diff options
Diffstat (limited to 'contrib/llvm/lib/Target/X86/X86InstrAVX512.td')
| -rw-r--r-- | contrib/llvm/lib/Target/X86/X86InstrAVX512.td | 3526 |
1 files changed, 3526 insertions, 0 deletions
diff --git a/contrib/llvm/lib/Target/X86/X86InstrAVX512.td b/contrib/llvm/lib/Target/X86/X86InstrAVX512.td new file mode 100644 index 0000000..cb19fbd --- /dev/null +++ b/contrib/llvm/lib/Target/X86/X86InstrAVX512.td @@ -0,0 +1,3526 @@ +// Bitcasts between 512-bit vector types. Return the original type since +// no instruction is needed for the conversion +let Predicates = [HasAVX512] in { + def : Pat<(v8f64 (bitconvert (v16f32 VR512:$src))), (v8f64 VR512:$src)>; + def : Pat<(v8f64 (bitconvert (v16i32 VR512:$src))), (v8f64 VR512:$src)>; + def : Pat<(v8f64 (bitconvert (v8i64 VR512:$src))), (v8f64 VR512:$src)>; + def : Pat<(v16f32 (bitconvert (v16i32 VR512:$src))), (v16f32 VR512:$src)>; + def : Pat<(v16f32 (bitconvert (v8i64 VR512:$src))), (v16f32 VR512:$src)>; + def : Pat<(v16f32 (bitconvert (v8f64 VR512:$src))), (v16f32 VR512:$src)>; + def : Pat<(v8i64 (bitconvert (v16f32 VR512:$src))), (v8i64 VR512:$src)>; + def : Pat<(v8i64 (bitconvert (v16i32 VR512:$src))), (v8i64 VR512:$src)>; + def : Pat<(v8i64 (bitconvert (v8f64 VR512:$src))), (v8i64 VR512:$src)>; + def : Pat<(v16i32 (bitconvert (v16f32 VR512:$src))), (v16i32 VR512:$src)>; + def : Pat<(v16i32 (bitconvert (v8i64 VR512:$src))), (v16i32 VR512:$src)>; + def : Pat<(v16i32 (bitconvert (v8f64 VR512:$src))), (v16i32 VR512:$src)>; + def : Pat<(v8f64 (bitconvert (v8i64 VR512:$src))), (v8f64 VR512:$src)>; + + def : Pat<(v2i64 (bitconvert (v4i32 VR128X:$src))), (v2i64 VR128X:$src)>; + def : Pat<(v2i64 (bitconvert (v8i16 VR128X:$src))), (v2i64 VR128X:$src)>; + def : Pat<(v2i64 (bitconvert (v16i8 VR128X:$src))), (v2i64 VR128X:$src)>; + def : Pat<(v2i64 (bitconvert (v2f64 VR128X:$src))), (v2i64 VR128X:$src)>; + def : Pat<(v2i64 (bitconvert (v4f32 VR128X:$src))), (v2i64 VR128X:$src)>; + def : Pat<(v4i32 (bitconvert (v2i64 VR128X:$src))), (v4i32 VR128X:$src)>; + def : Pat<(v4i32 (bitconvert (v8i16 VR128X:$src))), (v4i32 VR128X:$src)>; + def : Pat<(v4i32 (bitconvert (v16i8 VR128X:$src))), (v4i32 VR128X:$src)>; + def : Pat<(v4i32 (bitconvert (v2f64 VR128X:$src))), (v4i32 VR128X:$src)>; + def : Pat<(v4i32 (bitconvert (v4f32 VR128X:$src))), (v4i32 VR128X:$src)>; + def : Pat<(v8i16 (bitconvert (v2i64 VR128X:$src))), (v8i16 VR128X:$src)>; + def : Pat<(v8i16 (bitconvert (v4i32 VR128X:$src))), (v8i16 VR128X:$src)>; + def : Pat<(v8i16 (bitconvert (v16i8 VR128X:$src))), (v8i16 VR128X:$src)>; + def : Pat<(v8i16 (bitconvert (v2f64 VR128X:$src))), (v8i16 VR128X:$src)>; + def : Pat<(v8i16 (bitconvert (v4f32 VR128X:$src))), (v8i16 VR128X:$src)>; + def : Pat<(v16i8 (bitconvert (v2i64 VR128X:$src))), (v16i8 VR128X:$src)>; + def : Pat<(v16i8 (bitconvert (v4i32 VR128X:$src))), (v16i8 VR128X:$src)>; + def : Pat<(v16i8 (bitconvert (v8i16 VR128X:$src))), (v16i8 VR128X:$src)>; + def : Pat<(v16i8 (bitconvert (v2f64 VR128X:$src))), (v16i8 VR128X:$src)>; + def : Pat<(v16i8 (bitconvert (v4f32 VR128X:$src))), (v16i8 VR128X:$src)>; + def : Pat<(v4f32 (bitconvert (v2i64 VR128X:$src))), (v4f32 VR128X:$src)>; + def : Pat<(v4f32 (bitconvert (v4i32 VR128X:$src))), (v4f32 VR128X:$src)>; + def : Pat<(v4f32 (bitconvert (v8i16 VR128X:$src))), (v4f32 VR128X:$src)>; + def : Pat<(v4f32 (bitconvert (v16i8 VR128X:$src))), (v4f32 VR128X:$src)>; + def : Pat<(v4f32 (bitconvert (v2f64 VR128X:$src))), (v4f32 VR128X:$src)>; + def : Pat<(v2f64 (bitconvert (v2i64 VR128X:$src))), (v2f64 VR128X:$src)>; + def : Pat<(v2f64 (bitconvert (v4i32 VR128X:$src))), (v2f64 VR128X:$src)>; + def : Pat<(v2f64 (bitconvert (v8i16 VR128X:$src))), (v2f64 VR128X:$src)>; + def : Pat<(v2f64 (bitconvert (v16i8 VR128X:$src))), (v2f64 VR128X:$src)>; + def : Pat<(v2f64 (bitconvert (v4f32 VR128X:$src))), (v2f64 VR128X:$src)>; + +// Bitcasts between 256-bit vector types. Return the original type since +// no instruction is needed for the conversion + def : Pat<(v4f64 (bitconvert (v8f32 VR256X:$src))), (v4f64 VR256X:$src)>; + def : Pat<(v4f64 (bitconvert (v8i32 VR256X:$src))), (v4f64 VR256X:$src)>; + def : Pat<(v4f64 (bitconvert (v4i64 VR256X:$src))), (v4f64 VR256X:$src)>; + def : Pat<(v4f64 (bitconvert (v16i16 VR256X:$src))), (v4f64 VR256X:$src)>; + def : Pat<(v4f64 (bitconvert (v32i8 VR256X:$src))), (v4f64 VR256X:$src)>; + def : Pat<(v8f32 (bitconvert (v8i32 VR256X:$src))), (v8f32 VR256X:$src)>; + def : Pat<(v8f32 (bitconvert (v4i64 VR256X:$src))), (v8f32 VR256X:$src)>; + def : Pat<(v8f32 (bitconvert (v4f64 VR256X:$src))), (v8f32 VR256X:$src)>; + def : Pat<(v8f32 (bitconvert (v32i8 VR256X:$src))), (v8f32 VR256X:$src)>; + def : Pat<(v8f32 (bitconvert (v16i16 VR256X:$src))), (v8f32 VR256X:$src)>; + def : Pat<(v4i64 (bitconvert (v8f32 VR256X:$src))), (v4i64 VR256X:$src)>; + def : Pat<(v4i64 (bitconvert (v8i32 VR256X:$src))), (v4i64 VR256X:$src)>; + def : Pat<(v4i64 (bitconvert (v4f64 VR256X:$src))), (v4i64 VR256X:$src)>; + def : Pat<(v4i64 (bitconvert (v32i8 VR256X:$src))), (v4i64 VR256X:$src)>; + def : Pat<(v4i64 (bitconvert (v16i16 VR256X:$src))), (v4i64 VR256X:$src)>; + def : Pat<(v32i8 (bitconvert (v4f64 VR256X:$src))), (v32i8 VR256X:$src)>; + def : Pat<(v32i8 (bitconvert (v4i64 VR256X:$src))), (v32i8 VR256X:$src)>; + def : Pat<(v32i8 (bitconvert (v8f32 VR256X:$src))), (v32i8 VR256X:$src)>; + def : Pat<(v32i8 (bitconvert (v8i32 VR256X:$src))), (v32i8 VR256X:$src)>; + def : Pat<(v32i8 (bitconvert (v16i16 VR256X:$src))), (v32i8 VR256X:$src)>; + def : Pat<(v8i32 (bitconvert (v32i8 VR256X:$src))), (v8i32 VR256X:$src)>; + def : Pat<(v8i32 (bitconvert (v16i16 VR256X:$src))), (v8i32 VR256X:$src)>; + def : Pat<(v8i32 (bitconvert (v8f32 VR256X:$src))), (v8i32 VR256X:$src)>; + def : Pat<(v8i32 (bitconvert (v4i64 VR256X:$src))), (v8i32 VR256X:$src)>; + def : Pat<(v8i32 (bitconvert (v4f64 VR256X:$src))), (v8i32 VR256X:$src)>; + def : Pat<(v16i16 (bitconvert (v8f32 VR256X:$src))), (v16i16 VR256X:$src)>; + def : Pat<(v16i16 (bitconvert (v8i32 VR256X:$src))), (v16i16 VR256X:$src)>; + def : Pat<(v16i16 (bitconvert (v4i64 VR256X:$src))), (v16i16 VR256X:$src)>; + def : Pat<(v16i16 (bitconvert (v4f64 VR256X:$src))), (v16i16 VR256X:$src)>; + def : Pat<(v16i16 (bitconvert (v32i8 VR256X:$src))), (v16i16 VR256X:$src)>; +} + +// +// AVX-512: VPXOR instruction writes zero to its upper part, it's safe build zeros. +// + +let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1, + isPseudo = 1, Predicates = [HasAVX512] in { +def AVX512_512_SET0 : I<0, Pseudo, (outs VR512:$dst), (ins), "", + [(set VR512:$dst, (v16f32 immAllZerosV))]>; +} + +def : Pat<(v8i64 immAllZerosV), (AVX512_512_SET0)>; +def : Pat<(v16i32 immAllZerosV), (AVX512_512_SET0)>; +def : Pat<(v8f64 immAllZerosV), (AVX512_512_SET0)>; +def : Pat<(v16f32 immAllZerosV), (AVX512_512_SET0)>; + +//===----------------------------------------------------------------------===// +// AVX-512 - VECTOR INSERT +// +// -- 32x8 form -- +let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in { +def VINSERTF32x4rr : AVX512AIi8<0x18, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src1, VR128X:$src2, i8imm:$src3), + "vinsertf32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512; +let mayLoad = 1 in +def VINSERTF32x4rm : AVX512AIi8<0x18, MRMSrcMem, (outs VR512:$dst), + (ins VR512:$src1, f128mem:$src2, i8imm:$src3), + "vinsertf32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VT4>; +} + +// -- 64x4 fp form -- +let neverHasSideEffects = 1, ExeDomain = SSEPackedDouble in { +def VINSERTF64x4rr : AVX512AIi8<0x1a, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src1, VR256X:$src2, i8imm:$src3), + "vinsertf64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, VEX_W; +let mayLoad = 1 in +def VINSERTF64x4rm : AVX512AIi8<0x1a, MRMSrcMem, (outs VR512:$dst), + (ins VR512:$src1, i256mem:$src2, i8imm:$src3), + "vinsertf64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>; +} +// -- 32x4 integer form -- +let neverHasSideEffects = 1 in { +def VINSERTI32x4rr : AVX512AIi8<0x38, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src1, VR128X:$src2, i8imm:$src3), + "vinserti32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512; +let mayLoad = 1 in +def VINSERTI32x4rm : AVX512AIi8<0x38, MRMSrcMem, (outs VR512:$dst), + (ins VR512:$src1, i128mem:$src2, i8imm:$src3), + "vinserti32x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VT4>; + +} + +let neverHasSideEffects = 1 in { +// -- 64x4 form -- +def VINSERTI64x4rr : AVX512AIi8<0x3a, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src1, VR256X:$src2, i8imm:$src3), + "vinserti64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, VEX_W; +let mayLoad = 1 in +def VINSERTI64x4rm : AVX512AIi8<0x3a, MRMSrcMem, (outs VR512:$dst), + (ins VR512:$src1, i256mem:$src2, i8imm:$src3), + "vinserti64x4\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + []>, EVEX_4V, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>; +} + +def : Pat<(vinsert128_insert:$ins (v16f32 VR512:$src1), (v4f32 VR128X:$src2), + (iPTR imm)), (VINSERTF32x4rr VR512:$src1, VR128X:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v8f64 VR512:$src1), (v2f64 VR128X:$src2), + (iPTR imm)), (VINSERTF32x4rr VR512:$src1, VR128X:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (v2i64 VR128X:$src2), + (iPTR imm)), (VINSERTI32x4rr VR512:$src1, VR128X:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1), (v4i32 VR128X:$src2), + (iPTR imm)), (VINSERTI32x4rr VR512:$src1, VR128X:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; + +def : Pat<(vinsert128_insert:$ins (v16f32 VR512:$src1), (loadv4f32 addr:$src2), + (iPTR imm)), (VINSERTF32x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1), + (bc_v4i32 (loadv2i64 addr:$src2)), + (iPTR imm)), (VINSERTI32x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v8f64 VR512:$src1), (loadv2f64 addr:$src2), + (iPTR imm)), (VINSERTF32x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (loadv2i64 addr:$src2), + (iPTR imm)), (VINSERTI32x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert128_imm VR512:$ins))>; + +def : Pat<(vinsert256_insert:$ins (v16f32 VR512:$src1), (v8f32 VR256X:$src2), + (iPTR imm)), (VINSERTF64x4rr VR512:$src1, VR256X:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert256_insert:$ins (v8f64 VR512:$src1), (v4f64 VR256X:$src2), + (iPTR imm)), (VINSERTF64x4rr VR512:$src1, VR256X:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v8i64 VR512:$src1), (v4i64 VR256X:$src2), + (iPTR imm)), (VINSERTI64x4rr VR512:$src1, VR256X:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert128_insert:$ins (v16i32 VR512:$src1), (v8i32 VR256X:$src2), + (iPTR imm)), (VINSERTI64x4rr VR512:$src1, VR256X:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; + +def : Pat<(vinsert256_insert:$ins (v16f32 VR512:$src1), (loadv8f32 addr:$src2), + (iPTR imm)), (VINSERTF64x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert256_insert:$ins (v8f64 VR512:$src1), (loadv4f64 addr:$src2), + (iPTR imm)), (VINSERTF64x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert256_insert:$ins (v8i64 VR512:$src1), (loadv4i64 addr:$src2), + (iPTR imm)), (VINSERTI64x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; +def : Pat<(vinsert256_insert:$ins (v16i32 VR512:$src1), + (bc_v8i32 (loadv4i64 addr:$src2)), + (iPTR imm)), (VINSERTI64x4rm VR512:$src1, addr:$src2, + (INSERT_get_vinsert256_imm VR512:$ins))>; + +// vinsertps - insert f32 to XMM +def VINSERTPSzrr : AVX512AIi8<0x21, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src1, VR128X:$src2, u32u8imm:$src3), + "vinsertps{z}\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + [(set VR128X:$dst, (X86insrtps VR128X:$src1, VR128X:$src2, imm:$src3))]>, + EVEX_4V; +def VINSERTPSzrm: AVX512AIi8<0x21, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, f32mem:$src2, u32u8imm:$src3), + "vinsertps{z}\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}", + [(set VR128X:$dst, (X86insrtps VR128X:$src1, + (v4f32 (scalar_to_vector (loadf32 addr:$src2))), + imm:$src3))]>, EVEX_4V, EVEX_CD8<32, CD8VT1>; + +//===----------------------------------------------------------------------===// +// AVX-512 VECTOR EXTRACT +//--- +let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in { +// -- 32x4 form -- +def VEXTRACTF32x4rr : AVX512AIi8<0x19, MRMDestReg, (outs VR128X:$dst), + (ins VR512:$src1, i8imm:$src2), + "vextractf32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512; +def VEXTRACTF32x4mr : AVX512AIi8<0x19, MRMDestMem, (outs), + (ins f128mem:$dst, VR512:$src1, i8imm:$src2), + "vextractf32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VT4>; + +// -- 64x4 form -- +def VEXTRACTF64x4rr : AVX512AIi8<0x1b, MRMDestReg, (outs VR256X:$dst), + (ins VR512:$src1, i8imm:$src2), + "vextractf64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, VEX_W; +let mayStore = 1 in +def VEXTRACTF64x4mr : AVX512AIi8<0x1b, MRMDestMem, (outs), + (ins f256mem:$dst, VR512:$src1, i8imm:$src2), + "vextractf64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>; +} + +let neverHasSideEffects = 1 in { +// -- 32x4 form -- +def VEXTRACTI32x4rr : AVX512AIi8<0x39, MRMDestReg, (outs VR128X:$dst), + (ins VR512:$src1, i8imm:$src2), + "vextracti32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512; +def VEXTRACTI32x4mr : AVX512AIi8<0x39, MRMDestMem, (outs), + (ins i128mem:$dst, VR512:$src1, i8imm:$src2), + "vextracti32x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VT4>; + +// -- 64x4 form -- +def VEXTRACTI64x4rr : AVX512AIi8<0x3b, MRMDestReg, (outs VR256X:$dst), + (ins VR512:$src1, i8imm:$src2), + "vextracti64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, VEX_W; +let mayStore = 1 in +def VEXTRACTI64x4mr : AVX512AIi8<0x3b, MRMDestMem, (outs), + (ins i256mem:$dst, VR512:$src1, i8imm:$src2), + "vextracti64x4\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT4>; +} + +def : Pat<(vextract128_extract:$ext (v16f32 VR512:$src1), (iPTR imm)), + (v4f32 (VEXTRACTF32x4rr VR512:$src1, + (EXTRACT_get_vextract128_imm VR128X:$ext)))>; + +def : Pat<(vextract128_extract:$ext VR512:$src1, (iPTR imm)), + (v4i32 (VEXTRACTF32x4rr VR512:$src1, + (EXTRACT_get_vextract128_imm VR128X:$ext)))>; + +def : Pat<(vextract128_extract:$ext (v8f64 VR512:$src1), (iPTR imm)), + (v2f64 (VEXTRACTF32x4rr VR512:$src1, + (EXTRACT_get_vextract128_imm VR128X:$ext)))>; + +def : Pat<(vextract128_extract:$ext (v8i64 VR512:$src1), (iPTR imm)), + (v2i64 (VEXTRACTI32x4rr VR512:$src1, + (EXTRACT_get_vextract128_imm VR128X:$ext)))>; + + +def : Pat<(vextract256_extract:$ext (v16f32 VR512:$src1), (iPTR imm)), + (v8f32 (VEXTRACTF64x4rr VR512:$src1, + (EXTRACT_get_vextract256_imm VR256X:$ext)))>; + +def : Pat<(vextract256_extract:$ext (v16i32 VR512:$src1), (iPTR imm)), + (v8i32 (VEXTRACTI64x4rr VR512:$src1, + (EXTRACT_get_vextract256_imm VR256X:$ext)))>; + +def : Pat<(vextract256_extract:$ext (v8f64 VR512:$src1), (iPTR imm)), + (v4f64 (VEXTRACTF64x4rr VR512:$src1, + (EXTRACT_get_vextract256_imm VR256X:$ext)))>; + +def : Pat<(vextract256_extract:$ext (v8i64 VR512:$src1), (iPTR imm)), + (v4i64 (VEXTRACTI64x4rr VR512:$src1, + (EXTRACT_get_vextract256_imm VR256X:$ext)))>; + +// A 256-bit subvector extract from the first 512-bit vector position +// is a subregister copy that needs no instruction. +def : Pat<(v8i32 (extract_subvector (v16i32 VR512:$src), (iPTR 0))), + (v8i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_ymm))>; +def : Pat<(v8f32 (extract_subvector (v16f32 VR512:$src), (iPTR 0))), + (v8f32 (EXTRACT_SUBREG (v16f32 VR512:$src), sub_ymm))>; +def : Pat<(v4i64 (extract_subvector (v8i64 VR512:$src), (iPTR 0))), + (v4i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_ymm))>; +def : Pat<(v4f64 (extract_subvector (v8f64 VR512:$src), (iPTR 0))), + (v4f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_ymm))>; + +// zmm -> xmm +def : Pat<(v4i32 (extract_subvector (v16i32 VR512:$src), (iPTR 0))), + (v4i32 (EXTRACT_SUBREG (v16i32 VR512:$src), sub_xmm))>; +def : Pat<(v2i64 (extract_subvector (v8i64 VR512:$src), (iPTR 0))), + (v2i64 (EXTRACT_SUBREG (v8i64 VR512:$src), sub_xmm))>; +def : Pat<(v2f64 (extract_subvector (v8f64 VR512:$src), (iPTR 0))), + (v2f64 (EXTRACT_SUBREG (v8f64 VR512:$src), sub_xmm))>; +def : Pat<(v4f32 (extract_subvector (v16f32 VR512:$src), (iPTR 0))), + (v4f32 (EXTRACT_SUBREG (v16f32 VR512:$src), sub_xmm))>; + + +// A 128-bit subvector insert to the first 512-bit vector position +// is a subregister copy that needs no instruction. +def : Pat<(insert_subvector undef, (v2i64 VR128X:$src), (iPTR 0)), + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), + (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm), + sub_ymm)>; +def : Pat<(insert_subvector undef, (v2f64 VR128X:$src), (iPTR 0)), + (INSERT_SUBREG (v8f64 (IMPLICIT_DEF)), + (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128X:$src, sub_xmm), + sub_ymm)>; +def : Pat<(insert_subvector undef, (v4i32 VR128X:$src), (iPTR 0)), + (INSERT_SUBREG (v16i32 (IMPLICIT_DEF)), + (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128X:$src, sub_xmm), + sub_ymm)>; +def : Pat<(insert_subvector undef, (v4f32 VR128X:$src), (iPTR 0)), + (INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), + (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128X:$src, sub_xmm), + sub_ymm)>; + +def : Pat<(insert_subvector undef, (v4i64 VR256X:$src), (iPTR 0)), + (INSERT_SUBREG (v8i64 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>; +def : Pat<(insert_subvector undef, (v4f64 VR256X:$src), (iPTR 0)), + (INSERT_SUBREG (v8f64 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>; +def : Pat<(insert_subvector undef, (v8i32 VR256X:$src), (iPTR 0)), + (INSERT_SUBREG (v16i32 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>; +def : Pat<(insert_subvector undef, (v8f32 VR256X:$src), (iPTR 0)), + (INSERT_SUBREG (v16f32 (IMPLICIT_DEF)), VR256X:$src, sub_ymm)>; + +// vextractps - extract 32 bits from XMM +def VEXTRACTPSzrr : AVX512AIi8<0x17, MRMDestReg, (outs GR32:$dst), + (ins VR128X:$src1, u32u8imm:$src2), + "vextractps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [(set GR32:$dst, (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2))]>, + EVEX; + +def VEXTRACTPSzmr : AVX512AIi8<0x17, MRMDestMem, (outs), + (ins f32mem:$dst, VR128X:$src1, u32u8imm:$src2), + "vextractps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [(store (extractelt (bc_v4i32 (v4f32 VR128X:$src1)), imm:$src2), + addr:$dst)]>, EVEX; + +//===---------------------------------------------------------------------===// +// AVX-512 BROADCAST +//--- +multiclass avx512_fp_broadcast<bits<8> opc, string OpcodeStr, + RegisterClass DestRC, + RegisterClass SrcRC, X86MemOperand x86memop> { + def rr : AVX5128I<opc, MRMSrcReg, (outs DestRC:$dst), (ins SrcRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + []>, EVEX; + def rm : AVX5128I<opc, MRMSrcMem, (outs DestRC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),[]>, EVEX; +} +let ExeDomain = SSEPackedSingle in { + defm VBROADCASTSSZ : avx512_fp_broadcast<0x18, "vbroadcastss{z}", VR512, + VR128X, f32mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; +} + +let ExeDomain = SSEPackedDouble in { + defm VBROADCASTSDZ : avx512_fp_broadcast<0x19, "vbroadcastsd{z}", VR512, + VR128X, f64mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +} + +def : Pat<(v16f32 (X86VBroadcast (loadf32 addr:$src))), + (VBROADCASTSSZrm addr:$src)>; +def : Pat<(v8f64 (X86VBroadcast (loadf64 addr:$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:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + []>, EVEX, EVEX_V512; + def Zkrr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), + (ins KRC:$mask, SrcRC:$src), + !strconcat(OpcodeStr, + "\t{$src, $dst {${mask}} {z}|$dst {${mask}} {z}, $src}"), + []>, EVEX, EVEX_V512, EVEX_KZ; +} + +defm VPBROADCASTDr : avx512_int_broadcast_reg<0x7C, "vpbroadcastd", GR32, VK16WM>; +defm VPBROADCASTQr : avx512_int_broadcast_reg<0x7C, "vpbroadcastq", GR64, VK8WM>, + VEX_W; + +def : Pat <(v16i32 (X86vzext VK16WM:$mask)), + (VPBROADCASTDrZkrr VK16WM:$mask, (i32 (MOV32ri 0x1)))>; + +def : Pat <(v8i64 (X86vzext VK8WM:$mask)), + (VPBROADCASTQrZkrr VK8WM:$mask, (i64 (MOV64ri 0x1)))>; + +def : Pat<(v16i32 (X86VBroadcast (i32 GR32:$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:$src)>; + +multiclass avx512_int_broadcast_rm<bits<8> opc, string OpcodeStr, + X86MemOperand x86memop, PatFrag ld_frag, + RegisterClass DstRC, ValueType OpVT, ValueType SrcVT, + RegisterClass KRC> { + def rr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst), (ins VR128X:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set DstRC:$dst, + (OpVT (X86VBroadcast (SrcVT VR128X:$src))))]>, EVEX; + def krr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$dst), (ins KRC:$mask, + VR128X:$src), + !strconcat(OpcodeStr, + "\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"), + [(set DstRC:$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, + (OpVT (X86VBroadcast (ld_frag addr:$src))))]>, EVEX; + def krm : AVX5128I<opc, MRMSrcMem, (outs DstRC:$dst), (ins KRC:$mask, + x86memop:$src), + !strconcat(OpcodeStr, + "\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, + loadi32, VR512, v16i32, v4i32, VK16WM>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; +defm VPBROADCASTQZ : avx512_int_broadcast_rm<0x59, "vpbroadcastq", i64mem, + 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. +def : Pat<(v16f32 (X86VBroadcast FR32X:$src)), + (VBROADCASTSSZrr (COPY_TO_REGCLASS FR32X:$src, VR128X))>; +def : Pat<(v8f64 (X86VBroadcast FR64X:$src)), + (VBROADCASTSDZrr (COPY_TO_REGCLASS FR64X:$src, VR128X))>; + + +let Predicates = [HasAVX512] in { +def : Pat<(v8i32 (X86VBroadcastm (v8i1 VK8WM:$mask), (loadi32 addr:$src))), + (EXTRACT_SUBREG + (v16i32 (VPBROADCASTDZkrm (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), + addr:$src)), sub_ymm)>; +} +//===----------------------------------------------------------------------===// +// AVX-512 BROADCAST MASK TO VECTOR REGISTER +//--- + +multiclass avx512_mask_broadcast<bits<8> opc, string OpcodeStr, + RegisterClass DstRC, RegisterClass KRC, + ValueType OpVT, ValueType SrcVT> { +def rr : AVX512XS8I<opc, MRMDestReg, (outs DstRC:$dst), (ins KRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + []>, EVEX; +} + +defm VPBROADCASTMW2D : avx512_mask_broadcast<0x3A, "vpbroadcastmw2d", VR512, + VK16, v16i32, v16i1>, EVEX_V512; +defm VPBROADCASTMB2Q : avx512_mask_broadcast<0x2A, "vpbroadcastmb2q", VR512, + VK8, v8i64, v8i1>, EVEX_V512, VEX_W; + +//===----------------------------------------------------------------------===// +// AVX-512 - VPERM +// +// -- immediate form -- +multiclass avx512_perm_imm<bits<8> opc, string OpcodeStr, RegisterClass RC, + SDNode OpNode, PatFrag mem_frag, + X86MemOperand x86memop, ValueType OpVT> { + def ri : AVX512AIi8<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (OpNode RC:$src1, (i8 imm:$src2))))]>, + EVEX; + def mi : AVX512AIi8<opc, MRMSrcMem, (outs RC:$dst), + (ins x86memop:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (OpNode (mem_frag addr:$src1), + (i8 imm:$src2))))]>, EVEX; +} + +defm VPERMQZ : avx512_perm_imm<0x00, "vpermq", VR512, X86VPermi, memopv8i64, + i512mem, v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +let ExeDomain = SSEPackedDouble in +defm VPERMPDZ : avx512_perm_imm<0x01, "vpermpd", VR512, X86VPermi, memopv8f64, + f512mem, v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +// -- VPERM - register form -- +multiclass avx512_perm<bits<8> opc, string OpcodeStr, RegisterClass RC, + PatFrag mem_frag, X86MemOperand x86memop, ValueType OpVT> { + + def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (X86VPermv RC:$src1, RC:$src2)))]>, EVEX_4V; + + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (X86VPermv RC:$src1, (mem_frag addr:$src2))))]>, + EVEX_4V; +} + +defm VPERMDZ : avx512_perm<0x36, "vpermd", VR512, memopv16i32, i512mem, + v16i32>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPERMQZ : avx512_perm<0x36, "vpermq", VR512, memopv8i64, i512mem, + v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +let ExeDomain = SSEPackedSingle in +defm VPERMPSZ : avx512_perm<0x16, "vpermps", VR512, memopv16f32, f512mem, + v16f32>, EVEX_V512, EVEX_CD8<32, CD8VF>; +let ExeDomain = SSEPackedDouble in +defm VPERMPDZ : avx512_perm<0x16, "vpermpd", VR512, memopv8f64, f512mem, + v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +// -- VPERM2I - 3 source operands form -- +multiclass avx512_perm_3src<bits<8> opc, string OpcodeStr, RegisterClass RC, + PatFrag mem_frag, X86MemOperand x86memop, + ValueType OpVT> { +let Constraints = "$src1 = $dst" in { + def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2, RC:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, + (OpVT (X86VPermv3 RC:$src1, RC:$src2, RC:$src3)))]>, + EVEX_4V; + + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src2, x86memop:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, + (OpVT (X86VPermv3 RC:$src1, RC:$src2, + (mem_frag addr:$src3))))]>, EVEX_4V; + } +} +defm VPERMI2D : avx512_perm_3src<0x76, "vpermi2d", VR512, memopv16i32, i512mem, + v16i32>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPERMI2Q : avx512_perm_3src<0x76, "vpermi2q", VR512, memopv8i64, i512mem, + v8i64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPERMI2PS : avx512_perm_3src<0x77, "vpermi2ps", VR512, memopv16f32, i512mem, + v16f32>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPERMI2PD : avx512_perm_3src<0x77, "vpermi2pd", VR512, memopv8f64, i512mem, + v8f64>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 - BLEND using mask +// +multiclass avx512_blendmask<bits<8> opc, string OpcodeStr, Intrinsic Int, + RegisterClass KRC, RegisterClass RC, + X86MemOperand x86memop, PatFrag mem_frag, + SDNode OpNode, ValueType vt> { + def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, RC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, ${dst} {${mask}}|${dst} {${mask}}, $src1, $src2}"), + [(set RC:$dst, (OpNode KRC:$mask, (vt RC:$src2), + (vt RC:$src1)))]>, EVEX_4V, EVEX_K; + def rr_Int : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, RC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, ${dst} {${mask}}|${dst} {${mask}}, $src1, $src2}"), + [(set RC:$dst, (Int KRC:$mask, (vt RC:$src2), + (vt RC:$src1)))]>, EVEX_4V, EVEX_K; + + let mayLoad = 1 in { + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $mask, $dst|$dst, $mask, $src1, $src2}"), + []>, + EVEX_4V, EVEX_K; + + def rm_Int : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $mask, $dst|$dst, $mask, $src1, $src2}"), + [(set RC:$dst, (Int KRC:$mask, (vt RC:$src1), + (mem_frag addr:$src2)))]>, + EVEX_4V, EVEX_K; + } +} + +let ExeDomain = SSEPackedSingle in +defm VBLENDMPSZ : avx512_blendmask<0x65, "vblendmps", + int_x86_avx512_mskblend_ps_512, + VK16WM, VR512, f512mem, + memopv16f32, vselect, v16f32>, + EVEX_CD8<32, CD8VF>, EVEX_V512; +let ExeDomain = SSEPackedDouble in +defm VBLENDMPDZ : avx512_blendmask<0x65, "vblendmpd", + int_x86_avx512_mskblend_pd_512, + VK8WM, VR512, f512mem, + memopv8f64, vselect, v8f64>, + VEX_W, EVEX_CD8<64, CD8VF>, EVEX_V512; + +defm VPBLENDMDZ : avx512_blendmask<0x64, "vpblendmd", + int_x86_avx512_mskblend_d_512, + VK16WM, VR512, f512mem, + memopv16i32, vselect, v16i32>, + EVEX_CD8<32, CD8VF>, EVEX_V512; + +defm VPBLENDMQZ : avx512_blendmask<0x64, "vpblendmq", + int_x86_avx512_mskblend_q_512, + VK8WM, VR512, f512mem, + memopv8i64, vselect, v8i64>, + VEX_W, EVEX_CD8<64, CD8VF>, EVEX_V512; + +let Predicates = [HasAVX512] in { +def : Pat<(v8f32 (vselect (v8i1 VK8WM:$mask), (v8f32 VR256X:$src1), + (v8f32 VR256X:$src2))), + (EXTRACT_SUBREG + (v16f32 (VBLENDMPSZrr (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), + (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)), + (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>; + +def : Pat<(v8i32 (vselect (v8i1 VK8WM:$mask), (v8i32 VR256X:$src1), + (v8i32 VR256X:$src2))), + (EXTRACT_SUBREG + (v16i32 (VPBLENDMDZrr (COPY_TO_REGCLASS VK8WM:$mask, VK16WM), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>; +} + +multiclass avx512_icmp_packed<bits<8> opc, string OpcodeStr, RegisterClass KRC, + RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag, + SDNode OpNode, ValueType vt> { + def rr : AVX512BI<opc, MRMSrcReg, + (outs KRC:$dst), (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2)))], + IIC_SSE_ALU_F32P_RR>, EVEX_4V; + def rm : AVX512BI<opc, MRMSrcMem, + (outs KRC:$dst), (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set KRC:$dst, (OpNode (vt RC:$src1), (memop_frag addr:$src2)))], + IIC_SSE_ALU_F32P_RM>, EVEX_4V; +} + +defm VPCMPEQDZ : avx512_icmp_packed<0x76, "vpcmpeqd", VK16, VR512, i512mem, + memopv16i32, X86pcmpeqm, v16i32>, EVEX_V512; +defm VPCMPEQQZ : avx512_icmp_packed<0x29, "vpcmpeqq", VK8, VR512, i512mem, + memopv8i64, X86pcmpeqm, v8i64>, T8, EVEX_V512, VEX_W; + +defm VPCMPGTDZ : avx512_icmp_packed<0x66, "vpcmpgtd", VK16, VR512, i512mem, + memopv16i32, X86pcmpgtm, v16i32>, EVEX_V512; +defm VPCMPGTQZ : avx512_icmp_packed<0x37, "vpcmpgtq", VK8, VR512, i512mem, + memopv8i64, X86pcmpgtm, v8i64>, T8, EVEX_V512, VEX_W; + +def : Pat<(v8i1 (X86pcmpgtm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))), + (COPY_TO_REGCLASS (VPCMPGTDZrr + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>; + +def : Pat<(v8i1 (X86pcmpeqm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2))), + (COPY_TO_REGCLASS (VPCMPEQDZrr + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm))), VK8)>; + +multiclass avx512_icmp_cc<bits<8> opc, RegisterClass KRC, + RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag, + SDNode OpNode, ValueType vt, Operand CC, string asm, + string asm_alt> { + def rri : AVX512AIi8<opc, MRMSrcReg, + (outs KRC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm, + [(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2), imm:$cc))], + IIC_SSE_ALU_F32P_RR>, EVEX_4V; + def rmi : AVX512AIi8<opc, MRMSrcMem, + (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm, + [(set KRC:$dst, (OpNode (vt RC:$src1), (memop_frag addr:$src2), + imm:$cc))], IIC_SSE_ALU_F32P_RM>, EVEX_4V; + // Accept explicit immediate argument form instead of comparison code. + let neverHasSideEffects = 1 in { + def rri_alt : AVX512AIi8<opc, MRMSrcReg, + (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), + asm_alt, [], IIC_SSE_ALU_F32P_RR>, EVEX_4V; + def rmi_alt : AVX512AIi8<opc, MRMSrcMem, + (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), + asm_alt, [], IIC_SSE_ALU_F32P_RM>, EVEX_4V; + } +} + +defm VPCMPDZ : avx512_icmp_cc<0x1F, VK16, VR512, i512mem, memopv16i32, + X86cmpm, v16i32, AVXCC, + "vpcmp${cc}d\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vpcmpd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPCMPUDZ : avx512_icmp_cc<0x1E, VK16, VR512, i512mem, memopv16i32, + X86cmpmu, v16i32, AVXCC, + "vpcmp${cc}ud\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vpcmpud\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">, + EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VPCMPQZ : avx512_icmp_cc<0x1F, VK8, VR512, i512mem, memopv8i64, + X86cmpm, v8i64, AVXCC, + "vpcmp${cc}q\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vpcmpq\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">, + VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; +defm VPCMPUQZ : avx512_icmp_cc<0x1E, VK8, VR512, i512mem, memopv8i64, + X86cmpmu, v8i64, AVXCC, + "vpcmp${cc}uq\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vpcmpuq\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">, + VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; + +// avx512_cmp_packed - sse 1 & 2 compare packed instructions +multiclass avx512_cmp_packed<RegisterClass KRC, RegisterClass RC, + X86MemOperand x86memop, Operand CC, + SDNode OpNode, ValueType