index 34b3b855bf1df21dc6bca5cbc1f8e4940e3a2f81..af51ced3397cf07d2d630e40a1a995a4875d4d44 100644 (file)
SSE_DIV_F32P, SSE_DIV_F64P
>;
+let Sched = WriteVecLogic in
+def SSE_VEC_BIT_ITINS_P : OpndItins<
+ IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
+>;
+
def SSE_BIT_ITINS_P : OpndItins<
IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
>;
IIC_SSE_INSERTPS_RR, IIC_SSE_INSERTPS_RM
>;
+let Sched = WriteMPSAD in
def SSE_MPSADBW_ITINS : OpndItins<
IIC_SSE_MPSADBW_RR, IIC_SSE_MPSADBW_RM
>;
+let Sched = WriteVecIMul in
def SSE_PMULLD_ITINS : OpndItins<
IIC_SSE_PMULLD_RR, IIC_SSE_PMULLD_RM
>;
+// Definitions for backward compatibility.
+// The instructions mapped on these definitions uses a different itinerary
+// than the actual scheduling model.
+let Sched = WriteShuffle in
+def DEFAULT_ITINS_SHUFFLESCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteVecIMul in
+def DEFAULT_ITINS_VECIMULSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteShuffle in
+def SSE_INTALU_ITINS_SHUFF_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
+let Sched = WriteMPSAD in
+def DEFAULT_ITINS_MPSADSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteFBlend in
+def DEFAULT_ITINS_FBLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteBlend in
+def DEFAULT_ITINS_BLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteVarBlend in
+def DEFAULT_ITINS_VARBLENDSCHED : OpndItins<
+ IIC_ALU_NONMEM, IIC_ALU_MEM
+>;
+
+let Sched = WriteFBlend in
+def SSE_INTALU_ITINS_FBLEND_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
+let Sched = WriteBlend in
+def SSE_INTALU_ITINS_BLEND_P : OpndItins<
+ IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
+>;
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 Instructions Classes
//===----------------------------------------------------------------------===//
!strconcat(base_opc, asm_opr),
[(set VR128:$dst, (vt (OpNode VR128:$src1,
(scalar_to_vector RC:$src2))))],
- IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
+ IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>;
// For the disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0 in
def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
(ins VR128:$src1, RC:$src2),
!strconcat(base_opc, asm_opr),
- [], IIC_SSE_MOV_S_RR>, Sched<[WriteMove]>;
+ [], IIC_SSE_MOV_S_RR>, Sched<[WriteFShuffle]>;
}
multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
// Patterns
let Predicates = [UseAVX] in {
- let AddedComplexity = 15 in {
- // Move scalar to XMM zero-extended, zeroing a VR128 then do a
- // MOVS{S,D} to the lower bits.
- def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
- (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
- def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
- (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
- def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
- (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
- def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
- (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
-
- // Move low f32 and clear high bits.
- def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (v4f32 (V_SET0)),
- (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
- def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSSrr (v4i32 (V_SET0)),
- (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
- }
-
let AddedComplexity = 20 in {
// MOVSSrm zeros the high parts of the register; represent this
// with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
(v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
(SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
}
- def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
- (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0),
- (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
- sub_xmm)>;
- def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
- (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
- (SUBREG_TO_REG (i64 0),
- (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
- sub_xmm)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
(SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
- // Move low f64 and clear high bits.
- def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (v2f64 (V_SET0)),
- (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
-
- def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
- (SUBREG_TO_REG (i32 0),
- (VMOVSDrr (v2i64 (V_SET0)),
- (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
-
// Extract and store.
def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
addr:$dst),
(EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
sub_xmm)>;
-
// FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
// is during lowering, where it's not possible to recognize the fold cause
// it has two uses through a bitcast. One use disappears at isel time and the
}
let Predicates = [UseSSE1] in {
- let AddedComplexity = 15 in {
+ let Predicates = [NoSSE41], AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSS to the lower bits.
def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
}
let Predicates = [UseSSE2] in {
- let AddedComplexity = 15 in {
+ let Predicates = [NoSSE41], AddedComplexity = 15 in {
// Move scalar to XMM zero-extended, zeroing a VR128 then do a
// MOVSD to the lower bits.
def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
string asm, Domain d,
OpndItins itins,
bit IsReMaterializable = 1> {
-let neverHasSideEffects = 1 in
+let hasSideEffects = 0 in
def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
!strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>,
- Sched<[WriteMove]>;
+ Sched<[WriteFShuffle]>;
let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(asm, "\t{$src, $dst|$dst, $src}"),
Sched<[WriteLoad]>;
}
+let Predicates = [HasAVX, NoVLX] in {
defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB, VEX;
+ PS, VEX;
defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
"movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
PD, VEX;
defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
"movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB, VEX;
+ PS, VEX;
defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
PD, VEX;
defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB, VEX, VEX_L;
+ PS, VEX, VEX_L;
defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
"movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
PD, VEX, VEX_L;
defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
"movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB, VEX, VEX_L;
+ PS, VEX, VEX_L;
defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
PD, VEX, VEX_L;
+}
+
+let Predicates = [UseSSE1] in {
defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
"movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
- TB;
+ PS;
+defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
+ "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
+ PS;
+}
+let Predicates = [UseSSE2] in {
defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
"movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
PD;
-defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
- "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
- TB;
defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
"movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
PD;
+}
-let SchedRW = [WriteStore] in {
+let SchedRW = [WriteStore], Predicates = [HasAVX, NoVLX] in {
def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movaps\t{$src, $dst|$dst, $src}",
[(alignedstore (v4f32 VR128:$src), addr:$dst)],
// For disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
- SchedRW = [WriteMove] in {
+ SchedRW = [WriteFShuffle] in {
def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
(ins VR128:$src),
"movaps\t{$src, $dst|$dst, $src}", [],
// For disassembler
let isCodeGenOnly = 1, ForceDisassemble = 1, hasSideEffects = 0,
- SchedRW = [WriteMove] in {
+ SchedRW = [WriteFShuffle] in {
def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
"movaps\t{$src, $dst|$dst, $src}", [],
IIC_SSE_MOVA_P_RR>;
(MOVUPDmr addr:$dst, VR128:$src)>;
// Use vmovaps/vmovups for AVX integer load/store.
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
// 128-bit load/store
def : Pat<(alignedloadv2i64 addr:$src),
(VMOVAPSrm addr:$src)>;
@@ -1142,8 +1159,8 @@ multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
[(set VR128:$dst,
(psnode VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
- itin, SSEPackedSingle>, TB,
- Sched<[WriteShuffleLd, ReadAfterLd]>;
+ itin, SSEPackedSingle>, PS,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
def PDrm : PI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
@@ -1151,7 +1168,7 @@ multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
[(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))],
itin, SSEPackedDouble>, PD,
- Sched<[WriteShuffleLd, ReadAfterLd]>;
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
(VMOVLPDrm VR128:$src1, addr:$src2)>;
def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
(VMOVLPDrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v2f64 (X86Movsd VR128:$src1,
+ (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (VMOVLPDrm VR128:$src1, addr:$src2)>;
// Store patterns
def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
(MOVLPDrm VR128:$src1, addr:$src2)>;
def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
(MOVLPDrm VR128:$src1, addr:$src2)>;
+ def : Pat<(v2f64 (X86Movsd VR128:$src1,
+ (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
+ (MOVLPDrm VR128:$src1, addr:$src2)>;
// Store patterns
def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
(bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
(VMOVHPSrm VR128:$src1, addr:$src2)>;
+ // VMOVHPD patterns
+
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
// is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))),
(VMOVHPDrm VR128:$src1, addr:$src2)>;
+ // Also handle an i64 load because that may get selected as a faster way to
+ // load the data.
+ def : Pat<(v2f64 (X86Unpckl VR128:$src1,
+ (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
+ (VMOVHPDrm VR128:$src1, addr:$src2)>;
+
+ def : Pat<(store (f64 (vector_extract
+ (v2f64 (X86VPermilpi VR128:$src, (i8 1))),
+ (iPTR 0))), addr:$dst),
+ (VMOVHPDmr addr:$dst, VR128:$src)>;
}
let Predicates = [UseSSE1] in {
}
let Predicates = [UseSSE2] in {
+ // MOVHPD patterns
+
// FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
// is during lowering, where it's not possible to recognize the load fold
// cause it has two uses through a bitcast. One use disappears at isel time
def : Pat<(v2f64 (X86Unpckl VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))),
(MOVHPDrm VR128:$src1, addr:$src2)>;
+ // Also handle an i64 load because that may get selected as a faster way to
+ // load the data.
+ def : Pat<(v2f64 (X86Unpckl VR128:$src1,
+ (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src2)))))),
+ (MOVHPDrm VR128:$src1, addr:$src2)>;
+
+ def : Pat<(store (f64 (vector_extract
+ (v2f64 (X86Shufp VR128:$src, VR128:$src, (i8 1))),
+ (iPTR 0))), addr:$dst),
+ (MOVHPDmr addr:$dst, VR128:$src)>;
}
//===----------------------------------------------------------------------===//
[(set VR128:$dst,
(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
IIC_SSE_MOV_LH>,
- VEX_4V, Sched<[WriteShuffle]>;
+ VEX_4V, Sched<[WriteFShuffle]>;
def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
IIC_SSE_MOV_LH>,
- VEX_4V, Sched<[WriteShuffle]>;
+ VEX_4V, Sched<[WriteFShuffle]>;
}
let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
"movlhps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
- IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2),
"movhlps\t{$src2, $dst|$dst, $src2}",
[(set VR128:$dst,
(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
- IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
}
let Predicates = [UseAVX] in {
multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
X86MemOperand x86memop, string asm, Domain d,
OpndItins itins> {
-let neverHasSideEffects = 1 in {
+let hasSideEffects = 0 in {
def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
[], itins.rr, d>, Sched<[itins.Sched]>;
let mayLoad = 1 in
multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
X86MemOperand x86memop, string asm> {
-let neverHasSideEffects = 1, Predicates = [UseAVX] in {
+let hasSideEffects = 0, Predicates = [UseAVX] in {
def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
!strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
Sched<[WriteCvtI2F]>;
(ins DstRC:$src1, x86memop:$src),
!strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
Sched<[WriteCvtI2FLd, ReadAfterLd]>;
-} // neverHasSideEffects = 1
+} // hasSideEffects = 0
}
let Predicates = [UseAVX] in {
let Predicates = [UseAVX] in {
def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
- (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
+ (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src), 0>;
def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
- (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
+ (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src), 0>;
def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
(VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
(CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
- (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
+ (CVTSI2SSrm FR64:$dst, i32mem:$src), 0>;
def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
- (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
+ (CVTSI2SDrm FR64:$dst, i32mem:$src), 0>;
// Conversion Instructions Intrinsics - Match intrinsics which expect MM
// and/or XMM operand(s).
defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
"vcvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, VEX, Requires<[HasAVX]>;
+ PS, VEX, Requires<[HasAVX]>;
defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
"vcvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, VEX, VEX_L, Requires<[HasAVX]>;
+ PS, VEX, VEX_L, Requires<[HasAVX]>;
defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
"cvtdq2ps\t{$src, $dst|$dst, $src}",
SSEPackedSingle, SSE_CVT_PS>,
- TB, Requires<[UseSSE2]>;
+ PS, Requires<[UseSSE2]>;
let Predicates = [UseAVX] in {
def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
/// SSE 2 Only
// Convert scalar double to scalar single
-let neverHasSideEffects = 1, Predicates = [UseAVX] in {
+let hasSideEffects = 0, Predicates = [UseAVX] in {
def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
(ins FR64:$src1, FR64:$src2),
"cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
// Convert scalar single to scalar double
// SSE2 instructions with XS prefix
-let neverHasSideEffects = 1, Predicates = [UseAVX] in {
+let hasSideEffects = 0, Predicates = [UseAVX] in {
def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
(ins FR32:$src1, FR32:$src2),
"vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
// XMM only
def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
- (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
+ (VCVTPD2DQrr VR128:$dst, VR128:$src), 0>;
def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
@@ -1980,7 +2027,7 @@ def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
(int_x86_avx_cvt_pd2dq_256 (loadv4f64 addr:$src)))]>,
VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
- (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
+ (VCVTPD2DQYrr VR128:$dst, VR256:$src), 0>;
}
def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
// XMM only
def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
- (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
+ (VCVTTPD2DQrr VR128:$dst, VR128:$src), 0>;
def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvttpd2dqx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvttpd2dq
@@ -2102,7 +2149,7 @@ def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
(int_x86_avx_cvtt_pd2dq_256 (loadv4f64 addr:$src)))],
IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2ILd]>;
def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
- (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
+ (VCVTTPD2DQYrr VR128:$dst, VR256:$src), 0>;
let Predicates = [HasAVX] in {
def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
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:$src))],
- IIC_SSE_CVT_PD_RR>, TB, VEX, Sched<[WriteCvtF2F]>;
+ IIC_SSE_CVT_PD_RR>, PS, VEX, Sched<[WriteCvtF2F]>;
def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, TB, VEX, Sched<[WriteCvtF2FLd]>;
+ IIC_SSE_CVT_PD_RM>, PS, VEX, Sched<[WriteCvtF2FLd]>;
def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
- IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L, Sched<[WriteCvtF2F]>;
+ IIC_SSE_CVT_PD_RR>, PS, VEX, VEX_L, Sched<[WriteCvtF2F]>;
def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
"vcvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR256:$dst,
(int_x86_avx_cvt_ps2_pd_256 (loadv4f32 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
+ IIC_SSE_CVT_PD_RM>, PS, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
}
let Predicates = [UseSSE2] in {
def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
- IIC_SSE_CVT_PD_RR>, TB, Sched<[WriteCvtF2F]>;
+ IIC_SSE_CVT_PD_RR>, PS, Sched<[WriteCvtF2F]>;
def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
"cvtps2pd\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
- IIC_SSE_CVT_PD_RM>, TB, Sched<[WriteCvtF2FLd]>;
+ IIC_SSE_CVT_PD_RM>, PS, Sched<[WriteCvtF2FLd]>;
}
// Convert Packed DW Integers to Packed Double FP
let Predicates = [HasAVX] in {
-let neverHasSideEffects = 1, mayLoad = 1 in
+let hasSideEffects = 0, mayLoad = 1 in
def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"vcvtdq2pd\t{$src, $dst|$dst, $src}",
[]>, VEX, Sched<[WriteCvtI2FLd]>;
Sched<[WriteCvtI2F]>;
}
-let neverHasSideEffects = 1, mayLoad = 1 in
+let hasSideEffects = 0, mayLoad = 1 in
def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
"cvtdq2pd\t{$src, $dst|$dst, $src}", [],
IIC_SSE_CVT_PD_RR>, Sched<[WriteCvtI2FLd]>;
// XMM only
def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
- (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
+ (VCVTPD2PSrr VR128:$dst, VR128:$src), 0>;
def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
"cvtpd2psx\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
@@ -2227,7 +2274,7 @@ def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
(int_x86_avx_cvt_pd2_ps_256 (loadv4f64 addr:$src)))],
IIC_SSE_CVT_PD_RM>, VEX, VEX_L, Sched<[WriteCvtF2FLd]>;
def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
- (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
+ (VCVTPD2PSYrr VR128:$dst, VR256:$src), 0>;
def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"cvtpd2ps\t{$src, $dst|$dst, $src}",
multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
Operand CC, SDNode OpNode, ValueType VT,
PatFrag ld_frag, string asm, string asm_alt,
- OpndItins itins> {
+ OpndItins itins, ImmLeaf immLeaf> {
def rr : SIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
- [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
+ [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, immLeaf:$cc))],
itins.rr>, Sched<[itins.Sched]>;
def rm : SIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
[(set RC:$dst, (OpNode (VT RC:$src1),
- (ld_frag addr:$src2), imm:$cc))],
+ (ld_frag addr:$src2), immLeaf:$cc))],
itins.rm>,
Sched<[itins.Sched.Folded, ReadAfterLd]>;
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
+ (ins RC:$src1, RC:$src2, u8imm:$cc), asm_alt, [],
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, [],
+ (ins RC:$src1, x86memop:$src2, u8imm:$cc), asm_alt, [],
IIC_SSE_ALU_F32S_RM>,
Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmps, f32, loadf32,
"cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSE_ALU_F32S>,
- XS, VEX_4V, VEX_LIG;
+ SSE_ALU_F32S, i8immZExt5>, XS, VEX_4V, VEX_LIG;
defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmps, f64, loadf64,
"cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSE_ALU_F32S>, // same latency as 32 bit compare
+ SSE_ALU_F32S, i8immZExt5>, // same latency as 32 bit compare
XD, VEX_4V, VEX_LIG;
let Constraints = "$src1 = $dst" in {
defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmps, f32, loadf32,
"cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
- "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
- XS;
+ "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S,
+ i8immZExt3>, XS;
defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmps, f64, loadf64,
"cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
"cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSE_ALU_F64S>,
- XD;
+ SSE_ALU_F64S, i8immZExt3>, XD;
}
multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
- Intrinsic Int, string asm, OpndItins itins> {
+ Intrinsic Int, string asm, OpndItins itins,
+ ImmLeaf immLeaf> {
def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src, CC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
- VR128:$src, imm:$cc))],
+ VR128:$src, immLeaf:$cc))],
itins.rr>,
Sched<[itins.Sched]>;
def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, x86memop:$src, CC:$cc), asm,
[(set VR128:$dst, (Int VR128:$src1,
- (load addr:$src), imm:$cc))],
+ (load addr:$src), immLeaf:$cc))],
itins.rm>,
Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
// Aliases to match intrinsics which expect XMM operand(s).
defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
"cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
- SSE_ALU_F32S>,
+ SSE_ALU_F32S, i8immZExt5>,
XS, VEX_4V;
defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
"cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
- SSE_ALU_F32S>, // same latency as f32
+ SSE_ALU_F32S, i8immZExt5>, // same latency as f32
XD, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
"cmp${cc}ss\t{$src, $dst|$dst, $src}",
- SSE_ALU_F32S>, XS;
+ SSE_ALU_F32S, i8immZExt3>, XS;
defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
"cmp${cc}sd\t{$src, $dst|$dst, $src}",
- SSE_ALU_F64S>,
+ SSE_ALU_F64S, i8immZExt3>,
XD;
}
}
let Defs = [EFLAGS] in {
defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
- "ucomiss">, TB, VEX, VEX_LIG;
+ "ucomiss">, PS, VEX, VEX_LIG;
defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
"ucomisd">, PD, VEX, VEX_LIG;
let Pattern = []<dag> in {
defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
- "comiss">, TB, VEX, VEX_LIG;
+ "comiss">, PS, VEX, VEX_LIG;
defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
"comisd">, PD, VEX, VEX_LIG;
}
let isCodeGenOnly = 1 in {
defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
- load, "ucomiss">, TB, VEX;
+ load, "ucomiss">, PS, VEX;
defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
load, "ucomisd">, PD, VEX;
defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
- load, "comiss">, TB, VEX;
+ load, "comiss">, PS, VEX;
defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
load, "comisd">, PD, VEX;
}
defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
- "ucomiss">, TB;
+ "ucomiss">, PS;
defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
"ucomisd">, PD;
let Pattern = []<dag> in {
defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
- "comiss">, TB;
+ "comiss">, PS;
defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
"comisd">, PD;
}
let isCodeGenOnly = 1 in {
defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
- load, "ucomiss">, TB;
+ load, "ucomiss">, PS;
defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
load, "ucomisd">, PD;
defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
- "comiss">, TB;
+ "comiss">, PS;
defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
"comisd">, PD;
}
// sse12_cmp_packed - sse 1 & 2 compare packed instructions
multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
Operand CC, Intrinsic Int, string asm,
- string asm_alt, Domain d,
+ string asm_alt, Domain d, ImmLeaf immLeaf,
OpndItins itins = SSE_ALU_F32P> {
def rri : PIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
- [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
+ [(set RC:$dst, (Int RC:$src1, RC:$src2, immLeaf:$cc))],
itins.rr, d>,
Sched<[WriteFAdd]>;
def rmi : PIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
- [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
+ [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), immLeaf:$cc))],
itins.rm, d>,
Sched<[WriteFAddLd, ReadAfterLd]>;
// Accept explicit immediate argument form instead of comparison code.
let isAsmParserOnly = 1, hasSideEffects = 0 in {
def rri_alt : PIi8<0xC2, MRMSrcReg,
- (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
+ (outs RC:$dst), (ins RC:$src1, RC:$src2, u8imm:$cc),
asm_alt, [], itins.rr, d>, Sched<[WriteFAdd]>;
+ let mayLoad = 1 in
def rmi_alt : PIi8<0xC2, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
+ (outs RC:$dst), (ins RC:$src1, x86memop:$src2, u8imm:$cc),
asm_alt, [], itins.rm, d>,
Sched<[WriteFAddLd, ReadAfterLd]>;
}
defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
"cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedSingle>, TB, VEX_4V;
+ SSEPackedSingle, i8immZExt5>, PS, VEX_4V;
defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
"cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedDouble>, PD, VEX_4V;
+ SSEPackedDouble, i8immZExt5>, PD, VEX_4V;
defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
"cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedSingle>, TB, VEX_4V, VEX_L;
+ SSEPackedSingle, i8immZExt5>, PS, VEX_4V, VEX_L;
defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
"cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedDouble>, PD, VEX_4V, VEX_L;
+ SSEPackedDouble, i8immZExt5>, PD, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in {
defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
"cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
"cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSEPackedSingle, SSE_ALU_F32P>, TB;
+ SSEPackedSingle, i8immZExt5, SSE_ALU_F32P>, PS;
defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
"cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
"cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
- SSEPackedDouble, SSE_ALU_F64P>, PD;
+ SSEPackedDouble, i8immZExt5, SSE_ALU_F64P>, PD;
}
let Predicates = [HasAVX] in {
@@ -2530,29 +2577,28 @@ def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
// SSE 1 & 2 - Shuffle Instructions
//===----------------------------------------------------------------------===//
-/// sse12_shuffle - sse 1 & 2 shuffle instructions
+/// sse12_shuffle - sse 1 & 2 fp shuffle instructions
multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
ValueType vt, string asm, PatFrag mem_frag,
- Domain d, bit IsConvertibleToThreeAddress = 0> {
+ Domain d> {
def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
+ (ins RC:$src1, x86memop:$src2, u8imm:$src3), asm,
[(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
(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:$src3))))], IIC_SSE_SHUFP, d>,
- Sched<[WriteShuffle]>;
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
+ def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2, u8imm:$src3), asm,
+ [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
+ (i8 imm:$src3))))], IIC_SSE_SHUFP, d>,
+ Sched<[WriteFShuffle]>;
}
defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
"shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- loadv4f32, SSEPackedSingle>, TB, VEX_4V;
+ loadv4f32, SSEPackedSingle>, PS, VEX_4V;
defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
"shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- loadv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
+ loadv8f32, SSEPackedSingle>, PS, VEX_4V, VEX_L;
defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
"shufpd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
loadv2f64, SSEPackedDouble>, PD, VEX_4V;
let Constraints = "$src1 = $dst" in {
defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
"shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>, TB;
+ memopv4f32, SSEPackedSingle>, PS;
defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
"shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
- memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>, PD;
+ memopv2f64, SSEPackedDouble>, PD;
}
let Predicates = [HasAVX] in {
}
//===----------------------------------------------------------------------===//
-// SSE 1 & 2 - Unpack Instructions
+// SSE 1 & 2 - Unpack FP Instructions
//===----------------------------------------------------------------------===//
-/// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
+/// sse12_unpack_interleave - sse 1 & 2 fp unpack and interleave
multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
PatFrag mem_frag, RegisterClass RC,
X86MemOperand x86memop, string asm,
@@ -2626,38 +2672,38 @@ multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
(outs RC:$dst), (ins RC:$src1, RC:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1, RC:$src2)))],
- IIC_SSE_UNPCK, d>, Sched<[WriteShuffle]>;
+ IIC_SSE_UNPCK, d>, Sched<[WriteFShuffle]>;
def rm : PI<opc, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1,
(mem_frag addr:$src2))))],
IIC_SSE_UNPCK, d>,
- Sched<[WriteShuffleLd, ReadAfterLd]>;
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, loadv4f32,
VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
+ SSEPackedSingle>, PS, VEX_4V;
defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, loadv2f64,
VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, VEX_4V;
defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, loadv4f32,
VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V;
+ SSEPackedSingle>, PS, VEX_4V;
defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, loadv2f64,
VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, VEX_4V;
defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, loadv8f32,
VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V, VEX_L;
+ SSEPackedSingle>, PS, VEX_4V, VEX_L;
defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, loadv4f64,
VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, VEX_4V, VEX_L;
defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, loadv8f32,
VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- SSEPackedSingle>, TB, VEX_4V, VEX_L;
+ SSEPackedSingle>, PS, VEX_4V, VEX_L;
defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, loadv4f64,
VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
SSEPackedDouble>, PD, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst" in {
defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
SSEPackedDouble>, PD;
defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
SSEPackedDouble>, PD;
@@ -2730,11 +2776,11 @@ multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
let Predicates = [HasAVX] in {
defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
- "movmskps", SSEPackedSingle>, TB, VEX;
+ "movmskps", SSEPackedSingle>, PS, VEX;
defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
"movmskpd", SSEPackedDouble>, PD, VEX;
defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
- "movmskps", SSEPackedSingle>, TB,
+ "movmskps", SSEPackedSingle>, PS,
VEX, VEX_L;
defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
"movmskpd", SSEPackedDouble>, PD,
}
defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
- SSEPackedSingle>, TB;
+ SSEPackedSingle>, PS;
defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
SSEPackedDouble>, PD;
multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
ValueType OpVT128, ValueType OpVT256,
OpndItins itins, bit IsCommutable = 0> {
-let Predicates = [HasAVX] in
+let Predicates = [HasAVX, NoVLX] in
defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
VR128, loadv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
memopv2i64, i128mem, itins, IsCommutable, 1>;
-let Predicates = [HasAVX2] in
+let Predicates = [HasAVX2, NoVLX] in
defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
OpVT256, VR256, loadv4i64, i256mem, itins,
IsCommutable, 0>, VEX_4V, VEX_L;
// These are ordered here for pattern ordering requirements with the fp versions
-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 PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
+defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
+defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64,
+ SSE_VEC_BIT_ITINS_P, 1>;
defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
- SSE_BIT_ITINS_P, 0>;
+ SSE_VEC_BIT_ITINS_P, 0>;
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Logical Instructions
SDNode OpNode, OpndItins itins> {
defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
- TB, VEX_4V;
+ PS, VEX_4V;
defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
- TB;
+ PS;
defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
///
multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
SDNode OpNode> {
+ let Predicates = [HasAVX, NoVLX] in {
defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f256mem,
[(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
[(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
- (loadv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
+ (loadv4i64 addr:$src2)))], 0>, PS, VEX_4V, VEX_L;
defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f256mem,
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)),
- (loadv2i64 addr:$src2)))], 0>, TB, VEX_4V;
+ (loadv2i64 addr:$src2)))], 0>, PS, VEX_4V;
defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f128mem,
[(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
(loadv2i64 addr:$src2)))], 0>,
PD, VEX_4V;
+ }
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
!strconcat(OpcodeStr, "ps"), f128mem,
[(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
[(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
- (memopv2i64 addr:$src2)))]>, TB;
+ (memopv2i64 addr:$src2)))]>, PS;
defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
!strconcat(OpcodeStr, "pd"), f128mem,
let isCommutable = 0 in
defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
+// AVX1 requires type coercions in order to fold loads directly into logical
+// operations.
+let Predicates = [HasAVX1Only] in {
+ def : Pat<(bc_v8f32 (and VR256:$src1, (loadv4i64 addr:$src2))),
+ (VANDPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (or VR256:$src1, (loadv4i64 addr:$src2))),
+ (VORPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (xor VR256:$src1, (loadv4i64 addr:$src2))),
+ (VXORPSYrm VR256:$src1, addr:$src2)>;
+ def : Pat<(bc_v8f32 (X86andnp VR256:$src1, (loadv4i64 addr:$src2))),
+ (VANDNPSYrm VR256:$src1, addr:$src2)>;
+}
+
//===----------------------------------------------------------------------===//
// SSE 1 & 2 - Arithmetic Instructions
//===----------------------------------------------------------------------===//
/// classes below
multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
SDNode OpNode, SizeItins itins> {
+ let Predicates = [HasAVX, NoVLX] in {
defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
VR128, v4f32, f128mem, loadv4f32,
- SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
+ SSEPackedSingle, itins.s, 0>, PS, VEX_4V;
defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
VR128, v2f64, f128mem, loadv2f64,
SSEPackedDouble, itins.d, 0>, PD, VEX_4V;
defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
OpNode, VR256, v8f32, f256mem, loadv8f32,
- SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
+ SSEPackedSingle, itins.s, 0>, PS, VEX_4V, VEX_L;
defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
OpNode, VR256, v4f64, f256mem, loadv4f64,
SSEPackedDouble, itins.d, 0>, PD, VEX_4V, VEX_L;
+ }
let Constraints = "$src1 = $dst" in {
defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
v4f32, f128mem, memopv4f32, SSEPackedSingle,
- itins.s>, TB;
+ itins.s>, PS;
defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
v2f64, f128mem, memopv2f64, SSEPackedDouble,
itins.d>, PD;
// previously we generated:
// addss %xmm0, %xmm1
// movss %xmm1, %xmm0
-//
+//
// we now generate:
// addss %xmm1, %xmm0
let Predicates = [UseSSE2] in {
// SSE2 patterns to select scalar double-precision fp arithmetic instructions
-
def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
(f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
FR64:$src))))),
}
let Predicates = [UseSSE41] in {
- // If the subtarget has SSE4.1 but not AVX, the vector insert
- // instruction is lowered into a X86insrtps rather than a X86Movss.
- // When selecting SSE scalar single-precision fp arithmetic instructions,
- // make sure that we correctly match the X86insrtps.
+ // If the subtarget has SSE4.1 but not AVX, the vector insert instruction is
+ // lowered into a X86insertps or a X86Blendi rather than a X86Movss. When
+ // selecting SSE scalar single-precision fp arithmetic instructions, make
+ // sure that we correctly match them.
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (ADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (SUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (MULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (DIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (ADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (SUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (MULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (DIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
}
let Predicates = [HasAVX] in {
(f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
FR64:$src))))),
(VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fadd (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fsub (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fmul (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
- def : Pat<(v4f32 (X86insrtps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$dst), (v4f32 (scalar_to_vector
(fdiv (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
FR32:$src))), (iPTR 0))),
(VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fadd
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VADDSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fsub
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VSUBSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fmul
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VMULSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst), (v4f32 (scalar_to_vector (fdiv
+ (f32 (vector_extract (v4f32 VR128:$dst), (iPTR 0))),
+ FR32:$src))), (i8 1))),
+ (VDIVSSrr_Int v4f32:$dst, (COPY_TO_REGCLASS FR32:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst), (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (i8 1))),
+ (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fadd
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VADDSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fsub
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VSUBSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fmul
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VMULSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 (scalar_to_vector (fdiv
+ (f64 (vector_extract (v2f64 VR128:$dst), (iPTR 0))),
+ FR64:$src))), (v2f64 VR128:$dst), (i8 2))),
+ (VDIVSDrr_Int v2f64:$dst, (COPY_TO_REGCLASS FR64:$src, VR128))>;
}
// Patterns used to select SSE scalar fp arithmetic instructions from
// previously we generated:
// addps %xmm0, %xmm1
// movss %xmm1, %xmm0
-//
+//
// we now generate:
// addss %xmm1, %xmm0
def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
(fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
(ADDSSrr_Int v4f32:$dst, v4f32:$src)>;
- def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
(fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
(SUBSSrr_Int v4f32:$dst, v4f32:$src)>;
def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
(fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
(MULSSrr_Int v4f32:$dst, v4f32:$src)>;
- def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
+ def : Pat<(v4f32 (X86Movss (v4f32 VR128:$dst),
(fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)))),
(DIVSSrr_Int v4f32:$dst, v4f32:$src)>;
}
(DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
}
+let Predicates = [UseSSE41] in {
+ // With SSE4.1 we may see these operations using X86Blendi rather than
+ // X86Movs{s,d}.
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (ADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (SUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (MULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (DIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (ADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (SUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (MULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (DIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+}
+
let Predicates = [HasAVX] in {
// The following patterns select AVX Scalar single/double precision fp
// arithmetic instructions from a packed single precision fp instruction
def : Pat<(v2f64 (X86Movsd (v2f64 VR128:$dst),
(fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)))),
(VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ // Also handle X86Blendi-based patterns.
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fadd (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VADDSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fsub (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VSUBSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fmul (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VMULSSrr_Int v4f32:$dst, v4f32:$src)>;
+ def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$dst),
+ (fdiv (v4f32 VR128:$dst), (v4f32 VR128:$src)), (i8 1))),
+ (VDIVSSrr_Int v4f32:$dst, v4f32:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$dst),
+ (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)), (i8 1))),
+ (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
+
+ def : Pat<(v2f64 (X86Blendi (fadd (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VADDSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fsub (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VSUBSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fmul (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VMULSDrr_Int v2f64:$dst, v2f64:$src)>;
+ def : Pat<(v2f64 (X86Blendi (fdiv (v2f64 VR128:$dst), (v2f64 VR128:$src)),
+ (v2f64 VR128:$dst), (i8 2))),
+ (VDIVSDrr_Int v2f64:$dst, v2f64:$src)>;
}
/// Unop Arithmetic
>;
}
+let Sched = WriteFRsqrt in {
+def SSE_RSQRTPS : OpndItins<
+ IIC_SSE_RSQRTPS_RR, IIC_SSE_RSQRTPS_RM
+>;
+
+def SSE_RSQRTSS : OpndItins<
+ IIC_SSE_RSQRTSS_RR, IIC_SSE_RSQRTSS_RM
+>;
+}
+
let Sched = WriteFRcp in {
def SSE_RCPP : OpndItins<
IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
>;
}
-/// sse1_fp_unop_s - SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
- SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
-let Predicates = [HasAVX], hasSideEffects = 0 in {
- def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
- (ins FR32:$src1, FR32:$src2),
- !strconcat("v", OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, VEX_4V, VEX_LIG, Sched<[itins.Sched]>;
- let mayLoad = 1 in {
- def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
- (ins FR32:$src1,f32mem:$src2),
- !strconcat("v", OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, VEX_4V, VEX_LIG,
- Sched<[itins.Sched.Folded, ReadAfterLd]>;
- let isCodeGenOnly = 1 in
- def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat("v", OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>, VEX_4V, VEX_LIG,
- Sched<[itins.Sched.Folded, ReadAfterLd]>;
- }
-}
-
- def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode FR32:$src))]>, Sched<[itins.Sched]>;
- // For scalar unary operations, fold a load into the operation
- // only in OptForSize mode. It eliminates an instruction, but it also
- // eliminates a whole-register clobber (the load), so it introduces a
- // partial register update condition.
- def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
- Requires<[UseSSE1, OptForSize]>, Sched<[itins.Sched.Folded]>;
-let isCodeGenOnly = 1 in {
- def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>,
- Sched<[itins.Sched]>;
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>,
- Sched<[itins.Sched.Folded]>;
-}
-}
-
-/// 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,
+/// sse1_fp_unop_s - SSE1 unops in scalar form
+/// For the non-AVX defs, we need $src1 to be tied to $dst because
+/// the HW instructions are 2 operand / destructive.
+multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
OpndItins itins> {
let Predicates = [HasAVX], hasSideEffects = 0 in {
def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
}
// Square root.
-defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
- SSE_SQRTSS>,
+defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, SSE_SQRTSS>,
sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPS>,
- sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
+ sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
SSE_SQRTSD>,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTPD>;
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
-defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTSS>,
- sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTPS>,
+defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, SSE_RSQRTSS>,
+ sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_RSQRTPS>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
- int_x86_avx_rsqrt_ps_256, SSE_SQRTPS>;
-defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
+ int_x86_avx_rsqrt_ps_256, SSE_RSQRTPS>;
+defm RCP : sse1_fp_unop_s<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>;
(VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
}
-// Reciprocal approximations. Note that these typically require refinement
-// in order to obtain suitable precision.
+// These are unary operations, but they are modeled as having 2 source operands
+// because the high elements of the destination are unchanged in SSE.
let Predicates = [UseSSE1] in {
def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
(RSQRTSSr_Int VR128:$src, VR128:$src)>;
def : Pat<(int_x86_sse_rcp_ss VR128:$src),
(RCPSSr_Int VR128:$src, VR128:$src)>;
+ def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
+ (SQRTSSr_Int VR128:$src, VR128:$src)>;
}
// There is no f64 version of the reciprocal approximation instructions.
let AddedComplexity = 400 in { // Prefer non-temporal versions
let SchedRW = [WriteStore] in {
+let Predicates = [HasAVX, NoVLX] in {
def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
(ins f128mem:$dst, VR128:$src),
"movntps\t{$src, $dst|$dst, $src}",
[(alignednontemporalstore (v4i64 VR256:$src),
addr:$dst)],
IIC_SSE_MOVNT>, VEX, VEX_L;
+}
def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
"movntps\t{$src, $dst|$dst, $src}",
"movnti{l}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i32 GR32:$src), addr:$dst)],
IIC_SSE_MOVNT>,
- TB, Requires<[HasSSE2]>;
+ PS, Requires<[HasSSE2]>;
def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
"movnti{q}\t{$src, $dst|$dst, $src}",
[(nontemporalstore (i64 GR64:$src), addr:$dst)],
IIC_SSE_MOVNT>,
- TB, Requires<[HasSSE2]>;
+ PS, Requires<[HasSSE2]>;
} // SchedRW = [WriteStore]
+let Predicates = [HasAVX, NoVLX] in {
+ def : Pat<(alignednontemporalstore (v4i32 VR128:$src), addr:$dst),
+ (VMOVNTPSmr addr:$dst, VR128:$src)>;
+}
+
+def : Pat<(alignednontemporalstore (v4i32 VR128:$src), addr:$dst),
+ (MOVNTPSmr addr:$dst, VR128:$src)>;
+
} // AddedComplexity
//===----------------------------------------------------------------------===//
IIC_SSE_PREFETCH>, TB;
}
-// FIXME: How should these memory instructions be modeled?
+// FIXME: How should flush instruction be modeled?
let SchedRW = [WriteLoad] in {
// Flush cache
def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
"clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
+}
+let SchedRW = [WriteNop] in {
// Pause. This "instruction" is encoded as "rep; nop", so even though it
// was introduced with SSE2, it's backward compatible.
-def PAUSE : I<0x90, RawFrm, (outs), (ins),
- "pause", [(int_x86_sse2_pause)], IIC_SSE_PAUSE>,
- REP, Requires<[HasSSE2]>;
+def PAUSE : I<0x90, RawFrm, (outs), (ins),
+ "pause", [(int_x86_sse2_pause)], IIC_SSE_PAUSE>,
+ OBXS, Requires<[HasSSE2]>;
+}
+let SchedRW = [WriteFence] in {
// Load, store, and memory fence
def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
"sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
"stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
IIC_SSE_STMXCSR>, VEX, Sched<[WriteStore]>;
-def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
- "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
- IIC_SSE_LDMXCSR>, Sched<[WriteLoad]>;
-def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
- "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
- IIC_SSE_STMXCSR>, Sched<[WriteStore]>;
+let Predicates = [UseSSE1] in {
+def LDMXCSR : I<0xAE, MRM2m, (outs), (ins i32mem:$src),
+ "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
+ IIC_SSE_LDMXCSR>, TB, Sched<[WriteLoad]>;
+def STMXCSR : I<0xAE, MRM3m, (outs), (ins i32mem:$dst),
+ "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
+ IIC_SSE_STMXCSR>, TB, Sched<[WriteStore]>;
+}
//===---------------------------------------------------------------------===//
// SSE2 - Move Aligned/Unaligned Packed Integer Instructions
let ExeDomain = SSEPackedInt in { // SSE integer instructions
-let neverHasSideEffects = 1, SchedRW = [WriteMove] in {
+let hasSideEffects = 0, SchedRW = [WriteMove] in {
def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
VEX;
@@ -3805,7 +4035,7 @@ def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
}
let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
- neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
+ hasSideEffects = 0, SchedRW = [WriteLoad] in {
def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
VEX;
}
}
-let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
+let mayStore = 1, hasSideEffects = 0, SchedRW = [WriteStore] in {
def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
(ins i128mem:$dst, VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
}
let SchedRW = [WriteMove] in {
-let neverHasSideEffects = 1 in
+let hasSideEffects = 0 in
def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
} // SchedRW
let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
- neverHasSideEffects = 1, SchedRW = [WriteLoad] in {
+ hasSideEffects = 0, SchedRW = [WriteLoad] in {
def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movdqa\t{$src, $dst|$dst, $src}",
[/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
XS, Requires<[UseSSE2]>;
}
-let mayStore = 1, neverHasSideEffects = 1, SchedRW = [WriteStore] in {
+let mayStore = 1, hasSideEffects = 0, SchedRW = [WriteStore] in {
def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
"movdqa\t{$src, $dst|$dst, $src}",
[/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
(bc_frag (memopv2i64 addr:$src2)))))], itins.rm>,
Sched<[WriteVecShiftLd, ReadAfterLd]>;
def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
- (ins RC:$src1, i8imm:$src2),
+ (ins RC:$src1, u8imm:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
SSE_INTALUQ_ITINS_P, 1>;
defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
SSE_INTMUL_ITINS_P, 1>;
+defm PMULHUW : PDI_binop_all<0xE4, "pmulhuw", mulhu, v8i16, v16i16,
+ SSE_INTMUL_ITINS_P, 1>;
+defm PMULHW : PDI_binop_all<0xE5, "pmulhw", mulhs, v8i16, v16i16,
+ SSE_INTMUL_ITINS_P, 1>;
defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
SSE_INTALU_ITINS_P, 0>;
defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
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,
SSE_INTALU_ITINS_P, 0>;
//===---------------------------------------------------------------------===//
-// SSE2 - Packed Integer Pack Instructions
-//===---------------------------------------------------------------------===//
-
-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
+// SSE2 - Packed Integer Shuffle Instructions
//===---------------------------------------------------------------------===//
let ExeDomain = SSEPackedInt in {
@@ -4292,14 +4511,14 @@ multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
SDNode OpNode> {
let Predicates = [HasAVX] in {
def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, i8imm:$src2),
+ (ins VR128:$src1, u8imm:$src2),
!strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
IIC_SSE_PSHUF_RI>, VEX, Sched<[WriteShuffle]>;
def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
- (ins i128mem:$src1, i8imm:$src2),
+ (ins i128mem:$src1, u8imm:$src2),
!strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
let Predicates = [HasAVX2] in {
def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, i8imm:$src2),
+ (ins VR256:$src1, u8imm:$src2),
!strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
IIC_SSE_PSHUF_RI>, VEX, VEX_L, Sched<[WriteShuffle]>;
def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
- (ins i256mem:$src1, i8imm:$src2),
+ (ins i256mem:$src1, u8imm:$src2),
!strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
let Predicates = [UseSSE2] in {
def ri : Ii8<0x70, MRMSrcReg,
- (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
+ (outs VR128:$dst), (ins VR128:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
IIC_SSE_PSHUF_RI>, Sched<[WriteShuffle]>;
def mi : Ii8<0x70, MRMSrcMem,
- (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
+ (outs VR128:$dst), (ins i128mem:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
(i8 imm:$src2))))], IIC_SSE_PSHUF_MI>,
- Sched<[WriteShuffleLd]>;
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
}
}
} // ExeDomain = SSEPackedInt
(PSHUFDri VR128:$src1, imm:$imm)>;
}
+//===---------------------------------------------------------------------===//
+// Packed Integer Pack Instructions (SSE & AVX)
+//===---------------------------------------------------------------------===//
+
+let ExeDomain = SSEPackedInt in {
+multiclass sse2_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag,
+ bit Is2Addr = 1> {
+ def rr : PDI<opc, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def rm : PDI<opc, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode VR128:$src1,
+ (bc_frag (memopv2i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+
+multiclass sse2_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag> {
+ def Yrr : PDI<opc, MRMSrcReg,
+ (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def Yrm : PDI<opc, MRMSrcMem,
+ (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode VR256:$src1,
+ (bc_frag (memopv4i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+
+multiclass sse4_pack<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag,
+ bit Is2Addr = 1> {
+ def rr : SS48I<opc, MRMSrcReg,
+ (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode (ArgVT VR128:$src1), VR128:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def rm : SS48I<opc, MRMSrcMem,
+ (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set VR128:$dst,
+ (OutVT (OpNode VR128:$src1,
+ (bc_frag (memopv2i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+
+multiclass sse4_pack_y<bits<8> opc, string OpcodeStr, ValueType OutVT,
+ ValueType ArgVT, SDNode OpNode, PatFrag bc_frag> {
+ def Yrr : SS48I<opc, MRMSrcReg,
+ (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode (ArgVT VR256:$src1), VR256:$src2)))]>,
+ Sched<[WriteShuffle]>;
+ def Yrm : SS48I<opc, MRMSrcMem,
+ (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
+ !strconcat(OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ [(set VR256:$dst,
+ (OutVT (OpNode VR256:$src1,
+ (bc_frag (memopv4i64 addr:$src2)))))]>,
+ Sched<[WriteShuffleLd, ReadAfterLd]>;
+}
+
+let Predicates = [HasAVX] in {
+ defm VPACKSSWB : sse2_pack<0x63, "vpacksswb", v16i8, v8i16, X86Packss,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPACKSSDW : sse2_pack<0x6B, "vpackssdw", v8i16, v4i32, X86Packss,
+ bc_v4i32, 0>, VEX_4V;
+
+ defm VPACKUSWB : sse2_pack<0x67, "vpackuswb", v16i8, v8i16, X86Packus,
+ bc_v8i16, 0>, VEX_4V;
+ defm VPACKUSDW : sse4_pack<0x2B, "vpackusdw", v8i16, v4i32, X86Packus,
+ bc_v4i32, 0>, VEX_4V;
+}
+
+let Predicates = [HasAVX2] in {
+ defm VPACKSSWB : sse2_pack_y<0x63, "vpacksswb", v32i8, v16i16, X86Packss,
+ bc_v16i16>, VEX_4V, VEX_L;
+ defm VPACKSSDW : sse2_pack_y<0x6B, "vpackssdw", v16i16, v8i32, X86Packss,
+ bc_v8i32>, VEX_4V, VEX_L;
+
+ defm VPACKUSWB : sse2_pack_y<0x67, "vpackuswb", v32i8, v16i16, X86Packus,
+ bc_v16i16>, VEX_4V, VEX_L;
+ defm VPACKUSDW : sse4_pack_y<0x2B, "vpackusdw", v16i16, v8i32, X86Packus,
+ bc_v8i32>, VEX_4V, VEX_L;
+}
+
+let Constraints = "$src1 = $dst" in {
+ defm PACKSSWB : sse2_pack<0x63, "packsswb", v16i8, v8i16, X86Packss,
+ bc_v8i16>;
+ defm PACKSSDW : sse2_pack<0x6B, "packssdw", v8i16, v4i32, X86Packss,
+ bc_v4i32>;
+
+ defm PACKUSWB : sse2_pack<0x67, "packuswb", v16i8, v8i16, X86Packus,
+ bc_v8i16>;
+
+ let Predicates = [HasSSE41] in
+ defm PACKUSDW : sse4_pack<0x2B, "packusdw", v8i16, v4i32, X86Packus,
+ bc_v4i32>;
+}
+} // ExeDomain = SSEPackedInt
+
//===---------------------------------------------------------------------===//
// SSE2 - Packed Integer Unpack Instructions
//===---------------------------------------------------------------------===//
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
- VEX, Sched<[WriteLoad]>;
+ VEX, Sched<[WriteStore]>;
def MOVPDI2DIrr : S2I<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
@@ -4674,7 +5023,7 @@ def MOVPDI2DImr : S2I<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
"movd\t{$src, $dst|$dst, $src}",
[(store (i32 (vector_extract (v4i32 VR128:$src),
(iPTR 0))), addr:$dst)],
- IIC_SSE_MOVDQ>, Sched<[WriteLoad]>;
+ IIC_SSE_MOVDQ>, Sched<[WriteStore]>;
def : Pat<(v8i32 (X86Vinsert (v8i32 immAllZerosV), GR32:$src2, (iPTR 0))),
(SUBREG_TO_REG (i32 0), (VMOVDI2PDIrr GR32:$src2), sub_xmm)>;
def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set RC:$dst, (vt (OpNode RC:$src)))],
- IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
+ IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set RC:$dst, (OpNode (mem_frag addr:$src)))],
- IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
+ IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
}
let Predicates = [HasAVX] in {
//===---------------------------------------------------------------------===//
multiclass sse3_replicate_dfp<string OpcodeStr> {
-let neverHasSideEffects = 1 in
+let hasSideEffects = 0 in
def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [], IIC_SSE_MOV_LH>, Sched<[WriteShuffle]>;
+ [], IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
(v2f64 (X86Movddup
(scalar_to_vector (loadf64 addr:$src)))))],
- IIC_SSE_MOV_LH>, Sched<[WriteShuffleLd]>;
+ IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
}
// FIXME: Merge with above classe when there're patterns for the ymm version
def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>,
- Sched<[WriteShuffle]>;
+ Sched<[WriteFShuffle]>;
def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
(v4f64 (X86Movddup
(scalar_to_vector (loadf64 addr:$src)))))]>,
- Sched<[WriteShuffleLd]>;
+ Sched<[WriteLoad]>;
}
let Predicates = [HasAVX] in {
(VMOVDDUPYrr VR256:$src)>;
}
+let Predicates = [UseAVX, OptForSize] in {
+ def : Pat<(v2f64 (X86VBroadcast (loadf64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>;
+ def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
+ (VMOVDDUPrm addr:$src)>;
+}
+
let Predicates = [UseSSE3] in {
def : Pat<(X86Movddup (memopv2f64 addr:$src)),
(MOVDDUPrm addr:$src)>;
f128mem, SSE_ALU_F64P>, PD;
}
+// Patterns used to select 'addsub' instructions.
