def : Pat<(v8f64 (X86VBroadcast (loadf64 addr:$src))),
(VBROADCASTSDZrm addr:$src)>;
+def : Pat<(int_x86_avx512_vbroadcast_ss_512 addr:$src),
+ (VBROADCASTSSZrm addr:$src)>;
+def : Pat<(int_x86_avx512_vbroadcast_sd_512 addr:$src),
+ (VBROADCASTSDZrm addr:$src)>;
+
multiclass avx512_int_broadcast_reg<bits<8> opc, string OpcodeStr,
RegisterClass SrcRC, RegisterClass KRC> {
def Zrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins SrcRC:$src),
(VPBROADCASTDrZrr GR32:$src)>;
def : Pat<(v8i64 (X86VBroadcast (i64 GR64:$src))),
(VPBROADCASTQrZrr GR64:$src)>;
+def : Pat<(v8i64 (X86VBroadcastm VK8WM:$mask, (i64 GR64:$src))),
+ (VPBROADCASTQrZkrr VK8WM:$mask, GR64:$src)>;
+
+def : Pat<(v16i32 (int_x86_avx512_pbroadcastd_i32_512 (i32 GR32:$src))),
+ (VPBROADCASTDrZrr GR32:$src)>;
+def : Pat<(v8i64 (int_x86_avx512_pbroadcastq_i64_512 (i64 GR64:$src))),
+ (VPBROADCASTQrZrr GR64:$src)>;
multiclass avx512_int_broadcast_rm<bits<8> opc, string OpcodeStr,
X86MemOperand x86memop, PatFrag ld_frag,
[(set DstRC:$dst,
(OpVT (X86VBroadcastm KRC:$mask, (SrcVT VR128X:$src))))]>,
EVEX, EVEX_KZ;
+ let mayLoad = 1 in {
def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
!strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
[(set DstRC:$dst,
"\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"),
[(set DstRC:$dst, (OpVT (X86VBroadcastm KRC:$mask,
(ld_frag addr:$src))))]>, EVEX, EVEX_KZ;
+ }
}
defm VPBROADCASTDZ : avx512_int_broadcast_rm<0x58, "vpbroadcastd", i32mem,
loadi64, VR512, v8i64, v2i64, VK8WM>, EVEX_V512, VEX_W,
EVEX_CD8<64, CD8VT1>;
+def : Pat<(v16i32 (int_x86_avx512_pbroadcastd_512 (v4i32 VR128X:$src))),
+ (VPBROADCASTDZrr VR128X:$src)>;
+def : Pat<(v8i64 (int_x86_avx512_pbroadcastq_512 (v2i64 VR128X:$src))),
+ (VPBROADCASTQZrr VR128X:$src)>;
+
def : Pat<(v16f32 (X86VBroadcast (v4f32 VR128X:$src))),
(VBROADCASTSSZrr VR128X:$src)>;
def : Pat<(v8f64 (X86VBroadcast (v2f64 VR128X:$src))),
(VBROADCASTSDZrr VR128X:$src)>;
+
+def : Pat<(v16f32 (int_x86_avx512_vbroadcast_ss_ps_512 (v4f32 VR128X:$src))),
+ (VBROADCASTSSZrr VR128X:$src)>;
+def : Pat<(v8f64 (int_x86_avx512_vbroadcast_sd_pd_512 (v2f64 VR128X:$src))),
+ (VBROADCASTSDZrr VR128X:$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.
memopv8i64, vselect, v8i64>, VEX_W,
EVEX_CD8<64, CD8VF>, EVEX_V512;
-
let Predicates = [HasAVX512] in {
def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1),
(v8f32 VR256X:$src2))),
// Accept explicit immediate argument form instead of comparison code.
let neverHasSideEffects = 1 in {
- def rri_alt : PIi8<0xC2, MRMSrcReg,
+ def rri_alt : AVX512PIi8<0xC2, MRMSrcReg,
(outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
- asm_alt, [], IIC_SSE_ALU_F32P_RR, d>;
- def rmi_alt : PIi8<0xC2, MRMSrcMem,
+ asm_alt, [], d>;
+ def rmi_alt : AVX512PIi8<0xC2, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
- asm_alt, [], IIC_SSE_ALU_F32P_RM, d>;
+ asm_alt, [], d>;
}
}
defm VCMPPSZ : avx512_cmp_packed<VK16, VR512, f512mem, AVXCC, X86cmpm, v16f32,
"vcmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"vcmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedSingle>, TB, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VF>;
+ SSEPackedSingle>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VCMPPDZ : avx512_cmp_packed<VK8, VR512, f512mem, AVXCC, X86cmpm, v8f64,
"vcmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
"vcmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
- SSEPackedDouble>, TB, OpSize, EVEX_4V, VEX_W, EVEX_V512,
+ SSEPackedDouble>, OpSize, EVEX_4V, VEX_W, EVEX_V512,
EVEX_CD8<64, CD8VF>;
def : Pat<(v8i1 (X86cmpm (v8f32 VR256X:$src1), (v8f32 VR256X:$src2), imm:$cc)),
EVEX_CD8<64, CD8VF>;
defm VMOVUPSZ : avx512_mov_packed<0x10, VR512, VK16WM, f512mem, loadv16f32,
"vmovups", SSEPackedSingle>,
- TB, EVEX_V512, EVEX_CD8<32, CD8VF>;
+ EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VMOVUPDZ : avx512_mov_packed<0x10, VR512, VK8WM, f512mem, loadv8f64,
"vmovupd", SSEPackedDouble>,
OpSize, EVEX_V512, VEX_W,
def VMOVAPSZmr : AVX512PI<0x29, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src),
"vmovaps\t{$src, $dst|$dst, $src}",
[(alignedstore512 (v16f32 VR512:$src), addr:$dst)],
- SSEPackedSingle>, EVEX, EVEX_V512, TB,
- EVEX_CD8<32, CD8VF>;
+ SSEPackedSingle>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>;
def VMOVAPDZmr : AVX512PI<0x29, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src),
"vmovapd\t{$src, $dst|$dst, $src}",
[(alignedstore512 (v8f64 VR512:$src), addr:$dst)],
SSEPackedDouble>, EVEX, EVEX_V512,
- OpSize, TB, VEX_W, EVEX_CD8<64, CD8VF>;
+ OpSize, VEX_W, EVEX_CD8<64, CD8VF>;
def VMOVUPSZmr : AVX512PI<0x11, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src),
"vmovups\t{$src, $dst|$dst, $src}",
[(store (v16f32 VR512:$src), addr:$dst)],
- SSEPackedSingle>, EVEX, EVEX_V512, TB,
- EVEX_CD8<32, CD8VF>;
+ SSEPackedSingle>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>;
def VMOVUPDZmr : AVX512PI<0x11, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src),
"vmovupd\t{$src, $dst|$dst, $src}",
[(store (v8f64 VR512:$src), addr:$dst)],
SSEPackedDouble>, EVEX, EVEX_V512,
- OpSize, TB, VEX_W, EVEX_CD8<64, CD8VF>;
-
-// Use vmovaps/vmovups for AVX-512 integer load/store.
