def VINSERTPSzrr : AVX512AIi8<0x21, MRMSrcReg, (outs VR128X:$dst),
(ins VR128X:$src1, VR128X:$src2, u32u8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- [(set VR128X:$dst, (X86insrtps VR128X:$src1, VR128X:$src2, imm:$src3))]>,
+ [(set VR128X:$dst, (X86insertps VR128X:$src1, VR128X:$src2, imm:$src3))]>,
EVEX_4V;
def VINSERTPSzrm: AVX512AIi8<0x21, MRMSrcMem, (outs VR128X:$dst),
(ins VR128X:$src1, f32mem:$src2, u32u8imm:$src3),
"vinsertps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
- [(set VR128X:$dst, (X86insrtps VR128X:$src1,
+ [(set VR128X:$dst, (X86insertps VR128X:$src1,
(v4f32 (scalar_to_vector (loadf32 addr:$src2))),
imm:$src3))]>, EVEX_4V, EVEX_CD8<32, CD8VT1>;
X86VPermv3, v16f32>, EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VPERMT2PD : avx512_perm_3src<0x7F, "vpermt2pd", VR512, memopv8f64, i512mem,
X86VPermv3, v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
+def : Pat<(v16f32 (int_x86_avx512_mask_vpermt_ps_512 (v16i32 VR512:$idx),
+ (v16f32 VR512:$src1), (v16f32 VR512:$src2), (i16 -1))),
+ (VPERMT2PSrr VR512:$src1, VR512:$idx, VR512:$src2)>;
+
+def : Pat<(v16i32 (int_x86_avx512_mask_vpermt_d_512 (v16i32 VR512:$idx),
+ (v16i32 VR512:$src1), (v16i32 VR512:$src2), (i16 -1))),
+ (VPERMT2Drr VR512:$src1, VR512:$idx, VR512:$src2)>;
+
+def : Pat<(v8f64 (int_x86_avx512_mask_vpermt_pd_512 (v8i64 VR512:$idx),
+ (v8f64 VR512:$src1), (v8f64 VR512:$src2), (i8 -1))),
+ (VPERMT2PDrr VR512:$src1, VR512:$idx, VR512:$src2)>;
+
+def : Pat<(v8i64 (int_x86_avx512_mask_vpermt_q_512 (v8i64 VR512:$idx),
+ (v8i64 VR512:$src1), (v8i64 VR512:$src2), (i8 -1))),
+ (VPERMT2Qrr VR512:$src1, VR512:$idx, VR512:$src2)>;
//===----------------------------------------------------------------------===//
// AVX-512 - BLEND using mask
//
(EXTRACT_SUBREG
(AND32ri (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16)), (i32 1)),
sub_16bit)>;
+ def : Pat<(v16i1 (scalar_to_vector VK1:$src)),
+ (COPY_TO_REGCLASS VK1:$src, VK16)>;
+ def : Pat<(v8i1 (scalar_to_vector VK1:$src)),
+ (COPY_TO_REGCLASS VK1:$src, VK8)>;
}
// With AVX-512 only, 8-bit mask is promoted to 16-bit mask.
