))], + IIC_SSE_MOV_S_RM>, Sched<[WriteLoad]>; } +defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS; +defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD; + let canFoldAsLoad = 1, isReMaterializable = 1 in { - def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS; + defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS; let AddedComplexity = 20 in - def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD; + defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD; } -def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src), - "movss\t{$src, $dst|$dst, $src}", - [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>; -def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src), - "movsd\t{$src, $dst|$dst, $src}", - [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>; - // Patterns let Predicates = [HasAVX] in { let AddedComplexity = 15 in { @@ -791,11 +771,13 @@ multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC, bit IsReMaterializable = 1> { let neverHasSideEffects = 1 in def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src), - !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>; + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>, + Sched<[WriteMove]>; let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>; + [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>, + Sched<[WriteLoad]>; } defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32, @@ -836,6 +818,7 @@ defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64, "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>, TB, OpSize; +let SchedRW = [WriteStore] in { def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [(alignedstore (v4f32 VR128:$src), addr:$dst)], @@ -868,9 +851,10 @@ def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src), "movupd\t{$src, $dst|$dst, $src}", [(store (v4f64 VR256:$src), addr:$dst)], IIC_SSE_MOVU_P_MR>, VEX, VEX_L; +} // SchedRW // For disassembler -let isCodeGenOnly = 1 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [], @@ -926,6 +910,7 @@ def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src), def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src), (VMOVUPDYmr addr:$dst, VR256:$src)>; +let SchedRW = [WriteStore] in { def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [(alignedstore (v4f32 VR128:$src), addr:$dst)], @@ -942,9 +927,10 @@ def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movupd\t{$src, $dst|$dst, $src}", [(store (v2f64 VR128:$src), addr:$dst)], IIC_SSE_MOVU_P_MR>; +} // SchedRW // For disassembler -let isCodeGenOnly = 1 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movaps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>; @@ -1055,7 +1041,7 @@ let Predicates = [HasAVX] in { (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>; def : Pat<(store (v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))), addr:$dst), - (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>; + (VMOVUPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>; def : Pat<(store (v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))), addr:$dst), (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>; @@ -1090,7 +1076,7 @@ let Predicates = [UseSSE1] in { // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper // bits are disregarded. FIXME: Set encoding to pseudo! -let neverHasSideEffects = 1 in { +let neverHasSideEffects = 1, SchedRW = [WriteMove] in { def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src), "movaps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, VEX; @@ -1107,7 +1093,7 @@ def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src), // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper // bits are disregarded. FIXME: Set encoding to pseudo! -let canFoldAsLoad = 1, isReMaterializable = 1 in { +let canFoldAsLoad = 1, isReMaterializable = 1, SchedRW = [WriteLoad] in { let isCodeGenOnly = 1 in { def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src), "movaps\t{$src, $dst|$dst, $src}", @@ -1132,36 +1118,46 @@ def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src), // SSE 1 & 2 - Move Low packed FP Instructions //===----------------------------------------------------------------------===// -multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC, - SDNode psnode, SDNode pdnode, string base_opc, - string asm_opr, InstrItinClass itin> { +multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode, + string base_opc, string asm_opr, + InstrItinClass itin> { def PSrm : PI<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2), !strconcat(base_opc, "s", asm_opr), - [(set RC:$dst, - (psnode RC:$src1, + [(set VR128:$dst, + (psnode VR128:$src1, (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))], - itin, SSEPackedSingle>, TB; + itin, SSEPackedSingle>, TB, + Sched<[WriteShuffleLd, ReadAfterLd]>; def PDrm : PI<opc, MRMSrcMem, - (outs RC:$dst), (ins RC:$src1, f64mem:$src2), + (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2), !strconcat(base_opc, "d", asm_opr), - [(set RC:$dst, (v2f64 (pdnode RC:$src1, + [(set VR128:$dst, (v2f64 (pdnode VR128:$src1, (scalar_to_vector (loadf64 addr:$src2)))))], - itin, SSEPackedDouble>, TB, OpSize; + itin, SSEPackedDouble>, TB, OpSize, + Sched<[WriteShuffleLd, ReadAfterLd]>; + } -let AddedComplexity = 20 in { - defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp", - "\t{$src2, $src1, $dst|$dst, $src1, $src2}", - IIC_SSE_MOV_LH>, VEX_4V; +multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode, + string base_opc, InstrItinClass itin> { + defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}", + itin>, VEX_4V; + +let Constraints = "$src1 = $dst" in + defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc, + "\t{$src2, $dst|$dst, $src2}", + itin>; } -let Constraints = "$src1 = $dst", AddedComplexity = 20 in { - defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp", - "\t{$src2, $dst|$dst, $src2}", - IIC_SSE_MOV_LH>; + +let AddedComplexity = 20 in { + defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp", + IIC_SSE_MOV_LH>; } +let SchedRW = [WriteStore] in { def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), "movlps\t{$src, $dst|$dst, $src}", [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)), @@ -1182,6 +1178,7 @@ def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), [(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>; +} // SchedRW let Predicates = [HasAVX] in { // Shuffle with VMOVLPS @@ -1257,16 +1254,11 @@ let Predicates = [UseSSE2] in { //===----------------------------------------------------------------------===// let AddedComplexity = 20 in { - defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp", - "\t{$src2, $src1, $dst|$dst, $src1, $src2}", - IIC_SSE_MOV_LH>, VEX_4V; -} -let Constraints = "$src1 = $dst", AddedComplexity = 20 in { - defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp", - "\t{$src2, $dst|$dst, $src2}", - IIC_SSE_MOV_LH>; + defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp", + IIC_SSE_MOV_LH>; } +let SchedRW = [WriteStore] in { // v2f64 extract element 1 is always custom lowered to unpack high to low // and extract element 0 so the non-store version isn't too horrible. def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), @@ -1291,6 +1283,7 @@ def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src), [(store (f64 (vector_extract (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>; +} // SchedRW let Predicates = [HasAVX] in { // VMOVHPS patterns @@ -1341,14 +1334,14 @@ let AddedComplexity = 20 in { [(set VR128:$dst, (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))], IIC_SSE_MOV_LH>, - VEX_4V; + VEX_4V, Sched<[WriteShuffle]>; def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))], IIC_SSE_MOV_LH>, - VEX_4V; + VEX_4V, Sched<[WriteShuffle]>; } let Constraints = "$src1 = $dst", AddedComplexity = 20 in { def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst), @@ -1356,13 +1349,13 @@ let Constraints = "$src1 = $dst", AddedComplexity = 20 in { "movlhps\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "movhlps\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; } let Predicates = [HasAVX] in { @@ -1397,22 +1390,27 @@ def SSE_CVT_PD : OpndItins< IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM >; +let Sched = WriteCvtI2F in def SSE_CVT_PS : OpndItins< IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM >; +let Sched = WriteCvtI2F in def SSE_CVT_Scalar : OpndItins< IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM >; +let Sched = WriteCvtF2I in def SSE_CVT_SS2SI_32 : OpndItins< IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM >; +let Sched = WriteCvtF2I in def SSE_CVT_SS2SI_64 : OpndItins< IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM >; +let Sched = WriteCvtF2I in def SSE_CVT_SD2SI : OpndItins< IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM >; @@ -1422,10 +1420,10 @@ multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, string asm, OpndItins itins> { def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, [(set DstRC:$dst, (OpNode SrcRC:$src))], - itins.rr>; + itins.rr>, Sched<[itins.Sched]>; def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], - itins.rm>; + itins.rm>, Sched<[itins.Sched.Folded]>; } multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, @@ -1433,10 +1431,10 @@ multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, OpndItins itins> { let neverHasSideEffects = 1 in { def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, - [], itins.rr, d>; + [], itins.rr, d>, Sched<[itins.Sched]>; let mayLoad = 1 in def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, - [], itins.rm, d>; + [], itins.rm, d>, Sched<[itins.Sched.Folded]>; } } @@ -1444,11 +1442,13 @@ multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, X86MemOperand x86memop, string asm> { let neverHasSideEffects = 1 in { def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src), - !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>; + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + Sched<[WriteCvtI2F]>; let mayLoad = 1 in def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins DstRC:$src1, x86memop:$src), - !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>; + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + Sched<[WriteCvtI2FLd, ReadAfterLd]>; } // neverHasSideEffects = 1 } @@ -1457,7 +1457,7 @@ defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32, SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG; defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32, - "cvttss2si{q}\t{$src, $dst|$dst, $src}", + "cvttss2si\t{$src, $dst|$dst, $src}", SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG; defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64, @@ -1465,26 +1465,43 @@ defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64, SSE_CVT_SD2SI>, XD, VEX, VEX_LIG; defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64, - "cvttsd2si{q}\t{$src, $dst|$dst, $src}", + "cvttsd2si\t{$src, $dst|$dst, $src}", SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG; +def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}", + (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>; +def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}", + (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>; +def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}", + (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>; +def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}", + (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>; +def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}", + (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>; +def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}", + (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>; +def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}", + (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>; +def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}", + (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>; + // The assembler can recognize rr 64-bit instructions by seeing a rxx // register, but the same isn't true when only using memory operands, // provide other assembly "l" and "q" forms to address this explicitly // where appropriate to do so. -defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, +defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">, XS, VEX_4V, VEX_LIG; defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS, VEX_4V, VEX_W, VEX_LIG; -defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, +defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD, VEX_4V, VEX_LIG; defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD, VEX_4V, VEX_W, VEX_LIG; -def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}", - (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>; -def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}", +def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}", + (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>; +def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}", (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>; let Predicates = [HasAVX] in { @@ -1511,27 +1528,49 @@ defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32, "cvttss2si\t{$src, $dst|$dst, $src}", SSE_CVT_SS2SI_32>, XS; defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32, - "cvttss2si{q}\t{$src, $dst|$dst, $src}", + "cvttss2si\t{$src, $dst|$dst, $src}", SSE_CVT_SS2SI_64>, XS, REX_W; defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64, "cvttsd2si\t{$src, $dst|$dst, $src}", SSE_CVT_SD2SI>, XD; defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64, - "cvttsd2si{q}\t{$src, $dst|$dst, $src}", + "cvttsd2si\t{$src, $dst|$dst, $src}", SSE_CVT_SD2SI>, XD, REX_W; defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32, - "cvtsi2ss\t{$src, $dst|$dst, $src}", + "cvtsi2ss{l}\t{$src, $dst|$dst, $src}", SSE_CVT_Scalar>, XS; defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64, "cvtsi2ss{q}\t{$src, $dst|$dst, $src}", SSE_CVT_Scalar>, XS, REX_W; defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32, - "cvtsi2sd\t{$src, $dst|$dst, $src}", + "cvtsi2sd{l}\t{$src, $dst|$dst, $src}", SSE_CVT_Scalar>, XD; defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64, "cvtsi2sd{q}\t{$src, $dst|$dst, $src}", SSE_CVT_Scalar>, XD, REX_W; +def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}", + (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>; +def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}", + (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>; +def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}", + (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>; +def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}", + (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>; +def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}", + (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>; +def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}", + (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>; +def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}", + (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>; +def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}", + (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>; + +def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}", + (CVTSI2SSrm FR64:$dst, i32mem:$src)>; +def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}", + (CVTSI2SDrm FR64:$dst, i32mem:$src)>; + // Conversion Instructions Intrinsics - Match intrinsics which expect MM // and/or XMM operand(s). @@ -1540,10 +1579,12 @@ multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, string asm, OpndItins itins> { def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>; + [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>, + Sched<[itins.Sched]>; def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>; + [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>, + Sched<[itins.Sched.Folded]>; } multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC, @@ -1555,38 +1596,38 @@ multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC, !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], - itins.rr>; + itins.rr>, Sched<[itins.Sched]>; def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins DstRC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"), !