+let Predicates = [HasAVX] in {
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 VR128:$rhs))),
+ (VADDSUBPSrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 (memop addr:$rhs)))),
+ (VADDSUBPSrm VR128:$lhs, f128mem:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 VR128:$rhs))),
+ (VADDSUBPDrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 (memop addr:$rhs)))),
+ (VADDSUBPDrm VR128:$lhs, f128mem:$rhs)>;
+
+ def : Pat<(v8f32 (X86Addsub (v8f32 VR256:$lhs), (v8f32 VR256:$rhs))),
+ (VADDSUBPSYrr VR256:$lhs, VR256:$rhs)>;
+ def : Pat<(v8f32 (X86Addsub (v8f32 VR256:$lhs), (v8f32 (memop addr:$rhs)))),
+ (VADDSUBPSYrm VR256:$lhs, f256mem:$rhs)>;
+ def : Pat<(v4f64 (X86Addsub (v4f64 VR256:$lhs), (v4f64 VR256:$rhs))),
+ (VADDSUBPDYrr VR256:$lhs, VR256:$rhs)>;
+ def : Pat<(v4f64 (X86Addsub (v4f64 VR256:$lhs), (v4f64 (memop addr:$rhs)))),
+ (VADDSUBPDYrm VR256:$lhs, f256mem:$rhs)>;
+}
+
+let Predicates = [UseSSE3] in {
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 VR128:$rhs))),
+ (ADDSUBPSrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v4f32 (X86Addsub (v4f32 VR128:$lhs), (v4f32 (memop addr:$rhs)))),
+ (ADDSUBPSrm VR128:$lhs, f128mem:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 VR128:$rhs))),
+ (ADDSUBPDrr VR128:$lhs, VR128:$rhs)>;
+ def : Pat<(v2f64 (X86Addsub (v2f64 VR128:$lhs), (v2f64 (memop addr:$rhs)))),
+ (ADDSUBPDrm VR128:$lhs, f128mem:$rhs)>;
+}
+
//===---------------------------------------------------------------------===//
// SSE3 Instructions
//===---------------------------------------------------------------------===//
}
multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId256> {
+ Intrinsic IntId256,
+ X86FoldableSchedWrite Sched> {
let isCommutable = 1 in
def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>;
+ [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
+ Sched<[Sched]>;
def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>;
+ (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
let ImmT = NoImm, Predicates = [HasAVX] in {
SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
loadv4i64, i256mem,
- SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
+ SSE_PSHUFB, 0>, VEX_4V, VEX_L;
defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
- int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
+ int_x86_avx2_phadd_sw,
+ WriteVecALU>, VEX_4V, VEX_L;
defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
- int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
+ int_x86_avx2_phsub_sw,
+ WriteVecALU>, VEX_4V, VEX_L;
defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
- int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
+ int_x86_avx2_pmadd_ub_sw,
+ WriteVecIMul>, VEX_4V, VEX_L;
}
defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
- int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
+ int_x86_avx2_pmul_hr_sw,
+ WriteVecIMul>, VEX_4V, VEX_L;
}
// None of these have i8 immediate fields.
//===---------------------------------------------------------------------===//
multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
- let neverHasSideEffects = 1 in {
+ let hasSideEffects = 0 in {
def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
[], IIC_SSE_PALIGNRR>, Sched<[WriteShuffle]>;
let mayLoad = 1 in
def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
}
multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
- let neverHasSideEffects = 1 in {
+ let hasSideEffects = 0 in {
def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+ (ins VR256:$src1, VR256:$src2, u8imm:$src3),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[]>, Sched<[WriteShuffle]>;
let mayLoad = 1 in
def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
- (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
+ (ins VR256:$src1, i256mem:$src2, u8imm:$src3),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
@@ -5673,524 +6068,240 @@ def : InstAlias<"monitor\t{%rax, %rcx, %rdx|rdx, rcx, rax}", (MONITORrrr)>,
// SSE4.1 - Packed Move with Sign/Zero Extend
//===----------------------------------------------------------------------===//
-multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId,
- OpndItins itins = DEFAULT_ITINS> {
- def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))], itins.rr>;
-
- def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst,
- (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))],
- itins.rm>;
-}
-
-multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
- def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>;
-
- def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId (load addr:$src)))]>;
-}
-
-let Predicates = [HasAVX] in {
-defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw",
- int_x86_sse41_pmovsxbw>, VEX;
-defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd",
- int_x86_sse41_pmovsxwd>, VEX;
-defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq",
- int_x86_sse41_pmovsxdq>, VEX;
-defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw",
- int_x86_sse41_pmovzxbw>, VEX;
-defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd",
- int_x86_sse41_pmovzxwd>, VEX;
-defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq",
- int_x86_sse41_pmovzxdq>, VEX;
-}
-
-let Predicates = [HasAVX2] in {
-defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
- int_x86_avx2_pmovsxbw>, VEX, VEX_L;
-defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
- int_x86_avx2_pmovsxwd>, VEX, VEX_L;
-defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
- int_x86_avx2_pmovsxdq>, VEX, VEX_L;
-defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
- int_x86_avx2_pmovzxbw>, VEX, VEX_L;
-defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
- int_x86_avx2_pmovzxwd>, VEX, VEX_L;
-defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
- int_x86_avx2_pmovzxdq>, VEX, VEX_L;
-}
-
-defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw, SSE_INTALU_ITINS_P>;
-defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd, SSE_INTALU_ITINS_P>;
-defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq, SSE_INTALU_ITINS_P>;
-defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw, SSE_INTALU_ITINS_P>;
-defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd, SSE_INTALU_ITINS_P>;
-defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq, SSE_INTALU_ITINS_P>;
-
-let Predicates = [HasAVX] in {
- // Common patterns involving scalar load.
- def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
- (VPMOVSXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
- (VPMOVSXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
- (VPMOVSXBWrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
- (VPMOVSXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
- (VPMOVSXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
- (VPMOVSXWDrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
- (VPMOVSXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
- (VPMOVSXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
- (VPMOVSXDQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
- (VPMOVZXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
- (VPMOVZXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
- (VPMOVZXBWrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
- (VPMOVZXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
- (VPMOVZXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
- (VPMOVZXWDrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
- (VPMOVZXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
- (VPMOVZXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
- (VPMOVZXDQrm addr:$src)>;
-}
-
-let Predicates = [UseSSE41] in {
- // Common patterns involving scalar load.
- def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
- (PMOVSXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
- (PMOVSXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
- (PMOVSXBWrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
- (PMOVSXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
- (PMOVSXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
- (PMOVSXWDrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
- (PMOVSXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
- (PMOVSXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
- (PMOVSXDQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
- (PMOVZXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
- (PMOVZXBWrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
- (PMOVZXBWrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
- (PMOVZXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
- (PMOVZXWDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
- (PMOVZXWDrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
- (PMOVZXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
- (PMOVZXDQrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
- (PMOVZXDQrm addr:$src)>;
-}
-
-multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId,
- OpndItins itins = DEFAULT_ITINS> {
- def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))], itins.rr>;
-
- def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst,
- (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))],
- itins.rm>;
-}
-
-multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
- def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>;
-
- def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst,
- (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>;
-}
-
-let Predicates = [HasAVX] in {
-defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
- VEX;
-defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
- VEX;
-defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
- VEX;
-defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
- VEX;
-}
-
-let Predicates = [HasAVX2] in {
-defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
- int_x86_avx2_pmovsxbd>, VEX, VEX_L;
-defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
- int_x86_avx2_pmovsxwq>, VEX, VEX_L;
-defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
- int_x86_avx2_pmovzxbd>, VEX, VEX_L;
-defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
- int_x86_avx2_pmovzxwq>, VEX, VEX_L;
-}
-
-defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd,
- SSE_INTALU_ITINS_P>;
-defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq,
- SSE_INTALU_ITINS_P>;
-defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd,
- SSE_INTALU_ITINS_P>;
-defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq,
- SSE_INTALU_ITINS_P>;
-
-let Predicates = [HasAVX] in {
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
- (VPMOVSXBDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
- (VPMOVSXWQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
- (VPMOVZXBDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
- (VPMOVZXWQrm addr:$src)>;
-}
-
-let Predicates = [UseSSE41] in {
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
- (PMOVSXBDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
- (PMOVSXWQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
- (PMOVZXBDrm addr:$src)>;
- def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
- (PMOVZXWQrm addr:$src)>;
-}
-
-multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId,
- OpndItins itins = DEFAULT_ITINS> {
- def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId VR128:$src))]>;
-
- // Expecting a i16 load any extended to i32 value.
- def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
+multiclass SS41I_pmovx_rrrm<bits<8> opc, string OpcodeStr, X86MemOperand MemOp,
+ RegisterClass OutRC, RegisterClass InRC,
+ OpndItins itins> {
+ def rr : SS48I<opc, MRMSrcReg, (outs OutRC:$dst), (ins InRC:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId (bitconvert
- (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>;
-}
+ [], itins.rr>,
+ Sched<[itins.Sched]>;
-multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId> {
- def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
+ def rm : SS48I<opc, MRMSrcMem, (outs OutRC:$dst), (ins MemOp:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId VR128:$src))]>;
-
- // Expecting a i16 load any extended to i32 value.
- def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (IntId (bitconvert
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>;
-}
-
-let Predicates = [HasAVX] in {
-defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
- VEX;
-defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
- VEX;
-}
-let Predicates = [HasAVX2] in {
-defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
- int_x86_avx2_pmovsxbq>, VEX, VEX_L;
-defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
- int_x86_avx2_pmovzxbq>, VEX, VEX_L;
-}
-defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq,
- SSE_INTALU_ITINS_P>;
-defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq,
- SSE_INTALU_ITINS_P>;
-
-let Predicates = [HasAVX2] in {
- def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
- def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
- def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
-
- def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
- def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
-
- def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
-
- def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
- (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
- (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
- (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
-
- def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
- (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
- (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
-
- def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
- (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
-
- def : Pat<(v8i32 (X86vsext (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
- (VPMOVSXWDYrm addr:$src)>;
- def : Pat<(v4i64 (X86vsext (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
- (VPMOVSXDQYrm addr:$src)>;
-
- def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
- (scalar_to_vector (loadi64 addr:$src))))))),
- (VPMOVSXBDYrm addr:$src)>;
- def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
- (scalar_to_vector (loadf64 addr:$src))))))),
- (VPMOVSXBDYrm addr:$src)>;
-
- def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
- (scalar_to_vector (loadi64 addr:$src))))))),
- (VPMOVSXWQYrm addr:$src)>;
- def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
- (scalar_to_vector (loadf64 addr:$src))))))),
- (VPMOVSXWQYrm addr:$src)>;
-
- def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
- (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVSXBQYrm addr:$src)>;
-}
-
-let Predicates = [HasAVX] in {
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVSXBQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVZXBQrm addr:$src)>;
+ [],
+ itins.rm>, Sched<[itins.Sched.Folded]>;
}
-let Predicates = [UseSSE41] in {
- def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
- def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
-
- def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
-
- def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
-
- // Common patterns involving scalar load
- def : Pat<(int_x86_sse41_pmovsxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVSXBQrm addr:$src)>;
-
- def : Pat<(int_x86_sse41_pmovzxbq
- (bitconvert (v4i32 (X86vzmovl
- (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVZXBQrm addr:$src)>;
-
- def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
- (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVSXWDrm addr:$src)>;
- def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
- (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVSXWDrm addr:$src)>;
- def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
- (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVSXBDrm addr:$src)>;
- def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
- (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVSXWQrm addr:$src)>;
- def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
- (scalar_to_vector (extloadi32i16 addr:$src))))))),
- (PMOVSXBQrm addr:$src)>;
- def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
- (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVSXDQrm addr:$src)>;
- def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
- (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVSXDQrm addr:$src)>;
- def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
- (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVSXBWrm addr:$src)>;
- def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
- (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVSXBWrm addr:$src)>;
+multiclass SS41I_pmovx_rm_all<bits<8> opc, string OpcodeStr,
+ X86MemOperand MemOp, X86MemOperand MemYOp,
+ OpndItins SSEItins, OpndItins AVXItins,
+ OpndItins AVX2Itins> {
+ defm NAME : SS41I_pmovx_rrrm<opc, OpcodeStr, MemOp, VR128, VR128, SSEItins>;
+ let Predicates = [HasAVX] in
+ defm V#NAME : SS41I_pmovx_rrrm<opc, !strconcat("v", OpcodeStr), MemOp,
+ VR128, VR128, AVXItins>, VEX;
+ let Predicates = [HasAVX2] in
+ defm V#NAME#Y : SS41I_pmovx_rrrm<opc, !strconcat("v", OpcodeStr), MemYOp,
+ VR256, VR128, AVX2Itins>, VEX, VEX_L;
+}
+
+multiclass SS41I_pmovx_rm<bits<8> opc, string OpcodeStr,
+ X86MemOperand MemOp, X86MemOperand MemYOp> {
+ defm PMOVSX#NAME : SS41I_pmovx_rm_all<opc, !strconcat("pmovsx", OpcodeStr),
+ MemOp, MemYOp,
+ SSE_INTALU_ITINS_SHUFF_P,
+ DEFAULT_ITINS_SHUFFLESCHED,
+ DEFAULT_ITINS_SHUFFLESCHED>;
+ defm PMOVZX#NAME : SS41I_pmovx_rm_all<!add(opc, 0x10),
+ !strconcat("pmovzx", OpcodeStr),
+ MemOp, MemYOp,
+ SSE_INTALU_ITINS_SHUFF_P,
+ DEFAULT_ITINS_SHUFFLESCHED,
+ DEFAULT_ITINS_SHUFFLESCHED>;
+}
+
+defm BW : SS41I_pmovx_rm<0x20, "bw", i64mem, i128mem>;
+defm WD : SS41I_pmovx_rm<0x23, "wd", i64mem, i128mem>;
+defm DQ : SS41I_pmovx_rm<0x25, "dq", i64mem, i128mem>;
+
+defm BD : SS41I_pmovx_rm<0x21, "bd", i32mem, i64mem>;
+defm WQ : SS41I_pmovx_rm<0x24, "wq", i32mem, i64mem>;
+
+defm BQ : SS41I_pmovx_rm<0x22, "bq", i16mem, i32mem>;
+
+// AVX2 Patterns
+multiclass SS41I_pmovx_avx2_patterns<string OpcPrefix, SDNode ExtOp> {
+ // Register-Register patterns
+ def : Pat<(v16i16 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BWYrr) VR128:$src)>;
+ def : Pat<(v8i32 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BDYrr) VR128:$src)>;
+ def : Pat<(v4i64 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BQYrr) VR128:$src)>;
+
+ def : Pat<(v8i32 (ExtOp (v8i16 VR128:$src))),
+ (!cast<I>(OpcPrefix#WDYrr) VR128:$src)>;
+ def : Pat<(v4i64 (ExtOp (v8i16 VR128:$src))),
+ (!cast<I>(OpcPrefix#WQYrr) VR128:$src)>;
+
+ def : Pat<(v4i64 (ExtOp (v4i32 VR128:$src))),
+ (!cast<I>(OpcPrefix#DQYrr) VR128:$src)>;
+
+ // On AVX2, we also support 256bit inputs.
+ // FIXME: remove these patterns when the old shuffle lowering goes away.