let Predicates = [HasAVX512] in {
"vmovdqu64", SSEPackedInt, v8i64>,
XS, VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
+def: Pat<(v16i32 (int_x86_avx512_mask_loadu_d_512 addr:$ptr,
+ (v16i32 immAllZerosV), GR16:$mask)),
+ (VMOVDQU32rmkz (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), addr:$ptr)>;
+
+def: Pat<(v8i64 (int_x86_avx512_mask_loadu_q_512 addr:$ptr,
+ (bc_v8i64 (v16i32 immAllZerosV)), GR8:$mask)),
+ (VMOVDQU64rmkz (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), addr:$ptr)>;
+
+def: Pat<(int_x86_avx512_mask_storeu_d_512 addr:$ptr, (v16i32 VR512:$src),
+ GR16:$mask),
+ (VMOVDQU32mrk addr:$ptr, (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)),
+ VR512:$src)>;
+def: Pat<(int_x86_avx512_mask_storeu_q_512 addr:$ptr, (v8i64 VR512:$src),
+ GR8:$mask),
+ (VMOVDQU64mrk addr:$ptr, (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)),
+ VR512:$src)>;
+
let AddedComplexity = 20 in {
def : Pat<(v8i64 (vselect VK8WM:$mask, (v8i64 VR512:$src),
(bc_v8i64 (v16i32 immAllZerosV)))),
def : Pat<(v8i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))),
(SUBREG_TO_REG (i32 0), (VMOV64toPQIZrr GR64:$src2), sub_xmm)>;
+//===----------------------------------------------------------------------===//
+// AVX-512 - Non-temporals
+//===----------------------------------------------------------------------===//
+
+def VMOVNTDQAZrm : AVX5128I<0x2A, MRMSrcMem, (outs VR512:$dst),
+ (ins i512mem:$src),
+ "vmovntdqa\t{$src, $dst|$dst, $src}",
+ [(set VR512:$dst,
+ (int_x86_avx512_movntdqa addr:$src))]>,
+ EVEX, EVEX_V512;
+
//===----------------------------------------------------------------------===//
// AVX-512 - Integer arithmetic
//
(EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
(v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+def : Pat<(v4i32 (fp_to_uint (v4f32 VR128X:$src1))),
+ (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr
+ (v16f32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
+
+def : Pat<(v8f32 (uint_to_fp (v8i32 VR256X:$src1))),
+ (EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
+ (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>;
+
+def : Pat<(v4f32 (uint_to_fp (v4i32 VR128X:$src1))),
+ (EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
+ (v16i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
def : Pat<(v16f32 (int_x86_avx512_mask_cvtdq2ps_512 (v16i32 VR512:$src),
(bc_v16f32 (v16i32 immAllZerosV)), (i16 -1), imm:$rc)),
EVEX_CD8<32, CD8VF>;
def : Pat<(v16f32 (int_x86_avx512_mask_rndscale_ps_512 (v16f32 VR512:$src1),
- imm:$src2, (bc_v16f32 (v16i32 immAllZerosV)), (i16 -1),
+ imm:$src2, (v16f32 VR512:$src1), (i16 -1),
FROUND_CURRENT)),
(VRNDSCALEPSZr VR512:$src1, imm:$src2)>;
VEX_W, EVEX_CD8<64, CD8VF>;
def : Pat<(v8f64 (int_x86_avx512_mask_rndscale_pd_512 (v8f64 VR512:$src1),
- imm:$src2, (bc_v8f64 (v16i32 immAllZerosV)), (i8 -1),
+ imm:$src2, (v8f64 VR512:$src1), (i8 -1),
FROUND_CURRENT)),
(VRNDSCALEPDZr VR512:$src1, imm:$src2)>;
!strconcat(OpcodeStr," \t{$src, $dst|$dst, $src}"),
[]>, EVEX;
- def krr : AVX512XS8I<opc, MRMDestReg, (outs dstRC:$dst),
+ def rrk : AVX512XS8I<opc, MRMDestReg, (outs dstRC:$dst),
+ (ins KRC:$mask, srcRC:$src),
+ !strconcat(OpcodeStr,