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], - itins.rm>; + itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>; } defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, - int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}", + int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, VEX, VEX_LIG; defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, - int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}", + int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG; defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si, - sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD; + sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD; defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64, - sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W; + sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W; defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128, - int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", + int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}", SSE_CVT_Scalar, 0>, XS, VEX_4V; defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128, int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}", SSE_CVT_Scalar, 0>, XS, VEX_4V, VEX_W; defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128, - int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", + int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}", SSE_CVT_Scalar, 0>, XD, VEX_4V; defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128, int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}", @@ -1596,13 +1637,13 @@ defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128, let Constraints = "$src1 = $dst" in { defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128, int_x86_sse_cvtsi2ss, i32mem, loadi32, - "cvtsi2ss", SSE_CVT_Scalar>, XS; + "cvtsi2ss{l}", SSE_CVT_Scalar>, XS; defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128, int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W; defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128, int_x86_sse2_cvtsi2sd, i32mem, loadi32, - "cvtsi2sd", SSE_CVT_Scalar>, XD; + "cvtsi2sd{l}", SSE_CVT_Scalar>, XD; defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128, int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W; @@ -1616,40 +1657,40 @@ defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si, SSE_CVT_SS2SI_32>, XS, VEX; defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64, int_x86_sse_cvttss2si64, ssmem, sse_load_f32, - "cvttss2si{q}", SSE_CVT_SS2SI_64>, + "cvttss2si", SSE_CVT_SS2SI_64>, XS, VEX, VEX_W; defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si, sdmem, sse_load_f64, "cvttsd2si", SSE_CVT_SD2SI>, XD, VEX; defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64, int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64, - "cvttsd2si{q}", SSE_CVT_SD2SI>, + "cvttsd2si", SSE_CVT_SD2SI>, XD, VEX, VEX_W; defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si, ssmem, sse_load_f32, "cvttss2si", SSE_CVT_SS2SI_32>, XS; defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64, int_x86_sse_cvttss2si64, ssmem, sse_load_f32, - "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W; + "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W; defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si, sdmem, sse_load_f64, "cvttsd2si", SSE_CVT_SD2SI>, XD; defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64, int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64, - "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W; + "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W; defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si, - ssmem, sse_load_f32, "cvtss2si{l}", + ssmem, sse_load_f32, "cvtss2si", SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG; defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64, - ssmem, sse_load_f32, "cvtss2si{q}", + ssmem, sse_load_f32, "cvtss2si", SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG; defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si, - ssmem, sse_load_f32, "cvtss2si{l}", + ssmem, sse_load_f32, "cvtss2si", SSE_CVT_SS2SI_32>, XS; defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64, - ssmem, sse_load_f32, "cvtss2si{q}", + ssmem, sse_load_f32, "cvtss2si", SSE_CVT_SS2SI_64>, XS, REX_W; defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem, @@ -1666,6 +1707,40 @@ defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem, SSEPackedSingle, SSE_CVT_PS>, TB, Requires<[UseSSE2]>; +def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}", + (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>; +def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}", + (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>; +def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}", + (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>; +def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}", + (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>; +def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}", + (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>; +def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}", + (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>; +def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}", + (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>; +def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}", + (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>; + +def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}", + (CVTSS2SIrr GR32:$dst, VR128:$src), 0>; +def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}", + (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>; +def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}", + (CVTSD2SIrr GR32:$dst, VR128:$src), 0>; +def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}", + (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>; +def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}", + (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>; +def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}", + (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>; +def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}", + (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>; +def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}", + (CVTSD2SI64rm GR64:$dst, sdmem:$src)>; + /// SSE 2 Only // Convert scalar double to scalar single @@ -1673,13 +1748,15 @@ let neverHasSideEffects = 1 in { def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src1, FR64:$src2), "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], - IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG; + IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2F]>; let mayLoad = 1 in def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins FR64:$src1, f64mem:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RM>, - XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG; + XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>, @@ -1688,26 +1765,28 @@ def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>, def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src), "cvtsd2ss\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (fround FR64:$src))], - IIC_SSE_CVT_Scalar_RR>; + IIC_SSE_CVT_Scalar_RR>, Sched<[WriteCvtF2F]>; def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src), "cvtsd2ss\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (fround (loadf64 addr:$src)))], IIC_SSE_CVT_Scalar_RM>, XD, - Requires<[UseSSE2, OptForSize]>; + Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>; def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2F]>; def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2), "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, sse_load_f64:$src2))], - IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; let Constraints = "$src1 = $dst" in { def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg, @@ -1715,13 +1794,15 @@ def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg, "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>, + Sched<[WriteCvtF2F]>; def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2), "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1, sse_load_f64:$src2))], - IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } // Convert scalar single to scalar double @@ -1731,13 +1812,15 @@ def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src1, FR32:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RR>, - XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG; + XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG, + Sched<[WriteCvtF2F]>; let mayLoad = 1 in def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins FR32:$src1, f32mem:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], IIC_SSE_CVT_Scalar_RM>, - XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>; + XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } def : Pat<(f64 (fextend FR32:$src)), @@ -1756,12 +1839,12 @@ def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src), "cvtss2sd\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (fextend FR32:$src))], IIC_SSE_CVT_Scalar_RR>, XS, - Requires<[UseSSE2]>; + Requires<[UseSSE2]>, Sched<[WriteCvtF2F]>; def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src), "cvtss2sd\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (extloadf32 addr:$src))], IIC_SSE_CVT_Scalar_RM>, XS, - Requires<[UseSSE2, OptForSize]>; + Requires<[UseSSE2, OptForSize]>, Sched<[WriteCvtF2FLd]>; // extload f32 -> f64. This matches load+fextend because we have a hack in // the isel (PreprocessForFPConvert) that can introduce loads after dag @@ -1778,57 +1861,61 @@ def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg, "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2F]>; def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2), "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))], - IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>; + IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2), "cvtss2sd\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))], - IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>, + Sched<[WriteCvtF2F]>; def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2), "cvtss2sd\t{$src2, $dst|$dst, $src2}", [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))], - IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>; + IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>, + Sched<[WriteCvtF2FLd, ReadAfterLd]>; } // Convert packed single/double fp to doubleword def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>, VEX; + IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>; def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX; + IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>; def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2dq_256 VR256:$src))], - IIC_SSE_CVT_PS_RR>, VEX, VEX_L; + IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX, VEX_L; + IIC_SSE_CVT_PS_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>; + IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>; def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>; + IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>; // Convert Packed Double FP to Packed DW Integers @@ -1839,7 +1926,7 @@ let Predicates = [HasAVX] in { def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>, - VEX; + VEX, Sched<[WriteCvtF2I]>; // XMM only def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}", @@ -1847,18 +1934,20 @@ def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}", def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "vcvtpd2dqx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX; + (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX, + Sched<[WriteCvtF2ILd]>; // YMM only def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L; + (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L, + Sched<[WriteCvtF2I]>; def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>, - VEX, VEX_L; + VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}", (VCVTPD2DQYrr VR128:$dst, VR256:$src)>; } @@ -1867,11 +1956,11 @@ def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2ILd]>; def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>; // Convert with truncation packed single/double fp to doubleword // SSE2 packed instructions with XS prefix @@ -1879,32 +1968,33 @@ def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>, VEX; + IIC_SSE_CVT_PS_RR>, VEX, Sched<[WriteCvtF2I]>; def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX; + IIC_SSE_CVT_PS_RM>, VEX, Sched<[WriteCvtF2ILd]>; def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256 VR256:$src))], - IIC_SSE_CVT_PS_RR>, VEX, VEX_L; + IIC_SSE_CVT_PS_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>, VEX, VEX_L; + IIC_SSE_CVT_PS_RM>, VEX, VEX_L, + Sched<[WriteCvtF2ILd]>; def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))], - IIC_SSE_CVT_PS_RR>; + IIC_SSE_CVT_PS_RR>, Sched<[WriteCvtF2I]>; def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttps2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PS_RM>; + IIC_SSE_CVT_PS_RM>, Sched<[WriteCvtF2ILd]>; let Predicates = [HasAVX] in { def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))), @@ -1954,7 +2044,7 @@ def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>, VEX; + IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2I]>; // The assembler can recognize rr 256-bit instructions by seeing a ymm // register, but the same isn't true when using memory operands instead. @@ -1967,19 +2057,19 @@ def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvttpd2dqx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX; + IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2ILd]>; // YMM only def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvtt_pd2dq_256 VR256:$src))], - IIC_SSE_CVT_PD_RR>, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2I]>; def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>; def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}", (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>; @@ -1993,12 +2083,13 @@ let Predicates = [HasAVX] in { def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2I]>; def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src), "cvttpd2dq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvttpd2dq (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, + Sched<[WriteCvtF2ILd]>; // Convert packed single to packed double let Predicates = [HasAVX] in { @@ -2006,32 +2097,32 @@ let Predicates = [HasAVX] in { def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB, VEX; + IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>; def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB, VEX; + IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>; def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2_pd_256 VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>; def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src), "vcvtps2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>; } let Predicates = [UseSSE2] in { def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))], - IIC_SSE_CVT_PD_RR>, TB; + IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>; def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), "cvtps2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))], - IIC_SSE_CVT_PD_RM>, TB; + IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>; } // Convert Packed DW Integers to Packed Double FP @@ -2039,30 +2130,33 @@ let Predicates = [HasAVX] in { let neverHasSideEffects = 1, mayLoad = 1 in def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", - []>, VEX; + []>, VEX, Sched<[WriteCvtI2FLd]>; def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX; + (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX, + Sched<[WriteCvtI2F]>; def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, (int_x86_avx_cvtdq2_pd_256 - (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L; + (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L, + Sched<[WriteCvtI2FLd]>; def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src), "vcvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR256:$dst, - (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L; + (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L, + Sched<[WriteCvtI2F]>; } let neverHasSideEffects = 1, mayLoad = 1 in def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "cvtdq2pd\t{$src, $dst|$dst, $src}", [], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>; def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtdq2pd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtI2F]>; // AVX 256-bit register conversion intrinsics let Predicates = [HasAVX] in { @@ -2079,7 +2173,7 @@ let Predicates = [HasAVX] in { def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))], - IIC_SSE_CVT_PD_RR>, VEX; + IIC_SSE_CVT_PD_RR>, VEX, Sched<[WriteCvtF2F]>; // XMM only def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}", @@ -2088,31 +2182,31 @@ def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2psx\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX; + IIC_SSE_CVT_PD_RM>, VEX, Sched<[WriteCvtF2FLd]>; // YMM only def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src), "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2_ps_256 VR256:$src))], - IIC_SSE_CVT_PD_RR>, VEX, VEX_L; + IIC_SSE_CVT_PD_RR>, VEX, VEX_L, Sched<[WriteCvtF2F]>; def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src), "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>, VEX, VEX_L; + IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>; def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}", (VCVTPD2PSYrr VR128:$dst, VR256:$src)>; def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))], - IIC_SSE_CVT_PD_RR>; + IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtF2F]>; def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), "cvtpd2ps\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))], - IIC_SSE_CVT_PD_RM>; + IIC_SSE_CVT_PD_RM>, Sched<[WriteCvtF2FLd]>; // AVX 256-bit register conversion intrinsics @@ -2165,22 +2259,24 @@ multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop, def rr : SIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm, [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))], - itins.rr>; + itins.rr>, Sched<[itins.Sched]>; def rm : SIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm, [(set RC:$dst, (OpNode (VT RC:$src1), (ld_frag addr:$src2), imm:$cc))], - itins.rm>; + itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; // Accept explicit immediate argument form instead of comparison code. let neverHasSideEffects = 1 in { def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [], - IIC_SSE_ALU_F32S_RR>; + IIC_SSE_ALU_F32S_RR>, Sched<[itins.Sched]>; let mayLoad = 1 in def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [], - IIC_SSE_ALU_F32S_RM>; + IIC_SSE_ALU_F32S_RM>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } } @@ -2213,12 +2309,14 @@ multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC, (ins VR128:$src1, VR128:$src, CC:$cc), asm, [(set VR128:$dst, (Int VR128:$src1, VR128:$src, imm:$cc))], - itins.rr>; + itins.rr>, + Sched<[itins.Sched]>; def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, x86memop:$src, CC:$cc), asm, [(set VR128:$dst, (Int VR128:$src1, (load addr:$src), imm:$cc))], - itins.rm>; + itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } // Aliases to match intrinsics which expect XMM operand(s). @@ -2248,12 +2346,14 @@ multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode, def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"), [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], - IIC_SSE_COMIS_RR, d>; + IIC_SSE_COMIS_RR, d>, + Sched<[WriteFAdd]>; def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2), !