def : Pat<(i32 (zext VK1:$src)),
(AND32ri (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16)), (i32 1))>;
+ def : Pat<(i32 (anyext VK1:$src)),
+ (KMOVWrk (COPY_TO_REGCLASS VK1:$src, VK16))>;
def : Pat<(i8 (zext VK1:$src)),
(EXTRACT_SUBREG
(AND32ri (KMOVWrk
def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))),
(v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>;
+def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 0))),
+ (v32i1 (COPY_TO_REGCLASS VK64:$src, VK32))>;
+
+def : Pat<(v32i1 (extract_subvector (v64i1 VK64:$src), (iPTR 32))),
+ (v32i1 (COPY_TO_REGCLASS (KSHIFTRQri VK64:$src, (i8 32)), VK32))>;
+
let Predicates = [HasVLX] in {
def : Pat<(v8i1 (insert_subvector undef, (v4i1 VK4:$src), (iPTR 0))),
(v8i1 (COPY_TO_REGCLASS VK4:$src, VK8))>;
(VPMOVQDrrkz VK8WM:$mask, VR512:$src)>;
-multiclass avx512_extend<bits<8> opc, string OpcodeStr, RegisterClass KRC,
- RegisterClass DstRC, RegisterClass SrcRC, SDNode OpNode,
- PatFrag mem_frag, X86MemOperand x86memop,
- ValueType OpVT, ValueType InVT> {
+multiclass avx512_extend_common<bits<8> opc, string OpcodeStr,
+ X86VectorVTInfo DestInfo, X86VectorVTInfo SrcInfo,
+ X86MemOperand x86memop, PatFrag LdFrag, SDNode OpNode>{
- def rr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
- (ins SrcRC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
- [(set DstRC:$dst, (OpVT (OpNode (InVT SrcRC:$src))))]>, EVEX;
+ defm rr : AVX512_maskable<opc, MRMSrcReg, DestInfo, (outs DestInfo.RC:$dst),
+ (ins SrcInfo.RC:$src), OpcodeStr ,"$src", "$src",
+ (DestInfo.VT (OpNode (SrcInfo.VT SrcInfo.RC:$src)))>,
+ EVEX;
- def rrk : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
- (ins KRC:$mask, SrcRC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst {${mask}} |$dst {${mask}}, $src}"),
- []>, EVEX, EVEX_K;
+ let mayLoad = 1 in {
+ defm rm : AVX512_maskable<opc, MRMSrcMem, DestInfo, (outs DestInfo.RC:$dst),
+ (ins x86memop:$src), OpcodeStr ,"$src", "$src",
+ (DestInfo.VT (LdFrag addr:$src))>,
+ EVEX;
+ }
+}
- def rrkz : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst),
- (ins KRC:$mask, SrcRC:$src),
- !strconcat(OpcodeStr, "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"),
- []>, EVEX, EVEX_KZ;
+multiclass avx512_extend_BW<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
+ let Predicates = [HasVLX, HasBWI] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v8i16x_info,
+ v16i8x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VH>, T8PD, EVEX_V128;
- let mayLoad = 1 in {
- def rm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$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 Z256: avx512_extend_common<opc, OpcodeStr, v16i16x_info,
+ v16i8x_info, i128mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VH>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasBWI] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v32i16_info,
+ v32i8x_info, i256mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VH>, T8PD, EVEX_V512;
+ }
+}
+
+multiclass avx512_extend_BD<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
+ let Predicates = [HasVLX, HasAVX512] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v4i32x_info,
+ v16i8x_info, i32mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V128;
+
+ defm Z256: avx512_extend_common<opc, OpcodeStr, v8i32x_info,
+ v16i8x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasAVX512] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v16i32_info,
+ v16i8x_info, i128mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VQ>, T8PD, EVEX_V512;
+ }
+}
+
+multiclass avx512_extend_BQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi8")> {
+ let Predicates = [HasVLX, HasAVX512] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v2i64x_info,
+ v16i8x_info, i16mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VO>, T8PD, EVEX_V128;
+
+ defm Z256: avx512_extend_common<opc, OpcodeStr, v4i64x_info,
+ v16i8x_info, i32mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VO>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasAVX512] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v8i64_info,
+ v16i8x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<8, CD8VO>, T8PD, EVEX_V512;
+ }
+}
+
+multiclass avx512_extend_WD<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi16")> {