-// 512-bit load/store
-def : Pat<(alignedloadv8i64 addr:$src),
- (VMOVAPSZrm addr:$src)>;
-def : Pat<(loadv8i64 addr:$src),
- (VMOVUPSZrm addr:$src)>;
-
-def : Pat<(alignedstore512 (v8i64 VR512:$src), addr:$dst),
- (VMOVAPSZmr addr:$dst, VR512:$src)>;
-def : Pat<(alignedstore512 (v16i32 VR512:$src), addr:$dst),
- (VMOVAPSZmr addr:$dst, VR512:$src)>;
-
-def : Pat<(store (v8i64 VR512:$src), addr:$dst),
- (VMOVUPDZmr addr:$dst, VR512:$src)>;
-def : Pat<(store (v16i32 VR512:$src), addr:$dst),
- (VMOVUPSZmr addr:$dst, VR512:$src)>;
+ OpSize, VEX_W, EVEX_CD8<64, CD8VF>;
let neverHasSideEffects = 1 in {
def VMOVDQA32rr : AVX512BI<0x6F, MRMSrcReg, (outs VR512:$dst),
}
}
-multiclass avx512_mov_int<bits<8> opc, string asm, RegisterClass RC,
- RegisterClass KRC,
+// 512-bit aligned load/store
+def : Pat<(alignedloadv8i64 addr:$src), (VMOVDQA64rm addr:$src)>;
+def : Pat<(alignedloadv16i32 addr:$src), (VMOVDQA32rm addr:$src)>;
+
+def : Pat<(alignedstore512 (v8i64 VR512:$src), addr:$dst),
+ (VMOVDQA64mr addr:$dst, VR512:$src)>;
+def : Pat<(alignedstore512 (v16i32 VR512:$src), addr:$dst),
+ (VMOVDQA32mr addr:$dst, VR512:$src)>;
+
+multiclass avx512_mov_int<bits<8> load_opc, bits<8> store_opc, string asm,
+ RegisterClass RC, RegisterClass KRC,
PatFrag ld_frag, X86MemOperand x86memop> {
let neverHasSideEffects = 1 in
- def rr : AVX512XSI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"), []>,
- EVEX;
+ def rr : AVX512XSI<load_opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"), []>, EVEX;
let canFoldAsLoad = 1 in
- def rm : AVX512XSI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
- !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
- [(set RC:$dst, (ld_frag addr:$src))]>,
- EVEX;
+ def rm : AVX512XSI<load_opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
+ [(set RC:$dst, (ld_frag addr:$src))]>, EVEX;
+let mayStore = 1 in
+ def mr : AVX512XSI<store_opc, MRMDestMem, (outs),
+ (ins x86memop:$dst, VR512:$src),
+ !strconcat(asm, "\t{$src, $dst|$dst, $src}"), []>, EVEX;
let Constraints = "$src1 = $dst" in {
- def rrk : AVX512XSI<opc, MRMSrcReg, (outs RC:$dst),
+ def rrk : AVX512XSI<load_opc, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, KRC:$mask, RC:$src2),
!strconcat(asm,
"\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), []>,
EVEX, EVEX_K;
- def rmk : AVX512XSI<opc, MRMSrcMem, (outs RC:$dst),
+ def rmk : AVX512XSI<load_opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, KRC:$mask, x86memop:$src2),
!strconcat(asm,
"\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"),
}
}
-defm VMOVDQU32 : avx512_mov_int<0x6F, "vmovdqu32", VR512, VK16WM, memopv16i32, i512mem>,
+defm VMOVDQU32 : avx512_mov_int<0x6F, 0x7F, "vmovdqu32", VR512, VK16WM,
+ memopv16i32, i512mem>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
-defm VMOVDQU64 : avx512_mov_int<0x6F, "vmovdqu64", VR512, VK8WM, memopv8i64, i512mem>,
+defm VMOVDQU64 : avx512_mov_int<0x6F, 0x7F, "vmovdqu64", VR512, VK8WM,
+ memopv8i64, i512mem>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+// 512-bit unaligned load/store
+def : Pat<(loadv8i64 addr:$src), (VMOVDQU64rm addr:$src)>;
+def : Pat<(loadv16i32 addr:$src), (VMOVDQU32rm addr:$src)>;
+
+def : Pat<(store (v8i64 VR512:$src), addr:$dst),
+ (VMOVDQU64mr addr:$dst, VR512:$src)>;
+def : Pat<(store (v16i32 VR512:$src), addr:$dst),
+ (VMOVDQU32mr addr:$dst, VR512:$src)>;
+
let AddedComplexity = 20 in {
def : Pat<(v16f32 (vselect VK16WM:$mask, (v16f32 VR512:$src1),
(v16f32 VR512:$src2))),
[(set VR128X:$dst,
(v2i64 (scalar_to_vector GR64:$src)))],
IIC_SSE_MOVDQ>, EVEX, VEX_W, VEX_LIG;
+let isCodeGenOnly = 1 in {
def VMOV64toSDZrr : AVX512SI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
"vmovq{z}\t{$src, $dst|$dst, $src}",
[(set FR64:$dst, (bitconvert GR64:$src))],
"vmovq{z}\t{$src, $dst|$dst, $src}",
[(set GR64:$dst, (bitconvert FR64:$src))],
IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteMove]>;
+}
def VMOVSDto64Zmr : AVX512SI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
"vmovq{z}\t{$src, $dst|$dst, $src}",
[(store (i64 (bitconvert FR64:$src)), addr:$dst)],
// Move Int Doubleword to Single Scalar
//
+let isCodeGenOnly = 1 in {
def VMOVDI2SSZrr : AVX512SI<0x6E, MRMSrcReg, (outs FR32X:$dst), (ins GR32:$src),
"vmovd{z}\t{$src, $dst|$dst, $src}",
[(set FR32X:$dst, (bitconvert GR32:$src))],
"vmovd{z}\t{$src, $dst|$dst, $src}",