+ " \t{$src, ${dst} {${mask}}|${dst} {${mask}}, $src}"),
+ []>, EVEX, EVEX_K;
+
+ def rrkz : AVX512XS8I<opc, MRMDestReg, (outs dstRC:$dst),
(ins KRC:$mask, srcRC:$src),
!strconcat(OpcodeStr,
" \t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"),
def mr : AVX512XS8I<opc, MRMDestMem, (outs), (ins x86memop:$dst, srcRC:$src),
!strconcat(OpcodeStr, " \t{$src, $dst|$dst, $src}"),
[]>, EVEX;
+
+ def mrk : AVX512XS8I<opc, MRMDestMem, (outs),
+ (ins x86memop:$dst, KRC:$mask, srcRC:$src),
+ !strconcat(OpcodeStr, " \t{$src, $dst {${mask}}|${dst} {${mask}}, $src}"),
+ []>, EVEX, EVEX_K;
+
}
defm VPMOVQB : avx512_trunc_sat<0x32, "vpmovqb", VR128X, VR512, VK8WM,
i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>;
def : Pat<(v8i32 (X86vtrunc (v8i64 VR512:$src))), (VPMOVQDrr VR512:$src)>;
def : Pat<(v16i8 (X86vtruncm VK16WM:$mask, (v16i32 VR512:$src))),
- (VPMOVDBkrr VK16WM:$mask, VR512:$src)>;
+ (VPMOVDBrrkz VK16WM:$mask, VR512:$src)>;
def : Pat<(v16i16 (X86vtruncm VK16WM:$mask, (v16i32 VR512:$src))),
- (VPMOVDWkrr VK16WM:$mask, VR512:$src)>;
+ (VPMOVDWrrkz VK16WM:$mask, VR512:$src)>;
def : Pat<(v8i16 (X86vtruncm VK8WM:$mask, (v8i64 VR512:$src))),
- (VPMOVQWkrr VK8WM:$mask, VR512:$src)>;
+ (VPMOVQWrrkz VK8WM:$mask, VR512:$src)>;
def : Pat<(v8i32 (X86vtruncm VK8WM:$mask, (v8i64 VR512:$src))),
- (VPMOVQDkrr VK8WM:$mask, VR512:$src)>;
+ (VPMOVQDrrkz VK8WM:$mask, VR512:$src)>;
-multiclass avx512_extend<bits<8> opc, string OpcodeStr, RegisterClass DstRC,
- RegisterClass SrcRC, SDNode OpNode, PatFrag mem_frag,
- X86MemOperand x86memop, ValueType OpVT, ValueType InVT> {
+multiclass avx512_extend<bits<8> opc, string OpcodeStr, RegisterClass KRC,
+ RegisterClass DstRC, RegisterClass SrcRC, SDNode OpNode,
+ PatFrag mem_frag, X86MemOperand x86memop,
+ ValueType OpVT, ValueType InVT> {
def rr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
(ins SrcRC:$src),
!strconcat(OpcodeStr, " \t{$src, $dst|$dst, $src}"),
[(set DstRC:$dst, (OpVT (OpNode (InVT SrcRC:$src))))]>, EVEX;
- def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst),
+
+ def rrk : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
+ (ins KRC:$mask, SrcRC:$src),
+ !strconcat(OpcodeStr, " \t{$src, $dst {${mask}} |$dst {${mask}}, $src}"),
+ []>, EVEX, EVEX_K;
+
+ def rrkz : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
+ (ins KRC:$mask, SrcRC:$src),
+ !strconcat(OpcodeStr, " \t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"),
+ []>, EVEX, EVEX_KZ;
+
+ let mayLoad = 1 in {
+ def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst),
(ins x86memop:$src),
!strconcat(OpcodeStr," \t{$src, $dst|$dst, $src}"),
[(set DstRC:$dst,
(OpVT (OpNode (InVT (bitconvert (mem_frag addr:$src))))))]>,
EVEX;
+
+ def rmk : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst),
+ (ins KRC:$mask, x86memop:$src),
+ !strconcat(OpcodeStr," \t{$src, $dst {${mask}} |$dst {${mask}}, $src}"),
+ []>,
+ EVEX, EVEX_K;
+
+ def rmkz : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst),
+ (ins KRC:$mask, x86memop:$src),
+ !strconcat(OpcodeStr," \t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"),
+ []>,
+ EVEX, EVEX_KZ;
+ }
}
-defm VPMOVZXBDZ: avx512_extend<0x31, "vpmovzxbd", VR512, VR128X, X86vzext,
+defm VPMOVZXBDZ: avx512_extend<0x31, "vpmovzxbd", VK16WM, VR512, VR128X, X86vzext,
memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VQ>;
-defm VPMOVZXBQZ: avx512_extend<0x32, "vpmovzxbq", VR512, VR128X, X86vzext,
+defm VPMOVZXBQZ: avx512_extend<0x32, "vpmovzxbq", VK8WM, VR512, VR128X, X86vzext,
memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VO>;
-defm VPMOVZXWDZ: avx512_extend<0x33, "vpmovzxwd", VR512, VR256X, X86vzext,
+defm VPMOVZXWDZ: avx512_extend<0x33, "vpmovzxwd", VK16WM, VR512, VR256X, X86vzext,
memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
EVEX_CD8<16, CD8VH>;
-defm VPMOVZXWQZ: avx512_extend<0x34, "vpmovzxwq", VR512, VR128X, X86vzext,
+defm VPMOVZXWQZ: avx512_extend<0x34, "vpmovzxwq", VK8WM, VR512, VR128X, X86vzext,
memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
EVEX_CD8<16, CD8VQ>;
-defm VPMOVZXDQZ: avx512_extend<0x35, "vpmovzxdq", VR512, VR256X, X86vzext,
+defm VPMOVZXDQZ: avx512_extend<0x35, "vpmovzxdq", VK8WM, VR512, VR256X, X86vzext,
memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
EVEX_CD8<32, CD8VH>;
-
-defm VPMOVSXBDZ: avx512_extend<0x21, "vpmovsxbd", VR512, VR128X, X86vsext,
+
+defm VPMOVSXBDZ: avx512_extend<0x21, "vpmovsxbd", VK16WM, VR512, VR128X, X86vsext,
memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VQ>;
-defm VPMOVSXBQZ: avx512_extend<0x22, "vpmovsxbq", VR512, VR128X, X86vsext,
+defm VPMOVSXBQZ: avx512_extend<0x22, "vpmovsxbq", VK8WM, VR512, VR128X, X86vsext,
memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
EVEX_CD8<8, CD8VO>;
-defm VPMOVSXWDZ: avx512_extend<0x23, "vpmovsxwd", VR512, VR256X, X86vsext,
+defm VPMOVSXWDZ: avx512_extend<0x23, "vpmovsxwd", VK16WM, VR512, VR256X, X86vsext,
memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
EVEX_CD8<16, CD8VH>;
-defm VPMOVSXWQZ: avx512_extend<0x24, "vpmovsxwq", VR512, VR128X, X86vsext,
+defm VPMOVSXWQZ: avx512_extend<0x24, "vpmovsxwq", VK8WM, VR512, VR128X, X86vsext,
memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
EVEX_CD8<16, CD8VQ>;
-defm VPMOVSXDQZ: avx512_extend<0x25, "vpmovsxdq", VR512, VR256X, X86vsext,
+defm VPMOVSXDQZ: avx512_extend<0x25, "vpmovsxdq", VK8WM, VR512, VR256X, X86vsext,
memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
EVEX_CD8<32, CD8VH>;
defm VPSCATTERQDZ : avx512_scatter<0xA1, "vpscatterqd", VK8WM, VR256X, vz64mem>,
EVEX_V512, EVEX_CD8<32, CD8VT1>;
+// prefetch
+multiclass avx512_gather_scatter_prefetch<bits<8> opc, Format F, string OpcodeStr,
+ RegisterClass KRC, X86MemOperand memop> {
+ let Predicates = [HasPFI], hasSideEffects = 1 in
+ def m : AVX5128I<opc, F, (outs), (ins KRC:$mask, memop:$src),
+ !strconcat(OpcodeStr, " \t{$src {${mask}}|{${mask}}, $src}"),
+ []>, EVEX, EVEX_K;
+}
+
+defm VGATHERPF0DPS: avx512_gather_scatter_prefetch<0xC6, MRM1m, "vgatherpf0dps",
+ VK16WM, vz32mem>, EVEX_V512, EVEX_CD8<32, CD8VT1>;
+
+defm VGATHERPF0QPS: avx512_gather_scatter_prefetch<0xC7, MRM1m, "vgatherpf0qps",
+ VK8WM, vz64mem>, EVEX_V512, EVEX_CD8<64, CD8VT1>;
+
+defm VGATHERPF0DPD: avx512_gather_scatter_prefetch<0xC6, MRM1m, "vgatherpf0dpd",
+ VK8WM, vy32mem>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VT1>;
+
+defm VGATHERPF0QPD: avx512_gather_scatter_prefetch<0xC7, MRM1m, "vgatherpf0qpd",
+ VK8WM, vz64mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
