[]>, EVEX;
}
+let Predicates = [HasCDI] in {
defm VPBROADCASTMW2D : avx512_mask_broadcast<0x3A, "vpbroadcastmw2d", VR512,
VK16, v16i32, v16i1>, EVEX_V512;
defm VPBROADCASTMB2Q : avx512_mask_broadcast<0x2A, "vpbroadcastmb2q", VR512,
VK8, v8i64, v8i1>, EVEX_V512, VEX_W;
+}
//===----------------------------------------------------------------------===//
// AVX-512 - VPERM
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
//
(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:$src))]>,
+ EVEX, EVEX_V512, EVEX_CD8<64, CD8VF>;
+
//===----------------------------------------------------------------------===//
// AVX-512 - Integer arithmetic
//
(EXTRACT_SUBREG (v16f32 (VCVTUDQ2PSZrr
(v16i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_xmm)>;
+def : Pat<(v4f64 (uint_to_fp (v4i32 VR128X:$src1))),
+ (EXTRACT_SUBREG (v8f64 (VCVTUDQ2PDZrr
+ (v8i32 (SUBREG_TO_REG (i32 0), VR128X:$src1, sub_xmm)))), sub_ymm)>;
+
def : Pat<(v16f32 (int_x86_avx512_mask_cvtdq2ps_512 (v16i32 VR512:$src),
(bc_v16f32 (v16i32 immAllZerosV)), (i16 -1), imm:$rc)),
(VCVTDQ2PSZrrb VR512:$src, imm:$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)>;
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
(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<(v16i32 (ctlz (memopv16i32 addr:$src))),
+ (VPLZCNTDrm addr:$src)>;
+def : Pat<(v16i32 (ctlz (v16i32 VR512:$src))),
+ (VPLZCNTDrr VR512:$src)>;
+def : Pat<(v8i64 (ctlz (memopv8i64 addr:$src))),
+ (VPLZCNTQrm addr:$src)>;
+def : Pat<(v8i64 (ctlz (v8i64 VR512:$src))),
+ (VPLZCNTQrr VR512:$src)>;
+
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)>;
+
projects / oota-llvm.git / blobdiff