// in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
//===----------------------------------------------------------------------===//
-class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
- SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
- [(set VR128:$dst, (vt (OpNode VR128:$src1,
- (scalar_to_vector RC:$src2))))],
- IIC_SSE_MOV_S_RR>;
+multiclass sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt,
+ X86MemOperand x86memop, string base_opc,
+ string asm_opr> {
+ def rr : SI<0x10, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, RC:$src2),
+ !strconcat(base_opc, asm_opr),
+ [(set VR128:$dst, (vt (OpNode VR128:$src1,
+ (scalar_to_vector RC:$src2))))],
+ IIC_SSE_MOV_S_RR>;
-// Loading from memory automatically zeroing upper bits.
-class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
- PatFrag mem_pat, string OpcodeStr> :
- SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (mem_pat addr:$src))],
- IIC_SSE_MOV_S_RM>;
-
-// AVX
-def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
- "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
- VEX_LIG;
-def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
- "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
- VEX_LIG;
-
-// For the disassembler
-let isCodeGenOnly = 1, hasSideEffects = 0 in {
- def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR32:$src2),
- "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
- IIC_SSE_MOV_S_RR>,
- XS, VEX_4V, VEX_LIG;
- def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR64:$src2),
- "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
- IIC_SSE_MOV_S_RR>,
- XD, VEX_4V, VEX_LIG;
+ // For the disassembler
+ let isCodeGenOnly = 1, hasSideEffects = 0 in
+ def rr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
+ (ins VR128:$src1, RC:$src2),
+ !strconcat(base_opc, asm_opr),
+ [], IIC_SSE_MOV_S_RR>;
}
-let canFoldAsLoad = 1, isReMaterializable = 1 in {
- def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
- VEX_LIG;
- let AddedComplexity = 20 in
- def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
- VEX_LIG;
-}
+multiclass sse12_move<RegisterClass RC, SDNode OpNode, ValueType vt,
+ X86MemOperand x86memop, string OpcodeStr> {
+ // AVX
+ defm V#NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
+ VEX_4V, VEX_LIG;
-def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
- XS, VEX, VEX_LIG;
-def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
- XD, VEX, VEX_LIG;
+ def V#NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
+ VEX, VEX_LIG;
+ // SSE1 & 2
+ let Constraints = "$src1 = $dst" in {
+ defm NAME : sse12_move_rr<RC, OpNode, vt, x86memop, OpcodeStr,
+ "\t{$src2, $dst|$dst, $src2}">;
+ }
-// SSE1 & 2
-let Constraints = "$src1 = $dst" in {
- def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
- "movss\t{$src2, $dst|$dst, $src2}">, XS;
- def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
- "movsd\t{$src2, $dst|$dst, $src2}">, XD;
+ def NAME#mr : SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
+}
- // For the disassembler
- let isCodeGenOnly = 1, hasSideEffects = 0 in {
- def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR32:$src2),
- "movss\t{$src2, $dst|$dst, $src2}", [],
- IIC_SSE_MOV_S_RR>, XS;
- def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
- (ins VR128:$src1, FR64:$src2),
- "movsd\t{$src2, $dst|$dst, $src2}", [],
- IIC_SSE_MOV_S_RR>, XD;
- }
+// Loading from memory automatically zeroing upper bits.