vt, string asm, + string asm_alt, Domain d> { + def rri : AVX512PIi8<0xC2, MRMSrcReg, + (outs KRC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm, + [(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2), imm:$cc))], d>; + def rmi : AVX512PIi8<0xC2, MRMSrcMem, + (outs KRC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm, + [(set KRC:$dst, + (OpNode (vt RC:$src1), (memop addr:$src2), imm:$cc))], d>; + + // Accept explicit immediate argument form instead of comparison code. + let neverHasSideEffects = 1 in { + def rri_alt : AVX512PIi8<0xC2, MRMSrcReg, + (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc), + asm_alt, [], d>; + def rmi_alt : AVX512PIi8<0xC2, MRMSrcMem, + (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc), + asm_alt, [], d>; + } +} + +defm VCMPPSZ : avx512_cmp_packed<VK16, VR512, f512mem, AVXCC, X86cmpm, v16f32, + "vcmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vcmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}", + SSEPackedSingle>, EVEX_4V, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VCMPPDZ : avx512_cmp_packed<VK8, VR512, f512mem, AVXCC, X86cmpm, v8f64, + "vcmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}", + "vcmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}", + SSEPackedDouble>, OpSize, EVEX_4V, VEX_W, EVEX_V512, + EVEX_CD8<64, CD8VF>; + +def : Pat<(v8i1 (X86cmpm (v8f32 VR256X:$src1), (v8f32 VR256X:$src2), imm:$cc)), + (COPY_TO_REGCLASS (VCMPPSZrri + (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)), + (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)), + imm:$cc), VK8)>; +def : Pat<(v8i1 (X86cmpm (v8i32 VR256X:$src1), (v8i32 VR256X:$src2), imm:$cc)), + (COPY_TO_REGCLASS (VPCMPDZrri + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)), + imm:$cc), VK8)>; +def : Pat<(v8i1 (X86cmpmu (v8i32 VR256X:$src1), (v8i32 VR256X:$src2), imm:$cc)), + (COPY_TO_REGCLASS (VPCMPUDZrri + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)), + (v16i32 (SUBREG_TO_REG (i32 0), VR256X:$src2, sub_ymm)), + imm:$cc), VK8)>; + +// Mask register copy, including +// - copy between mask registers +// - load/store mask registers +// - copy from GPR to mask register and vice versa +// +multiclass avx512_mask_mov<bits<8> opc_kk, bits<8> opc_km, bits<8> opc_mk, + string OpcodeStr, RegisterClass KRC, + ValueType vt, X86MemOperand x86memop> { + let neverHasSideEffects = 1 in { + def kk : I<opc_kk, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>; + let mayLoad = 1 in + def km : I<opc_km, MRMSrcMem, (outs KRC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set KRC:$dst, (vt (load addr:$src)))]>; + let mayStore = 1 in + def mk : I<opc_mk, MRMDestMem, (outs), (ins x86memop:$dst, KRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>; + } +} + +multiclass avx512_mask_mov_gpr<bits<8> opc_kr, bits<8> opc_rk, + string OpcodeStr, + RegisterClass KRC, RegisterClass GRC> { + let neverHasSideEffects = 1 in { + def kr : I<opc_kr, MRMSrcReg, (outs KRC:$dst), (ins GRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>; + def rk : I<opc_rk, MRMSrcReg, (outs GRC:$dst), (ins KRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>; + } +} + +let Predicates = [HasAVX512] in { + defm KMOVW : avx512_mask_mov<0x90, 0x90, 0x91, "kmovw", VK16, v16i1, i16mem>, + VEX, TB; + defm KMOVW : avx512_mask_mov_gpr<0x92, 0x93, "kmovw", VK16, GR32>, + VEX, TB; +} + +let Predicates = [HasAVX512] in { + // GR16 from/to 16-bit mask + def : Pat<(v16i1 (bitconvert (i16 GR16:$src))), + (KMOVWkr (SUBREG_TO_REG (i32 0), GR16:$src, sub_16bit))>; + def : Pat<(i16 (bitconvert (v16i1 VK16:$src))), + (EXTRACT_SUBREG (KMOVWrk VK16:$src), sub_16bit)>; + + // Store kreg in memory + def : Pat<(store (v16i1 VK16:$src), addr:$dst), + (KMOVWmk addr:$dst, VK16:$src)>; + + def : Pat<(store (v8i1 VK8:$src), addr:$dst), + (KMOVWmk addr:$dst, (v16i1 (COPY_TO_REGCLASS VK8:$src, VK16)))>; +} +// With AVX-512 only, 8-bit mask is promoted to 16-bit mask. +let Predicates = [HasAVX512] in { + // GR from/to 8-bit mask without native support + def : Pat<(v8i1 (bitconvert (i8 GR8:$src))), + (COPY_TO_REGCLASS + (KMOVWkr (SUBREG_TO_REG (i32 0), GR8:$src, sub_8bit)), + VK8)>; + def : Pat<(i8 (bitconvert (v8i1 VK8:$src))), + (EXTRACT_SUBREG + (KMOVWrk (COPY_TO_REGCLASS VK8:$src, VK16)), + sub_8bit)>; +} + +// Mask unary operation +// - KNOT +multiclass avx512_mask_unop<bits<8> opc, string OpcodeStr, + RegisterClass KRC, SDPatternOperator OpNode> { + let Predicates = [HasAVX512] in + def rr : I<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set KRC:$dst, (OpNode KRC:$src))]>; +} + +multiclass avx512_mask_unop_w<bits<8> opc, string OpcodeStr, + SDPatternOperator OpNode> { + defm W : avx512_mask_unop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>, + VEX, TB; +} + +defm KNOT : avx512_mask_unop_w<0x44, "knot", not>; + +def : Pat<(xor VK16:$src1, (v16i1 immAllOnesV)), (KNOTWrr VK16:$src1)>; +def : Pat<(xor VK8:$src1, (v8i1 immAllOnesV)), + (COPY_TO_REGCLASS (KNOTWrr (COPY_TO_REGCLASS VK8:$src1, VK16)), VK8)>; + +// With AVX-512, 8-bit mask is promoted to 16-bit mask. +def : Pat<(not VK8:$src), + (COPY_TO_REGCLASS + (KNOTWrr (COPY_TO_REGCLASS VK8:$src, VK16)), VK8)>; + +// Mask binary operation +// - KADD, KAND, KANDN, KOR, KXNOR, KXOR +multiclass avx512_mask_binop<bits<8> opc, string OpcodeStr, + RegisterClass KRC, SDPatternOperator OpNode> { + let Predicates = [HasAVX512] in + def rr : I<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src1, KRC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set KRC:$dst, (OpNode KRC:$src1, KRC:$src2))]>; +} + +multiclass avx512_mask_binop_w<bits<8> opc, string OpcodeStr, + SDPatternOperator OpNode> { + defm W : avx512_mask_binop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>, + VEX_4V, VEX_L, TB; +} + +def andn : PatFrag<(ops node:$i0, node:$i1), (and (not node:$i0), node:$i1)>; +def xnor : PatFrag<(ops node:$i0, node:$i1), (not (xor node:$i0, node:$i1))>; + +let isCommutable = 1 in { + defm KADD : avx512_mask_binop_w<0x4a, "kadd", add>; + defm KAND : avx512_mask_binop_w<0x41, "kand", and>; + let isCommutable = 0 in + defm KANDN : avx512_mask_binop_w<0x42, "kandn", andn>; + defm KOR : avx512_mask_binop_w<0x45, "kor", or>; + defm KXNOR : avx512_mask_binop_w<0x46, "kxnor", xnor>; + defm KXOR : avx512_mask_binop_w<0x47, "kxor", xor>; +} + +multiclass avx512_mask_binop_int<string IntName, string InstName> { + let Predicates = [HasAVX512] in + def : Pat<(!cast<Intrinsic>("int_x86_"##IntName##"_v16i1") + VK16:$src1, VK16:$src2), + (!cast<Instruction>(InstName##"Wrr") VK16:$src1, VK16:$src2)>; +} + +defm : avx512_mask_binop_int<"kadd", "KADD">; +defm : avx512_mask_binop_int<"kand", "KAND">; +defm : avx512_mask_binop_int<"kandn", "KANDN">; +defm : avx512_mask_binop_int<"kor", "KOR">; +defm : avx512_mask_binop_int<"kxnor", "KXNOR">; +defm : avx512_mask_binop_int<"kxor", "KXOR">; +// With AVX-512, 8-bit mask is promoted to 16-bit mask. +multiclass avx512_binop_pat<SDPatternOperator OpNode, Instruction Inst> { + let Predicates = [HasAVX512] in + def : Pat<(OpNode VK8:$src1, VK8:$src2), + (COPY_TO_REGCLASS + (Inst (COPY_TO_REGCLASS VK8:$src1, VK16), + (COPY_TO_REGCLASS VK8:$src2, VK16)), VK8)>; +} + +defm : avx512_binop_pat<and, KANDWrr>; +defm : avx512_binop_pat<andn, KANDNWrr>; +defm : avx512_binop_pat<or, KORWrr>; +defm : avx512_binop_pat<xnor, KXNORWrr>; +defm : avx512_binop_pat<xor, KXORWrr>; + +// Mask unpacking +multiclass avx512_mask_unpck<bits<8> opc, string OpcodeStr, + RegisterClass KRC1, RegisterClass KRC2> { + let Predicates = [HasAVX512] in + def rr : I<opc, MRMSrcReg, (outs KRC1:$dst), (ins KRC2:$src1, KRC2:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>; +} + +multiclass avx512_mask_unpck_bw<bits<8> opc, string OpcodeStr> { + defm BW : avx512_mask_unpck<opc, !strconcat(OpcodeStr, "bw"), VK16, VK8>, + VEX_4V, VEX_L, OpSize, TB; +} + +defm KUNPCK : avx512_mask_unpck_bw<0x4b, "kunpck">; + +multiclass avx512_mask_unpck_int<string IntName, string InstName> { + let Predicates = [HasAVX512] in + def : Pat<(!cast<Intrinsic>("int_x86_"##IntName##"_v16i1") + VK8:$src1, VK8:$src2), + (!cast<Instruction>(InstName##"BWrr") VK8:$src1, VK8:$src2)>; +} + +defm : avx512_mask_unpck_int<"kunpck", "KUNPCK">; +// Mask bit testing +multiclass avx512_mask_testop<bits<8> opc, string OpcodeStr, RegisterClass KRC, + SDNode OpNode> { + let Predicates = [HasAVX512], Defs = [EFLAGS] in + def rr : I<opc, MRMSrcReg, (outs), (ins KRC:$src1, KRC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"), + [(set EFLAGS, (OpNode KRC:$src1, KRC:$src2))]>; +} + +multiclass avx512_mask_testop_w<bits<8> opc, string OpcodeStr, SDNode OpNode> { + defm W : avx512_mask_testop<opc, !strconcat(OpcodeStr, "w"), VK16, OpNode>, + VEX, TB; +} + +defm KORTEST : avx512_mask_testop_w<0x98, "kortest", X86kortest>; +defm KTEST : avx512_mask_testop_w<0x99, "ktest", X86ktest>; + +// Mask shift +multiclass avx512_mask_shiftop<bits<8> opc, string OpcodeStr, RegisterClass KRC, + SDNode OpNode> { + let Predicates = [HasAVX512] in + def ri : Ii8<opc, MRMSrcReg, (outs KRC:$dst), (ins KRC:$src, i8imm:$imm), + !strconcat(OpcodeStr, + "\t{$imm, $src, $dst|$dst, $src, $imm}"), + [(set KRC:$dst, (OpNode KRC:$src, (i8 imm:$imm)))]>; +} + +multiclass avx512_mask_shiftop_w<bits<8> opc1, bits<8> opc2, string OpcodeStr, + SDNode OpNode> { + defm W : avx512_mask_shiftop<opc1, !strconcat(OpcodeStr, "w"), VK16, OpNode>, + VEX, OpSize, TA, VEX_W; +} + +defm KSHIFTL : avx512_mask_shiftop_w<0x32, 0x33, "kshiftl", shl>; +defm KSHIFTR : avx512_mask_shiftop_w<0x30, 0x31, "kshiftr", srl>; + +// Mask setting all 0s or 1s +multiclass avx512_mask_setop<RegisterClass KRC, ValueType VT, PatFrag Val> { + let Predicates = [HasAVX512] in + let isReMaterializable = 1, isAsCheapAsAMove = 1, isPseudo = 1 in + def #NAME# : I<0, Pseudo, (outs KRC:$dst), (ins), "", + [(set KRC:$dst, (VT Val))]>; +} + +multiclass avx512_mask_setop_w<PatFrag Val> { + defm B : avx512_mask_setop<VK8, v8i1, Val>; + defm W : avx512_mask_setop<VK16, v16i1, Val>; +} + +defm KSET0 : avx512_mask_setop_w<immAllZerosV>; +defm KSET1 : avx512_mask_setop_w<immAllOnesV>; + +// With AVX-512 only, 8-bit mask is promoted to 16-bit mask. +let Predicates = [HasAVX512] in { + def : Pat<(v8i1 immAllZerosV), (COPY_TO_REGCLASS (KSET0W), VK8)>; + def : Pat<(v8i1 immAllOnesV), (COPY_TO_REGCLASS (KSET1W), VK8)>; +} +def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 0))), + (v8i1 (COPY_TO_REGCLASS VK16:$src, VK8))>; + +def : Pat<(v16i1 (insert_subvector undef, (v8i1 VK8:$src), (iPTR 0))), + (v16i1 (COPY_TO_REGCLASS VK8:$src, VK16))>; + +def : Pat<(v8i1 (extract_subvector (v16i1 VK16:$src), (iPTR 8))), + (v8i1 (COPY_TO_REGCLASS (KSHIFTRWri VK16:$src, (i8 8)), VK8))>; + +//===----------------------------------------------------------------------===// +// AVX-512 - Aligned and unaligned load and store +// + +multiclass avx512_mov_packed<bits<8> opc, RegisterClass RC, RegisterClass KRC, + X86MemOperand x86memop, PatFrag ld_frag, + string asm, Domain d> { +let neverHasSideEffects = 1 in + def rr : AVX512PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, + EVEX; +let canFoldAsLoad = 1 in + def rm : AVX512PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (ld_frag addr:$src))], d>, EVEX; +let Constraints = "$src1 = $dst" in { + def rrk : AVX512PI<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, RC:$src2), + !strconcat(asm, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), [], d>, + EVEX, EVEX_K; + def rmk : AVX512PI<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, x86memop:$src2), + !strconcat(asm, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + [], d>, EVEX, EVEX_K; +} +} + +defm VMOVAPSZ : avx512_mov_packed<0x28, VR512, VK16WM, f512mem, alignedloadv16f32, + "vmovaps", SSEPackedSingle>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VMOVAPDZ : avx512_mov_packed<0x28, VR512, VK8WM, f512mem, alignedloadv8f64, + "vmovapd", SSEPackedDouble>, + OpSize, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +defm VMOVUPSZ : avx512_mov_packed<0x10, VR512, VK16WM, f512mem, loadv16f32, + "vmovups", SSEPackedSingle>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VMOVUPDZ : avx512_mov_packed<0x10, VR512, VK8WM, f512mem, loadv8f64, + "vmovupd", SSEPackedDouble>, + OpSize, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +def VMOVAPSZmr : AVX512PI<0x29, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src), + "vmovaps\t{$src, $dst|$dst, $src}", + [(alignedstore512 (v16f32 VR512:$src), addr:$dst)], + SSEPackedSingle>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; +def VMOVAPDZmr : AVX512PI<0x29, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src), + "vmovapd\t{$src, $dst|$dst, $src}", + [(alignedstore512 (v8f64 VR512:$src), addr:$dst)], + SSEPackedDouble>, EVEX, EVEX_V512, + OpSize, VEX_W, EVEX_CD8<64, CD8VF>; +def VMOVUPSZmr : AVX512PI<0x11, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src), + "vmovups\t{$src, $dst|$dst, $src}", + [(store (v16f32 VR512:$src), addr:$dst)], + SSEPackedSingle>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; +def VMOVUPDZmr : AVX512PI<0x11, MRMDestMem, (outs), (ins f512mem:$dst, VR512:$src), + "vmovupd\t{$src, $dst|$dst, $src}", + [(store (v8f64 VR512:$src), addr:$dst)], + SSEPackedDouble>, EVEX, EVEX_V512, + OpSize, VEX_W, EVEX_CD8<64, CD8VF>; + +let neverHasSideEffects = 1 in { + def VMOVDQA32rr : AVX512BI<0x6F, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src), + "vmovdqa32\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512; + def VMOVDQA64rr : AVX512BI<0x6F, MRMSrcReg, (outs VR512:$dst), + (ins VR512:$src), + "vmovdqa64\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512, VEX_W; +let mayStore = 1 in { + def VMOVDQA32mr : AVX512BI<0x7F, MRMDestMem, (outs), + (ins i512mem:$dst, VR512:$src), + "vmovdqa32\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; + def VMOVDQA64mr : AVX512BI<0x7F, MRMDestMem, (outs), + (ins i512mem:$dst, VR512:$src), + "vmovdqa64\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} +let mayLoad = 1 in { +def VMOVDQA32rm : AVX512BI<0x6F, MRMSrcMem, (outs VR512:$dst), + (ins i512mem:$src), + "vmovdqa32\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; +def VMOVDQA64rm : AVX512BI<0x6F, MRMSrcMem, (outs VR512:$dst), + (ins i512mem:$src), + "vmovdqa64\t{$src, $dst|$dst, $src}", []>, + EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} +} + +// 512-bit aligned load/store +def : Pat<(alignedloadv8i64 addr:$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:$src)>; + +multiclass avx512_mov_int<bits<8> load_opc, bits<8> store_opc, string asm, + RegisterClass RC, RegisterClass KRC, + PatFrag ld_frag, X86MemOperand x86memop> { +let neverHasSideEffects = 1 in + def rr : AVX512XSI<load_opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), []>, EVEX; +let canFoldAsLoad = 1 in + def rm : AVX512XSI<load_opc, MRMSrcMem, (outs RC:$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 = $dst" in { + def rrk : AVX512XSI<load_opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, RC:$src2), + !strconcat(asm, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), []>, + EVEX, EVEX_K; + def rmk : AVX512XSI<load_opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, x86memop:$src2), + !strconcat(asm, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + []>, EVEX, EVEX_K; +} +} + +defm VMOVDQU32 : avx512_mov_int<0x6F, 0x7F, "vmovdqu32", VR512, VK16WM, + memopv16i32, i512mem>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VMOVDQU64 : avx512_mov_int<0x6F, 0x7F, "vmovdqu64", VR512, VK8WM, + memopv8i64, i512mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +// 512-bit unaligned load/store +def : Pat<(loadv8i64 addr:$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))), + (VMOVUPSZrrk VR512:$src2, VK16WM:$mask, VR512:$src1)>; +def : Pat<(v8f64 (vselect VK8WM:$mask, (v8f64 