[(set FR32X:$dst, (bitconvert (loadi32 addr:$src)))],
IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+}
// Move Packed Doubleword Int to Packed Double Int
//
IIC_SSE_MOVD_ToGP>, TB, OpSize, EVEX, VEX_LIG, VEX_W,
Requires<[HasAVX512, In64BitMode]>;
-def VMOVPQIto64Zmr : I<0x7E, MRMDestMem, (outs),
+def VMOVPQIto64Zmr : I<0xD6, MRMDestMem, (outs),
(ins i64mem:$dst, VR128X:$src),
"vmovq{z}\t{$src, $dst|$dst, $src}",
[(store (extractelt (v2i64 VR128X:$src), (iPTR 0)),
addr:$dst)], IIC_SSE_MOVDQ>,
- EVEX, VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>,
+ EVEX, OpSize, VEX_LIG, VEX_W, TB, EVEX_CD8<64, CD8VT1>,
Sched<[WriteStore]>, Requires<[HasAVX512, In64BitMode]>;
// Move Scalar Single to Double Int
//
+let isCodeGenOnly = 1 in {
def VMOVSS2DIZrr : AVX512SI<0x7E, MRMDestReg, (outs GR32:$dst),
(ins FR32X:$src),
"vmovd{z}\t{$src, $dst|$dst, $src}",
"vmovd{z}\t{$src, $dst|$dst, $src}",
[(store (i32 (bitconvert FR32X:$src)), addr:$dst)],
IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+}
// Move Quadword Int to Packed Quadword Int
//
-def VMOVQI2PQIZrm : AVX512SI<0x7E, MRMSrcMem, (outs VR128X:$dst),
+def VMOVQI2PQIZrm : AVX512SI<0x6E, MRMSrcMem, (outs VR128X:$dst),
(ins i64mem:$src),
"vmovq{z}\t{$src, $dst|$dst, $src}",
[(set VR128X:$dst,
// 256-bit types
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
(v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
- (SUBREG_TO_REG (i32 0), (VMOVSSZrm addr:$src), sub_xmm)>;
+ (SUBREG_TO_REG (i32 0), (VMOVDI2PDIZrm addr:$src), sub_xmm)>;
def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
(v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
(SUBREG_TO_REG (i32 0), (VMOVSSZrm addr:$src), sub_xmm)>;
FR64X:$src)), sub_xmm)>;
def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
(v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
- (SUBREG_TO_REG (i64 0), (VMOVSDZrm addr:$src), sub_xmm)>;
+ (SUBREG_TO_REG (i64 0), (VMOVQI2PQIZrm addr:$src), sub_xmm)>;
// Move low f64 and clear high bits.
def : Pat<(v4f64 (X86vzmovl (v4f64 VR256X:$src))),
let AddedComplexity = 20 in {
def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$src))))),
(VMOVDI2PDIZrm addr:$src)>;
+ def : Pat<(v2i64 (X86vzmovl (v2i64 (scalar_to_vector GR64:$src)))),
+ (VMOV64toPQIZrr GR64:$src)>;
+ def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector GR32:$src)))),
+ (VMOVDI2PDIZrr GR32:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
(VMOVDI2PDIZrm addr:$src)>;
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128X:$src))),
(VMOVZPQILo2PQIZrr VR128X:$src)>;
}
+
// Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
(v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))),
(VPMULUDQZrr VR512:$src1, VR512:$src2)>;
+defm VPMAXUDZ : avx512_binop_rm<0x3F, "vpmaxud", X86umax, v16i32, VR512, memopv16i32,
+ i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>,
+ T8, EVEX_V512, EVEX_CD8<32, CD8VF>;
+defm VPMAXUQZ : avx512_binop_rm<0x3F, "vpmaxuq", X86umax, v8i64, VR512, memopv8i64,
+ i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>,
+ T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
+defm VPMAXSDZ : avx512_binop_rm<0x3D, "vpmaxsd", X86smax, v16i32, VR512, memopv16i32,
+ i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>,
+ EVEX_V512, EVEX_CD8<32, CD8VF>;
+defm VPMAXSQZ : avx512_binop_rm<0x3D, "vpmaxsq", X86smax, v8i64, VR512, memopv8i64,
+ i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>,
+ T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
+defm VPMINUDZ : avx512_binop_rm<0x3B, "vpminud", X86umin, v16i32, VR512, memopv16i32,
+ i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>,
+ T8, EVEX_V512, EVEX_CD8<32, CD8VF>;
+defm VPMINUQZ : avx512_binop_rm<0x3B, "vpminuq", X86umin, v8i64, VR512, memopv8i64,
+ i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>,
+ T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
+defm VPMINSDZ : avx512_binop_rm<0x39, "vpminsd", X86smin, v16i32, VR512, memopv16i32,
+ i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>,
+ T8, EVEX_V512, EVEX_CD8<32, CD8VF>;
+defm VPMINSQZ : avx512_binop_rm<0x39, "vpminsq", X86smin, v8i64, VR512, memopv8i64,
+ i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>,
+ T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
//===----------------------------------------------------------------------===//
// AVX-512 - Unpack Instructions