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"), [(set EFLAGS, (OpNode (vt RC:$src1), (ld_frag addr:$src2)))], - IIC_SSE_COMIS_RM, d>; + IIC_SSE_COMIS_RM, d>, + Sched<[WriteFAddLd, ReadAfterLd]>; } let Defs = [EFLAGS] in { @@ -2310,20 +2410,23 @@ multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop, def rri : PIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm, [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], - IIC_SSE_CMPP_RR, d>; + IIC_SSE_CMPP_RR, d>, + Sched<[WriteFAdd]>; def rmi : PIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm, [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], - IIC_SSE_CMPP_RM, d>; + IIC_SSE_CMPP_RM, d>, + Sched<[WriteFAddLd, ReadAfterLd]>; // Accept explicit immediate argument form instead of comparison code. let neverHasSideEffects = 1 in { def rri_alt : PIi8<0xC2, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), - asm_alt, [], IIC_SSE_CMPP_RR, d>; + asm_alt, [], IIC_SSE_CMPP_RR, d>, Sched<[WriteFAdd]>; def rmi_alt : PIi8<0xC2, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), - asm_alt, [], IIC_SSE_CMPP_RM, d>; + asm_alt, [], IIC_SSE_CMPP_RM, d>, + Sched<[WriteFAddLd, ReadAfterLd]>; } } @@ -2399,12 +2502,14 @@ multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop, def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm, [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2), - (i8 imm:$src3))))], IIC_SSE_SHUFP, d>; + (i8 imm:$src3))))], IIC_SSE_SHUFP, d>, + Sched<[WriteShuffleLd, ReadAfterLd]>; let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$src3), asm, [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2, - (i8 imm:$src3))))], IIC_SSE_SHUFP, d>; + (i8 imm:$src3))))], IIC_SSE_SHUFP, d>, + Sched<[WriteShuffle]>; } defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32, @@ -2488,13 +2593,14 @@ multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt, (outs RC:$dst), (ins RC:$src1, RC:$src2), asm, [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], - IIC_SSE_UNPCK, d>; + IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>; def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), asm, [(set RC:$dst, (vt (OpNode RC:$src1, (mem_frag addr:$src2))))], - IIC_SSE_UNPCK, d>; + IIC_SSE_UNPCK, d>, + Sched<[WriteShuffleLd, ReadAfterLd]>; } defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32, @@ -2585,10 +2691,11 @@ multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm, Domain d> { def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), - [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>; + [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>, + Sched<[WriteVecLogic]>; def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], - IIC_SSE_MOVMSK, d>, REX_W; + IIC_SSE_MOVMSK, d>, REX_W, Sched<[WriteVecLogic]>; } let Predicates = [HasAVX] in { @@ -2616,18 +2723,18 @@ let Predicates = [HasAVX] in { // Assembler Only def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src), "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, - SSEPackedSingle>, TB, VEX; + SSEPackedSingle>, TB, VEX, Sched<[WriteVecLogic]>; def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src), "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, SSEPackedDouble>, TB, - OpSize, VEX; + OpSize, VEX, Sched<[WriteVecLogic]>; def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src), "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, - SSEPackedSingle>, TB, VEX, VEX_L; + SSEPackedSingle>, TB, VEX, VEX_L, Sched<[WriteVecLogic]>; def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src), "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK, SSEPackedDouble>, TB, - OpSize, VEX, VEX_L; + OpSize, VEX, VEX_L, Sched<[WriteVecLogic]>; } defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps", @@ -2657,17 +2764,16 @@ let ExeDomain = SSEPackedInt in { // SSE integer instructions /// PDI_binop_rm - Simple SSE2 binary operator. multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, ValueType OpVT, RegisterClass RC, PatFrag memop_frag, - X86MemOperand x86memop, - OpndItins itins, - bit IsCommutable = 0, - bit Is2Addr = 1> { + X86MemOperand x86memop, OpndItins itins, + bit IsCommutable, bit Is2Addr> { let isCommutable = IsCommutable in def rr : PDI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>; + [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>, + Sched<[itins.Sched]>; def rm : PDI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, @@ -2675,44 +2781,35 @@ multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpVT (OpNode RC:$src1, (bitconvert (memop_frag addr:$src2)))))], - itins.rm>; + itins.rm>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } } // ExeDomain = SSEPackedInt -// These are ordered here for pattern ordering requirements with the fp versions +multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode, + ValueType OpVT128, ValueType OpVT256, + OpndItins itins, bit IsCommutable = 0> { +let Predicates = [HasAVX] in + defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128, + VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V; -let Predicates = [HasAVX] in { -defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V; -defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V; -defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V; -defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V; +let Constraints = "$src1 = $dst" in + defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128, + memopv2i64, i128mem, itins, IsCommutable, 1>; + +let Predicates = [HasAVX2] in + defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, + OpVT256, VR256, memopv4i64, i256mem, itins, + IsCommutable, 0>, VEX_4V, VEX_L; } -let Constraints = "$src1 = $dst" in { -defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1>; -defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1>; -defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 1>; -defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64, - i128mem, SSE_BIT_ITINS_P, 0>; -} // Constraints = "$src1 = $dst" +// These are ordered here for pattern ordering requirements with the fp versions -let Predicates = [HasAVX2] in { -defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64, - i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64, - i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64, - i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64, - i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V, VEX_L; -} +defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>; +defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>; +defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>; +defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64, + SSE_BIT_ITINS_P, 0>; //===----------------------------------------------------------------------===// // SSE 1 & 2 - Logical Instructions @@ -2757,6 +2854,20 @@ let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in /// multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr, SDNode OpNode> { + defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle, + !strconcat(OpcodeStr, "ps"), f256mem, + [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))], + [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)), + (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L; + + defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble, + !strconcat(OpcodeStr, "pd"), f256mem, + [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)), + (bc_v4i64 (v4f64 VR256:$src2))))], + [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)), + (memopv4i64 addr:$src2)))], 0>, + TB, OpSize, VEX_4V, VEX_L; + // In AVX no need to add a pattern for 128-bit logical rr ps, because they // are all promoted to v2i64, and the patterns are covered by the int // version. This is needed in SSE only, because v2i64 isn't supported on @@ -2764,7 +2875,7 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr, defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle, !strconcat(OpcodeStr, "ps"), f128mem, [], [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)), - (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V; + (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V; defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble, !strconcat(OpcodeStr, "pd"), f128mem, @@ -2773,6 +2884,7 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr, [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)), (memopv2i64 addr:$src2)))], 0>, TB, OpSize, VEX_4V; + let Constraints = "$src1 = $dst" in { defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle, !strconcat(OpcodeStr, "ps"), f128mem, @@ -2789,31 +2901,6 @@ multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr, } } -/// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms -/// -multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr, - SDNode OpNode> { - defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle, - !strconcat(OpcodeStr, "ps"), f256mem, - [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))], - [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)), - (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L; - - defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble, - !strconcat(OpcodeStr, "pd"), f256mem, - [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)), - (bc_v4i64 (v4f64 VR256:$src2))))], - [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)), - (memopv4i64 addr:$src2)))], 0>, - TB, OpSize, VEX_4V, VEX_L; -} - -// AVX 256-bit packed logical ops forms -defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>; -defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>; -defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>; -defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>; - defm AND : sse12_fp_packed_logical<0x54, "and", and>; defm OR : sse12_fp_packed_logical<0x56, "or", or>; defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>; @@ -2848,26 +2935,32 @@ multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode, itins.d, Is2Addr>, XD; } -multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode, - SizeItins itins, - bit Is2Addr = 1> { +multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, + SDNode OpNode, SizeItins itins> { +let Predicates = [HasAVX] in { + defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, + VR128, v4f32, f128mem, memopv4f32, + SSEPackedSingle, itins.s, 0>, TB, VEX_4V; + defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, + VR128, v2f64, f128mem, memopv2f64, + SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V; + + defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), + OpNode, VR256, v8f32, f256mem, memopv8f32, + SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L; + defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), + OpNode, VR256, v4f64, f256mem, memopv4f64, + SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L; +} + +let Constraints = "$src1 = $dst" in { defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128, - v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>, - TB; + v4f32, f128mem, memopv4f32, SSEPackedSingle, + itins.s, 1>, TB; defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128, - v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>, - TB, OpSize; + v2f64, f128mem, memopv2f64, SSEPackedDouble, + itins.d, 1>, TB, OpSize; } - -multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr, - SDNode OpNode, - SizeItins itins> { - defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256, - v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>, - TB, VEX_L; - defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256, - v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>, - TB, OpSize, VEX_L; } multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr, @@ -2881,116 +2974,69 @@ multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr, itins.d, Is2Addr>, XD; } -multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr, - SizeItins itins, - bit Is2Addr = 1> { - defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128, - !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32, - SSEPackedSingle, itins.s, Is2Addr>, - TB; - - defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128, - !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64, - SSEPackedDouble, itins.d, Is2Addr>, - TB, OpSize; +// Binary Arithmetic instructions +defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>; +defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>; +let isCommutable = 0 in { + defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>; + defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>; + defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>; + defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>; } -multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr, - SizeItins itins> { - defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256, - !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32, - SSEPackedSingle, itins.s, 0>, TB, VEX_L; - - defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256, - !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64, - SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_L; +let isCodeGenOnly = 1 in { + defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>; + defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>; } -// Binary Arithmetic instructions defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; -defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>, - VEX_4V; defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>, VEX_4V, VEX_LIG; -defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>, - VEX_4V; let isCommutable = 0 in { defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, - VEX_4V; defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>, - VEX_4V; defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>, - basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>, - VEX_4V; defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>, basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>, - basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>, - VEX_4V; } let Constraints = "$src1 = $dst" in { defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>, basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>; defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>, - basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>, basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>; let isCommutable = 0 in { defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>; defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>, - basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>, basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>; defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>, - basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>; + basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>; defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>, - basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>; + basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>; } } let isCodeGenOnly = 1 in { defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>, VEX_4V; defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>, VEX_4V, VEX_LIG; - defm VMINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P, 0>, - basic_sse12_fp_binop_p_y<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>, VEX_4V; let Constraints = "$src1 = $dst" in { - defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>; - defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>, - basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>; + defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>; + defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>; } } @@ -3002,6 +3048,7 @@ let isCodeGenOnly = 1 in { /// /// And, we have a special variant form for a full-vector intrinsic form. +let Sched = WriteFSqrt in { def SSE_SQRTP : OpndItins< IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM >; @@ -3009,7 +3056,9 @@ def SSE_SQRTP : OpndItins< def SSE_SQRTS : OpndItins< IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM >; +} +let Sched = WriteFRcp in { def SSE_RCPP : OpndItins< IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM >; @@ -3017,13 +3066,36 @@ def SSE_RCPP : OpndItins< def SSE_RCPS : OpndItins< IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM >; +} /// sse1_fp_unop_s - SSE1 unops in scalar form. multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr, SDNode OpNode, Intrinsic F32Int, OpndItins itins> { +let Predicates = [HasAVX], hasSideEffects = 0 in { + def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), + (ins FR32:$src1, FR32:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>; + let mayLoad = 1 in { + def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), + (ins FR32:$src1,f32mem:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, ssmem:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + } +} + def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src), !