+ def : Pat<(v16i16 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BWYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BDYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (ExtOp (v32i8 VR256:$src))),
+ (!cast<I>(OpcPrefix#BQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (ExtOp (v16i16 VR256:$src))),
+ (!cast<I>(OpcPrefix#WDYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (ExtOp (v16i16 VR256:$src))),
+ (!cast<I>(OpcPrefix#WQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (ExtOp (v8i32 VR256:$src))),
+ (!cast<I>(OpcPrefix#DQYrr) (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ // AVX2 Register-Memory patterns
+ def : Pat<(v16i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWYrm) addr:$src)>;
+ def : Pat<(v16i16 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWYrm) addr:$src)>;
+ def : Pat<(v16i16 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWYrm) addr:$src)>;
+ def : Pat<(v16i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWYrm) addr:$src)>;
+
+ def : Pat<(v8i32 (ExtOp (bc_v16i8 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDYrm) addr:$src)>;
+
+ def : Pat<(v4i64 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQYrm) addr:$src)>;
+
+ def : Pat<(v8i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDYrm) addr:$src)>;
+ def : Pat<(v8i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDYrm) addr:$src)>;
+
+ def : Pat<(v4i64 (ExtOp (bc_v8i16 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#WQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQYrm) addr:$src)>;
+
+ def : Pat<(v4i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v4i32 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (v4i32 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQYrm) addr:$src)>;
+ def : Pat<(v4i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQYrm) addr:$src)>;
}
let Predicates = [HasAVX2] in {
- def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
- def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
- def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
-
- def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
- def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
-
- def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
-
- def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
- (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
- (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
- (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
-
- def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
- (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
- def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
- (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
-
- def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
- (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ defm : SS41I_pmovx_avx2_patterns<"VPMOVSX", X86vsext>;
+ defm : SS41I_pmovx_avx2_patterns<"VPMOVZX", X86vzext>;
+}
+
+// SSE4.1/AVX patterns.
+multiclass SS41I_pmovx_patterns<string OpcPrefix, SDNode ExtOp,
+ PatFrag ExtLoad16> {
+ def : Pat<(v8i16 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BWrr) VR128:$src)>;
+ def : Pat<(v4i32 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BDrr) VR128:$src)>;
+ def : Pat<(v2i64 (ExtOp (v16i8 VR128:$src))),
+ (!cast<I>(OpcPrefix#BQrr) VR128:$src)>;
+
+ def : Pat<(v4i32 (ExtOp (v8i16 VR128:$src))),
+ (!cast<I>(OpcPrefix#WDrr) VR128:$src)>;
+ def : Pat<(v2i64 (ExtOp (v8i16 VR128:$src))),
+ (!cast<I>(OpcPrefix#WQrr) VR128:$src)>;
+
+ def : Pat<(v2i64 (ExtOp (v4i32 VR128:$src))),
+ (!cast<I>(OpcPrefix#DQrr) VR128:$src)>;
+
+ def : Pat<(v8i16 (ExtOp (bc_v16i8 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BWrm) addr:$src)>;
+ def : Pat<(v8i16 (ExtOp (bc_v16i8 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BWrm) addr:$src)>;
+ def : Pat<(v8i16 (ExtOp (v16i8 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWrm) addr:$src)>;
+ def : Pat<(v8i16 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWrm) addr:$src)>;
+ def : Pat<(v8i16 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BWrm) addr:$src)>;
+
+ def : Pat<(v4i32 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BDrm) addr:$src)>;
+
+ def : Pat<(v2i64 (ExtOp (bc_v16i8 (v4i32 (scalar_to_vector (ExtLoad16 addr:$src)))))),
+ (!cast<I>(OpcPrefix#BQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v16i8 (vzmovl_v4i32 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v16i8 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (bc_v16i8 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#BQrm) addr:$src)>;
+
+ def : Pat<(v4i32 (ExtOp (bc_v8i16 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#WDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (bc_v8i16 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#WDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (v8i16 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDrm) addr:$src)>;
+ def : Pat<(v4i32 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WDrm) addr:$src)>;
+
+ def : Pat<(v2i64 (ExtOp (bc_v8i16 (v4i32 (scalar_to_vector (loadi32 addr:$src)))))),
+ (!cast<I>(OpcPrefix#WQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v8i16 (vzmovl_v4i32 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v8i16 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (bc_v8i16 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#WQrm) addr:$src)>;
+
+ def : Pat<(v2i64 (ExtOp (bc_v4i32 (v2i64 (scalar_to_vector (loadi64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#DQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (bc_v4i32 (v2f64 (scalar_to_vector (loadf64 addr:$src)))))),
+ (!cast<I>(OpcPrefix#DQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v4i32 (vzmovl_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (v4i32 (vzload_v2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQrm) addr:$src)>;
+ def : Pat<(v2i64 (ExtOp (bc_v4i32 (loadv2i64 addr:$src)))),
+ (!cast<I>(OpcPrefix#DQrm) addr:$src)>;
}
let Predicates = [HasAVX] in {
- def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
- def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
-
- def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
-
- def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
-
- def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (VPMOVZXBWrm addr:$src)>;
- def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (VPMOVZXBWrm addr:$src)>;
- def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVZXBDrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
- (VPMOVZXBQrm addr:$src)>;
-
- def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (VPMOVZXWDrm addr:$src)>;
- def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (VPMOVZXWDrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (VPMOVZXWQrm addr:$src)>;
-
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (VPMOVZXDQrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (VPMOVZXDQrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
- (VPMOVZXDQrm addr:$src)>;
-
- def : Pat<(v8i16 (X86vsext (v16i8 VR128:$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:$src)>;
+ defm : SS41I_pmovx_patterns<"VPMOVSX", X86vsext, extloadi32i16>;
+ defm : SS41I_pmovx_patterns<"VPMOVZX", X86vzext, loadi16_anyext>;
}
let Predicates = [UseSSE41] in {
- def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
- def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
-
- def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
- def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
-
- def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
-
- def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVZXBWrm addr:$src)>;
- def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVZXBWrm addr:$src)>;
- def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVZXBDrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
- (PMOVZXBQrm addr:$src)>;
-
- def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVZXWDrm addr:$src)>;
- def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVZXWDrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
- (PMOVZXWQrm addr:$src)>;
-
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
- (PMOVZXDQrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
- (PMOVZXDQrm addr:$src)>;
- def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
- (PMOVZXDQrm addr:$src)>;
+ defm : SS41I_pmovx_patterns<"PMOVSX", X86vsext, extloadi32i16>;
+ defm : SS41I_pmovx_patterns<"PMOVZX", X86vzext, loadi16_anyext>;
}
//===----------------------------------------------------------------------===//
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32orGR64:$dst, (X86pextrb (v16i8 VR128:$src1),
- imm:$src2))]>;
- let neverHasSideEffects = 1, mayStore = 1 in
+ imm:$src2))]>,
+ Sched<[WriteShuffle]>;
+ let hasSideEffects = 0, mayStore = 1,
+ SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>;
-// FIXME:
-// There's an AssertZext in the way of writing the store pattern
-// (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
+ [(store (i8 (trunc (assertzext (X86pextrb (v16i8 VR128:$src1),
+ imm:$src2)))), addr:$dst)]>;
}
let Predicates = [HasAVX] in
(ins VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>;
+ []>, Sched<[WriteShuffle]>;
- let neverHasSideEffects = 1, mayStore = 1 in
+ let hasSideEffects = 0, mayStore = 1,
+ SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- []>;
-// FIXME:
-// There's an AssertZext in the way of writing the store pattern
-// (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
+ [(store (i16 (trunc (assertzext (X86pextrw (v8i16 VR128:$src1),
+ imm:$src2)))), addr:$dst)]>;
}
let Predicates = [HasAVX] in
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32:$dst,
- (extractelt (v4i32 VR128:$src1), imm:$src2))]>;
+ (extractelt (v4i32 VR128:$src1), imm:$src2))]>,
+ Sched<[WriteShuffle]>;
+ let SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR64:$dst,
- (extractelt (v2i64 VR128:$src1), imm:$src2))]>, REX_W;
+ (extractelt (v2i64 VR128:$src1), imm:$src2))]>,
+ Sched<[WriteShuffle]>, REX_W;
+ let SchedRW = [WriteShuffleLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set GR32orGR64:$dst,
(extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))],
- itins.rr>;
+ itins.rr>, Sched<[WriteFBlend]>;
+ let SchedRW = [WriteFBlendLd, WriteRMW] in
def mr : SS4AIi8<opc, MRMDestMem, (outs),
(ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
!strconcat(OpcodeStr,
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>;
+ (X86pinsrb VR128:$src1, GR32orGR64:$src2, imm:$src3))]>,
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
- imm:$src3))]>;
+ imm:$src3))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>;
+ (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
- imm:$src3)))]>;
+ imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>;
+ (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
+ Sched<[WriteShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
!if(Is2Addr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
(v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
- imm:$src3)))]>;
+ imm:$src3)))]>, Sched<[WriteShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in
multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1,
OpndItins itins = DEFAULT_ITINS> {
def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>;
+ (X86insertps VR128:$src1, VR128:$src2, imm:$src3))], itins.rr>,
+ Sched<[WriteFShuffle]>;
def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
+ (ins VR128:$src1, f32mem:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(asm,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (X86insrtps VR128:$src1,
+ (X86insertps VR128:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
- imm:$src3))], itins.rm>;
+ imm:$src3))], itins.rm>,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
let ExeDomain = SSEPackedSingle in {
defm INSERTPS : SS41I_insertf32<0x21, "insertps", 1, SSE_INSERT_ITINS>;
}
+let Predicates = [UseSSE41] in {
+ // If we're inserting an element from a load or a null pshuf of a load,
+ // fold the load into the insertps instruction.
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1), (X86PShufd (v4f32
+ (scalar_to_vector (loadf32 addr:$src2))), (i8 0)),
+ imm:$src3)),
+ (INSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1), (X86PShufd
+ (loadv4f32 addr:$src2), (i8 0)), imm:$src3)),
+ (INSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+}
+
+let Predicates = [UseAVX] in {
+ // If we're inserting an element from a vbroadcast of a load, fold the
+ // load into the X86insertps instruction.
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1),
+ (X86VBroadcast (loadf32 addr:$src2)), imm:$src3)),
+ (VINSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+ def : Pat<(v4f32 (X86insertps (v4f32 VR128:$src1),
+ (X86VBroadcast (loadv4f32 addr:$src2)), imm:$src3)),
+ (VINSERTPSrm VR128:$src1, addr:$src2, imm:$src3)>;
+}
+
//===----------------------------------------------------------------------===//
// SSE4.1 - Round Instructions
//===----------------------------------------------------------------------===//
!strconcat(OpcodeStr,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))],
- IIC_SSE_ROUNDPS_REG>;
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAdd]>;
// Vector intrinsic operation, mem
def PSm : SS4AIi8<opcps, MRMSrcMem,
"ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
(V4F32Int (mem_frag32 addr:$src1),imm:$src2))],
- IIC_SSE_ROUNDPS_MEM>;
+ IIC_SSE_ROUNDPS_MEM>, Sched<[WriteFAddLd]>;
} // ExeDomain = SSEPackedSingle
let ExeDomain = SSEPackedDouble in {
!strconcat(OpcodeStr,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))],
- IIC_SSE_ROUNDPS_REG>;
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAdd]>;
// Vector intrinsic operation, mem
def PDm : SS4AIi8<opcpd, MRMSrcMem,
"pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
(V2F64Int (mem_frag64 addr:$src1),imm:$src2))],
- IIC_SSE_ROUNDPS_REG>;
+ IIC_SSE_ROUNDPS_REG>, Sched<[WriteFAddLd]>;
} // ExeDomain = SSEPackedDouble
}
"ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>;
+ []>, Sched<[WriteFAdd]>;
// Intrinsic operation, reg.
let isCodeGenOnly = 1 in
"ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>;
+ [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
+ Sched<[WriteFAdd]>;
// Intrinsic operation, mem.
def SSm : SS4AIi8<opcss, MRMSrcMem,
!strconcat(OpcodeStr,
"ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>;
+ (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
+ Sched<[WriteFAddLd, ReadAfterLd]>;
// Operation, reg.
let hasSideEffects = 0 in
"sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- []>;
+ []>, Sched<[WriteFAdd]>;
// Intrinsic operation, reg.
let isCodeGenOnly = 1 in
"sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>;
+ [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
+ Sched<[WriteFAdd]>;
// Intrinsic operation, mem.
def SDm : SS4AIi8<opcsd, MRMSrcMem,
!strconcat(OpcodeStr,
"sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set VR128:$dst,
- (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>;
+ (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
+ Sched<[WriteFAddLd, ReadAfterLd]>;
} // ExeDomain = GenericDomain
}
def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
- VEX;
+ Sched<[WriteVecLogic]>, VEX;
def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS,(X86ptest VR128:$src1, (loadv2i64 addr:$src2)))]>,
- VEX;
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX;
def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
- VEX, VEX_L;
+ Sched<[WriteVecLogic]>, VEX, VEX_L;
def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
"vptest\t{$src2, $src1|$src1, $src2}",
[(set EFLAGS,(X86ptest VR256:$src1, (loadv4i64 addr:$src2)))]>,
- VEX, VEX_L;
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX, VEX_L;
}
let Defs = [EFLAGS] in {
def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
"ptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>;
+ [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
+ Sched<[WriteVecLogic]>;
def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
"ptest\t{$src2, $src1|$src1, $src2}",
- [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>;
+ [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
+ Sched<[WriteVecLogicLd, ReadAfterLd]>;
}
// The bit test instructions below are AVX only
X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
- [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, VEX;
+ [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>,
+ Sched<[WriteVecLogic]>, VEX;
def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
- [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>, VEX;
+ [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
+ Sched<[WriteVecLogicLd, ReadAfterLd]>, VEX;
}
let Defs = [EFLAGS], Predicates = [HasAVX] in {
def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
"popcnt{w}\t{$src, $dst|$dst, $src}",
[(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)],
- IIC_SSE_POPCNT_RR>,
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>,
OpSize16, XS;
def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
"popcnt{w}\t{$src, $dst|$dst, $src}",
[(set GR16:$dst, (ctpop (loadi16 addr:$src))),
- (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, OpSize16, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, OpSize16, XS;
def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
"popcnt{l}\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)],
- IIC_SSE_POPCNT_RR>,
- XS;
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>,
+ OpSize32, XS;
+
def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
"popcnt{l}\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (ctpop (loadi32 addr:$src))),
- (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, OpSize32, XS;
def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
"popcnt{q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)],
- IIC_SSE_POPCNT_RR>,
- XS;
+ IIC_SSE_POPCNT_RR>, Sched<[WriteFAdd]>, XS;
def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
"popcnt{q}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (ctpop (loadi64 addr:$src))),
- (implicit EFLAGS)], IIC_SSE_POPCNT_RM>, XS;
+ (implicit EFLAGS)], IIC_SSE_POPCNT_RM>,
+ Sched<[WriteFAddLd]>, XS;
}
// SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
- Intrinsic IntId128> {
+ Intrinsic IntId128,
+ X86FoldableSchedWrite Sched> {
def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (IntId128 VR128:$src))]>;
+ [(set VR128:$dst, (IntId128 VR128:$src))]>,
+ Sched<[Sched]>;
def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (IntId128 (bitconvert (memopv2i64 addr:$src))))]>;
+ (IntId128 (bitconvert (memopv2i64 addr:$src))))]>,
+ Sched<[Sched.Folded]>;
}
+// PHMIN has the same profile as PSAD, thus we use the same scheduling
+// model, although the naming is misleading.
let Predicates = [HasAVX] in
defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
- int_x86_sse41_phminposuw>, VEX;
+ int_x86_sse41_phminposuw,
+ WriteVecIMul>, VEX;
defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
- int_x86_sse41_phminposuw>;
+ int_x86_sse41_phminposuw,
+ WriteVecIMul>;
/// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
(IntId128 VR128:$src1, (bitconvert (memopv2i64 addr:$src2))))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
/// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
- Intrinsic IntId256> {
+ Intrinsic IntId256,
+ X86FoldableSchedWrite Sched> {
let isCommutable = 1 in
def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>;
+ [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
+ Sched<[Sched]>;
def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
- (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>;
+ (IntId256 VR256:$src1, (bitconvert (loadv4i64 addr:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
X86MemOperand x86memop, bit Is2Addr = 1,
- OpndItins itins = DEFAULT_ITINS> {
+ OpndItins itins = SSE_INTALU_ITINS_P> {
let isCommutable = 1 in
def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
+ [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
+ Sched<[itins.Sched]>;
def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set RC:$dst,
- (OpVT (OpNode RC:$src1, (bitconvert (memop_frag addr:$src2)))))]>;
+ (OpVT (OpNode RC:$src1, (bitconvert (memop_frag addr:$src2)))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
-let Predicates = [HasAVX] in {
+/// SS48I_binop_rm2 - Simple SSE41 binary operator with different src and dst
+/// types.