//===----------------------------------------------------------------------===//
(outs RC:$dst), (ins RC:$src1, RC:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1, RC:$src2)))],
- d>, EVEX_4V, TB;
+ d>, EVEX_4V;
def rm : AVX512PI<opc, MRMSrcMem,
(outs RC:$dst), (ins RC:$src1, x86memop:$src2),
asm, [(set RC:$dst,
(vt (OpNode RC:$src1,
(bitconvert (mem_frag addr:$src2)))))],
- d>, EVEX_4V, TB;
+ d>, EVEX_4V;
}
defm VUNPCKHPSZ: avx512_unpack_fp<0x15, X86Unpckh, v16f32, memopv8f64,
def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>,
- EVEX_4V;
+ EVEX_4V, TB;
let mayLoad = 1 in {
def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
- itins.rm, d>, EVEX_4V;
+ itins.rm, d>, EVEX_4V, TB;
def rmb : PI<opc, MRMSrcMem, (outs RC:$dst),
(ins RC:$src1, x86scalar_mop:$src2),
!strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
", $src1, $dst|$dst, $src1, ${src2}", BrdcstStr, "}"),
[(set RC:$dst, (OpNode RC:$src1,
(vt (X86VBroadcast (scalar_mfrag addr:$src2)))))],
- itins.rm, d>, EVEX_4V, EVEX_B;
+ itins.rm, d>, EVEX_4V, EVEX_B, TB;
}
}
ValueType vt, X86MemOperand x86memop, PatFrag mem_frag,
RegisterClass KRC> {
def ri : AVX512BIi8<opc, ImmFormR, (outs RC:$dst),
- (ins RC:$src1, i32i8imm:$src2),
+ (ins RC:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set RC:$dst, (vt (OpNode RC:$src1, (i32 imm:$src2))))],
+ [(set RC:$dst, (vt (OpNode RC:$src1, (i8 imm:$src2))))],
SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V;
def rik : AVX512BIi8<opc, ImmFormR, (outs RC:$dst),
- (ins KRC:$mask, RC:$src1, i32i8imm:$src2),
+ (ins KRC:$mask, RC:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
[], SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V, EVEX_K;
def mi: AVX512BIi8<opc, ImmFormM, (outs RC:$dst),
- (ins x86memop:$src1, i32i8imm:$src2),
+ (ins x86memop:$src1, i8imm:$src2),
!strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
[(set RC:$dst, (OpNode (mem_frag addr:$src1),
- (i32 imm:$src2)))], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V;
+ (i8 imm:$src2)))], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V;
def mik: AVX512BIi8<opc, ImmFormM, (outs RC:$dst),
- (ins KRC:$mask, x86memop:$src1, i32i8imm:$src2),
+ (ins KRC:$mask, x86memop:$src1, i8imm:$src2),
!strconcat(OpcodeStr,
"\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
[], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V, EVEX_K;
X86MemOperand x86memop, string asm> {
let neverHasSideEffects = 1 in {
def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
- !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, EVEX_4V;
+ !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+ EVEX_4V;
let mayLoad = 1 in
def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
(ins DstRC:$src1, x86memop:$src),
- !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, EVEX_4V;
+ !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>,
+ EVEX_4V;
} // neverHasSideEffects = 1
}
-
+let Predicates = [HasAVX512] in {
defm VCVTSI2SSZ : avx512_vcvtsi<0x2A, GR32, FR32X, i32mem, "cvtsi2ss{l}{z}">,
XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
-defm VCVTSI2SS64Z : avx512_vcvtsi<0x2A, GR64, FR32X, i64mem, "cvtsi2ss{q}{z}">,
+defm VCVTSI642SSZ : avx512_vcvtsi<0x2A, GR64, FR32X, i64mem, "cvtsi2ss{q}{z}">,
XS, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>;
defm VCVTSI2SDZ : avx512_vcvtsi<0x2A, GR32, FR64X, i32mem, "cvtsi2sd{l}{z}">,
XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
-defm VCVTSI2SD64Z : avx512_vcvtsi<0x2A, GR64, FR64X, i64mem, "cvtsi2sd{q}{z}">,
+defm VCVTSI642SDZ : avx512_vcvtsi<0x2A, GR64, FR64X, i64mem, "cvtsi2sd{q}{z}">,
XD, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>;
def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
(VCVTSI2SSZrm (f32 (IMPLICIT_DEF)), addr:$src)>;
def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
- (VCVTSI2SS64Zrm (f32 (IMPLICIT_DEF)), addr:$src)>;
+ (VCVTSI642SSZrm (f32 (IMPLICIT_DEF)), addr:$src)>;
def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
(VCVTSI2SDZrm (f64 (IMPLICIT_DEF)), addr:$src)>;
def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
- (VCVTSI2SD64Zrm (f64 (IMPLICIT_DEF)), addr:$src)>;
+ (VCVTSI642SDZrm (f64 (IMPLICIT_DEF)), addr:$src)>;
def : Pat<(f32 (sint_to_fp GR32:$src)),
(VCVTSI2SSZrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
def : Pat<(f32 (sint_to_fp GR64:$src)),
- (VCVTSI2SS64Zrr (f32 (IMPLICIT_DEF)), GR64:$src)>;