+multiclass sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
+ PatFrag mem_pat, string OpcodeStr> {
+ def V#NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (mem_pat addr:$src))],
+ IIC_SSE_MOV_S_RM>, VEX, VEX_LIG;
+ def NAME#rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (mem_pat addr:$src))],
+ IIC_SSE_MOV_S_RM>;
}
+defm MOVSS : sse12_move<FR32, X86Movss, v4f32, f32mem, "movss">, XS;
+defm MOVSD : sse12_move<FR64, X86Movsd, v2f64, f64mem, "movsd">, XD;
+
let canFoldAsLoad = 1, isReMaterializable = 1 in {
- def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
+ defm MOVSS : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
let AddedComplexity = 20 in
- def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
+ defm MOVSD : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
}
-def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
- "movss\t{$src, $dst|$dst, $src}",
- [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
- "movsd\t{$src, $dst|$dst, $src}",
- [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
-
// Patterns
let Predicates = [HasAVX] in {
let AddedComplexity = 15 in {
// SSE 1 & 2 - Move Low packed FP Instructions
//===----------------------------------------------------------------------===//
-multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
- SDNode psnode, SDNode pdnode, string base_opc,
- string asm_opr, InstrItinClass itin> {
+multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
+ string base_opc, string asm_opr,
+ InstrItinClass itin> {
def PSrm : PI<opc, MRMSrcMem,
(outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
!strconcat(base_opc, "s", asm_opr),
- [(set RC:$dst,
- (psnode RC:$src1,
+ [(set VR128:$dst,
+ (psnode VR128:$src1,
(bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
itin, SSEPackedSingle>, TB;
def PDrm : PI<opc, MRMSrcMem,
- (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
+ (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
!strconcat(base_opc, "d", asm_opr),
- [(set RC:$dst, (v2f64 (pdnode RC:$src1,
+ [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
(scalar_to_vector (loadf64 addr:$src2)))))],
itin, SSEPackedDouble>, TB, OpSize;
+
}
-let AddedComplexity = 20 in {
- defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- IIC_SSE_MOV_LH>, VEX_4V;
+multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
+ string base_opc, InstrItinClass itin> {
+ defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ itin>, VEX_4V;
+
+let Constraints = "$src1 = $dst" in
+ defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
+ "\t{$src2, $dst|$dst, $src2}",
+ itin>;
}
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
- defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
- "\t{$src2, $dst|$dst, $src2}",
- IIC_SSE_MOV_LH>;
+
+let AddedComplexity = 20 in {
+ defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
+ IIC_SSE_MOV_LH>;
}
def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
//===----------------------------------------------------------------------===//
let AddedComplexity = 20 in {
- defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
- "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
- IIC_SSE_MOV_LH>, VEX_4V;
-}
-let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
- defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
- "\t{$src2, $dst|$dst, $src2}",
- IIC_SSE_MOV_LH>;
+ defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
+ IIC_SSE_MOV_LH>;
}
// v2f64 extract element 1 is always custom lowered to unpack high to low
SSE_CVT_SS2SI_32>,
XS, VEX, VEX_LIG;
defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
- "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_64>,
XS, VEX, VEX_W, VEX_LIG;
defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
SSE_CVT_SD2SI>,
XD, VEX, VEX_LIG;
defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
- "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>,
XD, VEX, VEX_W, VEX_LIG;
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+
// The assembler can recognize rr 64-bit instructions by seeing a rxx
// register, but the same isn't true when only using memory operands,
// provide other assembly "l" and "q" forms to address this explicitly
// where appropriate to do so.
-defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
+defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
XS, VEX_4V, VEX_LIG;
defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
XS, VEX_4V, VEX_W, VEX_LIG;
-defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
+defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
XD, VEX_4V, VEX_LIG;
defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
XD, VEX_4V, VEX_W, VEX_LIG;
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
- (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
-def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
+def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
+ (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
+def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
(VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
let Predicates = [HasAVX] in {
"cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_32>, XS;
defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
- "cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttss2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SS2SI_64>, XS, REX_W;
defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
"cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>, XD;
defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
- "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ "cvttsd2si\t{$src, $dst|$dst, $src}",
SSE_CVT_SD2SI>, XD, REX_W;
defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
- "cvtsi2ss\t{$src, $dst|$dst, $src}",
+ "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XS;
defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
"cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XS, REX_W;
defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
- "cvtsi2sd\t{$src, $dst|$dst, $src}",
+ "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XD;
defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
"cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
SSE_CVT_Scalar>, XD, REX_W;
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
+def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
+def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
+
+def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
+ (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
+def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
+ (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
+
// Conversion Instructions Intrinsics - Match intrinsics which expect MM
// and/or XMM operand(s).