+ let Predicates = [HasVLX, HasAVX512] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v4i32x_info,
+ v8i16x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VH>, T8PD, EVEX_V128;
+
+ defm Z256: avx512_extend_common<opc, OpcodeStr, v8i32x_info,
+ v8i16x_info, i128mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VH>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasAVX512] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v16i32_info,
+ v16i16x_info, i256mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VH>, T8PD, EVEX_V512;
}
}
-defm VPMOVZXBDZ: avx512_extend<0x31, "vpmovzxbd", VK16WM, VR512, VR128X, X86vzext,
- loadv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
- EVEX_CD8<8, CD8VQ>;
-defm VPMOVZXBQZ: avx512_extend<0x32, "vpmovzxbq", VK8WM, VR512, VR128X, X86vzext,
- loadv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
- EVEX_CD8<8, CD8VO>;
-defm VPMOVZXWDZ: avx512_extend<0x33, "vpmovzxwd", VK16WM, VR512, VR256X, X86vzext,
- loadv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
- EVEX_CD8<16, CD8VH>;
-defm VPMOVZXWQZ: avx512_extend<0x34, "vpmovzxwq", VK8WM, VR512, VR128X, X86vzext,
- loadv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
- EVEX_CD8<16, CD8VQ>;
-defm VPMOVZXDQZ: avx512_extend<0x35, "vpmovzxdq", VK8WM, VR512, VR256X, X86vzext,
- loadv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
- EVEX_CD8<32, CD8VH>;
-
-defm VPMOVSXBDZ: avx512_extend<0x21, "vpmovsxbd", VK16WM, VR512, VR128X, X86vsext,
- loadv2i64, i128mem, v16i32, v16i8>, EVEX_V512,
- EVEX_CD8<8, CD8VQ>;
-defm VPMOVSXBQZ: avx512_extend<0x22, "vpmovsxbq", VK8WM, VR512, VR128X, X86vsext,
- loadv2i64, i128mem, v8i64, v16i8>, EVEX_V512,
- EVEX_CD8<8, CD8VO>;
-defm VPMOVSXWDZ: avx512_extend<0x23, "vpmovsxwd", VK16WM, VR512, VR256X, X86vsext,
- loadv4i64, i256mem, v16i32, v16i16>, EVEX_V512,
- EVEX_CD8<16, CD8VH>;
-defm VPMOVSXWQZ: avx512_extend<0x24, "vpmovsxwq", VK8WM, VR512, VR128X, X86vsext,
- loadv2i64, i128mem, v8i64, v8i16>, EVEX_V512,
- EVEX_CD8<16, CD8VQ>;
-defm VPMOVSXDQZ: avx512_extend<0x25, "vpmovsxdq", VK8WM, VR512, VR256X, X86vsext,
- loadv4i64, i256mem, v8i64, v8i32>, EVEX_V512,
- EVEX_CD8<32, CD8VH>;
+multiclass avx512_extend_WQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi16")> {
+ let Predicates = [HasVLX, HasAVX512] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v2i64x_info,
+ v8i16x_info, i32mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V128;
+
+ defm Z256: avx512_extend_common<opc, OpcodeStr, v4i64x_info,
+ v8i16x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasAVX512] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v8i64_info,
+ v8i16x_info, i128mem, LdFrag, OpNode>,
+ EVEX_CD8<16, CD8VQ>, T8PD, EVEX_V512;
+ }
+}
+
+multiclass avx512_extend_DQ<bits<8> opc, string OpcodeStr, SDNode OpNode,
+ string ExtTy,PatFrag LdFrag = !cast<PatFrag>(ExtTy#"extloadvi32")> {
+
+ let Predicates = [HasVLX, HasAVX512] in {
+ defm Z128: avx512_extend_common<opc, OpcodeStr, v2i64x_info,
+ v4i32x_info, i64mem, LdFrag, OpNode>,
+ EVEX_CD8<32, CD8VH>, T8PD, EVEX_V128;
+
+ defm Z256: avx512_extend_common<opc, OpcodeStr, v4i64x_info,
+ v4i32x_info, i128mem, LdFrag, OpNode>,
+ EVEX_CD8<32, CD8VH>, T8PD, EVEX_V256;
+ }
+ let Predicates = [HasAVX512] in {
+ defm Z : avx512_extend_common<opc, OpcodeStr, v8i64_info,
+ v8i32x_info, i256mem, LdFrag, OpNode>,
+ EVEX_CD8<32, CD8VH>, T8PD, EVEX_V512;
+ }
+}
+
+defm VPMOVZXBW : avx512_extend_BW<0x30, "vpmovzxbw", X86vzext, "z">;
+defm VPMOVZXBD : avx512_extend_BD<0x31, "vpmovzxbd", X86vzext, "z">;
+defm VPMOVZXBQ : avx512_extend_BQ<0x32, "vpmovzxbq", X86vzext, "z">;
+defm VPMOVZXWD : avx512_extend_WD<0x33, "vpmovzxwd", X86vzext, "z">;
+defm VPMOVZXWQ : avx512_extend_WQ<0x34, "vpmovzxwq", X86vzext, "z">;
+defm VPMOVZXDQ : avx512_extend_DQ<0x35, "vpmovzxdq", X86vzext, "z">;
+
+
+defm VPMOVSXBW: avx512_extend_BW<0x20, "vpmovsxbw", X86vsext, "s">;
+defm VPMOVSXBD: avx512_extend_BD<0x21, "vpmovsxbd", X86vsext, "s">;
+defm VPMOVSXBQ: avx512_extend_BQ<0x22, "vpmovsxbq", X86vsext, "s">;
+defm VPMOVSXWD: avx512_extend_WD<0x23, "vpmovsxwd", X86vsext, "s">;
+defm VPMOVSXWQ: avx512_extend_WQ<0x24, "vpmovsxwq", X86vsext, "s">;
+defm VPMOVSXDQ: avx512_extend_DQ<0x25, "vpmovsxdq", X86vsext, "s">;
//===----------------------------------------------------------------------===//
// GATHER - SCATTER Operations
projects / oota-llvm.git / blobdiff