+ (VCVTSI642SSZrr (f32 (IMPLICIT_DEF)), GR64:$src)>;
def : Pat<(f64 (sint_to_fp GR32:$src)),
(VCVTSI2SDZrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
def : Pat<(f64 (sint_to_fp GR64:$src)),
- (VCVTSI2SD64Zrr (f64 (IMPLICIT_DEF)), GR64:$src)>;
+ (VCVTSI642SDZrr (f64 (IMPLICIT_DEF)), GR64:$src)>;
+
+defm VCVTUSI2SSZ : avx512_vcvtsi<0x7B, GR32, FR32X, i32mem, "cvtusi2ss{l}{z}">,
+ XS, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+defm VCVTUSI642SSZ : avx512_vcvtsi<0x7B, GR64, FR32X, i64mem, "cvtusi2ss{q}{z}">,
+ XS, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>;
+defm VCVTUSI2SDZ : avx512_vcvtsi<0x7B, GR32, FR64X, i32mem, "cvtusi2sd{l}{z}">,
+ XD, VEX_LIG, EVEX_CD8<32, CD8VT1>;
+defm VCVTUSI642SDZ : avx512_vcvtsi<0x7B, GR64, FR64X, i64mem, "cvtusi2sd{q}{z}">,
+ XD, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>;
+def : Pat<(f32 (uint_to_fp (loadi32 addr:$src))),
+ (VCVTUSI2SSZrm (f32 (IMPLICIT_DEF)), addr:$src)>;
+def : Pat<(f32 (uint_to_fp (loadi64 addr:$src))),
+ (VCVTUSI642SSZrm (f32 (IMPLICIT_DEF)), addr:$src)>;
+def : Pat<(f64 (uint_to_fp (loadi32 addr:$src))),
+ (VCVTUSI2SDZrm (f64 (IMPLICIT_DEF)), addr:$src)>;
+def : Pat<(f64 (uint_to_fp (loadi64 addr:$src))),
+ (VCVTUSI642SDZrm (f64 (IMPLICIT_DEF)), addr:$src)>;
+
+def : Pat<(f32 (uint_to_fp GR32:$src)),
+ (VCVTUSI2SSZrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
+def : Pat<(f32 (uint_to_fp GR64:$src)),
+ (VCVTUSI642SSZrr (f32 (IMPLICIT_DEF)), GR64:$src)>;
+def : Pat<(f64 (uint_to_fp GR32:$src)),
+ (VCVTUSI2SDZrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
+def : Pat<(f64 (uint_to_fp GR64:$src)),
+ (VCVTUSI642SDZrr (f64 (IMPLICIT_DEF)), GR64:$src)>;
+}
+//===----------------------------------------------------------------------===//
+// AVX-512 Scalar convert from float/double to integer
+//===----------------------------------------------------------------------===//
+multiclass avx512_cvt_s_int<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
+ Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
+ string asm> {
+let neverHasSideEffects = 1 in {
+ def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
+ !strconcat(asm,"\t{$src, $dst|$dst, $src}"),
+ [(set DstRC:$dst, (Int SrcRC:$src))]>, EVEX, VEX_LIG;
+ let mayLoad = 1 in
+ def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
+ !strconcat(asm,"\t{$src, $dst|$dst, $src}"), []>, EVEX, VEX_LIG;
+} // neverHasSideEffects = 1
+}
+let Predicates = [HasAVX512] in {
+// Convert float/double to signed/unsigned int 32/64
+defm VCVTSS2SIZ: avx512_cvt_s_int<0x2D, VR128X, GR32, int_x86_sse_cvtss2si,
+ ssmem, sse_load_f32, "cvtss2si{z}">,
+ XS, EVEX_CD8<32, CD8VT1>;
+defm VCVTSS2SI64Z: avx512_cvt_s_int<0x2D, VR128X, GR64, int_x86_sse_cvtss2si64,
+ ssmem, sse_load_f32, "cvtss2si{z}">,
+ XS, VEX_W, EVEX_CD8<32, CD8VT1>;
+defm VCVTSS2USIZ: avx512_cvt_s_int<0x79, VR128X, GR32, int_x86_avx512_cvtss2usi,
+ ssmem, sse_load_f32, "cvtss2usi{z}">,
+ XS, EVEX_CD8<32, CD8VT1>;
+defm VCVTSS2USI64Z: avx512_cvt_s_int<0x79, VR128X, GR64,
+ int_x86_avx512_cvtss2usi64, ssmem,
+ sse_load_f32, "cvtss2usi{z}">, XS, VEX_W,
+ EVEX_CD8<32, CD8VT1>;
+defm VCVTSD2SIZ: avx512_cvt_s_int<0x2D, VR128X, GR32, int_x86_sse2_cvtsd2si,
+ sdmem, sse_load_f64, "cvtsd2si{z}">,
+ XD, EVEX_CD8<64, CD8VT1>;
+defm VCVTSD2SI64Z: avx512_cvt_s_int<0x2D, VR128X, GR64, int_x86_sse2_cvtsd2si64,
+ sdmem, sse_load_f64, "cvtsd2si{z}">,
+ XD, VEX_W, EVEX_CD8<64, CD8VT1>;
+defm VCVTSD2USIZ: avx512_cvt_s_int<0x79, VR128X, GR32, int_x86_avx512_cvtsd2usi,
+ sdmem, sse_load_f64, "cvtsd2usi{z}">,
+ XD, EVEX_CD8<64, CD8VT1>;
+defm VCVTSD2USI64Z: avx512_cvt_s_int<0x79, VR128X, GR64,
+ int_x86_avx512_cvtsd2usi64, sdmem,
+ sse_load_f64, "cvtsd2usi{z}">, XD, VEX_W,
+ EVEX_CD8<64, CD8VT1>;
+
+defm Int_VCVTSI2SSZ : sse12_cvt_sint_3addr<0x2A, GR32, VR128X,
+ int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}{z}",
+ SSE_CVT_Scalar, 0>, XS, EVEX_4V;
+defm Int_VCVTSI2SS64Z : sse12_cvt_sint_3addr<0x2A, GR64, VR128X,
+ int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}{z}",
+ SSE_CVT_Scalar, 0>, XS, EVEX_4V, VEX_W;
+defm Int_VCVTSI2SDZ : sse12_cvt_sint_3addr<0x2A, GR32, VR128X,
+ int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}{z}",
+ SSE_CVT_Scalar, 0>, XD, EVEX_4V;
+defm Int_VCVTSI2SD64Z : sse12_cvt_sint_3addr<0x2A, GR64, VR128X,
+ int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}{z}",