VR512:$src1), + (v8f64 VR512:$src2))), + (VMOVUPDZrrk VR512:$src2, VK8WM:$mask, VR512:$src1)>; +def : Pat<(v16i32 (vselect VK16WM:$mask, (v16i32 VR512:$src1), + (v16i32 VR512:$src2))), + (VMOVDQU32rrk VR512:$src2, VK16WM:$mask, VR512:$src1)>; +def : Pat<(v8i64 (vselect VK8WM:$mask, (v8i64 VR512:$src1), + (v8i64 VR512:$src2))), + (VMOVDQU64rrk VR512:$src2, VK8WM:$mask, VR512:$src1)>; +} +// Move Int Doubleword to Packed Double Int +// +def VMOVDI2PDIZrr : AVX512SI<0x6E, MRMSrcReg, (outs VR128X:$dst), (ins GR32:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$dst, + (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>, + EVEX, VEX_LIG; +def VMOVDI2PDIZrm : AVX512SI<0x6E, MRMSrcMem, (outs VR128X:$dst), (ins i32mem:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$dst, + (v4i32 (scalar_to_vector (loadi32 addr:$src))))], + IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>; +def VMOV64toPQIZrr : AVX512SI<0x6E, MRMSrcReg, (outs VR128X:$dst), (ins GR64:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$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:$src))], + IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteMove]>; +def VMOVSDto64Zrr : AVX512SI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src), + "vmovq{z}\t{$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:$dst)], + IIC_SSE_MOVDQ>, EVEX, VEX_W, Sched<[WriteStore]>, + EVEX_CD8<64, CD8VT1>; + +// Move Int Doubleword to Single Scalar +// +let isCodeGenOnly = 1 in { +def VMOVDI2SSZrr : AVX512SI<0x6E, MRMSrcReg, (outs FR32X:$dst), (ins GR32:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set FR32X:$dst, (bitconvert GR32:$src))], + IIC_SSE_MOVDQ>, EVEX, VEX_LIG; + +def VMOVDI2SSZrm : AVX512SI<0x6E, MRMSrcMem, (outs FR32X:$dst), (ins i32mem:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set FR32X:$dst, (bitconvert (loadi32 addr:$src)))], + IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>; +} + +// Move Packed Doubleword Int to Packed Double Int +// +def VMOVPDI2DIZrr : AVX512SI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128X:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set GR32:$dst, (vector_extract (v4i32 VR128X:$src), + (iPTR 0)))], IIC_SSE_MOVD_ToGP>, + EVEX, VEX_LIG; +def VMOVPDI2DIZmr : AVX512SI<0x7E, MRMDestMem, (outs), + (ins i32mem:$dst, VR128X:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(store (i32 (vector_extract (v4i32 VR128X:$src), + (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>, + EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>; + +// Move Packed Doubleword Int first element to Doubleword Int +// +def VMOVPQIto64Zrr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128X:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(set GR64:$dst, (extractelt (v2i64 VR128X:$src), + (iPTR 0)))], + IIC_SSE_MOVD_ToGP>, TB, OpSize, EVEX, VEX_LIG, VEX_W, + Requires<[HasAVX512, In64BitMode]>; + +def VMOVPQIto64Zmr : I<0xD6, MRMDestMem, (outs), + (ins i64mem:$dst, VR128X:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(store (extractelt (v2i64 VR128X:$src), (iPTR 0)), + addr:$dst)], IIC_SSE_MOVDQ>, + EVEX, OpSize, VEX_LIG, VEX_W, TB, EVEX_CD8<64, CD8VT1>, + Sched<[WriteStore]>, Requires<[HasAVX512, In64BitMode]>; + +// Move Scalar Single to Double Int +// +let isCodeGenOnly = 1 in { +def VMOVSS2DIZrr : AVX512SI<0x7E, MRMDestReg, (outs GR32:$dst), + (ins FR32X:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(set GR32:$dst, (bitconvert FR32X:$src))], + IIC_SSE_MOVD_ToGP>, EVEX, VEX_LIG; +def VMOVSS2DIZmr : AVX512SI<0x7E, MRMDestMem, (outs), + (ins i32mem:$dst, FR32X:$src), + "vmovd{z}\t{$src, $dst|$dst, $src}", + [(store (i32 (bitconvert FR32X:$src)), addr:$dst)], + IIC_SSE_MOVDQ>, EVEX, VEX_LIG, EVEX_CD8<32, CD8VT1>; +} + +// Move Quadword Int to Packed Quadword Int +// +def VMOVQI2PQIZrm : AVX512SI<0x6E, MRMSrcMem, (outs VR128X:$dst), + (ins i64mem:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$dst, + (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, + EVEX, VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>; + +//===----------------------------------------------------------------------===// +// AVX-512 MOVSS, MOVSD +//===----------------------------------------------------------------------===// + +multiclass avx512_move_scalar <string asm, RegisterClass RC, + SDNode OpNode, ValueType vt, + X86MemOperand x86memop, PatFrag mem_pat> { + def rr : SI<0x10, MRMSrcReg, (outs VR128X:$dst), (ins VR128X:$src1, RC:$src2), + !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128X:$dst, (vt (OpNode VR128X:$src1, + (scalar_to_vector RC:$src2))))], + IIC_SSE_MOV_S_RR>, EVEX_4V, VEX_LIG; + def rm : SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (mem_pat addr:$src))], IIC_SSE_MOV_S_RM>, + EVEX, VEX_LIG; + def mr: SI<0x11, MRMDestMem, (outs), (ins x86memop:$dst, RC:$src), + !strconcat(asm, "\t{$src, $dst|$dst, $src}"), + [(store RC:$src, addr:$dst)], IIC_SSE_MOV_S_MR>, + EVEX, VEX_LIG; +} + +let ExeDomain = SSEPackedSingle in +defm VMOVSSZ : avx512_move_scalar<"movss{z}", FR32X, X86Movss, v4f32, f32mem, + loadf32>, XS, EVEX_CD8<32, CD8VT1>; + +let ExeDomain = SSEPackedDouble in +defm VMOVSDZ : avx512_move_scalar<"movsd{z}", FR64X, X86Movsd, v2f64, f64mem, + loadf64>, XD, VEX_W, EVEX_CD8<64, CD8VT1>; + + +// For the disassembler +let isCodeGenOnly = 1 in { + def VMOVSSZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst), + (ins VR128X:$src1, FR32X:$src2), + "movss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], + IIC_SSE_MOV_S_RR>, + XS, EVEX_4V, VEX_LIG; + def VMOVSDZrr_REV : SI<0x11, MRMDestReg, (outs VR128X:$dst), + (ins VR128X:$src1, FR64X:$src2), + "movsd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", [], + IIC_SSE_MOV_S_RR>, + XD, EVEX_4V, VEX_LIG, VEX_W; +} + +let Predicates = [HasAVX512] in { + let AddedComplexity = 15 in { + // Move scalar to XMM zero-extended, zeroing a VR128X then do a + // MOVS{S,D} to the lower bits. + def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32X:$src)))), + (VMOVSSZrr (v4f32 (V_SET0)), FR32X:$src)>; + def : Pat<(v4f32 (X86vzmovl (v4f32 VR128X:$src))), + (VMOVSSZrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128X:$src, FR32X))>; + def : Pat<(v4i32 (X86vzmovl (v4i32 VR128X:$src))), + (VMOVSSZrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128X:$src, FR32X))>; + def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64X:$src)))), + (VMOVSDZrr (v2f64 (V_SET0)), FR64X:$src)>; + + // Move low f32 and clear high bits. + def : Pat<(v8f32 (X86vzmovl (v8f32 VR256X:$src))), + (SUBREG_TO_REG (i32 0), + (VMOVSSZrr (v4f32 (V_SET0)), + (EXTRACT_SUBREG (v8f32 VR256X:$src), sub_xmm)), sub_xmm)>; + def : Pat<(v8i32 (X86vzmovl (v8i32 VR256X:$src))), + (SUBREG_TO_REG (i32 0), + (VMOVSSZrr (v4i32 (V_SET0)), + (EXTRACT_SUBREG (v8i32 VR256X:$src), sub_xmm)), sub_xmm)>; + } + + let AddedComplexity = 20 in { + // MOVSSrm zeros the high parts of the register; represent this + // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0 + def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))), + (COPY_TO_REGCLASS (VMOVSSZrm addr:$src), VR128X)>; + def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))), + (COPY_TO_REGCLASS (VMOVSSZrm addr:$src), VR128X)>; + def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))), + (COPY_TO_REGCLASS (VMOVSSZrm addr:$src), VR128X)>; + + // MOVSDrm zeros the high parts of the register; represent this + // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0 + def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))), + (COPY_TO_REGCLASS (VMOVSDZrm addr:$src), VR128X)>; + def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))), + (COPY_TO_REGCLASS (VMOVSDZrm addr:$src), VR128X)>; + def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))), + (COPY_TO_REGCLASS (VMOVSDZrm addr:$src), VR128X)>; + def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))), + (COPY_TO_REGCLASS (VMOVSDZrm addr:$src), VR128X)>; + def : Pat<(v2f64 (X86vzload addr:$src)), + (COPY_TO_REGCLASS (VMOVSDZrm addr:$src), VR128X)>; + + // Represent the same patterns above but in the form they appear for + // 256-bit types + def : Pat<(v8i32 (X86vzmovl (insert_subvector undef, + (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))), + (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:$src), sub_xmm)>; + def : Pat<(v4f64 (X86vzmovl (insert_subvector undef, + (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))), + (SUBREG_TO_REG (i32 0), (VMOVSDZrm addr:$src), sub_xmm)>; + } + def : Pat<(v8f32 (X86vzmovl (insert_subvector undef, + (v4f32 (scalar_to_vector FR32X:$src)), (iPTR 0)))), + (SUBREG_TO_REG (i32 0), (v4f32 (VMOVSSZrr (v4f32 (V_SET0)), + FR32X:$src)), sub_xmm)>; + def : Pat<(v4f64 (X86vzmovl (insert_subvector undef, + (v2f64 (scalar_to_vector FR64X:$src)), (iPTR 0)))), + (SUBREG_TO_REG (i64 0), (v2f64 (VMOVSDZrr (v2f64 (V_SET0)), + 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), (VMOVQI2PQIZrm addr:$src), sub_xmm)>; + + // Move low f64 and clear high bits. + def : Pat<(v4f64 (X86vzmovl (v4f64 VR256X:$src))), + (SUBREG_TO_REG (i32 0), + (VMOVSDZrr (v2f64 (V_SET0)), + (EXTRACT_SUBREG (v4f64 VR256X:$src), sub_xmm)), sub_xmm)>; + + def : Pat<(v4i64 (X86vzmovl (v4i64 VR256X:$src))), + (SUBREG_TO_REG (i32 0), (VMOVSDZrr (v2i64 (V_SET0)), + (EXTRACT_SUBREG (v4i64 VR256X:$src), sub_xmm)), sub_xmm)>; + + // Extract and store. + def : Pat<(store (f32 (vector_extract (v4f32 VR128X:$src), (iPTR 0))), + addr:$dst), + (VMOVSSZmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128X:$src), FR32X))>; + def : Pat<(store (f64 (vector_extract (v2f64 VR128X:$src), (iPTR 0))), + addr:$dst), + (VMOVSDZmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128X:$src), FR64X))>; + + // Shuffle with VMOVSS + def : Pat<(v4i32 (X86Movss VR128X:$src1, VR128X:$src2)), + (VMOVSSZrr (v4i32 VR128X:$src1), + (COPY_TO_REGCLASS (v4i32 VR128X:$src2), FR32X))>; + def : Pat<(v4f32 (X86Movss VR128X:$src1, VR128X:$src2)), + (VMOVSSZrr (v4f32 VR128X:$src1), + (COPY_TO_REGCLASS (v4f32 VR128X:$src2), FR32X))>; + + // 256-bit variants + def : Pat<(v8i32 (X86Movss VR256X:$src1, VR256X:$src2)), + (SUBREG_TO_REG (i32 0), + (VMOVSSZrr (EXTRACT_SUBREG (v8i32 VR256X:$src1), sub_xmm), + (EXTRACT_SUBREG (v8i32 VR256X:$src2), sub_xmm)), + sub_xmm)>; + def : Pat<(v8f32 (X86Movss VR256X:$src1, VR256X:$src2)), + (SUBREG_TO_REG (i32 0), + (VMOVSSZrr (EXTRACT_SUBREG (v8f32 VR256X:$src1), sub_xmm), + (EXTRACT_SUBREG (v8f32 VR256X:$src2), sub_xmm)), + sub_xmm)>; + + // Shuffle with VMOVSD + def : Pat<(v2i64 (X86Movsd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v2f64 (X86Movsd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v4f32 (X86Movsd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v4i32 (X86Movsd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + + // 256-bit variants + def : Pat<(v4i64 (X86Movsd VR256X:$src1, VR256X:$src2)), + (SUBREG_TO_REG (i32 0), + (VMOVSDZrr (EXTRACT_SUBREG (v4i64 VR256X:$src1), sub_xmm), + (EXTRACT_SUBREG (v4i64 VR256X:$src2), sub_xmm)), + sub_xmm)>; + def : Pat<(v4f64 (X86Movsd VR256X:$src1, VR256X:$src2)), + (SUBREG_TO_REG (i32 0), + (VMOVSDZrr (EXTRACT_SUBREG (v4f64 VR256X:$src1), sub_xmm), + (EXTRACT_SUBREG (v4f64 VR256X:$src2), sub_xmm)), + sub_xmm)>; + + def : Pat<(v2f64 (X86Movlpd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v2i64 (X86Movlpd VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v4f32 (X86Movlps VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; + def : Pat<(v4i32 (X86Movlps VR128X:$src1, VR128X:$src2)), + (VMOVSDZrr VR128X:$src1, (COPY_TO_REGCLASS VR128X:$src2, FR64X))>; +} + +let AddedComplexity = 15 in +def VMOVZPQILo2PQIZrr : AVX512XSI<0x7E, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$dst, (v2i64 (X86vzmovl + (v2i64 VR128X:$src))))], + IIC_SSE_MOVQ_RR>, EVEX, VEX_W; + +let AddedComplexity = 20 in +def VMOVZPQILo2PQIZrm : AVX512XSI<0x7E, MRMSrcMem, (outs VR128X:$dst), + (ins i128mem:$src), + "vmovq{z}\t{$src, $dst|$dst, $src}", + [(set VR128X:$dst, (v2i64 (X86vzmovl + (loadv2i64 addr:$src))))], + IIC_SSE_MOVDQ>, EVEX, VEX_W, + EVEX_CD8<8, CD8VT8>; + +let Predicates = [HasAVX512] in { + // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part. + let AddedComplexity = 20 in { + def : Pat<(v4i32 (X86vzmovl (v4i32 (scalar_to_vector (loadi32 addr:$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:$src)>; + def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))), + (VMOVDI2PDIZrm addr:$src)>; + def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))), + (VMOVZPQILo2PQIZrm 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)))), + (SUBREG_TO_REG (i32 0), (VMOVDI2PDIZrr GR32:$src), sub_xmm)>; + def : Pat<(v4i64 (X86vzmovl (insert_subvector undef, + (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))), + (SUBREG_TO_REG (i64 0), (VMOV64toPQIZrr GR64:$src), sub_xmm)>; +} + +def : Pat<(v16i32 (X86Vinsert (v16i32 immAllZerosV), GR32:$src2, (iPTR 0))), + (SUBREG_TO_REG (i32 0), (VMOVDI2PDIZrr GR32:$src2), sub_xmm)>; + +def : Pat<(v8i64 (X86Vinsert (bc_v8i64 (v16i32 immAllZerosV)), GR64:$src2, (iPTR 0))), + (SUBREG_TO_REG (i32 0), (VMOV64toPQIZrr GR64:$src2), sub_xmm)>; + +def : Pat<(v16i32 (X86Vinsert undef, GR32:$src2, (iPTR 0))), + (SUBREG_TO_REG (i32 0), (VMOVDI2PDIZrr GR32:$src2), sub_xmm)>; + +def : Pat<(v8i64 (X86Vinsert undef, GR64:$src2, (iPTR 0))), + (SUBREG_TO_REG (i32 0), (VMOV64toPQIZrr GR64:$src2), sub_xmm)>; + +//===----------------------------------------------------------------------===// +// AVX-512 - Integer arithmetic +// +multiclass avx512_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, + ValueType OpVT, RegisterClass RC, PatFrag memop_frag, + X86MemOperand x86memop, PatFrag scalar_mfrag, + X86MemOperand x86scalar_mop, string BrdcstStr, + OpndItins itins, bit IsCommutable = 0> { + let isCommutable = IsCommutable in + def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), (OpVT RC:$src2))))], + itins.rr>, EVEX_4V; + def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), (memop_frag addr:$src2))))], + itins.rm>, EVEX_4V; + def rmb : AVX512BI<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:$src1, + (OpVT (X86VBroadcast (scalar_mfrag addr:$src2)))))], + itins.rm>, EVEX_4V, EVEX_B; +} +multiclass avx512_binop_rm2<bits<8> opc, string OpcodeStr, + ValueType DstVT, ValueType SrcVT, RegisterClass RC, + PatFrag memop_frag, X86MemOperand x86memop, + OpndItins itins, + bit IsCommutable = 0> { + let isCommutable = IsCommutable in + def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V, VEX_W; + def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V, VEX_W; +} + +defm VPADDDZ : avx512_binop_rm<0xFE, "vpaddd", add, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VPSUBDZ : avx512_binop_rm<0xFA, "vpsubd", sub, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 0>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VPMULLDZ : avx512_binop_rm<0x40, "vpmulld", mul, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VPADDQZ : avx512_binop_rm<0xD4, "vpaddq", add, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 1>, + EVEX_CD8<64, CD8VF>, EVEX_V512, VEX_W; + +defm VPSUBQZ : avx512_binop_rm<0xFB, "vpsubq", sub, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", v8i64, v16i32, + VR512, memopv8i64, i512mem, SSE_INTALU_ITINS_P, 1>, T8, + EVEX_V512, EVEX_CD8<64, CD8VF>; + +defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", v8i64, v16i32, + VR512, memopv8i64, i512mem, SSE_INTMUL_ITINS_P, 1>, EVEX_V512, + EVEX_CD8<64, CD8VF>; + +def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))), + (VPMULUDQZrr VR512:$src1, VR512:$src2)>; + +defm VPMAXUDZ : avx512_binop_rm<0x3F, "vpmaxud", X86umax, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMAXUQZ : avx512_binop_rm<0x3F, "vpmaxuq", X86umax, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMAXSDZ : avx512_binop_rm<0x3D, "vpmaxsd", X86smax, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMAXSQZ : avx512_binop_rm<0x3D, "vpmaxsq", X86smax, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMINUDZ : avx512_binop_rm<0x3B, "vpminud", X86umin, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMINUQZ : avx512_binop_rm<0x3B, "vpminuq", X86umin, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMINSDZ : avx512_binop_rm<0x39, "vpminsd", X86smin, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMINSQZ : avx512_binop_rm<0x39, "vpminsq", X86smin, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 - Unpack Instructions +//===----------------------------------------------------------------------===// + +multiclass avx512_unpack_fp<bits<8> opc, SDNode OpNode, ValueType vt, + PatFrag mem_frag, RegisterClass RC, + X86MemOperand x86memop, string asm, + Domain d> { + def rr : AVX512PI<opc, MRMSrcReg, + (outs RC:$dst), (ins RC:$src1, RC:$src2), + asm, [(set RC:$dst, + (vt (OpNode RC:$src1, RC:$src2)))], + d>, EVEX_4V; + def rm : AVX512PI<opc, MRMSrcMem, + (outs RC:$dst), (ins RC:$src1, x86memop:$src2), + asm, [(set RC:$dst, + (vt (OpNode RC:$src1, + (bitconvert (mem_frag addr:$src2)))))], + d>, EVEX_4V; +} + +defm VUNPCKHPSZ: avx512_unpack_fp<0x15, X86Unpckh, v16f32, memopv8f64, + VR512, f512mem, "vunpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}", + SSEPackedSingle>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VUNPCKHPDZ: avx512_unpack_fp<0x15, X86Unpckh, v8f64, memopv8f64, + VR512, f512mem, "vunpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}", + SSEPackedDouble>, OpSize, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +defm VUNPCKLPSZ: avx512_unpack_fp<0x14, X86Unpckl, v16f32, memopv8f64, + VR512, f512mem, "vunpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}", + SSEPackedSingle>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VUNPCKLPDZ: avx512_unpack_fp<0x14, X86Unpckl, v8f64, memopv8f64, + VR512, f512mem, "vunpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}", + SSEPackedDouble>, OpSize, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +multiclass avx512_unpack_int<bits<8> opc, string OpcodeStr, SDNode OpNode, + ValueType OpVT, RegisterClass RC, PatFrag memop_frag, + X86MemOperand x86memop> { + def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), (OpVT RC:$src2))))], + IIC_SSE_UNPCK>, EVEX_4V; + def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (OpVT (OpNode (OpVT RC:$src1), + (bitconvert (memop_frag addr:$src2)))))], + IIC_SSE_UNPCK>, EVEX_4V; +} +defm VPUNPCKLDQZ : avx512_unpack_int<0x62, "vpunpckldq", X86Unpckl, v16i32, + VR512, memopv16i32, i512mem>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPUNPCKLQDQZ : avx512_unpack_int<0x6C, "vpunpcklqdq", X86Unpckl, v8i64, + VR512, memopv8i64, i512mem>, EVEX_V512, + VEX_W, EVEX_CD8<64, CD8VF>; +defm VPUNPCKHDQZ : avx512_unpack_int<0x6A, "vpunpckhdq", X86Unpckh, v16i32, + VR512, memopv16i32, i512mem>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPUNPCKHQDQZ : avx512_unpack_int<0x6D, "vpunpckhqdq", X86Unpckh, v8i64, + VR512, memopv8i64, i512mem>, EVEX_V512, + VEX_W, EVEX_CD8<64, CD8VF>; +//===----------------------------------------------------------------------===// +// AVX-512 - PSHUFD +// + +multiclass avx512_pshuf_imm<bits<8> opc, string OpcodeStr, RegisterClass RC, + SDNode OpNode, PatFrag mem_frag, + X86MemOperand x86memop, ValueType OpVT> { + def ri : AVX512Ii8<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (OpNode RC:$src1, (i8 imm:$src2))))]>, + EVEX; + def mi : AVX512Ii8<opc, MRMSrcMem, (outs RC:$dst), + (ins x86memop:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (OpVT (OpNode (mem_frag addr:$src1), + (i8 imm:$src2))))]>, EVEX; +} + +defm VPSHUFDZ : avx512_pshuf_imm<0x70, "vpshufd", VR512, X86PShufd, memopv16i32, + i512mem, v16i32>, OpSize, EVEX_V512, EVEX_CD8<32, CD8VF>; + +let ExeDomain = SSEPackedSingle in +defm VPERMILPSZ : avx512_pshuf_imm<0x04, "vpermilps", VR512, X86VPermilp, + memopv16f32, i512mem, v16f32>, OpSize, TA, EVEX_V512, + EVEX_CD8<32, CD8VF>; +let ExeDomain = SSEPackedDouble in +defm VPERMILPDZ : avx512_pshuf_imm<0x05, "vpermilpd", VR512, X86VPermilp, + memopv8f64, i512mem, v8f64>, OpSize, TA, EVEX_V512, + VEX_W, EVEX_CD8<32, CD8VF>; + +def : Pat<(v16i32 (X86VPermilp VR512:$src1, (i8 imm:$imm))), + (VPERMILPSZri VR512:$src1, imm:$imm)>; +def : Pat<(v8i64 (X86VPermilp VR512:$src1, (i8 imm:$imm))), + (VPERMILPDZri VR512:$src1, imm:$imm)>; + +//===----------------------------------------------------------------------===// +// AVX-512 Logical Instructions +//===----------------------------------------------------------------------===// + +defm VPANDDZ : avx512_binop_rm<0xDB, "vpandd", and, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPANDQZ : avx512_binop_rm<0xDB, "vpandq", and, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPORDZ : avx512_binop_rm<0xEB, "vpord", or, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPORQZ : avx512_binop_rm<0xEB, "vporq", or, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPXORDZ : avx512_binop_rm<0xEF, "vpxord", xor, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPXORQZ : avx512_binop_rm<0xEF, "vpxorq", xor, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_BIT_ITINS_P, 1>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +defm VPANDNDZ : avx512_binop_rm<0xDF, "vpandnd", X86andnp, v16i32, VR512, + memopv16i32, i512mem, loadi32, i32mem, "{1to16}", + SSE_BIT_ITINS_P, 0>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPANDNQZ : avx512_binop_rm<0xDF, "vpandnq", X86andnp, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_BIT_ITINS_P, 0>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 FP arithmetic +//===----------------------------------------------------------------------===// + +multiclass avx512_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode, + SizeItins itins> { + defm SSZ : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss{z}"), OpNode, FR32X, + f32mem, itins.s, 0>, XS, EVEX_4V, VEX_LIG, + EVEX_CD8<32, CD8VT1>; + defm SDZ : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd{z}"), OpNode, FR64X, + f64mem, itins.d, 0>, XD, VEX_W, EVEX_4V, VEX_LIG, + EVEX_CD8<64, CD8VT1>; +} + +let isCommutable = 1 in { +defm VADD : avx512_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>; +defm VMUL : avx512_binop_s<0x59, "mul", fmul, SSE_ALU_ITINS_S>; +defm VMIN : avx512_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>; +defm VMAX : avx512_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>; +} +let isCommutable = 0 in { +defm VSUB : avx512_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>; +defm VDIV : avx512_binop_s<0x5E, "div", fdiv, SSE_ALU_ITINS_S>; +} + +multiclass avx512_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode, + RegisterClass RC, ValueType vt, + X86MemOperand x86memop, PatFrag mem_frag, + X86MemOperand x86scalar_mop, PatFrag scalar_mfrag, + string BrdcstStr, + Domain d, OpndItins itins, bit commutable> { + let isCommutable = commutable in + 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, 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, 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, TB; + } +} + +defm VADDPSZ : avx512_fp_packed<0x58, "addps", fadd, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 1>, EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VADDPDZ : avx512_fp_packed<0x58, "addpd", fadd, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 1>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VMULPSZ : avx512_fp_packed<0x59, "mulps", fmul, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 1>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VMULPDZ : avx512_fp_packed<0x59, "mulpd", fmul, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 1>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VMINPSZ : avx512_fp_packed<0x5D, "minps", X86fmin, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VMAXPSZ : avx512_fp_packed<0x5F, "maxps", X86fmax, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VMINPDZ : avx512_fp_packed<0x5D, "minpd", X86fmin, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 1>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; +defm VMAXPDZ : avx512_fp_packed<0x5F, "maxpd", X86fmax, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 1>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VSUBPSZ : avx512_fp_packed<0x5C, "subps", fsub, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 0>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VDIVPSZ : avx512_fp_packed<0x5E, "divps", fdiv, VR512, v16f32, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", SSEPackedSingle, + SSE_ALU_ITINS_P.s, 0>, EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VSUBPDZ : avx512_fp_packed<0x5C, "subpd", fsub, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 0>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; +defm VDIVPDZ : avx512_fp_packed<0x5E, "divpd", fdiv, VR512, v8f64, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", SSEPackedDouble, + SSE_ALU_ITINS_P.d, 0>, + EVEX_V512, OpSize, VEX_W, EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 VPTESTM instructions +//===----------------------------------------------------------------------===// + +multiclass avx512_vptest<bits<8> opc, string OpcodeStr, RegisterClass KRC, + RegisterClass RC, X86MemOperand x86memop, PatFrag memop_frag, + SDNode OpNode, ValueType vt> { + def rr : AVX5128I<opc, MRMSrcReg, + (outs KRC:$dst), (ins RC:$src1, RC:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set KRC:$dst, (OpNode (vt RC:$src1), (vt RC:$src2)))]>, EVEX_4V; + def rm : AVX5128I<opc, MRMSrcMem, + (outs KRC:$dst), (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set KRC:$dst, (OpNode (vt RC:$src1), + (bitconvert (memop_frag addr:$src2))))]>, EVEX_4V; +} + +defm VPTESTMDZ : avx512_vptest<0x27, "vptestmd", VK16, VR512, f512mem, + memopv16i32, X86testm, v16i32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPTESTMQZ : avx512_vptest<0x27, "vptestmq", VK8, VR512, f512mem, + memopv8i64, X86testm, v8i64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 Shift instructions +//===----------------------------------------------------------------------===// +multiclass avx512_shift_rmi<bits<8> opc, Format ImmFormR, Format ImmFormM, + string OpcodeStr, SDNode OpNode, RegisterClass RC, + ValueType vt, X86MemOperand x86memop, PatFrag mem_frag, + RegisterClass KRC> { + def ri : AVX512BIi8<opc, ImmFormR, (outs RC:$dst), + (ins RC:$src1, i8imm:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$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:$src1, i8imm:$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:$src1, i8imm:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (OpNode (mem_frag addr:$src1), + (i8 imm:$src2)))], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V; + def mik: AVX512BIi8<opc, ImmFormM, (outs RC:$dst), + (ins KRC:$mask, x86memop:$src1, i8imm:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"), + [], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V, EVEX_K; +} + +multiclass avx512_shift_rrm<bits<8> opc, string OpcodeStr, SDNode OpNode, + RegisterClass RC, ValueType vt, ValueType SrcVT, + PatFrag bc_frag, RegisterClass KRC> { + // src2 is always 128-bit + def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, VR128X:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (vt (OpNode RC:$src1, (SrcVT VR128X:$src2))))], + SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V; + def rrk : AVX512BI<opc, MRMSrcReg, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, VR128X:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"), + [], SSE_INTSHIFT_ITINS_P.rr>, EVEX_4V, EVEX_K; + def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, i128mem:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (vt (OpNode RC:$src1, + (bc_frag (memopv2i64 addr:$src2)))))], + SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V; + def rmk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst), + (ins KRC:$mask, RC:$src1, i128mem:$src2), + !strconcat(OpcodeStr, + "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"), + [], SSE_INTSHIFT_ITINS_P.rm>, EVEX_4V, EVEX_K; +} + +defm VPSRLDZ : avx512_shift_rmi<0x72, MRM2r, MRM2m, "vpsrld", X86vsrli, + VR512, v16i32, i512mem, memopv16i32, VK16WM>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPSRLDZ : avx512_shift_rrm<0xD2, "vpsrld", X86vsrl, + VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512, + EVEX_CD8<32, CD8VQ>; + +defm VPSRLQZ : avx512_shift_rmi<0x73, MRM2r, MRM2m, "vpsrlq", X86vsrli, + VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VF>, VEX_W; +defm VPSRLQZ : avx512_shift_rrm<0xD3, "vpsrlq", X86vsrl, + VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VQ>, VEX_W; + +defm VPSLLDZ : avx512_shift_rmi<0x72, MRM6r, MRM6m, "vpslld", X86vshli, + VR512, v16i32, i512mem, memopv16i32, VK16WM>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPSLLDZ : avx512_shift_rrm<0xF2, "vpslld", X86vshl, + VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512, + EVEX_CD8<32, CD8VQ>; + +defm VPSLLQZ : avx512_shift_rmi<0x73, MRM6r, MRM6m, "vpsllq", X86vshli, + VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VF>, VEX_W; +defm VPSLLQZ : avx512_shift_rrm<0xF3, "vpsllq", X86vshl, + VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VQ>, VEX_W; + +defm VPSRADZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsrad", X86vsrai, + VR512, v16i32, i512mem, memopv16i32, VK16WM>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPSRADZ : avx512_shift_rrm<0xE2, "vpsrad", X86vsra, + VR512, v16i32, v4i32, bc_v4i32, VK16WM>, EVEX_V512, + EVEX_CD8<32, CD8VQ>; + +defm VPSRAQZ : avx512_shift_rmi<0x72, MRM4r, MRM4m, "vpsraq", X86vsrai, + VR512, v8i64, i512mem, memopv8i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VF>, VEX_W; +defm VPSRAQZ : avx512_shift_rrm<0xE2, "vpsraq", X86vsra, + VR512, v8i64, v2i64, bc_v2i64, VK8WM>, EVEX_V512, + EVEX_CD8<64, CD8VQ>, VEX_W; + +//===-------------------------------------------------------------------===// +// Variable Bit Shifts +//===-------------------------------------------------------------------===// +multiclass avx512_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode, + RegisterClass RC, ValueType vt, + X86MemOperand x86memop, PatFrag mem_frag> { + def rr : AVX5128I<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:$src1, (vt RC:$src2))))]>, + EVEX_4V; + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (vt (OpNode RC:$src1, (mem_frag addr:$src2))))]>, + EVEX_4V; +} + +defm VPSLLVDZ : avx512_var_shift<0x47, "vpsllvd", shl, VR512, v16i32, + i512mem, memopv16i32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPSLLVQZ : avx512_var_shift<0x47, "vpsllvq", shl, VR512, v8i64, + i512mem, memopv8i64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +defm VPSRLVDZ : avx512_var_shift<0x45, "vpsrlvd", srl, VR512, v16i32, + i512mem, memopv16i32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPSRLVQZ : avx512_var_shift<0x45, "vpsrlvq", srl, VR512, v8i64, + i512mem, memopv8i64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +defm VPSRAVDZ : avx512_var_shift<0x46, "vpsravd", sra, VR512, v16i32, + i512mem, memopv16i32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPSRAVQZ : avx512_var_shift<0x46, "vpsravq", sra, VR512, v8i64, + i512mem, memopv8i64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + +//===----------------------------------------------------------------------===// +// AVX-512 - MOVDDUP +//===----------------------------------------------------------------------===// + +multiclass avx512_movddup<string OpcodeStr, RegisterClass RC, ValueType VT, + X86MemOperand x86memop, PatFrag memop_frag> { +def rr : AVX512PDI<0x12, MRMSrcReg, (outs RC:$dst), (ins RC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (VT (X86Movddup RC:$src)))]>, EVEX; +def rm : AVX512PDI<0x12, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, + (VT (X86Movddup (memop_frag addr:$src))))]>, EVEX; +} + +defm VMOVDDUPZ : avx512_movddup<"vmovddup", VR512, v8f64, f512mem, memopv8f64>, + VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; +def : Pat<(X86Movddup (v8f64 (scalar_to_vector (loadf64 addr:$src)))), + (VMOVDDUPZrm addr:$src)>; + +//===---------------------------------------------------------------------===// +// Replicate Single FP - MOVSHDUP and MOVSLDUP +//===---------------------------------------------------------------------===// +multiclass avx512_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr, + ValueType vt, RegisterClass RC, PatFrag mem_frag, + X86MemOperand x86memop> { + def rr : AVX512XSI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (vt (OpNode RC:$src)))]>, EVEX; + let mayLoad = 1 in + def rm : AVX512XSI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>, EVEX; +} + +defm VMOVSHDUPZ : avx512_replicate_sfp<0x16, X86Movshdup, "vmovshdup", + v16f32, VR512, memopv16f32, f512mem>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VMOVSLDUPZ : avx512_replicate_sfp<0x12, X86Movsldup, "vmovsldup", + v16f32, VR512, memopv16f32, f512mem>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + +def : Pat<(v16i32 (X86Movshdup VR512:$src)), (VMOVSHDUPZrr VR512:$src)>; +def : Pat<(v16i32 (X86Movshdup (memopv16i32 addr:$src))), + (VMOVSHDUPZrm addr:$src)>; +def : Pat<(v16i32 (X86Movsldup VR512:$src)), (VMOVSLDUPZrr VR512:$src)>; +def : Pat<(v16i32 (X86Movsldup (memopv16i32 addr:$src))), + (VMOVSLDUPZrm addr:$src)>; + +//===----------------------------------------------------------------------===// +// Move Low to High and High to Low packed FP Instructions +//===----------------------------------------------------------------------===// +def VMOVLHPSZrr : AVX512PSI<0x16, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src1, VR128X:$src2), + "vmovlhps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [(set VR128X:$dst, (v4f32 (X86Movlhps VR128X:$src1, VR128X:$src2)))], + IIC_SSE_MOV_LH>, EVEX_4V; +def VMOVHLPSZrr : AVX512PSI<0x12, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src1, VR128X:$src2), + "vmovhlps{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + [(set VR128X:$dst, (v4f32 (X86Movhlps VR128X:$src1, VR128X:$src2)))], + IIC_SSE_MOV_LH>, EVEX_4V; + +let Predicates = [HasAVX512] in { + // MOVLHPS patterns + def : Pat<(v4i32 (X86Movlhps VR128X:$src1, VR128X:$src2)), + (VMOVLHPSZrr VR128X:$src1, VR128X:$src2)>; + def : Pat<(v2i64 (X86Movlhps VR128X:$src1, VR128X:$src2)), + (VMOVLHPSZrr (v2i64 VR128X:$src1), VR128X:$src2)>; + + // MOVHLPS patterns + def : Pat<(v4i32 (X86Movhlps VR128X:$src1, VR128X:$src2)), + (VMOVHLPSZrr VR128X:$src1, VR128X:$src2)>; +} + +//===----------------------------------------------------------------------===// +// FMA - Fused Multiply Operations +// +let Constraints = "$src1 = $dst" in { +multiclass avx512_fma3p_rm<bits<8> opc, string OpcodeStr, + RegisterClass RC, X86MemOperand x86memop, + PatFrag mem_frag, X86MemOperand x86scalar_mop, PatFrag scalar_mfrag, + string BrdcstStr, SDNode OpNode, ValueType OpVT> { + def r: AVX512FMA3<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2, RC:$src3), + !strconcat(OpcodeStr,"\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, (OpVT(OpNode RC:$src1, RC:$src2, RC:$src3)))]>; + + let mayLoad = 1 in + def m: AVX512FMA3<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src2, x86memop:$src3), + !strconcat(OpcodeStr, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2, + (mem_frag addr:$src3))))]>; + def mb: AVX512FMA3<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src2, x86scalar_mop:$src3), + !strconcat(OpcodeStr, "\t{${src3}", BrdcstStr, + ", $src2, $dst|$dst, $src2, ${src3}", BrdcstStr, "}"), + [(set RC:$dst, (OpNode RC:$src1, RC:$src2, + (OpVT (X86VBroadcast (scalar_mfrag addr:$src3)))))]>, EVEX_B; +} +} // Constraints = "$src1 = $dst" + +let ExeDomain = SSEPackedSingle in { + defm VFMADD213PSZ : avx512_fma3p_rm<0xA8, "vfmadd213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmadd, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFMSUB213PSZ : avx512_fma3p_rm<0xAA, "vfmsub213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmsub, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFMADDSUB213PSZ : avx512_fma3p_rm<0xA6, "vfmaddsub213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmaddsub, v16f32>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + defm VFMSUBADD213PSZ : avx512_fma3p_rm<0xA7, "vfmsubadd213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmsubadd, v16f32>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + defm VFNMADD213PSZ : avx512_fma3p_rm<0xAC, "vfnmadd213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fnmadd, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFNMSUB213PSZ : avx512_fma3p_rm<0xAE, "vfnmsub213ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fnmsub, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +} +let ExeDomain = SSEPackedDouble in { + defm VFMADD213PDZ : avx512_fma3p_rm<0xA8, "vfmadd213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmadd, v8f64>, EVEX_V512, + VEX_W, EVEX_CD8<64, CD8VF>; + defm VFMSUB213PDZ : avx512_fma3p_rm<0xAA, "vfmsub213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFMADDSUB213PDZ : avx512_fma3p_rm<0xA6, "vfmaddsub213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmaddsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFMSUBADD213PDZ : avx512_fma3p_rm<0xA7, "vfmsubadd213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmsubadd, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFNMADD213PDZ : avx512_fma3p_rm<0xAC, "vfnmadd213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fnmadd, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFNMSUB213PDZ : avx512_fma3p_rm<0xAE, "vfnmsub213pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fnmsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +} + +let Constraints = "$src1 = $dst" in { +multiclass avx512_fma3p_m132<bits<8> opc, string OpcodeStr, + RegisterClass RC, X86MemOperand x86memop, + PatFrag mem_frag, X86MemOperand x86scalar_mop, PatFrag scalar_mfrag, + string BrdcstStr, SDNode OpNode, ValueType OpVT> { + let mayLoad = 1 in + def m: AVX512FMA3<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src3, x86memop:$src2), + !strconcat(OpcodeStr, "\t{$src2, $src3, $dst|$dst, $src3, $src2}"), + [(set RC:$dst, (OpVT (OpNode RC:$src1, (mem_frag addr:$src2), RC:$src3)))]>; + def mb: AVX512FMA3<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src3, x86scalar_mop:$src2), + !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr, + ", $src3, $dst|$dst, $src3, ${src2}", BrdcstStr, "}"), + [(set RC:$dst, (OpNode RC:$src1, + (OpVT (X86VBroadcast (scalar_mfrag addr:$src2))), RC:$src3))]>, EVEX_B; +} +} // Constraints = "$src1 = $dst" + + +let ExeDomain = SSEPackedSingle in { + defm VFMADD132PSZ : avx512_fma3p_m132<0x98, "vfmadd132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmadd, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFMSUB132PSZ : avx512_fma3p_m132<0x9A, "vfmsub132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmsub, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFMADDSUB132PSZ : avx512_fma3p_m132<0x96, "vfmaddsub132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmaddsub, v16f32>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + defm VFMSUBADD132PSZ : avx512_fma3p_m132<0x97, "vfmsubadd132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fmsubadd, v16f32>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + defm VFNMADD132PSZ : avx512_fma3p_m132<0x9C, "vfnmadd132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fnmadd, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + defm VFNMSUB132PSZ : avx512_fma3p_m132<0x9E, "vfnmsub132ps", VR512, f512mem, + memopv16f32, f32mem, loadf32, "{1to16}", + X86Fnmsub, v16f32>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +} +let ExeDomain = SSEPackedDouble in { + defm VFMADD132PDZ : avx512_fma3p_m132<0x98, "vfmadd132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmadd, v8f64>, EVEX_V512, + VEX_W, EVEX_CD8<64, CD8VF>; + defm VFMSUB132PDZ : avx512_fma3p_m132<0x9A, "vfmsub132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFMADDSUB132PDZ : avx512_fma3p_m132<0x96, "vfmaddsub132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmaddsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFMSUBADD132PDZ : avx512_fma3p_m132<0x97, "vfmsubadd132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fmsubadd, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFNMADD132PDZ : avx512_fma3p_m132<0x9C, "vfnmadd132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fnmadd, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + defm VFNMSUB132PDZ : avx512_fma3p_m132<0x9E, "vfnmsub132pd", VR512, f512mem, + memopv8f64, f64mem, loadf64, "{1to8}", + X86Fnmsub, v8f64>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; +} + +// Scalar FMA +let Constraints = "$src1 = $dst" in { +multiclass avx512_fma3s_rm<bits<8> opc, string OpcodeStr, SDNode OpNode, + RegisterClass RC, ValueType OpVT, + X86MemOperand x86memop, Operand memop, + PatFrag mem_frag> { + let isCommutable = 1 in + def r : AVX512FMA3<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2, RC:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, + (OpVT (OpNode RC:$src2, RC:$src1, RC:$src3)))]>; + let mayLoad = 1 in + def m : AVX512FMA3<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, RC:$src2, f128mem:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $dst|$dst, $src2, $src3}"), + [(set RC:$dst, + (OpVT (OpNode RC:$src2, RC:$src1, + (mem_frag addr:$src3))))]>; +} + +} // Constraints = "$src1 = $dst" + +defm VFMADDSSZ : avx512_fma3s_rm<0xA9, "vfmadd213ss{z}", X86Fmadd, FR32X, + f32, f32mem, ssmem, loadf32>, EVEX_CD8<32, CD8VT1>; +defm VFMADDSDZ : avx512_fma3s_rm<0xA9, "vfmadd213sd{z}", X86Fmadd, FR64X, + f64, f64mem, sdmem, loadf64>, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VFMSUBSSZ : avx512_fma3s_rm<0xAB, "vfmsub213ss{z}", X86Fmsub, FR32X, + f32, f32mem, ssmem, loadf32>, EVEX_CD8<32, CD8VT1>; +defm VFMSUBSDZ : avx512_fma3s_rm<0xAB, "vfmsub213sd{z}", X86Fmsub, FR64X, + f64, f64mem, sdmem, loadf64>, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VFNMADDSSZ : avx512_fma3s_rm<0xAD, "vfnmadd213ss{z}", X86Fnmadd, FR32X, + f32, f32mem, ssmem, loadf32>, EVEX_CD8<32, CD8VT1>; +defm VFNMADDSDZ : avx512_fma3s_rm<0xAD, "vfnmadd213sd{z}", X86Fnmadd, FR64X, + f64, f64mem, sdmem, loadf64>, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VFNMSUBSSZ : avx512_fma3s_rm<0xAF, "vfnmsub213ss{z}", X86Fnmsub, FR32X, + f32, f32mem, ssmem, loadf32>, EVEX_CD8<32, CD8VT1>; +defm VFNMSUBSDZ : avx512_fma3s_rm<0xAF, "vfnmsub213sd{z}", X86Fnmsub, FR64X, + f64, f64mem, sdmem, loadf64>, VEX_W, EVEX_CD8<64, CD8VT1>; + +//===----------------------------------------------------------------------===// +// AVX-512 Scalar convert from sign integer to float/double +//===----------------------------------------------------------------------===// + +multiclass avx512_vcvtsi<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC, + X86MemOperand x86memop, string asm> { +let neverHasSideEffects = 1 in { + def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src), + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + EVEX_4V; + let mayLoad = 1 in + def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), + (ins DstRC:$src1, x86memop:$src), + !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>, + EVEX_4V; +} // neverHasSideEffects = 1 +} +let Predicates = [HasAVX512] in { +defm VCVTSI2SSZ : avx512_vcvtsi<0x2A, GR32, FR32X, i32mem, "cvtsi2ss{l}{z}">, + XS, VEX_LIG, EVEX_CD8<32, CD8VT1>; +defm VCVTSI642SSZ : avx512_vcvtsi<0x2A, GR64, FR32X, i64mem, "cvtsi2ss{q}{z}">, + XS, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>; +defm VCVTSI2SDZ : avx512_vcvtsi<0x2A, GR32, FR64X, i32mem, "cvtsi2sd{l}{z}">, + XD, VEX_LIG, EVEX_CD8<32, CD8VT1>; +defm VCVTSI642SDZ : avx512_vcvtsi<0x2A, GR64, FR64X, i64mem, "cvtsi2sd{q}{z}">, + XD, VEX_W, VEX_LIG, EVEX_CD8<64, CD8VT1>; + +def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))), + (VCVTSI2SSZrm (f32 (IMPLICIT_DEF)), addr:$src)>; +def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))), + (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))), + (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)), + (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)), + (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:$src)))]>, EVEX; +} + +defm VCVTTSS2SIZ : avx512_cvt_s<0x2C, FR32X, GR32, fp_to_sint, f32mem, + loadf32, "cvttss2si{z}">, XS, + EVEX_CD8<32, CD8VT1>; +defm VCVTTSS2USIZ : avx512_cvt_s<0x78, FR32X, GR32, fp_to_uint, f32mem, + loadf32, "cvttss2usi{z}">, XS, + EVEX_CD8<32, CD8VT1>; +defm VCVTTSS2SI64Z : avx512_cvt_s<0x2C, FR32X, GR64, fp_to_sint, f32mem, + loadf32, "cvttss2si{z}">, XS, VEX_W, + EVEX_CD8<32, CD8VT1>; +defm VCVTTSS2USI64Z : avx512_cvt_s<0x78, FR32X, GR64, fp_to_uint, f32mem, + loadf32, "cvttss2usi{z}">, XS, VEX_W, + EVEX_CD8<32, CD8VT1>; +defm VCVTTSD2SIZ : avx512_cvt_s<0x2C, FR64X, GR32, fp_to_sint, f64mem, + loadf64, "cvttsd2si{z}">, XD, + EVEX_CD8<64, CD8VT1>; +defm VCVTTSD2USIZ : avx512_cvt_s<0x78, FR64X, GR32, fp_to_uint, f64mem, + loadf64, "cvttsd2usi{z}">, XD, + EVEX_CD8<64, CD8VT1>; +defm VCVTTSD2SI64Z : avx512_cvt_s<0x2C, FR64X, GR64, fp_to_sint, f64mem, + loadf64, "cvttsd2si{z}">, XD, VEX_W, + EVEX_CD8<64, CD8VT1>; +defm VCVTTSD2USI64Z : avx512_cvt_s<0x78, FR64X, GR64, fp_to_uint, f64mem, + loadf64, "cvttsd2usi{z}">, XD, VEX_W, + EVEX_CD8<64, CD8VT1>; +//===----------------------------------------------------------------------===// +// AVX-512 Convert form float to double and back +//===----------------------------------------------------------------------===// +let neverHasSideEffects = 1 in { +def VCVTSS2SDZrr : AVX512XSI<0x5A, MRMSrcReg, (outs FR64X:$dst), + (ins FR32X:$src1, FR32X:$src2), + "vcvtss2sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX_4V, VEX_LIG, Sched<[WriteCvtF2F]>; +let mayLoad = 1 in +def VCVTSS2SDZrm : AVX512XSI<0x5A, MRMSrcMem, (outs FR64X:$dst), + (ins FR32X:$src1, f32mem:$src2), + "vcvtss2sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX_4V, VEX_LIG, Sched<[WriteCvtF2FLd, ReadAfterLd]>, + EVEX_CD8<32, CD8VT1>; + +// Convert scalar double to scalar single +def VCVTSD2SSZrr : AVX512XDI<0x5A, MRMSrcReg, (outs FR32X:$dst), + (ins FR64X:$src1, FR64X:$src2), + "vcvtsd2ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX_4V, VEX_LIG, VEX_W, Sched<[WriteCvtF2F]>; +let mayLoad = 1 in +def VCVTSD2SSZrm : AVX512XDI<0x5A, MRMSrcMem, (outs FR32X:$dst), + (ins FR64X:$src1, f64mem:$src2), + "vcvtsd2ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}", + []>, EVEX_4V, VEX_LIG, VEX_W, + Sched<[WriteCvtF2FLd, ReadAfterLd]>, EVEX_CD8<64, CD8VT1>; +} + +def : Pat<(f64 (fextend FR32X:$src)), (VCVTSS2SDZrr FR32X:$src, FR32X:$src)>, + Requires<[HasAVX512]>; +def : Pat<(fextend (loadf32 addr:$src)), + (VCVTSS2SDZrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX512]>; + +def : Pat<(extloadf32 addr:$src), + (VCVTSS2SDZrm (f32 (IMPLICIT_DEF)), addr:$src)>, + Requires<[HasAVX512, OptForSize]>; + +def : Pat<(extloadf32 addr:$src), + (VCVTSS2SDZrr (f32 (IMPLICIT_DEF)), (VMOVSSZrm addr:$src))>, + Requires<[HasAVX512, OptForSpeed]>; + +def : Pat<(f32 (fround FR64X:$src)), (VCVTSD2SSZrr FR64X:$src, FR64X:$src)>, + Requires<[HasAVX512]>; + +multiclass avx512_vcvt_fp<bits<8> opc, string asm, RegisterClass SrcRC, + RegisterClass DstRC, SDNode OpNode, PatFrag mem_frag, + X86MemOperand x86memop, ValueType OpVT, ValueType InVT, + Domain d> { +let neverHasSideEffects = 1 in { + def rr : AVX512PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), + !strconcat(asm,"\t{$src, $dst|$dst, $src}"), + [(set DstRC:$dst, + (OpVT (OpNode (InVT