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), - [(set FR32:$dst, (OpNode FR32:$src))]>; + [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>; // For scalar unary operations, fold a load into the operation // only in OptForSize mode. It eliminates an instruction, but it also // eliminates a whole-register clobber (the load), so it introduces a @@ -3031,204 +3103,238 @@ multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr, def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src), !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS, - Requires<[UseSSE1, OptForSize]>; + Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>; def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>; + [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>, + Sched<[itins.Sched]>; def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src), !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>; + [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>, + Sched<[itins.Sched.Folded]>; } -/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form. -multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> { - def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2), - !strconcat(OpcodeStr, - "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; +/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand. +multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode, + OpndItins itins> { +let Predicates = [HasAVX], hasSideEffects = 0 in { + def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), + (ins FR32:$src1, FR32:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>; let mayLoad = 1 in { - def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2), - !strconcat(OpcodeStr, - "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; - def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, ssmem:$src2), - !strconcat(OpcodeStr, - "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; + def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), + (ins FR32:$src1,f32mem:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, ssmem:$src2), + !strconcat("v", OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + } +} + + def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src), + !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), + [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>; + // For scalar unary operations, fold a load into the operation + // only in OptForSize mode. It eliminates an instruction, but it also + // eliminates a whole-register clobber (the load), so it introduces a + // partial register update condition. + def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src), + !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), + [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS, + Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>; + let Constraints = "$src1 = $dst" in { + def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, VR128:$src2), + !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"), + [], itins.rr>, Sched<[itins.Sched]>; + let mayLoad = 1, hasSideEffects = 0 in + def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, ssmem:$src2), + !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"), + [], itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>; } } /// sse1_fp_unop_p - SSE1 unops in packed form. multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode, OpndItins itins> { - def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>; - def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), - !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>; +let Predicates = [HasAVX] in { + def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], + itins.rr>, VEX, Sched<[itins.Sched]>; + def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], + itins.rm>, VEX, Sched<[itins.Sched.Folded]>; + def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))], + itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>; + def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))], + itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>; } -/// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form. -multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode, - OpndItins itins> { - def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), - !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))], - itins.rr>, VEX_L; - def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), + def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>, + Sched<[itins.Sched]>; + def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))], - itins.rm>, VEX_L; + [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>, + Sched<[itins.Sched.Folded]>; } /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms. multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr, - Intrinsic V4F32Int, OpndItins itins> { + Intrinsic V4F32Int, Intrinsic V8F32Int, + OpndItins itins> { +let Predicates = [HasAVX] in { + def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (V4F32Int VR128:$src))], + itins.rr>, VEX, Sched<[itins.Sched]>; + def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))], + itins.rm>, VEX, Sched<[itins.Sched.Folded]>; + def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (V8F32Int VR256:$src))], + itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>; + def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), + (ins f256mem:$src), + !strconcat("v", OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))], + itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>; +} + def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (V4F32Int VR128:$src))], - itins.rr>; + itins.rr>, Sched<[itins.Sched]>; def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))], - itins.rm>; -} - -/// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms. -multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr, - Intrinsic V4F32Int, OpndItins itins> { - def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), - !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (V4F32Int VR256:$src))], - itins.rr>, VEX_L; - def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), - !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))], - itins.rm>, VEX_L; + itins.rm>, Sched<[itins.Sched.Folded]>; } /// sse2_fp_unop_s - SSE2 unops in scalar form. multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr, SDNode OpNode, Intrinsic F64Int, OpndItins itins> { +let Predicates = [HasAVX], hasSideEffects = 0 in { + def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), + (ins FR64:$src1, FR64:$src2), + !strconcat("v", OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>; + let mayLoad = 1 in { + def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), + (ins FR64:$src1,f64mem:$src2), + !strconcat("v", OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), + (ins VR128:$src1, sdmem:$src2), + !strconcat("v", OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, VEX_4V, VEX_LIG, + Sched<[itins.Sched.Folded, ReadAfterLd]>; + } +} + def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src), !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"), - [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>; + [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>, + Sched<[itins.Sched]>; // See the comments in sse1_fp_unop_s for why this is OptForSize. def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src), !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"), [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD, - Requires<[UseSSE2, OptForSize]>; + Requires<[UseSSE2, OptForSize]>, Sched<[itins.Sched.Folded]>; def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>; + [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>, + Sched<[itins.Sched]>; def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src), !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>; -} - -/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form. -let hasSideEffects = 0 in -multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> { - def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2), - !strconcat(OpcodeStr, - "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; - let mayLoad = 1 in { - def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2), - !strconcat(OpcodeStr, - "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; - def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, sdmem:$src2), - !strconcat(OpcodeStr, - "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; - } + [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>, + Sched<[itins.Sched.Folded]>; } /// sse2_fp_unop_p - SSE2 unops in vector forms. multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode, OpndItins itins> { +let Predicates = [HasAVX] in { + def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), + !strconcat("v", OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], + itins.rr>, VEX, Sched<[itins.Sched]>; + def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), + !strconcat("v", OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], + itins.rm>, VEX, Sched<[itins.Sched.Folded]>; + def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), + !strconcat("v", OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))], + itins.rr>, VEX, VEX_L, Sched<[itins.Sched]>; + def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), + !strconcat("v", OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))], + itins.rm>, VEX, VEX_L, Sched<[itins.Sched.Folded]>; +} + def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>; + [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>, + Sched<[itins.Sched]>; def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>; + [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>, + Sched<[itins.Sched.Folded]>; } -/// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms. -multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode, - OpndItins itins> { - def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))], - itins.rr>, VEX_L; - def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))], - itins.rm>, VEX_L; -} - -/// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms. -multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr, - Intrinsic V2F64Int, OpndItins itins> { - def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (V2F64Int VR128:$src))], - itins.rr>; - def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))], - itins.rm>; -} - -/// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms. -multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr, - Intrinsic V2F64Int, OpndItins itins> { - def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (V2F64Int VR256:$src))], - itins.rr>, VEX_L; - def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), - !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))], - itins.rm>, VEX_L; -} +// Square root. +defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss, + SSE_SQRTS>, + sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>, + sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd, + SSE_SQRTS>, + sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>; -let Predicates = [HasAVX] in { - // Square root. - defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">, - sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG; - - defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>, - sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>, - sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>, - sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>, - sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps, - SSE_SQRTP>, - sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd, - SSE_SQRTP>, - sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256, - SSE_SQRTP>, - sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256, - SSE_SQRTP>, - VEX; - - // Reciprocal approximations. Note that these typically require refinement - // in order to obtain suitable precision. - defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG; - defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>, - sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>, - sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256, - SSE_SQRTP>, - sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps, - SSE_SQRTP>, VEX; - - defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG; - defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>, - sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>, - sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256, - SSE_RCPP>, - sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps, - SSE_RCPP>, VEX; -} +// Reciprocal approximations. Note that these typically require refinement +// in order to obtain suitable precision. +defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>, + sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>, + sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps, + int_x86_avx_rsqrt_ps_256, SSE_SQRTP>; +defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>, + sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>, + sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, + int_x86_avx_rcp_ps_256, SSE_RCPP>; def : Pat<(f32 (fsqrt FR32:$src)), (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>; @@ -3283,59 +3389,11 @@ let Predicates = [HasAVX] in { (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>; } -// Square root. -defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss, - SSE_SQRTS>, - sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>, - sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>, - sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd, - SSE_SQRTS>, - sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>, - sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>; - -/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand. -multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode, - Intrinsic F32Int, OpndItins itins> { - def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src), - !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), - [(set FR32:$dst, (OpNode FR32:$src))]>; - // For scalar unary operations, fold a load into the operation - // only in OptForSize mode. It eliminates an instruction, but it also - // eliminates a whole-register clobber (the load), so it introduces a - // partial register update condition. - def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src), - !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"), - [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS, - Requires<[UseSSE1, OptForSize]>; - let Constraints = "$src1 = $dst" in { - def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), - (ins VR128:$src1, VR128:$src2), - !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"), - [], itins.rr>; - def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), - (ins VR128:$src1, ssmem:$src2), - !