+ SSE_CVT_Scalar, 0>, XD, EVEX_4V, VEX_W;
+
+defm Int_VCVTUSI2SSZ : sse12_cvt_sint_3addr<0x2A, GR32, VR128X,
+ int_x86_avx512_cvtusi2ss, i32mem, loadi32, "cvtusi2ss{l}{z}",
+ SSE_CVT_Scalar, 0>, XS, EVEX_4V;
+defm Int_VCVTUSI2SS64Z : sse12_cvt_sint_3addr<0x2A, GR64, VR128X,
+ int_x86_avx512_cvtusi642ss, i64mem, loadi64, "cvtusi2ss{q}{z}",
+ SSE_CVT_Scalar, 0>, XS, EVEX_4V, VEX_W;
+defm Int_VCVTUSI2SDZ : sse12_cvt_sint_3addr<0x2A, GR32, VR128X,
+ int_x86_avx512_cvtusi2sd, i32mem, loadi32, "cvtusi2sd{l}{z}",
+ SSE_CVT_Scalar, 0>, XD, EVEX_4V;
+defm Int_VCVTUSI2SD64Z : sse12_cvt_sint_3addr<0x2A, GR64, VR128X,
+ int_x86_avx512_cvtusi642sd, i64mem, loadi64, "cvtusi2sd{q}{z}",
+ SSE_CVT_Scalar, 0>, XD, EVEX_4V, VEX_W;
+
+// Convert float/double to signed/unsigned int 32/64 with truncation
+defm Int_VCVTTSS2SIZ : avx512_cvt_s_int<0x2C, VR128X, GR32, int_x86_sse_cvttss2si,
+ ssmem, sse_load_f32, "cvttss2si{z}">,
+ XS, EVEX_CD8<32, CD8VT1>;
+defm Int_VCVTTSS2SI64Z : avx512_cvt_s_int<0x2C, VR128X, GR64,
+ int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
+ "cvttss2si{z}">, XS, VEX_W,
+ EVEX_CD8<32, CD8VT1>;
+defm Int_VCVTTSD2SIZ : avx512_cvt_s_int<0x2C, VR128X, GR32, int_x86_sse2_cvttsd2si,
+ sdmem, sse_load_f64, "cvttsd2si{z}">, XD,
+ EVEX_CD8<64, CD8VT1>;
+defm Int_VCVTTSD2SI64Z : avx512_cvt_s_int<0x2C, VR128X, GR64,
+ int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
+ "cvttsd2si{z}">, XD, VEX_W,
+ EVEX_CD8<64, CD8VT1>;
+defm Int_VCVTTSS2USIZ : avx512_cvt_s_int<0x78, VR128X, GR32,
+ int_x86_avx512_cvttss2usi, ssmem, sse_load_f32,
+ "cvttss2si{z}">, XS, EVEX_CD8<32, CD8VT1>;
+defm Int_VCVTTSS2USI64Z : avx512_cvt_s_int<0x78, VR128X, GR64,
+ int_x86_avx512_cvttss2usi64, ssmem,
+ sse_load_f32, "cvttss2usi{z}">, XS, VEX_W,
+ EVEX_CD8<32, CD8VT1>;
+defm Int_VCVTTSD2USIZ : avx512_cvt_s_int<0x78, VR128X, GR32,
+ int_x86_avx512_cvttsd2usi,
+ sdmem, sse_load_f64, "cvttsd2usi{z}">, XD,
+ EVEX_CD8<64, CD8VT1>;
+defm Int_VCVTTSD2USI64Z : avx512_cvt_s_int<0x78, VR128X, GR64,
+ int_x86_avx512_cvttsd2usi64, sdmem,
+ sse_load_f64, "cvttsd2usi{z}">, XD, VEX_W,
+ EVEX_CD8<64, CD8VT1>;
+}
+
+multiclass avx512_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
+ SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
+ string asm> {
+ def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
+ !strconcat(asm,"\t{$src, $dst|$dst, $src}"),
+ [(set DstRC:$dst, (OpNode SrcRC:$src))]>, EVEX;
+ def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
+ !strconcat(asm,"\t{$src, $dst|$dst, $src}"),
+ [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>, EVEX;
+}
+
+defm VCVTTSS2SIZ : avx512_cvt_s<0x2C, FR32X, GR32, fp_to_sint, f32mem,
+ loadf32, "cvttss2si{z}">, XS,
+ EVEX_CD8<32, CD8VT1>;
+defm VCVTTSS2USIZ : avx512_cvt_s<0x78, FR32X, GR32, fp_to_uint, f32mem,
+ loadf32, "cvttss2usi{z}">, XS,
+ EVEX_CD8<32, CD8VT1>;
+defm VCVTTSS2SI64Z : avx512_cvt_s<0x2C, FR32X, GR64, fp_to_sint, f32mem,
+ loadf32, "cvttss2si{z}">, XS, VEX_W,
+ EVEX_CD8<32, CD8VT1>;
+defm VCVTTSS2USI64Z : avx512_cvt_s<0x78, FR32X, GR64, fp_to_uint, f32mem,
+ loadf32, "cvttss2usi{z}">, XS, VEX_W,
+ EVEX_CD8<32, CD8VT1>;
+defm VCVTTSD2SIZ : avx512_cvt_s<0x2C, FR64X, GR32, fp_to_sint, f64mem,
+ loadf64, "cvttsd2si{z}">, XD,
+ EVEX_CD8<64, CD8VT1>;
+defm VCVTTSD2USIZ : avx512_cvt_s<0x78, FR64X, GR32, fp_to_uint, f64mem,
+ loadf64, "cvttsd2usi{z}">, XD,
+ EVEX_CD8<64, CD8VT1>;
+defm VCVTTSD2SI64Z : avx512_cvt_s<0x2C, FR64X, GR64, fp_to_sint, f64mem,
+ loadf64, "cvttsd2si{z}">, XD, VEX_W,
+ EVEX_CD8<64, CD8VT1>;
+defm VCVTTSD2USI64Z : avx512_cvt_s<0x78, FR64X, GR64, fp_to_uint, f64mem,
+ loadf64, "cvttsd2usi{z}">, XD, VEX_W,
+ EVEX_CD8<64, CD8VT1>;
//===----------------------------------------------------------------------===//
// AVX-512 Convert form float to double and back
//===----------------------------------------------------------------------===//
(VCVTPS2PDZrm addr:$src)>;
}
+//===----------------------------------------------------------------------===//
+// Half precision conversion instructions
+//===----------------------------------------------------------------------===//
+multiclass avx512_f16c_ph2ps<RegisterClass destRC, RegisterClass srcRC,
+ X86MemOperand x86memop, Intrinsic Int> {
+ def rr : AVX5128I<0x13, MRMSrcReg, (outs destRC:$dst), (ins srcRC:$src),
+ "vcvtph2ps\t{$src, $dst|$dst, $src}",
+ [(set destRC:$dst, (Int srcRC:$src))]>, EVEX;