+multiclass SS48I_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ ValueType DstVT, ValueType SrcVT, RegisterClass RC,
+ PatFrag memop_frag, X86MemOperand x86memop,
+ OpndItins itins,
+ bit IsCommutable = 0, bit Is2Addr = 1> {
+ let isCommutable = IsCommutable in
+ def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
+ (ins RC:$src1, RC:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>,
+ Sched<[itins.Sched]>;
+ def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
+ (ins RC:$src1, x86memop:$src2),
+ !if(Is2Addr,
+ !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
+ !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
+ [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
+ (bitconvert (memop_frag addr:$src2)))))]>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
+}
+
+let Predicates = [HasAVX, NoVLX] 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,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
- loadv2i64, i128mem, 0>, VEX_4V;
- defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
- 0>, VEX_4V;
+ loadv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
+ defm VPMULDQ : SS48I_binop_rm2<0x28, "vpmuldq", X86pmuldq, v2i64, v4i32,
+ VR128, loadv2i64, i128mem,
+ SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
}
-let Predicates = [HasAVX2] in {
+let Predicates = [HasAVX2, NoVLX] 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,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
- loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
- defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
- int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
+ loadv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
+ defm VPMULDQY : SS48I_binop_rm2<0x28, "vpmuldq", X86pmuldq, v4i64, v8i32,
+ VR256, loadv4i64, i256mem,
+ SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
}
let Constraints = "$src1 = $dst" in {
let isCommutable = 0 in
- defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
memopv2i64, i128mem, 1, SSE_INTALU_ITINS_P>;
- defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq,
- 1, SSE_INTMUL_ITINS_P>;
+ defm PMULDQ : SS48I_binop_rm2<0x28, "pmuldq", X86pmuldq, v2i64, v4i32,
+ VR128, memopv2i64, i128mem,
+ SSE_INTMUL_ITINS_P, 1>;
}
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
- memopv2i64, i128mem, 0>, VEX_4V;
+ memopv2i64, i128mem, 0, SSE_PMULLD_ITINS>,
+ VEX_4V;
defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
- memopv2i64, i128mem, 0>, VEX_4V;
+ memopv2i64, i128mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V;
}
let Predicates = [HasAVX2] in {
defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
- memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ memopv4i64, i256mem, 0, SSE_PMULLD_ITINS>,
+ VEX_4V, VEX_L;
defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
- memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ memopv4i64, i256mem, 0, SSE_INTALU_ITINS_P>,
+ VEX_4V, VEX_L;
}
let Constraints = "$src1 = $dst" in {
OpndItins itins = DEFAULT_ITINS> {
let isCommutable = 1 in
def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, u32u8imm:$src3),
+ (ins RC:$src1, RC:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
- [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>;
+ [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))], itins.rr>,
+ Sched<[itins.Sched]>;
def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, u8imm:$src3),
!if(Is2Addr,
!strconcat(OpcodeStr,
"\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
[(set RC:$dst,
(IntId RC:$src1,
- (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>;
+ (bitconvert (memop_frag addr:$src2)), imm:$src3))], itins.rm>,
+ Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let isCommutable = 0 in {
- let ExeDomain = SSEPackedSingle in {
- defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
- VR128, loadv4f32, f128mem, 0>, VEX_4V;
- defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
- int_x86_avx_blend_ps_256, VR256, loadv8f32,
- f256mem, 0>, VEX_4V, VEX_L;
- }
- let ExeDomain = SSEPackedDouble in {
- defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
- VR128, loadv2f64, f128mem, 0>, VEX_4V;
- defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
- int_x86_avx_blend_pd_256,VR256, loadv4f64,
- f256mem, 0>, VEX_4V, VEX_L;
- }
- defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
- VR128, loadv2i64, i128mem, 0>, VEX_4V;
- defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
- VR128, loadv2i64, i128mem, 0>, VEX_4V;
+ defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
+ VR128, loadv2i64, i128mem, 0,
+ DEFAULT_ITINS_MPSADSCHED>, VEX_4V;
+ }
+
+ let ExeDomain = SSEPackedSingle in {
+ defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
+ VR128, loadv4f32, f128mem, 0,
+ DEFAULT_ITINS_FBLENDSCHED>, VEX_4V;
+ defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
+ int_x86_avx_blend_ps_256, VR256, loadv8f32,
+ f256mem, 0, DEFAULT_ITINS_FBLENDSCHED>,
+ VEX_4V, VEX_L;
+ }
+ let ExeDomain = SSEPackedDouble in {
+ defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
+ VR128, loadv2f64, f128mem, 0,
+ DEFAULT_ITINS_FBLENDSCHED>, VEX_4V;
+ defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
+ int_x86_avx_blend_pd_256,VR256, loadv4f64,
+ f256mem, 0, DEFAULT_ITINS_FBLENDSCHED>,
+ VEX_4V, VEX_L;
}
+ defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
+ VR128, loadv2i64, i128mem, 0,
+ DEFAULT_ITINS_BLENDSCHED>, VEX_4V;
+
let ExeDomain = SSEPackedSingle in
defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
- VR128, loadv4f32, f128mem, 0>, VEX_4V;
+ VR128, loadv4f32, f128mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V;
let ExeDomain = SSEPackedDouble in
defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
- VR128, loadv2f64, f128mem, 0>, VEX_4V;
+ VR128, loadv2f64, f128mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V;
let ExeDomain = SSEPackedSingle in
defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
- VR256, loadv8f32, i256mem, 0>, VEX_4V, VEX_L;
+ VR256, loadv8f32, i256mem, 0,
+ SSE_DPPS_ITINS>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX2] in {
let isCommutable = 0 in {
- defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
- VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
- VR256, loadv4i64, i256mem, 0>, VEX_4V, VEX_L;
+ VR256, loadv4i64, i256mem, 0,
+ DEFAULT_ITINS_MPSADSCHED>, VEX_4V, VEX_L;
}
+ defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
+ VR256, loadv4i64, i256mem, 0,
+ DEFAULT_ITINS_BLENDSCHED>, VEX_4V, VEX_L;
}
let Constraints = "$src1 = $dst" in {
let isCommutable = 0 in {
+ defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
+ VR128, memopv2i64, i128mem,
+ 1, SSE_MPSADBW_ITINS>;
+ }
let ExeDomain = SSEPackedSingle in
defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
VR128, memopv4f32, f128mem,
- 1, SSE_INTALU_ITINS_P>;
+ 1, SSE_INTALU_ITINS_FBLEND_P>;
let ExeDomain = SSEPackedDouble in
defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
VR128, memopv2f64, f128mem,
- 1, SSE_INTALU_ITINS_P>;
+ 1, SSE_INTALU_ITINS_FBLEND_P>;
defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
VR128, memopv2i64, i128mem,
- 1, SSE_INTALU_ITINS_P>;
- defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
- VR128, memopv2i64, i128mem,
- 1, SSE_INTMUL_ITINS_P>;
- }
+ 1, SSE_INTALU_ITINS_BLEND_P>;
let ExeDomain = SSEPackedSingle in
defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
VR128, memopv4f32, f128mem, 1,
/// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
RegisterClass RC, X86MemOperand x86memop,
- PatFrag mem_frag, Intrinsic IntId> {
+ PatFrag mem_frag, Intrinsic IntId,
+ X86FoldableSchedWrite Sched> {
def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, RC:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
- NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM;
+ NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM,
+ Sched<[Sched]>;
def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop:$src2, RC:$src3),
[(set RC:$dst,
(IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
RC:$src3))],
- NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM;
+ NoItinerary, SSEPackedInt>, TAPD, VEX_4V, VEX_I8IMM,
+ Sched<[Sched.Folded, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
let ExeDomain = SSEPackedDouble in {
defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
- loadv2f64, int_x86_sse41_blendvpd>;
+ loadv2f64, int_x86_sse41_blendvpd,
+ WriteFVarBlend>;
defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
- loadv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
+ loadv4f64, int_x86_avx_blendv_pd_256,
+ WriteFVarBlend>, VEX_L;
} // ExeDomain = SSEPackedDouble
let ExeDomain = SSEPackedSingle in {
defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
- loadv4f32, int_x86_sse41_blendvps>;
+ loadv4f32, int_x86_sse41_blendvps,
+ WriteFVarBlend>;
defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
- loadv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
+ loadv8f32, int_x86_avx_blendv_ps_256,
+ WriteFVarBlend>, VEX_L;
} // ExeDomain = SSEPackedSingle
defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
- loadv2i64, int_x86_sse41_pblendvb>;
+ loadv2i64, int_x86_sse41_pblendvb,
+ WriteVarBlend>;
}
let Predicates = [HasAVX2] in {
defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
- loadv4i64, int_x86_avx2_pblendvb>, VEX_L;
+ loadv4i64, int_x86_avx2_pblendvb,
+ WriteVarBlend>, VEX_L;
}
let Predicates = [HasAVX] in {
(VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
}
+// Patterns
+let Predicates = [UseAVX] in {
+ let AddedComplexity = 15 in {
+ // Move scalar to XMM zero-extended, zeroing a VR128 then do a
+ // MOVS{S,D} to the lower bits.
+ def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
+ (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
+ def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
+ (VBLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
+ (VPBLENDWrri (v4i32 (V_SET0)), VR128:$src, (i8 3))>;
+ def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
+ (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
+
+ // Move low f32 and clear high bits.
+ def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
+ (VBLENDPSYrri (v8f32 (AVX_SET0)), VR256:$src, (i8 1))>;
+ def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
+ (VBLENDPSYrri (v8i32 (AVX_SET0)), VR256:$src, (i8 1))>;
+ }
+
+ def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
+ (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
+ (SUBREG_TO_REG (i32 0),
+ (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
+ sub_xmm)>;
+ def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
+ (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
+ (SUBREG_TO_REG (i64 0),
+ (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
+ sub_xmm)>;
+
+ // Move low f64 and clear high bits.
+ def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
+ (VBLENDPDYrri (v4f64 (AVX_SET0)), VR256:$src, (i8 1))>;
+
+ def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
+ (VBLENDPDYrri (v4i64 (AVX_SET0)), VR256:$src, (i8 1))>;
+}
+
+let Predicates = [UseSSE41] in {
+ // With SSE41 we can use blends for these patterns.
+ def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
+ (BLENDPSrri (v4f32 (V_SET0)), VR128:$src, (i8 1))>;
+ def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
+ (PBLENDWrri (v4i32 (V_SET0)), VR128:$src, (i8 3))>;
+ def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
+ (BLENDPDrri (v2f64 (V_SET0)), VR128:$src, (i8 1))>;
+}
+
+
/// SS41I_ternary_int - SSE 4.1 ternary operator
let Uses = [XMM0], Constraints = "$src1 = $dst" in {
multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
!strconcat(OpcodeStr,
"\t{$src2, $dst|$dst, $src2}"),
[(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))],
- itins.rr>;
+ itins.rr>, Sched<[itins.Sched]>;
def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, x86memop:$src2),
[(set VR128:$dst,
(IntId VR128:$src1,
(bitconvert (mem_frag addr:$src2)), XMM0))],
- itins.rm>;
+ itins.rm>, Sched<[itins.Sched.Folded, ReadAfterLd]>;
}
}
let ExeDomain = SSEPackedDouble in
defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
- int_x86_sse41_blendvpd>;
+ int_x86_sse41_blendvpd,
+ DEFAULT_ITINS_FBLENDSCHED>;
let ExeDomain = SSEPackedSingle in
defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
- int_x86_sse41_blendvps>;
+ int_x86_sse41_blendvps,
+ DEFAULT_ITINS_FBLENDSCHED>;
defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
- int_x86_sse41_pblendvb>;
+ int_x86_sse41_pblendvb,
+ DEFAULT_ITINS_VARBLENDSCHED>;
// Aliases with the implicit xmm0 argument
def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, xmm0}",
}
+let SchedRW = [WriteLoad] in {
let Predicates = [HasAVX] in
def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"vmovntdqa\t{$src, $dst|$dst, $src}",
@@ -7241,6 +7529,7 @@ def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
"movntdqa\t{$src, $dst|$dst, $src}",
[(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>;
+} // SchedRW
//===----------------------------------------------------------------------===//
// SSE4.2 - Compare Instructions
// Packed Compare Implicit Length Strings, Return Mask
multiclass pseudo_pcmpistrm<string asm> {
def REG : PseudoI<(outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
imm:$src3))]>;
def MEM : PseudoI<(outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
[(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
(bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
multiclass pcmpistrm_SS42AI<string asm> {
def rr : SS42AI<0x62, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>;
+ []>, Sched<[WritePCmpIStrM]>;
let mayLoad = 1 in
def rm :SS42AI<0x62, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>;
+ []>, Sched<[WritePCmpIStrMLd, ReadAfterLd]>;
}
-let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
+let Defs = [XMM0, EFLAGS], hasSideEffects = 0 in {
let Predicates = [HasAVX] in
defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
// Packed Compare Explicit Length Strings, Return Mask
multiclass pseudo_pcmpestrm<string asm> {
def REG : PseudoI<(outs VR128:$dst),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ (ins VR128:$src1, VR128:$src3, u8imm:$src5),
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128
VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
def MEM : PseudoI<(outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ (ins VR128:$src1, i128mem:$src3, u8imm:$src5),
[(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
(bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
}
multiclass SS42AI_pcmpestrm<string asm> {
def rr : SS42AI<0x60, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ (ins VR128:$src1, VR128:$src3, u8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>;
+ []>, Sched<[WritePCmpEStrM]>;
let mayLoad = 1 in
def rm : SS42AI<0x60, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ (ins VR128:$src1, i128mem:$src3, u8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>;
+ []>, Sched<[WritePCmpEStrMLd, ReadAfterLd]>;
}
-let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], hasSideEffects = 0 in {
let Predicates = [HasAVX] in
defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
// Packed Compare Implicit Length Strings, Return Index
multiclass pseudo_pcmpistri<string asm> {
def REG : PseudoI<(outs GR32:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
[(set GR32:$dst, EFLAGS,
(X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
def MEM : PseudoI<(outs GR32:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
[(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
(bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
}
multiclass SS42AI_pcmpistri<string asm> {
def rr : SS42AI<0x63, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>;
+ []>, Sched<[WritePCmpIStrI]>;
let mayLoad = 1 in
def rm : SS42AI<0x63, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
!strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
- []>;
+ []>, Sched<[WritePCmpIStrILd, ReadAfterLd]>;
}
-let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
+let Defs = [ECX, EFLAGS], hasSideEffects = 0 in {
let Predicates = [HasAVX] in
defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
// Packed Compare Explicit Length Strings, Return Index
multiclass pseudo_pcmpestri<string asm> {
def REG : PseudoI<(outs GR32:$dst),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ (ins VR128:$src1, VR128:$src3, u8imm:$src5),
[(set GR32:$dst, EFLAGS,
(X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
def MEM : PseudoI<(outs GR32:$dst),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ (ins VR128:$src1, i128mem:$src3, u8imm:$src5),
[(set GR32:$dst, EFLAGS,
(X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
imm:$src5))]>;
multiclass SS42AI_pcmpestri<string asm> {
def rr : SS42AI<0x61, MRMSrcReg, (outs),
- (ins VR128:$src1, VR128:$src3, i8imm:$src5),
+ (ins VR128:$src1, VR128:$src3, u8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>;
+ []>, Sched<[WritePCmpEStrI]>;
let mayLoad = 1 in
def rm : SS42AI<0x61, MRMSrcMem, (outs),
- (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
+ (ins VR128:$src1, i128mem:$src3, u8imm:$src5),
!strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
- []>;
+ []>, Sched<[WritePCmpEStrILd, ReadAfterLd]>;
}
-let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
+let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], hasSideEffects = 0 in {
let Predicates = [HasAVX] in
defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
RegisterClass RCIn, SDPatternOperator Int> :
SS42FI<opc, MRMSrcReg, (outs RCOut:$dst), (ins RCOut:$src1, RCIn:$src2),
!strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