}
defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
- int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
+ int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
- int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
+ int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
- sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
+ sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
- sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
+ sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
+ int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
SSE_CVT_Scalar, 0>, XS, VEX_4V;
defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
SSE_CVT_Scalar, 0>, XS, VEX_4V,
VEX_W;
defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
- int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
+ int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
SSE_CVT_Scalar, 0>, XD, VEX_4V;
defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
let Constraints = "$src1 = $dst" in {
defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
int_x86_sse_cvtsi2ss, i32mem, loadi32,
- "cvtsi2ss", SSE_CVT_Scalar>, XS;
+ "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
int_x86_sse_cvtsi642ss, i64mem, loadi64,
"cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
int_x86_sse2_cvtsi2sd, i32mem, loadi32,
- "cvtsi2sd", SSE_CVT_Scalar>, XD;
+ "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
int_x86_sse2_cvtsi642sd, i64mem, loadi64,
"cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
SSE_CVT_SS2SI_32>, XS, VEX;
defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
- "cvttss2si{q}", SSE_CVT_SS2SI_64>,
+ "cvttss2si", SSE_CVT_SS2SI_64>,
XS, VEX, VEX_W;
defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
sdmem, sse_load_f64, "cvttsd2si",
SSE_CVT_SD2SI>, XD, VEX;
defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
- "cvttsd2si{q}", SSE_CVT_SD2SI>,
+ "cvttsd2si", SSE_CVT_SD2SI>,
XD, VEX, VEX_W;
defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
ssmem, sse_load_f32, "cvttss2si",
SSE_CVT_SS2SI_32>, XS;
defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
- "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
+ "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
sdmem, sse_load_f64, "cvttsd2si",
SSE_CVT_SD2SI>, XD;
defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
- "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
+ "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- ssmem, sse_load_f32, "cvtss2si{l}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
- ssmem, sse_load_f32, "cvtss2si{q}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
- ssmem, sse_load_f32, "cvtss2si{l}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_32>, XS;
defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
- ssmem, sse_load_f32, "cvtss2si{q}",
+ ssmem, sse_load_f32, "cvtss2si",
SSE_CVT_SS2SI_64>, XS, REX_W;
defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
SSEPackedSingle, SSE_CVT_PS>,
TB, Requires<[UseSSE2]>;
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
+
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
+def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
+ (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
+
/// SSE 2 Only
// Convert scalar double to scalar single
ValueType OpVT128, ValueType OpVT256,
OpndItins itins, bit IsCommutable = 0> {
let Predicates = [HasAVX] in
- defm V#NAME# : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
+ defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
- defm #NAME# : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
- memopv2i64, i128mem, itins, IsCommutable, 1>;
+ defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
+ memopv2i64, i128mem, itins, IsCommutable, 1>;
let Predicates = [HasAVX2] in
defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
- OpVT256, VR256, memopv4i64, i256mem, itins,
- IsCommutable, 0>, VEX_4V, VEX_L;
+ OpVT256, VR256, memopv4i64, i256mem, itins,
+ IsCommutable, 0>, VEX_4V, VEX_L;
}
// These are ordered here for pattern ordering requirements with the fp versions
/// sse1_fp_unop_s - SSE1 unops in scalar form.
multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+ (ins FR32:$src1, FR32:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ let mayLoad = 1 in {
+ def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+ (ins FR32:$src1,f32mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ }
+}
+
def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
!strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
[(set FR32:$dst, (OpNode FR32:$src))]>;
[(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
}
-/// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
-multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
- def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
+multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
+ (ins FR32:$src1, FR32:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
let mayLoad = 1 in {
- def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat(OpcodeStr,
- "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
+ def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
+ (ins FR32:$src1,f32mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ }
+}
+
+ def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode FR32:$src))]>;
+ // For scalar unary operations, fold a load into the operation
+ // only in OptForSize mode. It eliminates an instruction, but it also
+ // eliminates a whole-register clobber (the load), so it introduces a
+ // partial register update condition.