+ let neverHasSideEffects = 1, mayLoad = 1 in
+ def rm : AVX5128I<0x13, MRMSrcMem, (outs destRC:$dst), (ins x86memop:$src),
+ "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, EVEX;
+}
+
+multiclass avx512_f16c_ps2ph<RegisterClass destRC, RegisterClass srcRC,
+ X86MemOperand x86memop, Intrinsic Int> {
+ def rr : AVX512AIi8<0x1D, MRMDestReg, (outs destRC:$dst),
+ (ins srcRC:$src1, i32i8imm:$src2),
+ "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
+ [(set destRC:$dst, (Int srcRC:$src1, imm:$src2))]>, EVEX;
+ let neverHasSideEffects = 1, mayStore = 1 in
+ def mr : AVX512AIi8<0x1D, MRMDestMem, (outs),
+ (ins x86memop:$dst, srcRC:$src1, i32i8imm:$src2),
+ "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, EVEX;
+}
+
+defm VCVTPH2PSZ : avx512_f16c_ph2ps<VR512, VR256X, f256mem,
+ int_x86_avx512_vcvtph2ps_512>, EVEX_V512,
+ EVEX_CD8<32, CD8VH>;
+defm VCVTPS2PHZ : avx512_f16c_ps2ph<VR256X, VR512, f256mem,
+ int_x86_avx512_vcvtps2ph_512>, EVEX_V512,
+ EVEX_CD8<32, CD8VH>;
+
let Defs = [EFLAGS], Predicates = [HasAVX512] in {
defm VUCOMISSZ : sse12_ord_cmp<0x2E, FR32X, X86cmp, f32, f32mem, loadf32,
"ucomiss{z}">, TB, EVEX, VEX_LIG,
}
/// avx512_fp_unop_s - AVX-512 unops in scalar form.
-multiclass avx512_fp_unop_s<bits<8> opc, string OpcodeStr,
- Intrinsic F32Int, Intrinsic F64Int> {
+multiclass avx512_fp_unop_s<bits<8> opc, string OpcodeStr> {
let hasSideEffects = 0 in {
def SSZr : AVX5128I<opc, MRMSrcReg, (outs FR32X:$dst),
(ins FR32X:$src1, FR32X:$src2),
(ins VR128X:$src1, ssmem:$src2),
!strconcat(OpcodeStr,
"ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst, (F32Int VR128X:$src1, sse_load_f32:$src2))]>,
- EVEX_4V, EVEX_CD8<32, CD8VT1>;
+ []>, EVEX_4V, EVEX_CD8<32, CD8VT1>;
}
def SDZr : AVX5128I<opc, MRMSrcReg, (outs FR64X:$dst),
(ins FR64X:$src1, FR64X:$src2),
(ins FR64X:$src1, f64mem:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>,
- EVEX_4V, VEX_W, EVEX_CD8<32, CD8VT1>;
+ EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>;
def SDZm_Int : AVX5128I<opc, MRMSrcMem, (outs VR128X:$dst),
(ins VR128X:$src1, sdmem:$src2),
!strconcat(OpcodeStr,
"sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
- [(set VR128X:$dst, (F64Int VR128X:$src1, sse_load_f64:$src2))]>,
- EVEX_4V, VEX_W, EVEX_CD8<32, CD8VT1>;
+ []>, EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>;
}
}
}
-defm VRCP14 : avx512_fp_unop_s<0x4D, "vrcp14", int_x86_avx512_rcp14_ss,
- int_x86_avx512_rcp14_sd>,
+defm VRCP14 : avx512_fp_unop_s<0x4D, "vrcp14">,
avx512_fp_unop_p<0x4C, "vrcp14", X86frcp>,
avx512_fp_unop_p_int<0x4C, "vrcp14",
int_x86_avx512_rcp14_ps_512, int_x86_avx512_rcp14_pd_512>;
-defm VRSQRT14 : avx512_fp_unop_s<0x4F, "vrsqrt14", int_x86_avx512_rsqrt14_ss,
- int_x86_avx512_rsqrt14_sd>,
+defm VRSQRT14 : avx512_fp_unop_s<0x4F, "vrsqrt14">,
avx512_fp_unop_p<0x4E, "vrsqrt14", X86frsqrt>,
avx512_fp_unop_p_int<0x4E, "vrsqrt14",
int_x86_avx512_rsqrt14_ps_512, int_x86_avx512_rsqrt14_pd_512>;
+def : Pat<(int_x86_avx512_rsqrt14_ss VR128X:$src),
+ (COPY_TO_REGCLASS (VRSQRT14SSZr (f32 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR32)),
+ VR128X)>;
+def : Pat<(int_x86_avx512_rsqrt14_ss sse_load_f32:$src),
+ (VRSQRT14SSZm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+def : Pat<(int_x86_avx512_rcp14_ss VR128X:$src),
+ (COPY_TO_REGCLASS (VRCP14SSZr (f32 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR32)),
+ VR128X)>;
+def : Pat<(int_x86_avx512_rcp14_ss sse_load_f32:$src),
+ (VRCP14SSZm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+let AddedComplexity = 20, Predicates = [HasERI] in {
+defm VRCP28 : avx512_fp_unop_s<0xCB, "vrcp28">,
+ avx512_fp_unop_p<0xCA, "vrcp28", X86frcp>,
+ avx512_fp_unop_p_int<0xCA, "vrcp28",
+ int_x86_avx512_rcp28_ps_512, int_x86_avx512_rcp28_pd_512>;
+
+defm VRSQRT28 : avx512_fp_unop_s<0xCD, "vrsqrt28">,
+ avx512_fp_unop_p<0xCC, "vrsqrt28", X86frsqrt>,
+ avx512_fp_unop_p_int<0xCC, "vrsqrt28",
+ int_x86_avx512_rsqrt28_ps_512, int_x86_avx512_rsqrt28_pd_512>;
+}
+
+let Predicates = [HasERI] in {
+ def : Pat<(int_x86_avx512_rsqrt28_ss VR128X:$src),
+ (COPY_TO_REGCLASS (VRSQRT28SSZr (f32 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR32)),
+ VR128X)>;
+ def : Pat<(int_x86_avx512_rsqrt28_ss sse_load_f32:$src),
+ (VRSQRT28SSZm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+ def : Pat<(int_x86_avx512_rcp28_ss VR128X:$src),