SrcRC:$src))))], d>, EVEX; + let mayLoad = 1 in + def rm : AVX512PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), + !strconcat(asm,"\t{$src, $dst|$dst, $src}"), + [(set DstRC:$dst, + (OpVT (OpNode (InVT (bitconvert (mem_frag addr:$src))))))], d>, EVEX; +} // neverHasSideEffects = 1 +} + +defm VCVTPD2PSZ : avx512_vcvt_fp<0x5A, "vcvtpd2ps", VR512, VR256X, fround, + memopv8f64, f512mem, v8f32, v8f64, + SSEPackedSingle>, EVEX_V512, VEX_W, OpSize, + EVEX_CD8<64, CD8VF>; + +defm VCVTPS2PDZ : avx512_vcvt_fp<0x5A, "vcvtps2pd", VR256X, VR512, fextend, + memopv4f64, f256mem, v8f64, v8f32, + SSEPackedDouble>, EVEX_V512, EVEX_CD8<32, CD8VH>; +def : Pat<(v8f64 (extloadv8f32 addr:$src)), + (VCVTPS2PDZrm addr:$src)>; + +//===----------------------------------------------------------------------===// +// AVX-512 Vector convert from sign integer to float/double +//===----------------------------------------------------------------------===// + +defm VCVTDQ2PSZ : avx512_vcvt_fp<0x5B, "vcvtdq2ps", VR512, VR512, sint_to_fp, + memopv8i64, i512mem, v16f32, v16i32, + SSEPackedSingle>, EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VCVTDQ2PDZ : avx512_vcvt_fp<0xE6, "vcvtdq2pd", VR256X, VR512, sint_to_fp, + memopv4i64, i256mem, v8f64, v8i32, + SSEPackedDouble>, EVEX_V512, XS, + EVEX_CD8<32, CD8VH>; + +defm VCVTTPS2DQZ : avx512_vcvt_fp<0x5B, "vcvttps2dq", VR512, VR512, fp_to_sint, + memopv16f32, f512mem, v16i32, v16f32, + SSEPackedSingle>, EVEX_V512, XS, + EVEX_CD8<32, CD8VF>; + +defm VCVTTPD2DQZ : avx512_vcvt_fp<0xE6, "vcvttpd2dq", VR512, VR256X, fp_to_sint, + memopv8f64, f512mem, v8i32, v8f64, + SSEPackedDouble>, EVEX_V512, OpSize, VEX_W, + EVEX_CD8<64, CD8VF>; + +defm VCVTTPS2UDQZ : avx512_vcvt_fp<0x78, "vcvttps2udq", VR512, VR512, fp_to_uint, + memopv16f32, f512mem, v16i32, v16f32, + SSEPackedSingle>, EVEX_V512, + EVEX_CD8<32, CD8VF>; + +defm VCVTTPD2UDQZ : avx512_vcvt_fp<0x78, "vcvttpd2udq", VR512, VR256X, fp_to_uint, + memopv8f64, f512mem, v8i32, v8f64, + SSEPackedDouble>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + +defm VCVTUDQ2PDZ : avx512_vcvt_fp<0x7A, "vcvtudq2pd", VR256X, VR512, uint_to_fp, + memopv4i64, f256mem, v8f64, v8i32, + SSEPackedDouble>, EVEX_V512, XS, + EVEX_CD8<32, CD8VH>; + +defm VCVTUDQ2PSZ : avx512_vcvt_fp<0x7A, "vcvtudq2ps", VR512, VR512, uint_to_fp, + memopv16i32, f512mem, v16f32, v16i32, + SSEPackedSingle>, EVEX_V512, XD, + EVEX_CD8<32, CD8VF>; + +def : Pat<(v8i32 (fp_to_uint (v8f32 VR256X:$src1))), + (EXTRACT_SUBREG (v16i32 (VCVTTPS2UDQZrr + (v16f32 (SUBREG_TO_REG (i32 0), VR256X:$src1, sub_ymm)))), sub_ymm)>; + + +def : Pat<(int_x86_avx512_cvtdq2_ps_512 VR512:$src), + (VCVTDQ2PSZrr VR512:$src)>; +def : Pat<(int_x86_avx512_cvtdq2_ps_512 (bitconvert (memopv8i64 addr:$src))), + (VCVTDQ2PSZrm addr:$src)>; + +def VCVTPS2DQZrr : AVX512BI<0x5B, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + "vcvtps2dq\t{$src, $dst|$dst, $src}", + [(set VR512:$dst, + (int_x86_avx512_cvt_ps2dq_512 VR512:$src))], + IIC_SSE_CVT_PS_RR>, EVEX, EVEX_V512; +def VCVTPS2DQZrm : AVX512BI<0x5B, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + "vcvtps2dq\t{$src, $dst|$dst, $src}", + [(set VR512:$dst, + (int_x86_avx512_cvt_ps2dq_512 (memopv16f32 addr:$src)))], + IIC_SSE_CVT_PS_RM>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; + + +let Predicates = [HasAVX512] in { + def : Pat<(v8f32 (fround (loadv8f64 addr:$src))), + (VCVTPD2PSZrm addr:$src)>; + def : Pat<(v8f64 (extloadv8f32 addr:$src)), + (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, $src2}", []>, EVEX; +} + +defm VCVTPH2PSZ : avx512_f16c_ph2ps<VR512, VR256X, f256mem, + int_x86_avx512_vcvtph2ps_512>, EVEX_V512, + EVEX_CD8<32, CD8VH>; +defm VCVTPS2PHZ : avx512_f16c_ps2ph<VR256X, VR512, f256mem, + int_x86_avx512_vcvtps2ph_512>, EVEX_V512, + EVEX_CD8<32, CD8VH>; + +let Defs = [EFLAGS], Predicates = [HasAVX512] in { + defm VUCOMISSZ : sse12_ord_cmp<0x2E, FR32X, X86cmp, f32, f32mem, loadf32, + "ucomiss{z}">, TB, EVEX, VEX_LIG, + EVEX_CD8<32, CD8VT1>; + defm VUCOMISDZ : sse12_ord_cmp<0x2E, FR64X, X86cmp, f64, f64mem, loadf64, + "ucomisd{z}">, TB, OpSize, EVEX, + VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>; + let Pattern = []<dag> in { + defm VCOMISSZ : sse12_ord_cmp<0x2F, VR128X, undef, v4f32, f128mem, load, + "comiss{z}">, TB, EVEX, VEX_LIG, + EVEX_CD8<32, CD8VT1>; + defm VCOMISDZ : sse12_ord_cmp<0x2F, VR128X, undef, v2f64, f128mem, load, + "comisd{z}">, TB, OpSize, EVEX, + VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>; + } + defm Int_VUCOMISSZ : sse12_ord_cmp<0x2E, VR128X, X86ucomi, v4f32, f128mem, + load, "ucomiss">, TB, EVEX, VEX_LIG, + EVEX_CD8<32, CD8VT1>; + defm Int_VUCOMISDZ : sse12_ord_cmp<0x2E, VR128X, X86ucomi, v2f64, f128mem, + load, "ucomisd">, TB, OpSize, EVEX, + VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>; + + defm Int_VCOMISSZ : sse12_ord_cmp<0x2F, VR128X, X86comi, v4f32, f128mem, + load, "comiss">, TB, EVEX, VEX_LIG, + EVEX_CD8<32, CD8VT1>; + defm Int_VCOMISDZ : sse12_ord_cmp<0x2F, VR128X, X86comi, v2f64, f128mem, + load, "comisd">, TB, OpSize, EVEX, + VEX_LIG, VEX_W, EVEX_CD8<64, CD8VT1>; +} + +/// avx512_unop_p - AVX-512 unops in packed form. +multiclass avx512_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> { + def PSZr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (v16f32 (OpNode VR512:$src)))]>, + EVEX, EVEX_V512; + def PSZm : AVX5128I<opc, MRMSrcMem, (outs VR512:$dst), (ins f256mem:$src), + !strconcat(OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (OpNode (memopv16f32 addr:$src)))]>, + EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; + def PDZr : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (v8f64 (OpNode VR512:$src)))]>, + EVEX, EVEX_V512, VEX_W; + def PDZm : AVX5128I<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (OpNode (memopv16f32 addr:$src)))]>, + EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} + +/// avx512_fp_unop_p_int - AVX-512 intrinsics unops in packed forms. +multiclass avx512_fp_unop_p_int<bits<8> opc, string OpcodeStr, + Intrinsic V16F32Int, Intrinsic V8F64Int> { + def PSZr_Int : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (V16F32Int VR512:$src))]>, + EVEX, EVEX_V512; + def PSZm_Int : AVX5128I<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, + (V16F32Int (memopv16f32 addr:$src)))]>, EVEX, + EVEX_V512, EVEX_CD8<32, CD8VF>; + def PDZr_Int : AVX5128I<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (V8F64Int VR512:$src))]>, + EVEX, EVEX_V512, VEX_W; + def PDZm_Int : AVX5128I<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, + (V8F64Int (memopv8f64 addr:$src)))]>, + EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} + +/// avx512_fp_unop_s - AVX-512 unops in scalar form. +multiclass avx512_fp_unop_s<bits<8> opc, string OpcodeStr> { + let hasSideEffects = 0 in { + def SSZr : AVX5128I<opc, MRMSrcReg, (outs FR32X:$dst), + (ins FR32X:$src1, FR32X:$src2), + !strconcat(OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V; + let mayLoad = 1 in { + def SSZm : AVX5128I<opc, MRMSrcMem, (outs FR32X:$dst), + (ins FR32X:$src1, f32mem:$src2), + !strconcat(OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V, EVEX_CD8<32, CD8VT1>; + def SSZm_Int : AVX5128I<opc, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, ssmem:$src2), + !strconcat(OpcodeStr, + "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V, EVEX_CD8<32, CD8VT1>; + } + def SDZr : AVX5128I<opc, MRMSrcReg, (outs FR64X:$dst), + (ins FR64X:$src1, FR64X:$src2), + !strconcat(OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, + EVEX_4V, VEX_W; + let mayLoad = 1 in { + def SDZm : AVX5128I<opc, MRMSrcMem, (outs FR64X:$dst), + (ins FR64X:$src1, f64mem:$src2), + !strconcat(OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, + EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>; + def SDZm_Int : AVX5128I<opc, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, sdmem:$src2), + !strconcat(OpcodeStr, + "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + []>, EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>; + } +} +} + +defm VRCP14 : avx512_fp_unop_s<0x4D, "vrcp14">, + avx512_fp_unop_p<0x4C, "vrcp14", X86frcp>, + avx512_fp_unop_p_int<0x4C, "vrcp14", + int_x86_avx512_rcp14_ps_512, int_x86_avx512_rcp14_pd_512>; + +defm VRSQRT14 : avx512_fp_unop_s<0x4F, "vrsqrt14">, + avx512_fp_unop_p<0x4E, "vrsqrt14", X86frsqrt>, + avx512_fp_unop_p_int<0x4E, "vrsqrt14", + int_x86_avx512_rsqrt14_ps_512, int_x86_avx512_rsqrt14_pd_512>; + +def : Pat<(int_x86_avx512_rsqrt14_ss VR128X:$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:$src)>; +} +multiclass avx512_sqrt_packed<bits<8> opc, string OpcodeStr, SDNode OpNode, + Intrinsic V16F32Int, Intrinsic V8F64Int, + OpndItins itins_s, OpndItins itins_d> { + def PSZrr :AVX512PSI<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (v16f32 (OpNode VR512:$src)))], itins_s.rr>, + EVEX, EVEX_V512; + + let mayLoad = 1 in + def PSZrm : AVX512PSI<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, + (OpNode (v16f32 (bitconvert (memopv16f32 addr:$src)))))], + itins_s.rm>, EVEX, EVEX_V512, EVEX_CD8<32, CD8VF>; + + def PDZrr : AVX512PDI<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (v8f64 (OpNode VR512:$src)))], itins_d.rr>, + EVEX, EVEX_V512; + + let mayLoad = 1 in + def PDZrm : AVX512PDI<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (OpNode + (v8f64 (bitconvert (memopv16f32 addr:$src)))))], + itins_d.rm>, EVEX, EVEX_V512, EVEX_CD8<64, CD8VF>; + + def PSZr_Int : AVX512PSI<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, + "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (V16F32Int VR512:$src))]>, + EVEX, EVEX_V512; + def PSZm_Int : AVX512PSI<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, + (V16F32Int (memopv16f32 addr:$src)))]>, EVEX, + EVEX_V512, EVEX_CD8<32, CD8VF>; + def PDZr_Int : AVX512PDI<opc, MRMSrcReg, (outs VR512:$dst), (ins VR512:$src), + !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (V8F64Int VR512:$src))]>, + EVEX, EVEX_V512, VEX_W; + def PDZm_Int : AVX512PDI<opc, MRMSrcMem, (outs VR512:$dst), (ins f512mem:$src), + !strconcat(OpcodeStr, + "pd\t{$src, $dst|$dst, $src}"), + [(set VR512:$dst, (V8F64Int (memopv8f64 addr:$src)))]>, + EVEX, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} + +multiclass avx512_sqrt_scalar<bits<8> opc, string OpcodeStr, + Intrinsic F32Int, Intrinsic F64Int, + OpndItins itins_s, OpndItins itins_d> { + def SSZr : SI<opc, MRMSrcReg, (outs FR32X:$dst), + (ins FR32X:$src1, FR32X:$src2), + !strconcat(OpcodeStr, + "ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [], itins_s.rr>, XS, EVEX_4V; + def SSZr_Int : SIi8<opc, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src1, VR128X:$src2), + !strconcat(OpcodeStr, + "ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128X:$dst, + (F32Int VR128X:$src1, VR128X:$src2))], + itins_s.rr>, XS, EVEX_4V; + let mayLoad = 1 in { + def SSZm : SI<opc, MRMSrcMem, (outs FR32X:$dst), + (ins FR32X:$src1, f32mem:$src2), + !strconcat(OpcodeStr, + "ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [], itins_s.rm>, XS, EVEX_4V, EVEX_CD8<32, CD8VT1>; + def SSZm_Int : SIi8<opc, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, ssmem:$src2), + !strconcat(OpcodeStr, + "ss{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128X:$dst, + (F32Int VR128X:$src1, sse_load_f32:$src2))], + itins_s.rm>, XS, EVEX_4V, EVEX_CD8<32, CD8VT1>; + } + def SDZr : SI<opc, MRMSrcReg, (outs FR64X:$dst), + (ins FR64X:$src1, FR64X:$src2), + !strconcat(OpcodeStr, + "sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, + XD, EVEX_4V, VEX_W; + def SDZr_Int : SIi8<opc, MRMSrcReg, (outs VR128X:$dst), + (ins VR128X:$src1, VR128X:$src2), + !strconcat(OpcodeStr, + "sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128X:$dst, + (F64Int VR128X:$src1, VR128X:$src2))], + itins_s.rr>, XD, EVEX_4V, VEX_W; + let mayLoad = 1 in { + def SDZm : SI<opc, MRMSrcMem, (outs FR64X:$dst), + (ins FR64X:$src1, f64mem:$src2), + !strconcat(OpcodeStr, + "sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>, + XD, EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>; + def SDZm_Int : SIi8<opc, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, sdmem:$src2), + !strconcat(OpcodeStr, + "sd{z}\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set VR128X:$dst, + (F64Int VR128X:$src1, sse_load_f64:$src2))]>, + XD, EVEX_4V, VEX_W, EVEX_CD8<64, CD8VT1>; + } +} + + +defm VSQRT : avx512_sqrt_scalar<0x51, "sqrt", + int_x86_avx512_sqrt_ss, int_x86_avx512_sqrt_sd, + SSE_SQRTSS, SSE_SQRTSD>, + avx512_sqrt_packed<0x51, "vsqrt", fsqrt, + int_x86_avx512_sqrt_ps_512, int_x86_avx512_sqrt_pd_512, + SSE_SQRTPS, SSE_SQRTPD>; + +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, + PatFrag mem_frag32, PatFrag mem_frag64, + Intrinsic V4F32Int, Intrinsic V2F64Int, + CD8VForm VForm> { +let ExeDomain = SSEPackedSingle in { + // Intrinsic operation, reg. + // Vector intrinsic operation, reg + def PSr : AVX512AIi8<opcps, MRMSrcReg, + (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>; + + // Vector intrinsic operation, mem + def PSm : AVX512AIi8<opcps, MRMSrcMem, + (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>, + EVEX_CD8<32, VForm>; +} // ExeDomain = SSEPackedSingle + +let ExeDomain = SSEPackedDouble in { + // Vector intrinsic operation, reg + def PDr : AVX512AIi8<opcpd, MRMSrcReg, + (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>; + + // Vector intrinsic operation, mem + def PDm : AVX512AIi8<opcpd, MRMSrcMem, + (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2), + !strconcat(OpcodeStr, + "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), + [(set RC:$dst, + (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>, + EVEX_CD8<64, VForm>; +} // ExeDomain = SSEPackedDouble +} + +multiclass avx512_fp_binop_rm<bits<8> opcss, bits<8> opcsd, + string OpcodeStr, + Intrinsic F32Int, + Intrinsic F64Int> { +let ExeDomain = GenericDomain in { + // Operation, reg. + let hasSideEffects = 0 in + def SSr : AVX512AIi8<opcss, MRMSrcReg, + (outs FR32X:$dst), (ins FR32X:$src1, FR32X:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + []>; + + // Intrinsic operation, reg. + def SSr_Int : AVX512AIi8<opcss, MRMSrcReg, + (outs VR128X:$dst), (ins VR128X:$src1, VR128X:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128X:$dst, (F32Int VR128X:$src1, VR128X:$src2, imm:$src3))]>; + + // Intrinsic operation, mem. + def SSm : AVX512AIi8<opcss, MRMSrcMem, (outs VR128X:$dst), + (ins VR128X:$src1, ssmem:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128X:$dst, (F32Int VR128X:$src1, + sse_load_f32:$src2, imm:$src3))]>, + EVEX_CD8<32, CD8VT1>; + + // Operation, reg. + let hasSideEffects = 0 in + def SDr : AVX512AIi8<opcsd, MRMSrcReg, + (outs FR64X:$dst), (ins FR64X:$src1, FR64X:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + []>, VEX_W; + + // Intrinsic operation, reg. + def SDr_Int : AVX512AIi8<opcsd, MRMSrcReg, + (outs VR128X:$dst), (ins VR128X:$src1, VR128X:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128X:$dst, (F64Int VR128X:$src1, VR128X:$src2, imm:$src3))]>, + VEX_W; + + // Intrinsic operation, mem. + def SDm : AVX512AIi8<opcsd, MRMSrcMem, + (outs VR128X:$dst), (ins VR128X:$src1, sdmem:$src2, i32i8imm:$src3), + !strconcat(OpcodeStr, + "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set VR128X:$dst, + (F64Int VR128X:$src1, sse_load_f64:$src2, imm:$src3))]>, + VEX_W, EVEX_CD8<64, CD8VT1>; +} // ExeDomain = GenericDomain +} + +let Predicates = [HasAVX512] in { + defm VRNDSCALE : avx512_fp_binop_rm<0x0A, 0x0B, "vrndscale", + int_x86_avx512_rndscale_ss, + int_x86_avx512_rndscale_sd>, EVEX_4V; + + defm VRNDSCALEZ : avx512_fp_unop_rm<0x08, 0x09, "vrndscale", f256mem, VR512, + memopv16f32, memopv8f64, + int_x86_avx512_rndscale_ps_512, + int_x86_avx512_rndscale_pd_512, CD8VF>, + EVEX, EVEX_V512; +} + +def : Pat<(ffloor FR32X:$src), + (VRNDSCALESSr (f32 (IMPLICIT_DEF)), FR32X:$src, (i32 0x1))>; +def : Pat<(f64 (ffloor FR64X:$src)), + (VRNDSCALESDr (f64 (IMPLICIT_DEF)), FR64X:$src, (i32 0x1))>; +def : Pat<(f32 (fnearbyint FR32X:$src)), + (VRNDSCALESSr (f32 (IMPLICIT_DEF)), FR32X:$src, (i32 0xC))>; +def : Pat<(f64 (fnearbyint FR64X:$src)), + (VRNDSCALESDr (f64 (IMPLICIT_DEF)), FR64X:$src, (i32 0xC))>; +def : Pat<(f32 (fceil FR32X:$src)), + (VRNDSCALESSr (f32 (IMPLICIT_DEF)), FR32X:$src, (i32 0x2))>; +def : Pat<(f64 (fceil FR64X:$src)), + (VRNDSCALESDr (f64 (IMPLICIT_DEF)), FR64X:$src, (i32 0x2))>; +def : Pat<(f32 (frint FR32X:$src)), + (VRNDSCALESSr (f32 (IMPLICIT_DEF)), FR32X:$src, (i32 0x4))>; +def : Pat<(f64 (frint FR64X:$src)), + (VRNDSCALESDr (f64 (IMPLICIT_DEF)), FR64X:$src, (i32 0x4))>; +def : Pat<(f32 (ftrunc FR32X:$src)), + (VRNDSCALESSr (f32 (IMPLICIT_DEF)), FR32X:$src, (i32 0x3))>; +def : Pat<(f64 (ftrunc FR64X:$src)), + (VRNDSCALESDr (f64 (IMPLICIT_DEF)), FR64X:$src, (i32 0x3))>; + +def : Pat<(v16f32 (ffloor VR512:$src)), + (VRNDSCALEZPSr VR512:$src, (i32 0x1))>; +def : Pat<(v16f32 (fnearbyint VR512:$src)), + (VRNDSCALEZPSr VR512:$src, (i32 0xC))>; +def : Pat<(v16f32 (fceil VR512:$src)), + (VRNDSCALEZPSr VR512:$src, (i32 0x2))>; +def : Pat<(v16f32 (frint VR512:$src)), + (VRNDSCALEZPSr VR512:$src, (i32 0x4))>; +def : Pat<(v16f32 (ftrunc VR512:$src)), + (VRNDSCALEZPSr VR512:$src, (i32 0x3))>; + +def : Pat<(v8f64 (ffloor VR512:$src)), + (VRNDSCALEZPDr VR512:$src, (i32 0x1))>; +def : Pat<(v8f64 (fnearbyint VR512:$src)), + (VRNDSCALEZPDr VR512:$src, (i32 0xC))>; +def : Pat<(v8f64 (fceil VR512:$src)), + (VRNDSCALEZPDr VR512:$src, (i32 0x2))>; +def : Pat<(v8f64 (frint VR512:$src)), + (VRNDSCALEZPDr VR512:$src, (i32 0x4))>; +def : Pat<(v8f64 (ftrunc VR512:$src)), + (VRNDSCALEZPDr VR512:$src, (i32 0x3))>; + +//------------------------------------------------- +// Integer truncate and extend operations +//------------------------------------------------- + +multiclass avx512_trunc_sat<bits<8> opc, string OpcodeStr, + RegisterClass dstRC, RegisterClass srcRC, + RegisterClass KRC, X86MemOperand x86memop> { + def rr : AVX512XS8I<opc, MRMDestReg, (outs dstRC:$dst), + (ins srcRC:$src), + !strconcat(OpcodeStr,"\t{$src, $dst|$dst, $src}"), + []>, EVEX; + + def krr : AVX512XS8I<opc, MRMDestReg, (outs dstRC:$dst), + (ins KRC:$mask, srcRC:$src), + !strconcat(OpcodeStr, + "\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"), + []>, EVEX, EVEX_KZ; + + def mr : AVX512XS8I<opc, MRMDestMem, (outs), (ins x86memop:$dst, srcRC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), + []>, EVEX; +} +defm VPMOVQB : avx512_trunc_sat<0x32, "vpmovqb", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>; +defm VPMOVSQB : avx512_trunc_sat<0x22, "vpmovsqb", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>; +defm VPMOVUSQB : avx512_trunc_sat<0x12, "vpmovusqb", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VO>; +defm VPMOVQW : avx512_trunc_sat<0x34, "vpmovqw", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<16, CD8VQ>; +defm VPMOVSQW : avx512_trunc_sat<0x24, "vpmovsqw", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<16, CD8VQ>; +defm VPMOVUSQW : avx512_trunc_sat<0x14, "vpmovusqw", VR128X, VR512, VK8WM, + i128mem>, EVEX_V512, EVEX_CD8<16, CD8VQ>; +defm VPMOVQD : avx512_trunc_sat<0x35, "vpmovqd", VR256X, VR512, VK8WM, + i256mem>, EVEX_V512, EVEX_CD8<32, CD8VH>; +defm VPMOVSQD : avx512_trunc_sat<0x25, "vpmovsqd", VR256X, VR512, VK8WM, + i256mem>, EVEX_V512, EVEX_CD8<32, CD8VH>; +defm VPMOVUSQD : avx512_trunc_sat<0x15, "vpmovusqd", VR256X, VR512, VK8WM, + i256mem>, EVEX_V512, EVEX_CD8<32, CD8VH>; +defm VPMOVDW : avx512_trunc_sat<0x33, "vpmovdw", VR256X, VR512, VK16WM, + i256mem>, EVEX_V512, EVEX_CD8<16, CD8VH>; +defm VPMOVSDW : avx512_trunc_sat<0x23, "vpmovsdw", VR256X, VR512, VK16WM, + i256mem>, EVEX_V512, EVEX_CD8<16, CD8VH>; +defm VPMOVUSDW : avx512_trunc_sat<0x13, "vpmovusdw", VR256X, VR512, VK16WM, + i256mem>, EVEX_V512, EVEX_CD8<16, CD8VH>; +defm VPMOVDB : avx512_trunc_sat<0x31, "vpmovdb", VR128X, VR512, VK16WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VQ>; +defm VPMOVSDB : avx512_trunc_sat<0x21, "vpmovsdb", VR128X, VR512, VK16WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VQ>; +defm VPMOVUSDB : avx512_trunc_sat<0x11, "vpmovusdb", VR128X, VR512, VK16WM, + i128mem>, EVEX_V512, EVEX_CD8<8, CD8VQ>; + +def : Pat<(v16i8 (X86vtrunc (v8i64 VR512:$src))), (VPMOVQBrr VR512:$src)>; +def : Pat<(v8i16 (X86vtrunc (v8i64 VR512:$src))), (VPMOVQWrr VR512:$src)>; +def : Pat<(v16i16 (X86vtrunc (v16i32 VR512:$src))), (VPMOVDWrr VR512:$src)>; +def : Pat<(v16i8 (X86vtrunc (v16i32 VR512:$src))), (VPMOVDBrr VR512:$src)>; +def : Pat<(v8i32 (X86vtrunc (v8i64 VR512:$src))), (VPMOVQDrr VR512:$src)>; + +def : Pat<(v16i8 (X86vtruncm VK16WM:$mask, (v16i32 VR512:$src))), + (VPMOVDBkrr VK16WM:$mask, VR512:$src)>; +def : Pat<(v16i16 (X86vtruncm VK16WM:$mask, (v16i32 VR512:$src))), + (VPMOVDWkrr VK16WM:$mask, VR512:$src)>; +def : Pat<(v8i16 (X86vtruncm VK8WM:$mask, (v8i64 VR512:$src))), + (VPMOVQWkrr VK8WM:$mask, VR512:$src)>; +def : Pat<(v8i32 (X86vtruncm VK8WM:$mask, (v8i64 VR512:$src))), + (VPMOVQDkrr VK8WM:$mask, VR512:$src)>; + + +multiclass avx512_extend<bits<8> opc, string OpcodeStr, RegisterClass DstRC, + RegisterClass SrcRC, SDNode OpNode, PatFrag mem_frag, + X86MemOperand x86memop, ValueType OpVT, ValueType InVT> { + + def rr : AVX5128I<opc, MRMSrcReg, (outs DstRC:$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), + (ins x86memop:$src), + !strconcat(OpcodeStr,"\t{$src, $dst|$dst, $src}"), + [(set DstRC:$dst, + (OpVT (OpNode (InVT (bitconvert (mem_frag addr:$src))))))]>, + EVEX; +} + +defm VPMOVZXBDZ: avx512_extend<0x31, "vpmovzxbd", VR512, VR128X, X86vzext, + memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512, + EVEX_CD8<8, CD8VQ>; +defm VPMOVZXBQZ: avx512_extend<0x32, "vpmovzxbq", VR512, VR128X, X86vzext, + memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512, + EVEX_CD8<8, CD8VO>; +defm VPMOVZXWDZ: avx512_extend<0x33, "vpmovzxwd", VR512, VR256X, X86vzext, + memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512, + EVEX_CD8<16, CD8VH>; +defm VPMOVZXWQZ: avx512_extend<0x34, "vpmovzxwq", VR512, VR128X, X86vzext, + memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512, + EVEX_CD8<16, CD8VQ>; +defm VPMOVZXDQZ: avx512_extend<0x35, "vpmovzxdq", VR512, VR256X, X86vzext, + memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512, + EVEX_CD8<32, CD8VH>; + +defm VPMOVSXBDZ: avx512_extend<0x21, "vpmovsxbd", VR512, VR128X, X86vsext, + memopv2i64, i128mem, v16i32, v16i8>, EVEX_V512, + EVEX_CD8<8, CD8VQ>; +defm VPMOVSXBQZ: avx512_extend<0x22, "vpmovsxbq", VR512, VR128X, X86vsext, + memopv2i64, i128mem, v8i64, v16i8>, EVEX_V512, + EVEX_CD8<8, CD8VO>; +defm VPMOVSXWDZ: avx512_extend<0x23, "vpmovsxwd", VR512, VR256X, X86vsext, + memopv4i64, i256mem, v16i32, v16i16>, EVEX_V512, + EVEX_CD8<16, CD8VH>; +defm VPMOVSXWQZ: avx512_extend<0x24, "vpmovsxwq", VR512, VR128X, X86vsext, + memopv2i64, i128mem, v8i64, v8i16>, EVEX_V512, + EVEX_CD8<16, CD8VQ>; +defm VPMOVSXDQZ: avx512_extend<0x25, "vpmovsxdq", VR512, VR256X, X86vsext, + memopv4i64, i256mem, v8i64, v8i32>, EVEX_V512, + EVEX_CD8<32, CD8VH>; + +//===----------------------------------------------------------------------===// +// GATHER - SCATTER Operations + +multiclass avx512_gather<bits<8> opc, string OpcodeStr, RegisterClass KRC, + RegisterClass RC, X86MemOperand memop> { +let mayLoad = 1, + Constraints = "@earlyclobber $dst, $src1 = $dst, $mask = $mask_wb" in + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst, KRC:$mask_wb), + (ins RC:$src1, KRC:$mask, memop:$src2), + !strconcat(OpcodeStr, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + []>, EVEX, EVEX_K; +} +defm VGATHERDPDZ : avx512_gather<0x92, "vgatherdpd", VK8WM, VR512, vy64xmem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VGATHERDPSZ : avx512_gather<0x92, "vgatherdps", VK16WM, VR512, vz32mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VGATHERQPDZ : avx512_gather<0x93, "vgatherqpd", VK8WM, VR512, vz64mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VGATHERQPSZ : avx512_gather<0x93, "vgatherqps", VK8WM, VR256X, vz64mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VPGATHERDQZ : avx512_gather<0x90, "vpgatherdq", VK8WM, VR512, vy64xmem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VPGATHERDDZ : avx512_gather<0x90, "vpgatherdd", VK16WM, VR512, vz32mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VPGATHERQQZ : avx512_gather<0x91, "vpgatherqq", VK8WM, VR512, vz64mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VPGATHERQDZ : avx512_gather<0x91, "vpgatherqd", VK8WM, VR256X, vz64mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +multiclass avx512_scatter<bits<8> opc, string OpcodeStr, RegisterClass KRC, + RegisterClass RC, X86MemOperand memop> { +let mayStore = 1, Constraints = "$mask = $mask_wb" in + def mr : AVX5128I<opc, MRMDestMem, (outs KRC:$mask_wb), + (ins memop:$dst, KRC:$mask, RC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + []>, EVEX, EVEX_K; +} + +defm VSCATTERDPDZ : avx512_scatter<0xA2, "vscatterdpd", VK8WM, VR512, vy64xmem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VSCATTERDPSZ : avx512_scatter<0xA2, "vscatterdps", VK16WM, VR512, vz32mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VSCATTERQPDZ : avx512_scatter<0xA3, "vscatterqpd", VK8WM, VR512, vz64mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VSCATTERQPSZ : avx512_scatter<0xA3, "vscatterqps", VK8WM, VR256X, vz64mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VPSCATTERDQZ : avx512_scatter<0xA0, "vpscatterdq", VK8WM, VR512, vy64xmem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VPSCATTERDDZ : avx512_scatter<0xA0, "vpscatterdd", VK16WM, VR512, vz32mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +defm VPSCATTERQQZ : avx512_scatter<0xA1, "vpscatterqq", VK8WM, VR512, vz64mem>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VT1>; +defm VPSCATTERQDZ : avx512_scatter<0xA1, "vpscatterqd", VK8WM, VR256X, vz64mem>, + EVEX_V512, EVEX_CD8<32, CD8VT1>; + +//===----------------------------------------------------------------------===// +// VSHUFPS - VSHUFPD Operations + +multiclass avx512_shufp<RegisterClass RC, X86MemOperand x86memop, + ValueType vt, string OpcodeStr, PatFrag mem_frag, + Domain d> { + def rmi : AVX512PIi8<0xC6, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2, i8imm:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2), + (i8 imm:$src3))))], d, IIC_SSE_SHUFP>, + EVEX_4V, Sched<[WriteShuffleLd, ReadAfterLd]>; + def rri : AVX512PIi8<0xC6, MRMSrcReg, (outs RC:$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:$src3))))], d, IIC_SSE_SHUFP>, + EVEX_4V, Sched<[WriteShuffle]>; +} + +defm VSHUFPSZ : avx512_shufp<VR512, f512mem, v16f32, "vshufps", memopv16f32, + SSEPackedSingle>, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VSHUFPDZ : avx512_shufp<VR512, f512mem, v8f64, "vshufpd", memopv8f64, + SSEPackedDouble>, OpSize, VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; + +def : Pat<(v16i32 (X86Shufp VR512:$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:$imm)>; + +multiclass avx512_alignr<string OpcodeStr, RegisterClass RC, + X86MemOperand x86memop> { + def rri : AVX512AIi8<0x03, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, RC:$src2, i8imm:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + []>, EVEX_4V; + let mayLoad = 1 in + def rmi : AVX512AIi8<0x03, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, x86memop:$src2, i8imm:$src3), + !strconcat(OpcodeStr, + "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"), + []>, EVEX_4V; +} +defm VALIGND : avx512_alignr<"valignd", VR512, i512mem>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VALIGNQ : avx512_alignr<"valignq", VR512, i512mem>, + VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>; + +def : Pat<(v16f32 (X86PAlignr VR512:$src1, VR512:$src2, (i8 imm:$imm))), + (VALIGNDrri VR512:$src2, VR512:$src1, imm:$imm)>; +def : Pat<(v8f64 (X86PAlignr VR512:$src1, VR512:$src2, (i8 imm:$imm))), + (VALIGNQrri VR512:$src2, VR512:$src1, imm:$imm)>; +def : Pat<(v16i32 (X86PAlignr VR512:$src1, VR512:$src2, (i8 imm:$imm))), + (VALIGNDrri VR512:$src2, VR512:$src1, imm:$imm)>; +def : Pat<(v8i64 (X86PAlignr VR512:$src1, VR512:$src2, (i8 imm:$imm))), + (VALIGNQrri VR512:$src2, VR512:$src1, imm:$imm)>; + +multiclass avx512_vpabs<bits<8> opc, string OpcodeStr, RegisterClass RC, + X86MemOperand x86memop> { + def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>, + EVEX; + def rm : AVX5128I<opc, MRMSrcMem, (outs VR512:$dst), + (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>, + EVEX; +} + +defm VPABSD : avx512_vpabs<0x1E, "vpabsd", VR512, i512mem>, EVEX_V512, + EVEX_CD8<32, CD8VF>; +defm VPABSQ : avx512_vpabs<0x1F, "vpabsq", VR512, i512mem>, EVEX_V512, VEX_W, + EVEX_CD8<64, CD8VF>; + +multiclass avx512_conflict<bits<8> opc, string OpcodeStr, + RegisterClass RC, RegisterClass KRC, PatFrag memop_frag, + X86MemOperand x86memop, PatFrag scalar_mfrag, + X86MemOperand x86scalar_mop, string BrdcstStr, + Intrinsic Int, Intrinsic maskInt, Intrinsic maskzInt> { + def rr : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src), + !strconcat(OpcodeStr, "\t{$src, ${dst} |${dst}, $src}"), + [(set RC:$dst, (Int RC:$src))]>, EVEX; + def rm : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins x86memop:$src), + !strconcat(OpcodeStr, "\t{$src, ${dst}|${dst}, $src}"), + [(set RC:$dst, (Int (memop_frag addr:$src)))]>, EVEX; + def rmb : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins x86scalar_mop:$src), + !strconcat(OpcodeStr, "\t{${src}", BrdcstStr, + ", ${dst}|${dst}, ${src}", BrdcstStr, "}"), + []>, EVEX, EVEX_B; + def rrkz : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins KRC:$mask, RC:$src), + !strconcat(OpcodeStr, + "\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"), + [(set RC:$dst, (maskzInt KRC:$mask, RC:$src))]>, EVEX, EVEX_KZ; + def rmkz : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins KRC:$mask, x86memop:$src), + !strconcat(OpcodeStr, + "\t{$src, ${dst} {${mask}} {z}|${dst} {${mask}} {z}, $src}"), + [(set RC:$dst, (maskzInt KRC:$mask, (memop_frag addr:$src)))]>, + EVEX, EVEX_KZ; + def rmbkz : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins KRC:$mask, x86scalar_mop:$src), + !strconcat(OpcodeStr, "\t{${src}", BrdcstStr, + ", ${dst} {${mask}} {z}|${dst} {${mask}} {z}, ${src}", + BrdcstStr, "}"), + []>, EVEX, EVEX_KZ, EVEX_B; + + let Constraints = "$src1 = $dst" in { + def rrk : AVX5128I<opc, MRMSrcReg, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, RC:$src2), + !strconcat(OpcodeStr, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + [(set RC:$dst, (maskInt RC:$src1, KRC:$mask, RC:$src2))]>, EVEX, EVEX_K; + def rmk : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, x86memop:$src2), + !strconcat(OpcodeStr, + "\t{$src2, ${dst} {${mask}}|${dst} {${mask}}, $src2}"), + [(set RC:$dst, (maskInt RC:$src1, KRC:$mask, (memop_frag addr:$src2)))]>, EVEX, EVEX_K; + def rmbk : AVX5128I<opc, MRMSrcMem, (outs RC:$dst), + (ins RC:$src1, KRC:$mask, x86scalar_mop:$src2), + !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr, + ", ${dst} {${mask}}|${dst} {${mask}}, ${src2}", BrdcstStr, "}"), + []>, EVEX, EVEX_K, EVEX_B; + } +} + +let Predicates = [HasCDI] in { +defm VPCONFLICTD : avx512_conflict<0xC4, "vpconflictd", VR512, VK16WM, + memopv16i32, i512mem, loadi32, i32mem, "{1to16}", + int_x86_avx512_conflict_d_512, + int_x86_avx512_conflict_d_mask_512, + int_x86_avx512_conflict_d_maskz_512>, + EVEX_V512, EVEX_CD8<32, CD8VF>; + +defm VPCONFLICTQ : avx512_conflict<0xC4, "vpconflictq", VR512, VK8WM, + memopv8i64, i512mem, loadi64, i64mem, "{1to8}", + int_x86_avx512_conflict_q_512, + int_x86_avx512_conflict_q_mask_512, + int_x86_avx512_conflict_q_maskz_512>, + EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; +} |