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"), - [], itins.rm>; - } -} - // Reciprocal approximations. Note that these typically require refinement // in order to obtain suitable precision. -defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss, - SSE_SQRTS>, - sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>, - sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps, - SSE_SQRTS>; let Predicates = [UseSSE1] in { def : Pat<(int_x86_sse_rsqrt_ss VR128:$src), (RSQRTSSr_Int VR128:$src, VR128:$src)>; -} - -defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss, - SSE_RCPS>, - sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>, - sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>; -let Predicates = [UseSSE1] in { def : Pat<(int_x86_sse_rcp_ss VR128:$src), (RCPSSr_Int VR128:$src, VR128:$src)>; } @@ -3347,52 +3405,48 @@ let Predicates = [UseSSE1] in { //===----------------------------------------------------------------------===// let AddedComplexity = 400 in { // Prefer non-temporal versions - def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs), - (ins f128mem:$dst, VR128:$src), - "movntps\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v4f32 VR128:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX; - def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs), - (ins f128mem:$dst, VR128:$src), - "movntpd\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v2f64 VR128:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX; - - let ExeDomain = SSEPackedInt in - def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs), - (ins f128mem:$dst, VR128:$src), - "movntdq\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v2i64 VR128:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX; - - def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), - (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>; - - def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs), - (ins f256mem:$dst, VR256:$src), - "movntps\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v8f32 VR256:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX, VEX_L; - def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs), - (ins f256mem:$dst, VR256:$src), - "movntpd\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v4f64 VR256:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX, VEX_L; - let ExeDomain = SSEPackedInt in - def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs), - (ins f256mem:$dst, VR256:$src), - "movntdq\t{$src, $dst|$dst, $src}", - [(alignednontemporalstore (v4i64 VR256:$src), - addr:$dst)], - IIC_SSE_MOVNT>, VEX, VEX_L; -} +let SchedRW = [WriteStore] in { +def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs), + (ins f128mem:$dst, VR128:$src), + "movntps\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v4f32 VR128:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX; +def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs), + (ins f128mem:$dst, VR128:$src), + "movntpd\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v2f64 VR128:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX; + +let ExeDomain = SSEPackedInt in +def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs), + (ins f128mem:$dst, VR128:$src), + "movntdq\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v2i64 VR128:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX; + +def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs), + (ins f256mem:$dst, VR256:$src), + "movntps\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v8f32 VR256:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX, VEX_L; +def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs), + (ins f256mem:$dst, VR256:$src), + "movntpd\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v4f64 VR256:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX, VEX_L; +let ExeDomain = SSEPackedInt in +def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs), + (ins f256mem:$dst, VR256:$src), + "movntdq\t{$src, $dst|$dst, $src}", + [(alignednontemporalstore (v4i64 VR256:$src), + addr:$dst)], + IIC_SSE_MOVNT>, VEX, VEX_L; -let AddedComplexity = 400 in { // Prefer non-temporal versions def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), "movntps\t{$src, $dst|$dst, $src}", [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)], @@ -3408,9 +3462,6 @@ def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src), [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)], IIC_SSE_MOVNT>; -def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), - (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>; - // There is no AVX form for instructions below this point def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src), "movnti{l}\t{$src, $dst|$dst, $src}", @@ -3422,14 +3473,21 @@ def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src), [(nontemporalstore (i64 GR64:$src), addr:$dst)], IIC_SSE_MOVNT>, TB, Requires<[HasSSE2]>; -} +} // SchedRW = [WriteStore] + +def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), + (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>; + +def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst), + (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>; +} // AddedComplexity //===----------------------------------------------------------------------===// // SSE 1 & 2 - Prefetch and memory fence //===----------------------------------------------------------------------===// // Prefetch intrinsic. -let Predicates = [HasSSE1] in { +let Predicates = [HasSSE1], SchedRW = [WriteLoad] in { def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src), "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))], IIC_SSE_PREFETCH>, TB; @@ -3444,6 +3502,8 @@ def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src), IIC_SSE_PREFETCH>, TB; } +// FIXME: How should these memory instructions be modeled? +let SchedRW = [WriteLoad] in { // Flush cache def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src), "clflush\t$src", [(int_x86_sse2_clflush addr:$src)], @@ -3463,6 +3523,7 @@ def LFENCE : I<0xAE, MRM_E8, (outs), (ins), def MFENCE : I<0xAE, MRM_F0, (outs), (ins), "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>, TB, Requires<[HasSSE2]>; +} // SchedRW def : Pat<(X86SFence), (SFENCE)>; def : Pat<(X86LFence), (LFENCE)>; @@ -3474,17 +3535,17 @@ def : Pat<(X86MFence), (MFENCE)>; def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src), "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)], - IIC_SSE_LDMXCSR>, VEX; + IIC_SSE_LDMXCSR>, VEX, Sched<[WriteLoad]>; def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst), "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)], - IIC_SSE_STMXCSR>, VEX; + IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>; def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src), "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)], - IIC_SSE_LDMXCSR>; + IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>; def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst), "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)], - IIC_SSE_STMXCSR>; + IIC_SSE_STMXCSR>, Sched<[WriteStore]>; //===---------------------------------------------------------------------===// // SSE2 - Move Aligned/Unaligned Packed Integer Instructions @@ -3492,23 +3553,23 @@ def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst), let ExeDomain = SSEPackedInt in { // SSE integer instructions -let neverHasSideEffects = 1 in { +let neverHasSideEffects = 1, SchedRW = [WriteMove] in { def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, VEX; def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, VEX, VEX_L; -} def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>, VEX; def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>, VEX, VEX_L; +} // For Disassembler -let isCodeGenOnly = 1 in { +let isCodeGenOnly = 1, hasSideEffects = 0, SchedRW = [WriteMove] in { def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>, @@ -3525,7 +3586,8 @@ def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src), IIC_SSE_MOVU_P_RR>, VEX, VEX_L; } -let canFoldAsLoad = 1, mayLoad = 1 in { +let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1, + neverHasSideEffects = 1, SchedRW = [WriteLoad] in { def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>, VEX; @@ -3542,7 +3604,7 @@ let Predicates = [HasAVX] in { } } -let mayStore = 1 in { +let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in { def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>, @@ -3561,6 +3623,7 @@ def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src), } } +let SchedRW = [WriteMove] in { let neverHasSideEffects = 1 in def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>; @@ -3570,7 +3633,7 @@ def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>; // For Disassembler -let isCodeGenOnly = 1 in { +let isCodeGenOnly = 1, hasSideEffects = 0 in { def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>; @@ -3579,8 +3642,10 @@ def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src), "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>; } +} // SchedRW -let canFoldAsLoad = 1, mayLoad = 1 in { +let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1, + neverHasSideEffects = 1, SchedRW = [WriteLoad] in { def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "movdqa\t{$src, $dst|$dst, $src}", [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/], @@ -3592,7 +3657,7 @@ def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), XS, Requires<[UseSSE2]>; } -let mayStore = 1 in { +let mayStore = 1, SchedRW = [WriteStore] in { def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), "movdqa\t{$src, $dst|$dst, $src}", [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/], @@ -3604,30 +3669,23 @@ def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), XS, Requires<[UseSSE2]>; } -// Intrinsic forms of MOVDQU load and store -def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), - "vmovdqu\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)], - IIC_SSE_MOVU_P_MR>, - XS, VEX, Requires<[HasAVX]>; - -def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src), - "movdqu\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)], - IIC_SSE_MOVU_P_MR>, - XS, Requires<[UseSSE2]>; - } // ExeDomain = SSEPackedInt let Predicates = [HasAVX] in { + def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src), + (VMOVDQUmr addr:$dst, VR128:$src)>; def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src), (VMOVDQUYmr addr:$dst, VR256:$src)>; } +let Predicates = [UseSSE2] in +def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src), + (MOVDQUmr addr:$dst, VR128:$src)>; //===---------------------------------------------------------------------===// // SSE2 - Packed Integer Arithmetic Instructions //===---------------------------------------------------------------------===// +let Sched = WriteVecIMul in def SSE_PMADD : OpndItins< IIC_SSE_PMADD, IIC_SSE_PMADD >; @@ -3646,14 +3704,33 @@ multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>; + [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm : PDI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))], - itins.rm>; + itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>; +} + +multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128, + Intrinsic IntId256, OpndItins itins, + bit IsCommutable = 0> { +let Predicates = [HasAVX] in + defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128, + VR128, memopv2i64, i128mem, itins, + IsCommutable, 0>, VEX_4V; + +let Constraints = "$src1 = $dst" in + defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64, + i128mem, itins, IsCommutable, 1>; + +let Predicates = [HasAVX2] in + defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256, + VR256, memopv4i64, i256mem, itins, + IsCommutable, 0>, VEX_4V, VEX_L; } multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm, @@ -3669,23 +3746,25 @@ multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))], - itins.rr>; + itins.rr>, Sched<[WriteVecShift]>; def rm : PDI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, i128mem:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (DstVT (OpNode RC:$src1, - (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>; + (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>, + Sched<[WriteVecShiftLd, ReadAfterLd]>; def ri : PDIi8<opc2, ImmForm, (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>; + [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>, + Sched<[WriteVecShift]>; } -/// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types +/// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode, ValueType DstVT, ValueType SrcVT, RegisterClass RC, PatFrag memop_frag, X86MemOperand x86memop, @@ -3697,260 +3776,88 @@ multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>; + [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>, + Sched<[itins.Sched]>; def rm : PDI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), - (bitconvert (memop_frag addr:$src2)))))]>; + (bitconvert (memop_frag addr:$src2)))))]>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } } // ExeDomain = SSEPackedInt -// 128-bit Integer Arithmetic +defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8, + SSE_INTALU_ITINS_P, 1>; +defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16, + SSE_INTALU_ITINS_P, 1>; +defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32, + SSE_INTALU_ITINS_P, 1>; +defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64, + SSE_INTALUQ_ITINS_P, 1>; +defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16, + SSE_INTMUL_ITINS_P, 1>; +defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8, + SSE_INTALU_ITINS_P, 0>; +defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16, + SSE_INTALU_ITINS_P, 0>; +defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32, + SSE_INTALU_ITINS_P, 0>; +defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64, + SSE_INTALUQ_ITINS_P, 0>; +defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8, + SSE_INTALU_ITINS_P, 0>; +defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16, + SSE_INTALU_ITINS_P, 0>; +defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8, + SSE_INTALU_ITINS_P, 1>; +defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16, + SSE_INTALU_ITINS_P, 1>; +defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8, + SSE_INTALU_ITINS_P, 1>; +defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16, + SSE_INTALU_ITINS_P, 1>; -let Predicates = [HasAVX] in { -defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>, - VEX_4V; -defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64, - i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V; -defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64, - i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V; -defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64, - i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V; +// Intrinsic forms +defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b, + int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>; +defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w, + int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>; +defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b, + int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>; +defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w, + int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>; +defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b, + int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>; +defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w, + int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>; +defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, + int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>; +defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, + int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>; +defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, + int_x86_avx2_pmadd_wd, SSE_PMADD, 1>; +defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b, + int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>; +defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w, + int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>; +defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw, + int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>; + +let Predicates = [HasAVX] in defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128, memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V; - -// Intrinsic forms -defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, - VR128, memopv2i64, i128mem, - SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V; -defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, - VR128, memopv2i64, i128mem, - SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V; -defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, - VR128, memopv2i64, i128mem, - SSE_PMADD, 1, 0>, VEX_4V; -defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; -} - -let Predicates = [HasAVX2] in { -defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64, - i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64, - i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64, - i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64, - i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V, VEX_L; +let Predicates = [HasAVX2] in defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32, VR256, memopv4i64, i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L; - -// Intrinsic forms -defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w, - VR256, memopv4i64, i256mem, - SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w, - VR256, memopv4i64, i256mem, - SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd, - VR256, memopv4i64, i256mem, - SSE_PMADD, 