+ (COPY_TO_REGCLASS (VRCP28SSZr (f32 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR32)),
+ VR128X)>;
+ def : Pat<(int_x86_avx512_rcp28_ss sse_load_f32:$src),
+ (VRCP28SSZm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+}
multiclass avx512_sqrt_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
Intrinsic V16F32Int, Intrinsic V8F64Int,
OpndItins itins_s, OpndItins itins_d> {
int_x86_avx512_sqrt_ps_512, int_x86_avx512_sqrt_pd_512,
SSE_SQRTPS, SSE_SQRTPD>;
-def : Pat<(f32 (fsqrt FR32X:$src)),
- (VSQRTSSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
-def : Pat<(f32 (fsqrt (load addr:$src))),
- (VSQRTSSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[OptForSize]>;
-def : Pat<(f64 (fsqrt FR64X:$src)),
- (VSQRTSDZr (f64 (IMPLICIT_DEF)), FR64X:$src)>;
-def : Pat<(f64 (fsqrt (load addr:$src))),
- (VSQRTSDZm (f64 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[OptForSize]>;
-
-def : Pat<(f32 (X86frsqrt FR32X:$src)),
- (VRSQRT14SSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
-def : Pat<(f32 (X86frsqrt (load addr:$src))),
- (VRSQRT14SSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[OptForSize]>;
-
-def : Pat<(f32 (X86frcp FR32X:$src)),
- (VRCP14SSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
-def : Pat<(f32 (X86frcp (load addr:$src))),
- (VRCP14SSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
- Requires<[OptForSize]>;
+let Predicates = [HasAVX512] in {
+ def : Pat<(f32 (fsqrt FR32X:$src)),
+ (VSQRTSSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
+ def : Pat<(f32 (fsqrt (load addr:$src))),
+ (VSQRTSSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[OptForSize]>;
+ def : Pat<(f64 (fsqrt FR64X:$src)),
+ (VSQRTSDZr (f64 (IMPLICIT_DEF)), FR64X:$src)>;
+ def : Pat<(f64 (fsqrt (load addr:$src))),
+ (VSQRTSDZm (f64 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[OptForSize]>;
+
+ def : Pat<(f32 (X86frsqrt FR32X:$src)),
+ (VRSQRT14SSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
+ def : Pat<(f32 (X86frsqrt (load addr:$src))),
+ (VRSQRT14SSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[OptForSize]>;
+
+ def : Pat<(f32 (X86frcp FR32X:$src)),
+ (VRCP14SSZr (f32 (IMPLICIT_DEF)), FR32X:$src)>;
+ def : Pat<(f32 (X86frcp (load addr:$src))),
+ (VRCP14SSZm (f32 (IMPLICIT_DEF)), addr:$src)>,
+ Requires<[OptForSize]>;
+
+ def : Pat<(int_x86_sse_sqrt_ss VR128X:$src),
+ (COPY_TO_REGCLASS (VSQRTSSZr (f32 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR32)),
+ VR128X)>;
+ def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
+ (VSQRTSSZm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
+
+ def : Pat<(int_x86_sse2_sqrt_sd VR128X:$src),
+ (COPY_TO_REGCLASS (VSQRTSDZr (f64 (IMPLICIT_DEF)),
+ (COPY_TO_REGCLASS VR128X:$src, FR64)),
+ VR128X)>;
+ def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
+ (VSQRTSDZm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
+}
+
multiclass avx512_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
X86MemOperand x86memop, RegisterClass RC,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
(i8 imm:$src3))))], d, IIC_SSE_SHUFP>,
- EVEX_4V, TB, Sched<[WriteShuffleLd, ReadAfterLd]>;
+ EVEX_4V, Sched<[WriteShuffleLd, ReadAfterLd]>;
def rri : AVX512PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
(ins RC:$src1, RC:$src2, i8imm:$src3),
!strconcat(OpcodeStr,
"\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
[(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
(i8 imm:$src3))))], d, IIC_SSE_SHUFP>,
- EVEX_4V, TB, Sched<[WriteShuffle]>;
+ EVEX_4V, Sched<[WriteShuffle]>;
}
defm VSHUFPSZ : avx512_shufp<VR512, f512mem, v16f32, "vshufps", memopv16f32,
defm VSHUFPDZ : avx512_shufp<VR512, f512mem, v8f64, "vshufpd", memopv8f64,
SSEPackedDouble>, OpSize, VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
+def : Pat<(v16i32 (X86Shufp VR512:$src1, VR512:$src2, (i8 imm:$imm))),
+ (VSHUFPSZrri VR512:$src1, VR512:$src2, imm:$imm)>;
+def : Pat<(v16i32 (X86Shufp VR512:$src1,
+ (memopv16i32 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPSZrmi VR512:$src1, addr:$src2, imm:$imm)>;
+
+def : Pat<(v8i64 (X86Shufp VR512:$src1, VR512:$src2, (i8 imm:$imm))),
+ (VSHUFPDZrri VR512:$src1, VR512:$src2, imm:$imm)>;
+def : Pat<(v8i64 (X86Shufp VR512:$src1,
+ (memopv8i64 addr:$src2), (i8 imm:$imm))),
+ (VSHUFPDZrmi VR512:$src1, addr:$src2, imm:$imm)>;
multiclass avx512_alignr<string OpcodeStr, RegisterClass RC,
X86MemOperand x86memop> {