- [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>;
+ [(set RCOut:$dst, (Int RCOut:$src1, RCIn:$src2))], IIC_CRC32_REG>,
+ Sched<[WriteFAdd]>;
class SS42I_crc32m<bits<8> opc, string asm, RegisterClass RCOut,
X86MemOperand x86memop, SDPatternOperator Int> :
SS42FI<opc, MRMSrcMem, (outs RCOut:$dst), (ins RCOut:$src1, x86memop:$src2),
!strconcat(asm, "\t{$src2, $src1|$src1, $src2}"),
[(set RCOut:$dst, (Int RCOut:$src1, (load addr:$src2)))],
- IIC_CRC32_MEM>;
+ IIC_CRC32_MEM>, Sched<[WriteFAddLd, ReadAfterLd]>;
let Constraints = "$src1 = $dst" in {
def CRC32r32m8 : SS42I_crc32m<0xF0, "crc32{b}", GR32, i8mem,
let Constraints = "$src1 = $dst", Predicates = [HasSHA] in {
def SHA1RNDS4rri : Ii8<0xCC, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
"sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst,
(int_x86_sha1rnds4 VR128:$src1, VR128:$src2,
(i8 imm:$src3)))]>, TA;
def SHA1RNDS4rmi : Ii8<0xCC, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
"sha1rnds4\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst,
(int_x86_sha1rnds4 VR128:$src1,
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
- [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>;
+ [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
+ Sched<[WriteAESDecEnc]>;
def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!if(Is2Addr,
!strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
[(set VR128:$dst,
- (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>;
+ (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>,
+ Sched<[WriteAESDecEncLd, ReadAfterLd]>;
}
// Perform One Round of an AES Encryption/Decryption Flow
(ins VR128:$src1),
"vaesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst,
- (int_x86_aesni_aesimc VR128:$src1))]>,
+ (int_x86_aesni_aesimc VR128:$src1))]>, Sched<[WriteAESIMC]>,
VEX;
def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1),
"vaesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst, (int_x86_aesni_aesimc (loadv2i64 addr:$src1)))]>,
- VEX;
+ Sched<[WriteAESIMCLd]>, VEX;
}
def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1),
"aesimc\t{$src1, $dst|$dst, $src1}",
[(set VR128:$dst,
- (int_x86_aesni_aesimc VR128:$src1))]>;
+ (int_x86_aesni_aesimc VR128:$src1))]>, Sched<[WriteAESIMC]>;
def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1),
"aesimc\t{$src1, $dst|$dst, $src1}",
- [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>;
+ [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
+ Sched<[WriteAESIMCLd]>;
// AES Round Key Generation Assist
let Predicates = [HasAVX, HasAES] in {
def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, i8imm:$src2),
+ (ins VR128:$src1, u8imm:$src2),
"vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
- VEX;
+ Sched<[WriteAESKeyGen]>, VEX;
def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
- (ins i128mem:$src1, i8imm:$src2),
+ (ins i128mem:$src1, u8imm:$src2),
"vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_aesni_aeskeygenassist (loadv2i64 addr:$src1), imm:$src2))]>,
- VEX;
+ Sched<[WriteAESKeyGenLd]>, VEX;
}
def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, i8imm:$src2),
+ (ins VR128:$src1, u8imm:$src2),
"aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
- (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>;
+ (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
+ Sched<[WriteAESKeyGen]>;
def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
- (ins i128mem:$src1, i8imm:$src2),
+ (ins i128mem:$src1, u8imm:$src2),
"aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
- (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>;
+ (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
+ Sched<[WriteAESKeyGenLd]>;
//===----------------------------------------------------------------------===//
// PCLMUL Instructions
// AVX carry-less Multiplication instructions
def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128:$dst,
- (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
+ (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>,
+ Sched<[WriteCLMul]>;
def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
"vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
- (loadv2i64 addr:$src2), imm:$src3))]>;
+ (loadv2i64 addr:$src2), imm:$src3))]>,
+ Sched<[WriteCLMulLd, ReadAfterLd]>;
// Carry-less Multiplication instructions
let Constraints = "$src1 = $dst" in {
def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR128:$src1, VR128:$src2, u8imm:$src3),
"pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst,
(int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))],
- IIC_SSE_PCLMULQDQ_RR>;
+ IIC_SSE_PCLMULQDQ_RR>, Sched<[WriteCLMul]>;
def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR128:$src1, i128mem:$src2, u8imm:$src3),
"pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
[(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
(memopv2i64 addr:$src2), imm:$src3))],
- IIC_SSE_PCLMULQDQ_RM>;
+ IIC_SSE_PCLMULQDQ_RM>,
+ Sched<[WriteCLMulLd, ReadAfterLd]>;
} // Constraints = "$src1 = $dst"
multiclass pclmul_alias<string asm, int immop> {
def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
- (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
+ (PCLMULQDQrr VR128:$dst, VR128:$src, immop), 0>;
def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
- (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
+ (PCLMULQDQrm VR128:$dst, i128mem:$src, immop), 0>;
def : InstAlias<!strconcat("vpclmul", asm,
"dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
- (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
+ (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop),
+ 0>;
def : InstAlias<!strconcat("vpclmul", asm,
"dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
- (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
+ (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop),
+ 0>;
}
defm : pclmul_alias<"hqhq", 0x11>;
defm : pclmul_alias<"hqlq", 0x01>;
let Predicates = [HasSSE4A] in {
let Constraints = "$src = $dst" in {
-def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
- (ins VR128:$src, i8imm:$len, i8imm:$idx),
+def EXTRQI : Ii8<0x78, MRMXr, (outs VR128:$dst),
+ (ins VR128:$src, u8imm:$len, u8imm:$idx),
"extrq\t{$idx, $len, $src|$src, $len, $idx}",
[(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
imm:$idx))]>, PD;
VR128:$mask))]>, PD;
def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
+ (ins VR128:$src, VR128:$src2, u8imm:$len, u8imm:$idx),
"insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
[(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
VR128:$src2, imm:$len, imm:$idx))]>, XD;
// destination operand
//
class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
- X86MemOperand x86memop, Intrinsic Int> :
+ X86MemOperand x86memop, Intrinsic Int, SchedWrite Sched> :
AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (Int addr:$src))]>, VEX;
+ [(set RC:$dst, (Int addr:$src))]>, Sched<[Sched]>, VEX;
+
+class avx_broadcast_no_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
+ X86MemOperand x86memop, ValueType VT,
+ PatFrag ld_frag, SchedWrite Sched> :
+ AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (VT (X86VBroadcast (ld_frag addr:$src))))]>,
+ Sched<[Sched]>, VEX {
+ let mayLoad = 1;
+}
// AVX2 adds register forms
class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
- Intrinsic Int> :
+ Intrinsic Int, SchedWrite Sched> :
AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (Int VR128:$src))]>, VEX;
+ [(set RC:$dst, (Int VR128:$src))]>, Sched<[Sched]>, VEX;
let ExeDomain = SSEPackedSingle in {
- def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
- int_x86_avx_vbroadcast_ss>;
- def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
- int_x86_avx_vbroadcast_ss_256>, VEX_L;
+ def VBROADCASTSSrm : avx_broadcast_no_int<0x18, "vbroadcastss", VR128,
+ f32mem, v4f32, loadf32, WriteLoad>;
+ def VBROADCASTSSYrm : avx_broadcast_no_int<0x18, "vbroadcastss", VR256,
+ f32mem, v8f32, loadf32,
+ WriteFShuffleLd>, VEX_L;
}
let ExeDomain = SSEPackedDouble in
-def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
- int_x86_avx_vbroadcast_sd_256>, VEX_L;
+def VBROADCASTSDYrm : avx_broadcast_no_int<0x19, "vbroadcastsd", VR256, f64mem,
+ v4f64, loadf64, WriteFShuffleLd>, VEX_L;
def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
- int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
+ int_x86_avx_vbroadcastf128_pd_256,
+ WriteFShuffleLd>, VEX_L;
let ExeDomain = SSEPackedSingle in {
def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
- int_x86_avx2_vbroadcast_ss_ps>;
+ int_x86_avx2_vbroadcast_ss_ps,
+ WriteFShuffle>;
def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
- int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
+ int_x86_avx2_vbroadcast_ss_ps_256,
+ WriteFShuffle256>, VEX_L;
}
let ExeDomain = SSEPackedDouble in
def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
- int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
+ int_x86_avx2_vbroadcast_sd_pd_256,
+ WriteFShuffle256>, VEX_L;
let Predicates = [HasAVX2] in
def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
- int_x86_avx2_vbroadcasti128>, VEX_L;
+ int_x86_avx2_vbroadcasti128, WriteLoad>,
+ VEX_L;
let Predicates = [HasAVX] in
def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
//===----------------------------------------------------------------------===//
// VINSERTF128 - Insert packed floating-point values
//
-let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
+let hasSideEffects = 0, ExeDomain = SSEPackedSingle in {
def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR256:$src1, VR128:$src2, u8imm:$src3),
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V, VEX_L;
+ []>, Sched<[WriteFShuffle]>, VEX_4V, VEX_L;
let mayLoad = 1 in
def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
- (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
+ (ins VR256:$src1, f128mem:$src2, u8imm:$src3),
"vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V, VEX_L;
+ []>, Sched<[WriteFShuffleLd, ReadAfterLd]>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX] in {
@@ -7800,6 +8116,49 @@ def : Pat<(vinsert128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
(INSERT_get_vinsert128_imm VR256:$ins))>;
}
+// Combine two consecutive 16-byte loads with a common destination register into
+// one 32-byte load to that register.
+let Predicates = [HasAVX, HasFastMem32] in {
+ def : Pat<(insert_subvector
+ (v8f32 (insert_subvector undef, (loadv4f32 addr:$src), (iPTR 0))),
+ (loadv4f32 (add addr:$src, (iPTR 16))),
+ (iPTR 4)),
+ (VMOVUPSYrm addr:$src)>;
+
+ def : Pat<(insert_subvector
+ (v4f64 (insert_subvector undef, (loadv2f64 addr:$src), (iPTR 0))),
+ (loadv2f64 (add addr:$src, (iPTR 16))),
+ (iPTR 2)),
+ (VMOVUPDYrm addr:$src)>;
+
+ def : Pat<(insert_subvector
+ (v32i8 (insert_subvector
+ undef, (bc_v16i8 (loadv2i64 addr:$src)), (iPTR 0))),
+ (bc_v16i8 (loadv2i64 (add addr:$src, (iPTR 16)))),
+ (iPTR 16)),
+ (VMOVDQUYrm addr:$src)>;
+
+ def : Pat<(insert_subvector
+ (v16i16 (insert_subvector
+ undef, (bc_v8i16 (loadv2i64 addr:$src)), (iPTR 0))),
+ (bc_v8i16 (loadv2i64 (add addr:$src, (iPTR 16)))),
+ (iPTR 8)),
+ (VMOVDQUYrm addr:$src)>;
+
+ def : Pat<(insert_subvector
+ (v8i32 (insert_subvector
+ undef, (bc_v4i32 (loadv2i64 addr:$src)), (iPTR 0))),
+ (bc_v4i32 (loadv2i64 (add addr:$src, (iPTR 16)))),
+ (iPTR 4)),
+ (VMOVDQUYrm addr:$src)>;
+
+ def : Pat<(insert_subvector
+ (v4i64 (insert_subvector undef, (loadv2i64 addr:$src), (iPTR 0))),
+ (loadv2i64 (add addr:$src, (iPTR 16))),
+ (iPTR 2)),
+ (VMOVDQUYrm addr:$src)>;
+}
+
let Predicates = [HasAVX1Only] in {
def : Pat<(vinsert128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
(iPTR imm)),
//===----------------------------------------------------------------------===//
// VEXTRACTF128 - Extract packed floating-point values
//
-let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
+let hasSideEffects = 0, ExeDomain = SSEPackedSingle in {
def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
- (ins VR256:$src1, i8imm:$src2),
+ (ins VR256:$src1, u8imm:$src2),
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- []>, VEX, VEX_L;
+ []>, Sched<[WriteFShuffle]>, VEX, VEX_L;
let mayStore = 1 in
def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
- (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
+ (ins f128mem:$dst, VR256:$src1, u8imm:$src2),
"vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- []>, VEX, VEX_L;
+ []>, Sched<[WriteStore]>, VEX, VEX_L;
}
// AVX1 patterns
def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
+ [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V,
+ Sched<[WriteFShuffle]>;
def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86memop_i:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (IntVar RC:$src1,
- (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
+ (bitconvert (i_frag addr:$src2))))]>, VEX_4V,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, i8imm:$src2),
+ (ins RC:$src1, u8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
+ [(set RC:$dst, (vt (X86VPermilpi RC:$src1, (i8 imm:$src2))))]>, VEX,
+ Sched<[WriteFShuffle]>;
def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
- (ins x86memop_f:$src1, i8imm:$src2),
+ (ins x86memop_f:$src1, u8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst,
- (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
+ (vt (X86VPermilpi (memop addr:$src1), (i8 imm:$src2))))]>, VEX,
+ Sched<[WriteFShuffleLd]>;
}
let ExeDomain = SSEPackedSingle in {
}
let Predicates = [HasAVX] in {
-def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v8f32 (X86VPermilpv VR256:$src1, (v8i32 VR256:$src2))),
+ (VPERMILPSYrr VR256:$src1, VR256:$src2)>;
+def : Pat<(v8f32 (X86VPermilpv VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)))),
+ (VPERMILPSYrm VR256:$src1, addr:$src2)>;
+def : Pat<(v4f64 (X86VPermilpv VR256:$src1, (v4i64 VR256:$src2))),
+ (VPERMILPDYrr VR256:$src1, VR256:$src2)>;
+def : Pat<(v4f64 (X86VPermilpv VR256:$src1, (loadv4i64 addr:$src2))),
+ (VPERMILPDYrm VR256:$src1, addr:$src2)>;
+
+def : Pat<(v8i32 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPSYri VR256:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi VR256:$src1, (i8 imm:$imm))),
(VPERMILPDYri VR256:$src1, imm:$imm)>;
-def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (loadv4i64 addr:$src1)),
+def : Pat<(v8i32 (X86VPermilpi (bc_v8i32 (loadv4i64 addr:$src1)),
(i8 imm:$imm))),
(VPERMILPSYmi addr:$src1, imm:$imm)>;
-def : Pat<(v4i64 (X86VPermilp (loadv4i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v4i64 (X86VPermilpi (loadv4i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDYmi addr:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86VPermilpv VR128:$src1, (v4i32 VR128:$src2))),
+ (VPERMILPSrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v4f32 (X86VPermilpv VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)))),
+ (VPERMILPSrm VR128:$src1, addr:$src2)>;
+def : Pat<(v2f64 (X86VPermilpv VR128:$src1, (v2i64 VR128:$src2))),
+ (VPERMILPDrr VR128:$src1, VR128:$src2)>;
+def : Pat<(v2f64 (X86VPermilpv VR128:$src1, (loadv2i64 addr:$src2))),
+ (VPERMILPDrm VR128:$src1, addr:$src2)>;
+
+def : Pat<(v2i64 (X86VPermilpi VR128:$src1, (i8 imm:$imm))),
(VPERMILPDri VR128:$src1, imm:$imm)>;
-def : Pat<(v2i64 (X86VPermilp (loadv2i64 addr:$src1), (i8 imm:$imm))),
+def : Pat<(v2i64 (X86VPermilpi (loadv2i64 addr:$src1), (i8 imm:$imm))),
(VPERMILPDmi addr:$src1, imm:$imm)>;
}
//
let ExeDomain = SSEPackedSingle in {
def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+ (ins VR256:$src1, VR256:$src2, u8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
- (i8 imm:$src3))))]>, VEX_4V, VEX_L;
+ (i8 imm:$src3))))]>, VEX_4V, VEX_L,
+ Sched<[WriteFShuffle]>;
def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
- (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
+ (ins VR256:$src1, f256mem:$src2, u8imm:$src3),
"vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv8f32 addr:$src2),
- (i8 imm:$src3)))]>, VEX_4V, VEX_L;
+ (i8 imm:$src3)))]>, VEX_4V, VEX_L,
+ Sched<[WriteFShuffleLd, ReadAfterLd]>;
}
let Predicates = [HasAVX] in {
YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
// Zero All YMM registers
def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
- [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
+ [(int_x86_avx_vzeroall)]>, PS, VEX, VEX_L, Requires<[HasAVX]>;
// Zero Upper bits of YMM registers
def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
- [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
+ [(int_x86_avx_vzeroupper)]>, PS, VEX, Requires<[HasAVX]>;
}
//===----------------------------------------------------------------------===//
@@ -8077,10 +8460,11 @@ multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
"vcvtph2ps\t{$src, $dst|$dst, $src}",
[(set RC:$dst, (Int VR128:$src))]>,
- T8PD, VEX;
- let neverHasSideEffects = 1, mayLoad = 1 in
+ T8PD, VEX, Sched<[WriteCvtF2F]>;
+ let hasSideEffects = 0, mayLoad = 1 in
def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8PD, VEX;
+ "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8PD, VEX,
+ Sched<[WriteCvtF2FLd]>;
}
multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
@@ -8088,8 +8472,9 @@ multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
(ins RC:$src1, i32i8imm:$src2),
"vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
- TAPD, VEX;
- let neverHasSideEffects = 1, mayStore = 1 in
+ TAPD, VEX, Sched<[WriteCvtF2F]>;
+ let hasSideEffects = 0, mayStore = 1,
+ SchedRW = [WriteCvtF2FLd, WriteRMW] in
def mr : Ii8<0x1D, MRMDestMem, (outs),
(ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
"vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
+
+ // Pattern match vcvtph2ps of a scalar i64 load.