1, 0>, VEX_4V, VEX_L; -defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; -} - -let Constraints = "$src1 = $dst" in { -defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1>; -defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1>; -defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P, 1>; -defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64, - i128mem, SSE_INTALUQ_ITINS_P, 1>; -defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64, - i128mem, SSE_INTMUL_ITINS_P, 1>; -defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P>; -defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P>; -defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64, - i128mem, SSE_INTALU_ITINS_P>; -defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64, - i128mem, SSE_INTALUQ_ITINS_P>; +let Constraints = "$src1 = $dst" in defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128, memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>; -// Intrinsic forms -defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, - VR128, memopv2i64, i128mem, - SSE_INTMUL_ITINS_P, 1>; -defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, - VR128, memopv2i64, i128mem, - SSE_INTMUL_ITINS_P, 1>; -defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, - VR128, memopv2i64, i128mem, - SSE_PMADD, 1>; -defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; -defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; - -} // Constraints = "$src1 = $dst" - //===---------------------------------------------------------------------===// // SSE2 - Packed Integer Logical Instructions //===---------------------------------------------------------------------===// @@ -3983,7 +3890,7 @@ defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai, VR128, v4i32, v4i32, bc_v4i32, SSE_INTSHIFT_ITINS_P, 0>, VEX_4V; -let ExeDomain = SSEPackedInt in { +let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in { // 128-bit logical shifts. def VPSLLDQri : PDIi8<0x73, MRM7r, (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2), @@ -4029,7 +3936,7 @@ defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai, VR256, v8i32, v4i32, bc_v4i32, SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L; -let ExeDomain = SSEPackedInt in { +let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in { // 256-bit logical shifts. def VPSLLDQYri : PDIi8<0x73, MRM7r, (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2), @@ -4075,7 +3982,7 @@ defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai, VR128, v4i32, v4i32, bc_v4i32, SSE_INTSHIFT_ITINS_P>; -let ExeDomain = SSEPackedInt in { +let ExeDomain = SSEPackedInt, SchedRW = [WriteVecShift] in { // 128-bit logical shifts. def PSLLDQri : PDIi8<0x73, MRM7r, (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2), @@ -4132,186 +4039,109 @@ let Predicates = [UseSSE2] in { // SSE2 - Packed Integer Comparison Instructions //===---------------------------------------------------------------------===// -let Predicates = [HasAVX] in { - defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; - defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; - defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V; - defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; - defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; - defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -} - -let Predicates = [HasAVX2] in { - defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; - defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; - defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 1, 0>, VEX_4V, VEX_L; - defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; - defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; - defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -} - -let Constraints = "$src1 = $dst" in { - defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; - defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; - defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 1>; - defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; - defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; - defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -} // Constraints = "$src1 = $dst" +defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8, + SSE_INTALU_ITINS_P, 1>; +defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16, + SSE_INTALU_ITINS_P, 1>; +defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32, + SSE_INTALU_ITINS_P, 1>; +defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8, + SSE_INTALU_ITINS_P, 0>; +defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16, + SSE_INTALU_ITINS_P, 0>; +defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32, + SSE_INTALU_ITINS_P, 0>; //===---------------------------------------------------------------------===// // SSE2 - Packed Integer Pack Instructions //===---------------------------------------------------------------------===// -let Predicates = [HasAVX] in { -defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V; -} - -let Predicates = [HasAVX2] in { -defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb, - VR256, memopv4i64, i256mem, - SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L; -} - -let Constraints = "$src1 = $dst" in { -defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128, - VR128, memopv2i64, i128mem, - SSE_INTALU_ITINS_P>; -} // Constraints = "$src1 = $dst" +defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128, + int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>; +defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128, + int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>; +defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128, + int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>; //===---------------------------------------------------------------------===// // SSE2 - Packed Integer Shuffle Instructions //===---------------------------------------------------------------------===// let ExeDomain = SSEPackedInt in { -multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> { -def ri : Ii8<0x70, MRMSrcReg, - (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))], - IIC_SSE_PSHUF>; -def mi : Ii8<0x70, MRMSrcMem, - (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR128:$dst, - (vt (OpNode (bitconvert (memopv2i64 addr:$src1)), - (i8 imm:$src2))))], - IIC_SSE_PSHUF>; -} - -multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> { -def Yri : Ii8<0x70, MRMSrcReg, - (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>; -def Ymi : Ii8<0x70, MRMSrcMem, - (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2), - !strconcat(OpcodeStr, - "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, - (vt (OpNode (bitconvert (memopv4i64 addr:$src1)), - (i8 imm:$src2))))]>; -} -} // ExeDomain = SSEPackedInt - +multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256, + SDNode OpNode> { let Predicates = [HasAVX] in { - let AddedComplexity = 5 in - defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX; - - // SSE2 with ImmT == Imm8 and XS prefix. - defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX; - - // SSE2 with ImmT == Imm8 and XD prefix. - defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX; - - def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))), - (VPSHUFDmi addr:$src1, imm:$imm)>; - def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))), - (VPSHUFDri VR128:$src1, imm:$imm)>; + def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst), + (ins VR128:$src1, i8imm:$src2), + !strconcat("v", OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, + (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))], + IIC_SSE_PSHUF>, VEX, Sched<[WriteShuffle]>; + def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst), + (ins i128mem:$src1, i8imm:$src2), + !strconcat("v", OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, + (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)), + (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, + Sched<[WriteShuffleLd]>; } let Predicates = [HasAVX2] in { - defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, - TB, OpSize, VEX,VEX_L; - defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, - XS, VEX, VEX_L; - defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, - XD, VEX, VEX_L; + def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst), + (ins VR256:$src1, i8imm:$src2), + !strconcat("v", OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR256:$dst, + (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))], + IIC_SSE_PSHUF>, VEX, VEX_L, Sched<[WriteShuffle]>; + def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst), + (ins i256mem:$src1, i8imm:$src2), + !strconcat("v", OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR256:$dst, + (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)), + (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, VEX_L, + Sched<[WriteShuffleLd]>; } let Predicates = [UseSSE2] in { - let AddedComplexity = 5 in - defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize; + def ri : Ii8<0x70, MRMSrcReg, + (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, + (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))], + IIC_SSE_PSHUF>, Sched<[WriteShuffle]>; + def mi : Ii8<0x70, MRMSrcMem, + (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128:$dst, + (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)), + (i8 imm:$src2))))], IIC_SSE_PSHUF>, + Sched<[WriteShuffleLd]>; +} +} +} // ExeDomain = SSEPackedInt - // SSE2 with ImmT == Imm8 and XS prefix. - defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS; +defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize; +defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS; +defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD; - // SSE2 with ImmT == Imm8 and XD prefix. - defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD; +let Predicates = [HasAVX] in { + def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))), + (VPSHUFDmi addr:$src1, imm:$imm)>; + def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))), + (VPSHUFDri VR128:$src1, imm:$imm)>; +} - def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))), - (PSHUFDmi addr:$src1, imm:$imm)>; - def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))), - (PSHUFDri VR128:$src1, imm:$imm)>; +let Predicates = [UseSSE2] in { + def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))), + (PSHUFDmi addr:$src1, imm:$imm)>; + def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))), + (PSHUFDri VR128:$src1, imm:$imm)>; } //===---------------------------------------------------------------------===// @@ -4327,7 +4157,7 @@ multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt, !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))], - IIC_SSE_UNPCK>; + IIC_SSE_UNPCK>, Sched<[WriteShuffle]>; def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2), !if(Is2Addr, @@ -4336,7 +4166,8 @@ multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt, [(set VR128:$dst, (OpNode VR128:$src1, (bc_frag (memopv2i64 addr:$src2))))], - IIC_SSE_UNPCK>; + IIC_SSE_UNPCK>, + Sched<[WriteShuffleLd, ReadAfterLd]>; } multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt, @@ -4344,12 +4175,14 @@ multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt, def Yrr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src1, VR256:$src2), !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"), - [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>; + [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>, + Sched<[WriteShuffle]>; def Yrm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2), !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"), [(set VR256:$dst, (OpNode VR256:$src1, - (bc_frag (memopv4i64 addr:$src2))))]>; + (bc_frag (memopv4i64 addr:$src2))))]>, + Sched<[WriteShuffleLd, ReadAfterLd]>; } let Predicates = [HasAVX] in { @@ -4426,7 +4259,8 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> { "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}", "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), [(set VR128:$dst, - (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>; + (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>, + Sched<[WriteShuffle]>; def rmi : Ii8<0xC4, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i16mem:$src2, i32i8imm:$src3), @@ -4435,7 +4269,8 @@ multiclass sse2_pinsrw<bit Is2Addr = 1> { "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), [(set VR128:$dst, (X86pinsrw VR128:$src1, (extloadi16 addr:$src2), - imm:$src3))], IIC_SSE_PINSRW>; + imm:$src3))], IIC_SSE_PINSRW>, + Sched<[WriteShuffleLd, ReadAfterLd]>; } // Extract @@ -4444,12 +4279,14 @@ def VPEXTRWri : Ii8<0xC5, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2), "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1), - imm:$src2))]>, TB, OpSize, VEX; + imm:$src2))]>, TB, OpSize, VEX, + Sched<[WriteShuffle]>; def PEXTRWri : PDIi8<0xC5, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2), "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}", [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1), - imm:$src2))], IIC_SSE_PEXTRW>; + imm:$src2))], IIC_SSE_PEXTRW>, + Sched<[WriteShuffleLd, ReadAfterLd]>; // Insert let Predicates = [HasAVX] in { @@ -4457,7 +4294,7 @@ let Predicates = [HasAVX] in { def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, GR64:$src2, i32i8imm:$src3), "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", - []>, TB, OpSize, VEX_4V; + []>, TB, OpSize, VEX_4V, Sched<[WriteShuffle]>; } let Constraints = "$src1 = $dst" in @@ -4469,7 +4306,7 @@ let Constraints = "$src1 = $dst" in // SSE2 - Packed Mask Creation //===---------------------------------------------------------------------===// -let ExeDomain = SSEPackedInt in { +let ExeDomain = SSEPackedInt, SchedRW = [WriteVecLogic] in { def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src), "pmovmskb\t{$src, $dst|$dst, $src}", @@ -4497,7 +4334,7 @@ def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src), // SSE2 - Conditional Store //===---------------------------------------------------------------------===// -let ExeDomain = SSEPackedInt in { +let ExeDomain = SSEPackedInt, SchedRW = [WriteStore] in { let Uses = [EDI] in def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs), @@ -4536,41 +4373,42 @@ def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>, - VEX; + VEX, Sched<[WriteMove]>; def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v4i32 (scalar_to_vector (loadi32 addr:$src))))], IIC_SSE_MOVDQ>, - VEX; + VEX, Sched<[WriteLoad]>; def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2i64 (scalar_to_vector GR64:$src)))], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>; def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (bitconvert GR64:$src))], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>; def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, - (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>; + (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>, + Sched<[WriteMove]>; def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v4i32 (scalar_to_vector (loadi32 addr:$src))))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteLoad]>; def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (v2i64 (scalar_to_vector GR64:$src)))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteMove]>; def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (bitconvert GR64:$src))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteMove]>; //===---------------------------------------------------------------------===// // Move Int Doubleword to Single Scalar @@ -4578,22 +4416,22 @@ def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src), def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (bitconvert GR32:$src))], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>; def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))], IIC_SSE_MOVDQ>, - VEX; + VEX, Sched<[WriteLoad]>; def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (bitconvert GR32:$src))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteMove]>; def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteLoad]>; //===---------------------------------------------------------------------===// // Move Packed Doubleword Int to Packed Double Int @@ -4601,26 +4439,29 @@ def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src), def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (vector_extract (v4i32 VR128:$src), - (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX; + (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX, + Sched<[WriteMove]>; def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (vector_extract (v4i32 VR128:$src), (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>, - VEX; + VEX, Sched<[WriteLoad]>; def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (vector_extract (v4i32 VR128:$src), - (iPTR 0)))], IIC_SSE_MOVD_ToGP>; + (iPTR 0)))], IIC_SSE_MOVD_ToGP>, + Sched<[WriteMove]>; def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (vector_extract (v4i32 