+ def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
+ !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
+ [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
+ Requires<[UseSSE1, OptForSize]>;
+ let Constraints = "$src1 = $dst" in {
+ def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
+ (ins VR128:$src1, VR128:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rr>;
+ let mayLoad = 1, hasSideEffects = 0 in
+ def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, ssmem:$src2),
+ !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
+ [], itins.rm>;
}
}
OpndItins itins> {
let Predicates = [HasAVX] in {
def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
itins.rr>, VEX;
def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
itins.rm>, VEX;
def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
itins.rr>, VEX, VEX_L;
def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
itins.rm>, VEX, VEX_L;
OpndItins itins> {
let Predicates = [HasAVX] in {
def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int VR128:$src))],
itins.rr>, VEX;
def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
itins.rm>, VEX;
def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V8F32Int VR256:$src))],
itins.rr>, VEX, VEX_L;
def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
(ins f256mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"ps\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
itins.rm>, VEX, VEX_L;
/// sse2_fp_unop_s - SSE2 unops in scalar form.
multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
+let Predicates = [HasAVX], hasSideEffects = 0 in {
+ def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
+ (ins FR64:$src1, FR64:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ let mayLoad = 1 in {
+ def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
+ (ins FR64:$src1,f64mem:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
+ (ins VR128:$src1, sdmem:$src2),
+ !strconcat("v", OpcodeStr,
+ "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
+ []>, VEX_4V, VEX_LIG;
+ }
+}
+
def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
!strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
[(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
[(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
}
-/// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
-let hasSideEffects = 0 in
-multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
- def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- let mayLoad = 1 in {
- def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, sdmem:$src2),
- !strconcat(OpcodeStr,
- "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
- }
-}
-
-/// sse2_fp_unop_p_new - SSE2 unops in vector forms.
+/// sse2_fp_unop_p - SSE2 unops in vector forms.
multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
SDNode OpNode, OpndItins itins> {
let Predicates = [HasAVX] in {
def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"pd\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
itins.rr>, VEX;
def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"pd\t{$src, $dst|$dst, $src}"),
[(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
itins.rm>, VEX;
def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"pd\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
itins.rr>, VEX, VEX_L;
def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"pd\t{$src, $dst|$dst, $src}"),
[(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
itins.rm>, VEX, VEX_L;
[(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
}
-defm SQRT : sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
+// Square root.
+defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
+ SSE_SQRTS>,
+ sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
+ sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
+ SSE_SQRTS>,
sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
-defm RSQRT : sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
+
+// Reciprocal approximations. Note that these typically require refinement
+// in order to obtain suitable precision.
+defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
+ sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
-defm RCP : sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
+defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
+ sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
int_x86_avx_rcp_ps_256, SSE_RCPP>;
-let Predicates = [HasAVX] in {
- // Square root.
- defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
- sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
-
- // Reciprocal approximations. Note that these typically require refinement
- // in order to obtain suitable precision.
- defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
- defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
-}
-
def : Pat<(f32 (fsqrt FR32:$src)),
(VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
def : Pat<(f32 (fsqrt (load addr:$src))),
(VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
}
-// Square root.
-defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
- SSE_SQRTS>,
- sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
- SSE_SQRTS>;
-
-/// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
-multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
- OpndItins itins> {
- def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode FR32:$src))]>;
- // For scalar unary operations, fold a load into the operation
- // only in OptForSize mode. It eliminates an instruction, but it also
- // eliminates a whole-register clobber (the load), so it introduces a
- // partial register update condition.