+ def : Pat<(int_x86_vcvtph2ps_128 (vzmovl_v2i64 addr:$src)),
+ (VCVTPH2PSrm addr:$src)>;
+ def : Pat<(int_x86_vcvtph2ps_128 (vzload_v2i64 addr:$src)),
+ (VCVTPH2PSrm addr:$src)>;
+}
+
+// Patterns for matching conversions from float to half-float and vice versa.
+let Predicates = [HasF16C] in {
+ def : Pat<(fp_to_f16 FR32:$src),
+ (i16 (EXTRACT_SUBREG (VMOVPDI2DIrr (VCVTPS2PHrr
+ (COPY_TO_REGCLASS FR32:$src, VR128), 0)), sub_16bit))>;
+
+ def : Pat<(f16_to_fp GR16:$src),
+ (f32 (COPY_TO_REGCLASS (VCVTPH2PSrr
+ (COPY_TO_REGCLASS (MOVSX32rr16 GR16:$src), VR128)), FR32)) >;
+
+ def : Pat<(f16_to_fp (i16 (fp_to_f16 FR32:$src))),
+ (f32 (COPY_TO_REGCLASS (VCVTPH2PSrr
+ (VCVTPS2PHrr (COPY_TO_REGCLASS FR32:$src, VR128), 0)), FR32)) >;
}
//===----------------------------------------------------------------------===//
X86MemOperand x86memop> {
let isCommutable = 1 in
def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
- (ins RC:$src1, RC:$src2, u32u8imm:$src3),
+ (ins RC:$src1, RC:$src2, u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
- VEX_4V;
+ Sched<[WriteBlend]>, VEX_4V;
def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
- (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
+ (ins RC:$src1, x86memop:$src2, u8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst,
(IntId RC:$src1,
(bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
- VEX_4V;
+ Sched<[WriteBlendLd, ReadAfterLd]>, VEX_4V;
}
-let isCommutable = 0 in {
defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
VR128, loadv2i64, i128mem>;
defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
VR256, loadv4i64, i256mem>, VEX_L;
-}
def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
imm:$mask)),
Intrinsic Int128, Intrinsic Int256> {
def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
+ [(set VR128:$dst, (Int128 VR128:$src))]>,
+ Sched<[WriteShuffle]>, VEX;
def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst,
- (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
+ (Int128 (scalar_to_vector (ld_frag addr:$src))))]>,
+ Sched<[WriteLoad]>, VEX;
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
+ [(set VR256:$dst, (Int256 VR128:$src))]>,
+ Sched<[WriteShuffle256]>, VEX, VEX_L;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst,
(Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
- VEX, VEX_L;
+ Sched<[WriteLoad]>, VEX, VEX_L;
}
defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
(VBROADCASTSDYrr VR128:$src)>;
+ // Provide aliases for broadcast from the same regitser class that
+ // automatically does the extract.
+ def : Pat<(v32i8 (X86VBroadcast (v32i8 VR256:$src))),
+ (VPBROADCASTBYrr (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v16i16 (X86VBroadcast (v16i16 VR256:$src))),
+ (VPBROADCASTWYrr (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v8i32 (X86VBroadcast (v8i32 VR256:$src))),
+ (VPBROADCASTDYrr (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v4i64 (X86VBroadcast (v4i64 VR256:$src))),
+ (VPBROADCASTQYrr (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v8f32 (X86VBroadcast (v8f32 VR256:$src))),
+ (VBROADCASTSSYrr (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src),
+ sub_xmm)))>;
+ def : Pat<(v4f64 (X86VBroadcast (v4f64 VR256:$src))),
+ (VBROADCASTSDYrr (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src),
+ sub_xmm)))>;
+
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
let AddedComplexity = 20 in {
(VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
(VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
+
+ def : Pat<(v16i8 (X86VBroadcast GR8:$src)),
+ (VPBROADCASTBrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+ VR128))>;
+ def : Pat<(v32i8 (X86VBroadcast GR8:$src)),
+ (VPBROADCASTBYrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)),
+ VR128))>;
+
+ def : Pat<(v8i16 (X86VBroadcast GR16:$src)),
+ (VPBROADCASTWrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+ VR128))>;
+ def : Pat<(v16i16 (X86VBroadcast GR16:$src)),
+ (VPBROADCASTWYrr (COPY_TO_REGCLASS
+ (i32 (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit)),
+ VR128))>;
+
+ // The patterns for VPBROADCASTD are not needed because they would match
+ // the exact same thing as VBROADCASTSS patterns.
+
+ def : Pat<(v2i64 (X86VBroadcast GR64:$src)),
+ (VPBROADCASTQrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
+ // The v4i64 pattern is not needed because VBROADCASTSDYrr already match.
}
}
}
let Predicates = [HasAVX] in {
-def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
- (VBROADCASTSSYrm addr:$src)>;
-def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
- (VBROADCASTSDYrm addr:$src)>;
-def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
- (VBROADCASTSSrm addr:$src)>;
-
// Provide fallback in case the load node that is used in the patterns above
// is used by additional users, which prevents the pattern selection.
let AddedComplexity = 20 in {
(VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
(VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
}
+
+ def : Pat<(v2f64 (X86VBroadcast f64:$src)),
+ (VMOVDDUPrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
}
//===----------------------------------------------------------------------===//
//
multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
- ValueType OpVT> {
+ ValueType OpVT, X86FoldableSchedWrite Sched> {
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
- VEX_4V, VEX_L;
+ Sched<[Sched]>, VEX_4V, VEX_L;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr,
[(set VR256:$dst,
(OpVT (X86VPermv VR256:$src1,
(bitconvert (mem_frag addr:$src2)))))]>,
- VEX_4V, VEX_L;
+ Sched<[Sched.Folded, ReadAfterLd]>, VEX_4V, VEX_L;
}
-defm VPERMD : avx2_perm<0x36, "vpermd", loadv4i64, v8i32>;
+defm VPERMD : avx2_perm<0x36, "vpermd", loadv4i64, v8i32, WriteShuffle256>;
let ExeDomain = SSEPackedSingle in
-defm VPERMPS : avx2_perm<0x16, "vpermps", loadv8f32, v8f32>;
+defm VPERMPS : avx2_perm<0x16, "vpermps", loadv8f32, v8f32, WriteFShuffle256>;
multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
- ValueType OpVT> {
+ ValueType OpVT, X86FoldableSchedWrite Sched> {
def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, i8imm:$src2),
+ (ins VR256:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
- VEX, VEX_L;
+ Sched<[Sched]>, VEX, VEX_L;
def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
- (ins i256mem:$src1, i8imm:$src2),
+ (ins i256mem:$src1, u8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(OpVT (X86VPermi (mem_frag addr:$src1),
- (i8 imm:$src2))))]>, VEX, VEX_L;
+ (i8 imm:$src2))))]>,
+ Sched<[Sched.Folded, ReadAfterLd]>, VEX, VEX_L;
}
-defm VPERMQ : avx2_perm_imm<0x00, "vpermq", loadv4i64, v4i64>, VEX_W;
+defm VPERMQ : avx2_perm_imm<0x00, "vpermq", loadv4i64, v4i64,
+ WriteShuffle256>, VEX_W;
let ExeDomain = SSEPackedDouble in
-defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", loadv4f64, v4f64>, VEX_W;
+defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", loadv4f64, v4f64,
+ WriteFShuffle256>, VEX_W;
//===----------------------------------------------------------------------===//
// VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
//
def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, VR256:$src2, i8imm:$src3),
+ (ins VR256:$src1, VR256:$src2, u8imm:$src3),
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
- (i8 imm:$src3))))]>, VEX_4V, VEX_L;
+ (i8 imm:$src3))))]>, Sched<[WriteShuffle256]>,
+ VEX_4V, VEX_L;
def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
- (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
+ (ins VR256:$src1, f256mem:$src2, u8imm:$src3),
"vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
[(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (loadv4i64 addr:$src2),
- (i8 imm:$src3)))]>, VEX_4V, VEX_L;
+ (i8 imm:$src3)))]>,
+ Sched<[WriteShuffle256Ld, ReadAfterLd]>, VEX_4V, VEX_L;
let Predicates = [HasAVX2] in {
def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
@@ -8373,16 +8827,16 @@ def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (loadv4i64 addr:$src2)),
//===----------------------------------------------------------------------===//
// VINSERTI128 - Insert packed integer values
//
-let neverHasSideEffects = 1 in {
+let hasSideEffects = 0 in {
def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
- (ins VR256:$src1, VR128:$src2, i8imm:$src3),
+ (ins VR256:$src1, VR128:$src2, u8imm:$src3),
"vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V, VEX_L;
+ []>, Sched<[WriteShuffle256]>, VEX_4V, VEX_L;
let mayLoad = 1 in
def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
- (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
+ (ins VR256:$src1, i128mem:$src2, u8imm:$src3),
"vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- []>, VEX_4V, VEX_L;
+ []>, Sched<[WriteShuffle256Ld, ReadAfterLd]>, VEX_4V, VEX_L;
}
let Predicates = [HasAVX2] in {
// VEXTRACTI128 - Extract packed integer values
//
def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
- (ins VR256:$src1, i8imm:$src2),
+ (ins VR256:$src1, u8imm:$src2),
"vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
[(set VR128:$dst,
(int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
- VEX, VEX_L;
-let neverHasSideEffects = 1, mayStore = 1 in
+ Sched<[WriteShuffle256]>, VEX, VEX_L;
+let hasSideEffects = 0, mayStore = 1 in
def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
- (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
+ (ins i128mem:$dst, VR256:$src1, u8imm:$src2),
"vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
- VEX, VEX_L;
+ Sched<[WriteStore]>, VEX, VEX_L;
let Predicates = [HasAVX2] in {
def : Pat<(vextract128_extract:$ext VR256:$src1, (iPTR imm)),
int_x86_avx2_maskstore_q,
int_x86_avx2_maskstore_q_256>, VEX_W;
+def: Pat<(masked_store addr:$ptr, (v8i32 VR256:$mask), (v8f32 VR256:$src)),
+ (VMASKMOVPSYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v8i32 VR256:$mask), (v8i32 VR256:$src)),
+ (VPMASKMOVDYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v4i32 VR128:$mask), (v4f32 VR128:$src)),
+ (VMASKMOVPSmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v4i32 VR128:$mask), (v4i32 VR128:$src)),
+ (VPMASKMOVDmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask), undef)),
+ (VMASKMOVPSYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask),
+ (bc_v8f32 (v8i32 immAllZerosV)))),
+ (VMASKMOVPSYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v8f32 (masked_load addr:$ptr, (v8i32 VR256:$mask), (v8f32 VR256:$src0))),
+ (VBLENDVPSYrr VR256:$src0, (VMASKMOVPSYrm VR256:$mask, addr:$ptr),
+ VR256:$mask)>;
+
+def: Pat<(v8i32 (masked_load addr:$ptr, (v8i32 VR256:$mask), undef)),
+ (VPMASKMOVDYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v8i32 (masked_load addr:$ptr, (v8i32 VR256:$mask), (v8i32 immAllZerosV))),
+ (VPMASKMOVDYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v8i32 (masked_load addr:$ptr, (v8i32 VR256:$mask), (v8i32 VR256:$src0))),
+ (VBLENDVPSYrr VR256:$src0, (VPMASKMOVDYrm VR256:$mask, addr:$ptr),
+ VR256:$mask)>;
+
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask), undef)),
+ (VMASKMOVPSrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask),
+ (bc_v4f32 (v4i32 immAllZerosV)))),
+ (VMASKMOVPSrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4f32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4f32 VR128:$src0))),
+ (VBLENDVPSrr VR128:$src0, (VMASKMOVPSrm VR128:$mask, addr:$ptr),
+ VR128:$mask)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), undef)),
+ (VPMASKMOVDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4i32 immAllZerosV))),
+ (VPMASKMOVDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v4i32 (masked_load addr:$ptr, (v4i32 VR128:$mask), (v4i32 VR128:$src0))),
+ (VBLENDVPSrr VR128:$src0, (VPMASKMOVDrm VR128:$mask, addr:$ptr),
+ VR128:$mask)>;
+
+def: Pat<(masked_store addr:$ptr, (v4i64 VR256:$mask), (v4f64 VR256:$src)),
+ (VMASKMOVPDYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v4i64 VR256:$mask), (v4i64 VR256:$src)),
+ (VPMASKMOVQYmr addr:$ptr, VR256:$mask, VR256:$src)>;
+
+def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask), undef)),
+ (VMASKMOVPDYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask),
+ (v4f64 immAllZerosV))),
+ (VMASKMOVPDYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v4f64 (masked_load addr:$ptr, (v4i64 VR256:$mask), (v4f64 VR256:$src0))),
+ (VBLENDVPDYrr VR256:$src0, (VMASKMOVPDYrm VR256:$mask, addr:$ptr),
+ VR256:$mask)>;
+
+def: Pat<(v4i64 (masked_load addr:$ptr, (v4i64 VR256:$mask), undef)),
+ (VPMASKMOVQYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v4i64 (masked_load addr:$ptr, (v4i64 VR256:$mask),
+ (bc_v4i64 (v8i32 immAllZerosV)))),
+ (VPMASKMOVQYrm VR256:$mask, addr:$ptr)>;
+
+def: Pat<(v4i64 (masked_load addr:$ptr, (v4i64 VR256:$mask), (v4i64 VR256:$src0))),
+ (VBLENDVPDYrr VR256:$src0, (VPMASKMOVQYrm VR256:$mask, addr:$ptr),
+ VR256:$mask)>;
+
+def: Pat<(masked_store addr:$ptr, (v2i64 VR128:$mask), (v2f64 VR128:$src)),
+ (VMASKMOVPDmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(masked_store addr:$ptr, (v2i64 VR128:$mask), (v2i64 VR128:$src)),
+ (VPMASKMOVQmr addr:$ptr, VR128:$mask, VR128:$src)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask), undef)),
+ (VMASKMOVPDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask),
+ (v2f64 immAllZerosV))),
+ (VMASKMOVPDrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2f64 (masked_load addr:$ptr, (v2i64 VR128:$mask), (v2f64 VR128:$src0))),
+ (VBLENDVPDrr VR128:$src0, (VMASKMOVPDrm VR128:$mask, addr:$ptr),
+ VR128:$mask)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask), undef)),
+ (VPMASKMOVQrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask),
+ (bc_v2i64 (v4i32 immAllZerosV)))),
+ (VPMASKMOVQrm VR128:$mask, addr:$ptr)>;
+
+def: Pat<(v2i64 (masked_load addr:$ptr, (v2i64 VR128:$mask), (v2i64 VR128:$src0))),
+ (VBLENDVPDrr VR128:$src0, (VPMASKMOVQrm VR128:$mask, addr:$ptr),
+ VR128:$mask)>;
//===----------------------------------------------------------------------===//
// Variable Bit Shifts
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
- VEX_4V;
+ VEX_4V, Sched<[WriteVarVecShift]>;
def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
(ins VR128:$src1, i128mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1,
(vt128 (bitconvert (loadv2i64 addr:$src2))))))]>,
- VEX_4V;
+ VEX_4V, Sched<[WriteVarVecShiftLd, ReadAfterLd]>;
def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
- VEX_4V, VEX_L;
+ VEX_4V, VEX_L, Sched<[WriteVarVecShift]>;
def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
(ins VR256:$src1, i256mem:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1,
(vt256 (bitconvert (loadv4i64 addr:$src2))))))]>,
- VEX_4V, VEX_L;
+ VEX_4V, VEX_L, Sched<[WriteVarVecShiftLd, ReadAfterLd]>;
}
defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
let mayLoad = 1, Constraints
= "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
in {
- defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
- defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
- defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
- defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;
+
+ let ExeDomain = SSEPackedDouble in {
+ defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
+ defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
+ }
+
+ let ExeDomain = SSEPackedSingle in {
+ defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
+ defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
+ }
}