+
+defm VGATHERPF1DPS: avx512_gather_scatter_prefetch<0xC6, MRM2m, "vgatherpf1dps",
+ VK16WM, vz32mem>, EVEX_V512, EVEX_CD8<32, CD8VT1>;
+
+defm VGATHERPF1QPS: avx512_gather_scatter_prefetch<0xC7, MRM2m, "vgatherpf1qps",
+ VK8WM, vz64mem>, EVEX_V512, EVEX_CD8<64, CD8VT1>;
+
+defm VGATHERPF1DPD: avx512_gather_scatter_prefetch<0xC6, MRM2m, "vgatherpf1dpd",
+ VK8WM, vy32mem>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VT1>;
+
+defm VGATHERPF1QPD: avx512_gather_scatter_prefetch<0xC7, MRM2m, "vgatherpf1qpd",
+ VK8WM, vz64mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
+
+defm VSCATTERPF0DPS: avx512_gather_scatter_prefetch<0xC6, MRM5m, "vscatterpf0dps",
+ VK16WM, vz32mem>, EVEX_V512, EVEX_CD8<32, CD8VT1>;
+
+defm VSCATTERPF0QPS: avx512_gather_scatter_prefetch<0xC7, MRM5m, "vscatterpf0qps",
+ VK8WM, vz64mem>, EVEX_V512, EVEX_CD8<64, CD8VT1>;
+
+defm VSCATTERPF0DPD: avx512_gather_scatter_prefetch<0xC6, MRM5m, "vscatterpf0dpd",
+ VK8WM, vy32mem>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VT1>;
+
+defm VSCATTERPF0QPD: avx512_gather_scatter_prefetch<0xC7, MRM5m, "vscatterpf0qpd",
+ VK8WM, vz64mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
+
+defm VSCATTERPF1DPS: avx512_gather_scatter_prefetch<0xC6, MRM6m, "vscatterpf1dps",
+ VK16WM, vz32mem>, EVEX_V512, EVEX_CD8<32, CD8VT1>;
+
+defm VSCATTERPF1QPS: avx512_gather_scatter_prefetch<0xC7, MRM6m, "vscatterpf1qps",
+ VK8WM, vz64mem>, EVEX_V512, EVEX_CD8<64, CD8VT1>;
+
+defm VSCATTERPF1DPD: avx512_gather_scatter_prefetch<0xC6, MRM6m, "vscatterpf1dpd",
+ VK8WM, vy32mem>, EVEX_V512, VEX_W, EVEX_CD8<32, CD8VT1>;
+
+defm VSCATTERPF1QPD: avx512_gather_scatter_prefetch<0xC7, MRM6m, "vscatterpf1qpd",
+ VK8WM, vz64mem>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>;
//===----------------------------------------------------------------------===//
// VSHUFPS - VSHUFPD Operations
GR8:$mask),
(VPCONFLICTQrrk VR512:$src1,
(v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), VR512:$src2)>;
+
+let Predicates = [HasCDI] in {
+defm VPLZCNTD : avx512_conflict<0x44, "vplzcntd", VR512, VK16WM,
+ i512mem, i32mem, "{1to16}">,
+ EVEX_V512, EVEX_CD8<32, CD8VF>;
+
+
+defm VPLZCNTQ : avx512_conflict<0x44, "vplzcntq", VR512, VK8WM,
+ i512mem, i64mem, "{1to8}">,
+ EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
+
+}
+
+def : Pat<(int_x86_avx512_mask_lzcnt_d_512 VR512:$src2, VR512:$src1,
+ GR16:$mask),
+ (VPLZCNTDrrk VR512:$src1,
+ (v16i1 (COPY_TO_REGCLASS GR16:$mask, VK16WM)), VR512:$src2)>;
+
+def : Pat<(int_x86_avx512_mask_lzcnt_q_512 VR512:$src2, VR512:$src1,
+ GR8:$mask),
+ (VPLZCNTQrrk VR512:$src1,
+ (v8i1 (COPY_TO_REGCLASS GR8:$mask, VK8WM)), VR512:$src2)>;
+
+def : Pat<(store (i1 -1), addr:$dst), (MOV8mi addr:$dst, (i8 1))>;
+def : Pat<(store (i1 1), addr:$dst), (MOV8mi addr:$dst, (i8 1))>;
+def : Pat<(store (i1 0), addr:$dst), (MOV8mi addr:$dst, (i8 0))>;
+
+def : Pat<(store VK1:$src, addr:$dst),
+ (KMOVWmk addr:$dst, (COPY_TO_REGCLASS VK1:$src, VK16))>;
+
+def truncstorei1 : PatFrag<(ops node:$val, node:$ptr),
+ (truncstore node:$val, node:$ptr), [{
+ return cast<StoreSDNode>(N)->getMemoryVT() == MVT::i1;
+}]>;
+
+def : Pat<(truncstorei1 GR8:$src, addr:$dst),
+ (MOV8mr addr:$dst, GR8:$src)>;
+