- def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
- !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
- [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
- Requires<[UseSSE1, OptForSize]>;
- let Constraints = "$src1 = $dst" in {
- def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
- (ins VR128:$src1, VR128:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [], itins.rr>;
- let mayLoad = 1, hasSideEffects = 0 in
- def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
- (ins VR128:$src1, ssmem:$src2),
- !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
- [], itins.rm>;
- }
-}
-
// Reciprocal approximations. Note that these typically require refinement
// in order to obtain suitable precision.
-defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>;
let Predicates = [UseSSE1] in {
def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
(RSQRTSSr_Int VR128:$src, VR128:$src)>;
-}
-
-defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>;
-let Predicates = [UseSSE1] in {
def : Pat<(int_x86_sse_rcp_ss VR128:$src),
(RCPSSr_Int VR128:$src, VR128:$src)>;
}
Intrinsic IntId256, OpndItins itins,
bit IsCommutable = 0> {
let Predicates = [HasAVX] in
- defm V#NAME# : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
- VR128, memopv2i64, i128mem, itins,
- IsCommutable, 0>, VEX_4V;
+ defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
+ VR128, memopv2i64, i128mem, itins,
+ IsCommutable, 0>, VEX_4V;
let Constraints = "$src1 = $dst" in
- defm #NAME# : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
- i128mem, itins, IsCommutable, 1>;
+ defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
+ i128mem, itins, IsCommutable, 1>;
let Predicates = [HasAVX2] in
defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
let Predicates = [HasAVX] in {
def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, i8imm:$src2),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
IIC_SSE_PSHUF>, VEX;
def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
(ins i128mem:$src1, i8imm:$src2),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR128:$dst,
(vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
let Predicates = [HasAVX2] in {
def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, i8imm:$src2),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
IIC_SSE_PSHUF>, VEX, VEX_L;
def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
(ins i256mem:$src1, i8imm:$src2),
- !strconcat(!strconcat("v", OpcodeStr),
+ !strconcat("v", OpcodeStr,
"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set VR256:$dst,
(vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
// SSSE3 - Packed Align Instruction Patterns
//===---------------------------------------------------------------------===//
-multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr<string asm, bit Is2Addr = 1> {
let neverHasSideEffects = 1 in {
def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
(ins VR128:$src1, VR128:$src2, i8imm:$src3),
}
}
-multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
+multiclass ssse3_palignr_y<string asm, bit Is2Addr = 1> {
let neverHasSideEffects = 1 in {
def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
(ins VR256:$src1, VR256:$src2, i8imm:$src3),
}
let Predicates = [HasAVX] in
- defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
+ defm VPALIGN : ssse3_palignr<"vpalignr", 0>, VEX_4V;
let Predicates = [HasAVX2] in
- defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
+ defm VPALIGN : ssse3_palignr_y<"vpalignr", 0>, VEX_4V, VEX_L;
let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
- defm PALIGN : ssse3_palign<"palignr">;
+ defm PALIGN : ssse3_palignr<"palignr">;
let Predicates = [HasAVX2] in {
-def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8i32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v8f32 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v16i16 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
-def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
+def : Pat<(v32i8 (X86PAlignr VR256:$src1, VR256:$src2, (i8 imm:$imm))),
(VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
}
let Predicates = [HasAVX] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
let Predicates = [UseSSSE3] in {
-def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4i32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v4f32 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v8i16 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
-def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
+def : Pat<(v16i8 (X86PAlignr VR128:$src1, VR128:$src2, (i8 imm:$imm))),
(PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
}
defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
let Predicates = [HasAVX2] in {
+ def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
+ def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
+
+ def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
+ def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
+
+ def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
+
+ def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
+ (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
+ (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+ def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
+ (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
+ def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
+ (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
+
def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
(VPMOVSXWDYrm addr:$src)>;
def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
}
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
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)>;
projects / oota-llvm.git / blobdiff