VR128:$src), (iPTR 0))), addr:$dst)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteLoad]>; //===---------------------------------------------------------------------===// // Move Packed Doubleword Int first element to Doubleword Int // +let SchedRW = [WriteMove] in { def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), "vmov{d|q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (vector_extract (v2i64 VR128:$src), @@ -4633,6 +4474,7 @@ def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), [(set GR64:$dst, (vector_extract (v2i64 VR128:$src), (iPTR 0)))], IIC_SSE_MOVD_ToGP>; +} //SchedRW //===---------------------------------------------------------------------===// // Bitcast FR64 <-> GR64 @@ -4641,28 +4483,28 @@ let Predicates = [HasAVX] in def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src), "vmovq\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>, - VEX; + VEX, Sched<[WriteLoad]>; def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (bitconvert FR64:$src))], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteMove]>; def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src), "movq\t{$src, $dst|$dst, $src}", [(store (i64 (bitconvert FR64:$src)), addr:$dst)], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>; def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src), "movq\t{$src, $dst|$dst, $src}", [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteLoad]>; def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src), "mov{d|q}\t{$src, $dst|$dst, $src}", [(set GR64:$dst, (bitconvert FR64:$src))], - IIC_SSE_MOVD_ToGP>; + IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>; def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src), "movq\t{$src, $dst|$dst, $src}", [(store (i64 (bitconvert FR64:$src)), addr:$dst)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteStore]>; //===---------------------------------------------------------------------===// // Move Scalar Single to Double Int @@ -4670,23 +4512,24 @@ def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src), def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (bitconvert FR32:$src))], - IIC_SSE_MOVD_ToGP>, VEX; + IIC_SSE_MOVD_ToGP>, VEX, Sched<[WriteMove]>; def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (bitconvert FR32:$src)), addr:$dst)], - IIC_SSE_MOVDQ>, VEX; + IIC_SSE_MOVDQ>, VEX, Sched<[WriteStore]>; def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(set GR32:$dst, (bitconvert FR32:$src))], - IIC_SSE_MOVD_ToGP>; + IIC_SSE_MOVD_ToGP>, Sched<[WriteMove]>; def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src), "movd\t{$src, $dst|$dst, $src}", [(store (i32 (bitconvert FR32:$src)), addr:$dst)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteStore]>; //===---------------------------------------------------------------------===// // Patterns and instructions to describe movd/movq to XMM register zero-extends // +let SchedRW = [WriteMove] in { let AddedComplexity = 15 in { def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src), "movd\t{$src, $dst|$dst, $src}", @@ -4712,8 +4555,9 @@ def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), (v2i64 (scalar_to_vector GR64:$src)))))], IIC_SSE_MOVDQ>; } +} // SchedRW -let AddedComplexity = 20 in { +let AddedComplexity = 20, SchedRW = [WriteLoad] in { def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), "movd\t{$src, $dst|$dst, $src}", [(set VR128:$dst, @@ -4726,7 +4570,7 @@ def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src), (v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))))], IIC_SSE_MOVDQ>; -} +} // AddedComplexity, SchedRW let Predicates = [HasAVX] in { // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part. @@ -4775,6 +4619,8 @@ def : InstAlias<"movq\t{$src, $dst|$dst, $src}", //===---------------------------------------------------------------------===// // Move Quadword Int to Packed Quadword Int // + +let SchedRW = [WriteLoad] in { def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), "vmovq\t{$src, $dst|$dst, $src}", [(set VR128:$dst, @@ -4786,10 +4632,12 @@ def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (scalar_to_vector (loadi64 addr:$src))))], IIC_SSE_MOVDQ>, XS, Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix +} // SchedRW //===---------------------------------------------------------------------===// // Move Packed Quadword Int to Quadword Int // +let SchedRW = [WriteStore] in { def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", [(store (i64 (vector_extract (v2i64 VR128:$src), @@ -4800,17 +4648,19 @@ def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), [(store (i64 (vector_extract (v2i64 VR128:$src), (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>; +} // SchedRW //===---------------------------------------------------------------------===// // Store / copy lower 64-bits of a XMM register. // def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", - [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX; + [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX, + Sched<[WriteStore]>; def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src), "movq\t{$src, $dst|$dst, $src}", [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)], - IIC_SSE_MOVDQ>; + IIC_SSE_MOVDQ>, Sched<[WriteStore]>; let AddedComplexity = 20 in def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), @@ -4819,7 +4669,7 @@ def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (X86vzmovl (v2i64 (scalar_to_vector (loadi64 addr:$src))))))], IIC_SSE_MOVDQ>, - XS, VEX, Requires<[HasAVX]>; + XS, VEX, Requires<[HasAVX]>, Sched<[WriteLoad]>; let AddedComplexity = 20 in def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), @@ -4828,7 +4678,7 @@ def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src), (v2i64 (X86vzmovl (v2i64 (scalar_to_vector (loadi64 addr:$src))))))], IIC_SSE_MOVDQ>, - XS, Requires<[UseSSE2]>; + XS, Requires<[UseSSE2]>, Sched<[WriteLoad]>; let Predicates = [HasAVX], AddedComplexity = 20 in { def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))), @@ -4858,6 +4708,7 @@ def : Pat<(v4i64 (X86vzload addr:$src)), // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in // IA32 document. movq xmm1, xmm2 does clear the high bits. // +let SchedRW = [WriteVecLogic] in { let AddedComplexity = 15 in def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vmovq\t{$src, $dst|$dst, $src}", @@ -4870,7 +4721,9 @@ def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))], IIC_SSE_MOVQ_RR>, XS, Requires<[UseSSE2]>; +} // SchedRW +let SchedRW = [WriteVecLogicLd] in { let AddedComplexity = 20 in def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "vmovq\t{$src, $dst|$dst, $src}", @@ -4886,6 +4739,7 @@ def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), IIC_SSE_MOVDQ>, XS, Requires<[UseSSE2]>; } +} // SchedRW let AddedComplexity = 20 in { let Predicates = [HasAVX] in { @@ -4903,6 +4757,7 @@ let AddedComplexity = 20 in { } // Instructions to match in the assembler +let SchedRW = [WriteMove] in { def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src), "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVDQ>, VEX, VEX_W; @@ -4913,16 +4768,19 @@ def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src), "movd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVDQ>, VEX, VEX_W; +} // SchedRW // Instructions for the disassembler // xr = XMM register // xm = mem64 +let SchedRW = [WriteMove] in { let Predicates = [HasAVX] in def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS; def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS; +} // SchedRW //===---------------------------------------------------------------------===// // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP @@ -4933,11 +4791,11 @@ multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr, def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set RC:$dst, (vt (OpNode RC:$src)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set RC:$dst, (OpNode (mem_frag addr:$src)))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>; } let Predicates = [HasAVX] in { @@ -4993,25 +4851,27 @@ multiclass sse3_replicate_dfp<string OpcodeStr> { let neverHasSideEffects = 1 in def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [], IIC_SSE_MOV_LH>; + [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>; def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (v2f64 (X86Movddup (scalar_to_vector (loadf64 addr:$src)))))], - IIC_SSE_MOV_LH>; + IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>; } // FIXME: Merge with above classe when there're patterns for the ymm version multiclass sse3_replicate_dfp_y<string OpcodeStr> { def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), - [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>; + [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>, + Sched<[WriteShuffle]>; def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (v4f64 (X86Movddup - (scalar_to_vector (loadf64 addr:$src)))))]>; + (scalar_to_vector (loadf64 addr:$src)))))]>, + Sched<[WriteShuffleLd]>; } let Predicates = [HasAVX] in { @@ -5059,6 +4919,7 @@ let Predicates = [UseSSE3] in { // SSE3 - Move Unaligned Integer //===---------------------------------------------------------------------===// +let SchedRW = [WriteLoad] in { let Predicates = [HasAVX] in { def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "vlddqu\t{$src, $dst|$dst, $src}", @@ -5072,6 +4933,7 @@ def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), "lddqu\t{$src, $dst|$dst, $src}", [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))], IIC_SSE_LDDQU>; +} //===---------------------------------------------------------------------===// // SSE3 - Arithmetic @@ -5085,13 +4947,15 @@ multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>; + [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>, + Sched<[itins.Sched]>; def rm : I<0xD0, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>; + [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } let Predicates = [HasAVX] in { @@ -5128,14 +4992,15 @@ multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>; + [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>, + Sched<[WriteFAdd]>; def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))], - IIC_SSE_HADDSUB_RM>; + IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>; } multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> { @@ -5143,14 +5008,15 @@ multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC, !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), - [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>; + [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>, + Sched<[WriteFAdd]>; def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))], - IIC_SSE_HADDSUB_RM>; + IIC_SSE_HADDSUB_RM>, Sched<[WriteFAddLd, ReadAfterLd]>; } let Predicates = [HasAVX] in { @@ -5199,7 +5065,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr, (ins VR128:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>, - OpSize; + OpSize, Sched<[WriteVecALU]>; def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src), @@ -5207,7 +5073,7 @@ multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr, [(set VR128:$dst, (IntId128 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>, - OpSize; + OpSize, Sched<[WriteVecALULd]>; } /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}. @@ -5217,14 +5083,15 @@ multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr, (ins VR256:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (IntId256 VR256:$src))]>, - OpSize; + OpSize, Sched<[WriteVecALU]>; def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src), !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), [(set VR256:$dst, (IntId256 - (bitconvert (memopv4i64 addr:$src))))]>, OpSize; + (bitconvert (memopv4i64 addr:$src))))]>, OpSize, + Sched<[WriteVecALULd]>; } let Predicates = [HasAVX] in { @@ -5256,6 +5123,7 @@ defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", // SSSE3 - Packed Binary Operator Instructions //===---------------------------------------------------------------------===// +let Sched = WriteVecALU in { def SSE_PHADDSUBD : OpndItins< IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM >; @@ -5265,12 +5133,16 @@ def SSE_PHADDSUBSW : OpndItins< def SSE_PHADDSUBW : OpndItins< IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM >; +} +let Sched = WriteShuffle in def SSE_PSHUFB : OpndItins< IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM >; +let Sched = WriteVecALU in def SSE_PSIGN : OpndItins< IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM >; +let Sched = WriteVecIMul in def SSE_PMULHRSW : OpndItins< IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW >; @@ -5287,7 +5159,7 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>, - OpSize; + OpSize, Sched<[itins.Sched]>; def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2), !if(Is2Addr, @@ -5295,7 +5167,8 @@ multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set RC:$dst, (OpVT (OpNode RC:$src1, - (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize; + (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}. @@ -5309,7 +5182,7 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"), !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, - OpSize; + OpSize, Sched<[itins.Sched]>; def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2), !if(Is2Addr, @@ -5317,7 +5190,8 @@ multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")), [(set VR128:$dst, (IntId128 VR128:$src1, - (bitconvert (memopv2i64 addr:$src2))))]>, OpSize; + (bitconvert (memopv2i64 addr:$src2))))]>, OpSize, + Sched<[itins.Sched.Folded, ReadAfterLd]>; } multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr, @@ -5451,7 +5325,7 @@ defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", // SSSE3 - Packed Align Instruction Patterns //===---------------------------------------------------------------------===// -multiclass ssse3_palign<string asm, bit Is2Addr = 1> { +multiclass ssse3_palignr<string asm, bit Is2Addr = 1> { let neverHasSideEffects = 1 in { def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i8imm:$src3), @@ -5459,7 +5333,7 @@ multiclass ssse3_palign<string asm, bit Is2Addr = 1> { !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [], IIC_SSE_PALIGNR>, OpSize; + [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffle]>; let mayLoad = 1 in def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2, i8imm:$src3), @@ -5467,63 +5341,63 @@ multiclass ssse3_palign<string asm, bit Is2Addr = 1> { !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")), - [], IIC_SSE_PALIGNR>, OpSize; + [], IIC_SSE_PALIGNR>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>; } } -multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> { +multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> { let neverHasSideEffects = 1 in { def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src1, VR256:$src2, i8imm:$src3), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, OpSize; + []>, OpSize, Sched<[WriteShuffle]>; let mayLoad = 1 in def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2, i8imm:$src3), !strconcat(asm, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), - []>, OpSize; + []>, OpSize, Sched<[WriteShuffleLd, ReadAfterLd]>; } } let Predicates = [HasAVX] in - defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V; + defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V; let Predicates = [HasAVX2] in - defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L; + defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L; let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in - defm PALIGN : ssse3_palign<"palignr">; + defm PALIGN : ssse3_palignr<"palignr">; let Predicates = [HasAVX2] in { -def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))), +def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))), (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>; -def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))), +def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))), (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>; -def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))), +def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))), (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>; -def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))), +def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))), (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>; } let Predicates = [HasAVX] in { -def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; } let Predicates = [UseSSSE3] in { -def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; -def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), +def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))), (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>; } @@ -5531,6 +5405,7 @@ def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))), // SSSE3 - Thread synchronization //===---------------------------------------------------------------------===// +let SchedRW = [WriteSystem] in { let usesCustomInserter = 1 in { def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3), [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>, @@ -5544,6 +5419,7 @@ let Uses = [ECX, EAX] in def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>, TB, Requires<[HasSSE3]>; +} // SchedRW def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>; def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>; @@ -5850,6 +5726,55 @@ defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq", defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>; defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>; +let Predicates = [HasAVX2] in { + def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>; + def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>; + def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>; + + def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>; + def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>; + + def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>; + + def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))), + (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))), + (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))), + (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + + def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))), + (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))), + (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + + def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))), + (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>; + + def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))), + (VPMOVSXWDYrm addr:$src)>; + def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))), + (VPMOVSXDQYrm addr:$src)>; + + def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (VPMOVSXBDYrm addr:$src)>; + def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (VPMOVSXBDYrm addr:$src)>; + + def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (VPMOVSXWQYrm addr:$src)>; + def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (VPMOVSXWQYrm addr:$src)>; + + def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32 + (scalar_to_vector (loadi32 addr:$src))))))), + (VPMOVSXBQYrm addr:$src)>; +} + let Predicates = [HasAVX] in { // Common patterns involving scalar load def : Pat<(int_x86_sse41_pmovsxbq @@ -5864,6 +5789,15 @@ let Predicates = [HasAVX] in { } let Predicates = [UseSSE41] in { + def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>; + def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>; + def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>; + + def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>; + def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>; + + def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>; + // Common patterns involving scalar load def : Pat<(int_x86_sse41_pmovsxbq (bitconvert (v4i32 (X86vzmovl @@ -5874,6 +5808,34 @@ let Predicates = [UseSSE41] in { (bitconvert (v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))))), (PMOVZXBQrm addr:$src)>; + + def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (PMOVSXWDrm addr:$src)>; + def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (PMOVSXWDrm addr:$src)>; + def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32 + (scalar_to_vector (loadi32 addr:$src))))))), + (PMOVSXBDrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32 + (scalar_to_vector (loadi32 addr:$src))))))), + (PMOVSXWQrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32 + (scalar_to_vector (extloadi32i16 addr:$src))))))), + (PMOVSXBQrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (PMOVSXDQrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (PMOVSXDQrm addr:$src)>; + def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (PMOVSXBWrm addr:$src)>; + def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (PMOVSXBWrm addr:$src)>; } let Predicates = [HasAVX2] in { @@ -5934,6 +5896,44 @@ let Predicates = [HasAVX] in { (VPMOVZXDQrm addr:$src)>; def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))), (VPMOVZXDQrm addr:$src)>; + + def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>; + def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>; + def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>; + + def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>; + def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>; + + def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>; + + def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (VPMOVSXWDrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (VPMOVSXDQrm addr:$src)>; + def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (VPMOVSXWDrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (VPMOVSXDQrm addr:$src)>; + def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64 + (scalar_to_vector (loadi64 addr:$src))))))), + (VPMOVSXBWrm addr:$src)>; + def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64 + (scalar_to_vector (loadf64 addr:$src))))))), + (VPMOVSXBWrm addr:$src)>; + + def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32 + (scalar_to_vector (loadi32 addr:$src))))))), + (VPMOVSXBDrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32 + (scalar_to_vector (loadi32 addr:$src))))))), + (VPMOVSXWQrm addr:$src)>; + def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32 + (scalar_to_vector (extloadi32i16 addr:$src))))))), + (VPMOVSXBQrm addr:$src)>; } let Predicates = [UseSSE41] in { @@ -6273,6 +6273,7 @@ multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd, Intrinsic F64Int, bit Is2Addr = 1> { let ExeDomain = GenericDomain in { // Operation, reg. + let hasSideEffects = 0 in def SSr : SS4AIi8<opcss, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3), !if(Is2Addr, @@ -6306,6 +6307,7 @@ let ExeDomain = GenericDomain in { OpSize; // Operation, reg. + let hasSideEffects = 0 in def SDr : SS4AIi8<opcsd, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3), !if(Is2Addr, @@ -6378,12 +6380,47 @@ let Predicates = [HasAVX] in { def : Pat<(v4f32 (ffloor VR128:$src)), (VROUNDPSr VR128:$src, (i32 0x1))>; + def : Pat<(v4f32 (fnearbyint VR128:$src)), + (VROUNDPSr VR128:$src, (i32 0xC))>; + def : Pat<(v4f32 (fceil VR128:$src)), + (VROUNDPSr VR128:$src, (i32 0x2))>; + def : Pat<(v4f32 (frint VR128:$src)), + (VROUNDPSr VR128:$src, (i32 0x4))>; + def : Pat<(v4f32 (ftrunc VR128:$src)), + (VROUNDPSr VR128:$src, (i32 0x3))>; + def : Pat<(v2f64 (ffloor VR128:$src)), (VROUNDPDr VR128:$src, (i32 0x1))>; + def : Pat<(v2f64 (fnearbyint VR128:$src)), + (VROUNDPDr VR128:$src, (i32 0xC))>; + def : Pat<(v2f64 (fceil VR128:$src)), + (VROUNDPDr VR128:$src, (i32 0x2))>; + def : Pat<(v2f64 (frint VR128:$src)), + (VROUNDPDr VR128:$src, (i32 0x4))>; + def : Pat<(v2f64 (ftrunc VR128:$src)), + (VROUNDPDr VR128:$src, (i32 0x3))>; + def : Pat<(v8f32 (ffloor VR256:$src)), (VROUNDYPSr VR256:$src, (i32 0x1))>; + def : Pat<(v8f32 (fnearbyint VR256:$src)), + (VROUNDYPSr VR256:$src, (i32 0xC))>; + def : Pat<(v8f32 (fceil VR256:$src)), + (VROUNDYPSr VR256:$src, (i32 0x2))>; + def : Pat<(v8f32 (frint VR256:$src)), + (VROUNDYPSr VR256:$src, (i32 0x4))>; + def : Pat<(v8f32 (ftrunc VR256:$src)), + (VROUNDYPSr VR256:$src, (i32 0x3))>; + def : Pat<(v4f64 (ffloor VR256:$src)), (VROUNDYPDr VR256:$src, (i32 0x1))>; + def : Pat<(v4f64 (fnearbyint VR256:$src)), + (VROUNDYPDr VR256:$src, (i32 0xC))>; + def : Pat<(v4f64 (fceil VR256:$src)), + (VROUNDYPDr VR256:$src, (i32 0x2))>; + def : Pat<(v4f64 (frint VR256:$src)), + (VROUNDYPDr VR256:$src, (i32 0x4))>; + def : Pat<(v4f64 (ftrunc VR256:$src)), + (VROUNDYPDr VR256:$src, (i32 0x3))>; } defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128, @@ -6417,8 +6454,25 @@ let Predicates = [UseSSE41] in { def : Pat<(v4f32 (ffloor VR128:$src)), (ROUNDPSr VR128:$src, (i32 0x1))>; + def : Pat<(v4f32 (fnearbyint VR128:$src)), + (ROUNDPSr VR128:$src, (i32 0xC))>; + def : Pat<(v4f32 (fceil VR128:$src)), + (ROUNDPSr VR128:$src, (i32 0x2))>; + def : Pat<(v4f32 (frint VR128:$src)), + (ROUNDPSr VR128:$src, (i32 0x4))>; + def : Pat<(v4f32 (ftrunc VR128:$src)), + (ROUNDPSr VR128:$src, (i32 0x3))>; + def : Pat<(v2f64 (ffloor VR128:$src)), (ROUNDPDr VR128:$src, (i32 0x1))>; + def : Pat<(v2f64 (fnearbyint VR128:$src)), + (ROUNDPDr VR128:$src, (i32 0xC))>; + def : Pat<(v2f64 (fceil VR128:$src)), + (ROUNDPDr VR128:$src, (i32 0x2))>; + def : Pat<(v2f64 (frint VR128:$src)), + (ROUNDPDr VR128:$src, (i32 0x4))>; + def : Pat<(v2f64 (ftrunc VR128:$src)), + (ROUNDPDr VR128:$src, (i32 0x3))>; } //===----------------------------------------------------------------------===// @@ -6575,67 +6629,6 @@ multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr, (bitconvert (memopv4i64 addr:$src2))))]>, OpSize; } -let Predicates = [HasAVX] in { - let isCommutable = 0 in - defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw, - 0>, VEX_4V; - defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb, - 0>, VEX_4V; - defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd, - 0>, VEX_4V; - defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud, - 0>, VEX_4V; - defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw, - 0>, VEX_4V; - defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb, - 0>, VEX_4V; - defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd, - 0>, VEX_4V; - defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud, - 0>, VEX_4V; - defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw, - 0>, VEX_4V; - defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq, - 0>, VEX_4V; -} - -let Predicates = [HasAVX2] in { - let isCommutable = 0 in - defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw", - int_x86_avx2_packusdw>, VEX_4V, VEX_L; - defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb", - int_x86_avx2_pmins_b>, VEX_4V, VEX_L; - defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd", - int_x86_avx2_pmins_d>, VEX_4V, VEX_L; - defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud", - int_x86_avx2_pminu_d>, VEX_4V, VEX_L; - defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw", - int_x86_avx2_pminu_w>, VEX_4V, VEX_L; - defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb", - int_x86_avx2_pmaxs_b>, VEX_4V, VEX_L; - defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd", - int_x86_avx2_pmaxs_d>, VEX_4V, VEX_L; - defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud", - int_x86_avx2_pmaxu_d>, VEX_4V, VEX_L; - defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw", - int_x86_avx2_pmaxu_w>, VEX_4V, VEX_L; - defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq", - int_x86_avx2_pmul_dq>, VEX_4V, VEX_L; -} - -let Constraints = "$src1 = $dst" in { - let isCommutable = 0 in - defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>; - defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>; - defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>; - defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>; - defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>; - defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>; - defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>; - defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>; - defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>; - defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>; -} /// SS48I_binop_rm - Simple SSE41 binary operator. multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, @@ -6659,6 +6652,76 @@ multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, } let Predicates = [HasAVX] in { + let isCommutable = 0 in + defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw, + 0>, VEX_4V; + defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128, + memopv2i64, i128mem, 0>, VEX_4V; + defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq, + 0>, VEX_4V; +} + +let Predicates = [HasAVX2] in { + let isCommutable = 0 in + defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw", + int_x86_avx2_packusdw>, VEX_4V, VEX_L; + defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256, + memopv4i64, i256mem, 0>, VEX_4V, VEX_L; + defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq", + int_x86_avx2_pmul_dq>, VEX_4V, VEX_L; +} + +let Constraints = "$src1 = $dst" in { + let isCommutable = 0 in + defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>; + defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128, + memopv2i64, i128mem>; + defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128, + memopv2i64, i128mem>; + defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128, + memopv2i64, i128mem>; + defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128, + memopv2i64, i128mem>; + defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128, + memopv2i64, i128mem>; + defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128, + memopv2i64, i128mem>; + defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128, + memopv2i64, i128mem>; + defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128, + memopv2i64, i128mem>; + defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>; +} + +let Predicates = [HasAVX] in { defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128, memopv2i64, i128mem, 0>, VEX_4V; defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128, @@ -6776,7 +6839,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr, !strconcat(OpcodeStr, "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))], - IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; + NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2, RC:$src3), @@ -6785,7 +6848,7 @@ multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr, [(set RC:$dst, (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)), RC:$src3))], - IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; + NoItinerary, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM; } let Predicates = [HasAVX] in { @@ -6839,31 +6902,31 @@ let Predicates = [HasAVX] in { (v4f64 VR256:$src2))), (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>; - def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2), + def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2), (imm:$mask))), - (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>; - def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2), + (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>; + def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2), (imm:$mask))), - (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>; + (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>; - def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2), + def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2), (imm:$mask))), - (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>; - def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2), + (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>; + def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2), (imm:$mask))), - (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>; - def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2), + (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>; + def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2), (imm:$mask))), - (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>; + (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>; } let Predicates = [HasAVX2] in { def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1), (v32i8 VR256:$src2))), - (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>; - def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2), + (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>; + def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2), (imm:$mask))), - (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>; + (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>; } /// SS41I_ternary_int - SSE 4.1 ternary operator @@ -6927,15 +6990,15 @@ let Predicates = [UseSSE41] in { (v2f64 VR128:$src2))), (BLENDVPDrr0 VR128:$src2, VR128:$src1)>; - def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2), + def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2), (imm:$mask))), - (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>; - def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2), + (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>; + def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2), (imm:$mask))), - (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>; - def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2), + (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>; + def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2), (imm:$mask))), - (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>; + (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>; } @@ -7821,6 +7884,13 @@ defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256, VR256, memopv4i64, i256mem>, VEX_L; } +def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2), + imm:$mask)), + (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>; +def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2), + imm:$mask)), + (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>; + //===----------------------------------------------------------------------===// // VPBROADCAST - Load from memory and broadcast to all elements of the // destination operand |
