318 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL,AVX512VL-FAST-ALL4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512VL,AVX512VL-FAST-PERLANE5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW,AVX512BW-FAST-ALL6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BW,AVX512BW-FAST-PERLANE7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL,AVX512BWVL-FAST-ALL8; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=AVX512,AVX512BWVL,AVX512BWVL-FAST-PERLANE9 10define void @shuffle_v64i8_to_v32i8_1(ptr %L, ptr %S) nounwind {11; AVX512F-LABEL: shuffle_v64i8_to_v32i8_1:12; AVX512F: # %bb.0:13; AVX512F-NEXT: vmovdqa (%rdi), %ymm014; AVX512F-NEXT: vmovdqa 32(%rdi), %ymm115; AVX512F-NEXT: vpbroadcastq {{.*#+}} ymm2 = [1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15]16; AVX512F-NEXT: vpshufb %ymm2, %ymm1, %ymm117; AVX512F-NEXT: vpshufb %ymm2, %ymm0, %ymm018; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]19; AVX512F-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]20; AVX512F-NEXT: vmovdqa %ymm0, (%rsi)21; AVX512F-NEXT: vzeroupper22; AVX512F-NEXT: retq23;24; AVX512VL-FAST-ALL-LABEL: shuffle_v64i8_to_v32i8_1:25; AVX512VL-FAST-ALL: # %bb.0:26; AVX512VL-FAST-ALL-NEXT: vmovdqa (%rdi), %ymm027; AVX512VL-FAST-ALL-NEXT: vmovdqa 32(%rdi), %ymm128; AVX512VL-FAST-ALL-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[u,u,u,u,u,u,u,u,1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31]29; AVX512VL-FAST-ALL-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[1,3,5,7,9,11,13,15,u,u,u,u,u,u,u,u,17,19,21,23,25,27,29,31,u,u,u,u,u,u,u,u]30; AVX512VL-FAST-ALL-NEXT: vpmovsxbq {{.*#+}} ymm2 = [0,2,5,7]31; AVX512VL-FAST-ALL-NEXT: vpermi2q %ymm1, %ymm0, %ymm232; AVX512VL-FAST-ALL-NEXT: vmovdqa %ymm2, (%rsi)33; AVX512VL-FAST-ALL-NEXT: vzeroupper34; AVX512VL-FAST-ALL-NEXT: retq35;36; AVX512VL-FAST-PERLANE-LABEL: shuffle_v64i8_to_v32i8_1:37; AVX512VL-FAST-PERLANE: # %bb.0:38; AVX512VL-FAST-PERLANE-NEXT: vmovdqa (%rdi), %ymm039; AVX512VL-FAST-PERLANE-NEXT: vmovdqa 32(%rdi), %ymm140; AVX512VL-FAST-PERLANE-NEXT: vpbroadcastq {{.*#+}} ymm2 = [1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15,1,3,5,7,9,11,13,15]41; AVX512VL-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm1, %ymm142; AVX512VL-FAST-PERLANE-NEXT: vpshufb %ymm2, %ymm0, %ymm043; AVX512VL-FAST-PERLANE-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]44; AVX512VL-FAST-PERLANE-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]45; AVX512VL-FAST-PERLANE-NEXT: vmovdqa %ymm0, (%rsi)46; AVX512VL-FAST-PERLANE-NEXT: vzeroupper47; AVX512VL-FAST-PERLANE-NEXT: retq48;49; AVX512BW-LABEL: shuffle_v64i8_to_v32i8_1:50; AVX512BW: # %bb.0:51; AVX512BW-NEXT: vpsrlw $8, (%rdi), %zmm052; AVX512BW-NEXT: vpmovwb %zmm0, (%rsi)53; AVX512BW-NEXT: vzeroupper54; AVX512BW-NEXT: retq55;56; AVX512BWVL-LABEL: shuffle_v64i8_to_v32i8_1:57; AVX512BWVL: # %bb.0:58; AVX512BWVL-NEXT: vpsrlw $8, (%rdi), %zmm059; AVX512BWVL-NEXT: vpmovwb %zmm0, (%rsi)60; AVX512BWVL-NEXT: vzeroupper61; AVX512BWVL-NEXT: retq62 %vec = load <64 x i8>, ptr %L63 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <32 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31, i32 33, i32 35, i32 37, i32 39, i32 41, i32 43, i32 45, i32 47, i32 49, i32 51, i32 53, i32 55, i32 57, i32 59, i32 61, i32 63>64 store <32 x i8> %strided.vec, ptr %S65 ret void66}67 68define void @shuffle_v32i16_to_v16i16_1(ptr %L, ptr %S) nounwind {69; AVX512-LABEL: shuffle_v32i16_to_v16i16_1:70; AVX512: # %bb.0:71; AVX512-NEXT: vpsrld $16, (%rdi), %zmm072; AVX512-NEXT: vpmovdw %zmm0, (%rsi)73; AVX512-NEXT: vzeroupper74; AVX512-NEXT: retq75 %vec = load <32 x i16>, ptr %L76 %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15, i32 17, i32 19, i32 21, i32 23, i32 25, i32 27, i32 29, i32 31>77 store <16 x i16> %strided.vec, ptr %S78 ret void79}80 81define void @shuffle_v16i32_to_v8i32_1(ptr %L, ptr %S) nounwind {82; AVX512F-LABEL: shuffle_v16i32_to_v8i32_1:83; AVX512F: # %bb.0:84; AVX512F-NEXT: vmovaps (%rdi), %ymm085; AVX512F-NEXT: vshufps {{.*#+}} ymm0 = ymm0[1,3],mem[1,3],ymm0[5,7],mem[5,7]86; AVX512F-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]87; AVX512F-NEXT: vmovaps %ymm0, (%rsi)88; AVX512F-NEXT: vzeroupper89; AVX512F-NEXT: retq90;91; AVX512VL-FAST-ALL-LABEL: shuffle_v16i32_to_v8i32_1:92; AVX512VL-FAST-ALL: # %bb.0:93; AVX512VL-FAST-ALL-NEXT: vpmovsxbd {{.*#+}} ymm0 = [1,3,5,7,9,11,13,15]94; AVX512VL-FAST-ALL-NEXT: vpermps (%rdi), %zmm0, %zmm095; AVX512VL-FAST-ALL-NEXT: vmovaps %ymm0, (%rsi)96; AVX512VL-FAST-ALL-NEXT: vzeroupper97; AVX512VL-FAST-ALL-NEXT: retq98;99; AVX512VL-FAST-PERLANE-LABEL: shuffle_v16i32_to_v8i32_1:100; AVX512VL-FAST-PERLANE: # %bb.0:101; AVX512VL-FAST-PERLANE-NEXT: vmovaps (%rdi), %ymm0102; AVX512VL-FAST-PERLANE-NEXT: vshufps {{.*#+}} ymm0 = ymm0[1,3],mem[1,3],ymm0[5,7],mem[5,7]103; AVX512VL-FAST-PERLANE-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]104; AVX512VL-FAST-PERLANE-NEXT: vmovaps %ymm0, (%rsi)105; AVX512VL-FAST-PERLANE-NEXT: vzeroupper106; AVX512VL-FAST-PERLANE-NEXT: retq107;108; AVX512BW-FAST-ALL-LABEL: shuffle_v16i32_to_v8i32_1:109; AVX512BW-FAST-ALL: # %bb.0:110; AVX512BW-FAST-ALL-NEXT: vpmovsxbd {{.*#+}} ymm0 = [1,3,5,7,9,11,13,15]111; AVX512BW-FAST-ALL-NEXT: vpermps (%rdi), %zmm0, %zmm0112; AVX512BW-FAST-ALL-NEXT: vmovaps %ymm0, (%rsi)113; AVX512BW-FAST-ALL-NEXT: vzeroupper114; AVX512BW-FAST-ALL-NEXT: retq115;116; AVX512BW-FAST-PERLANE-LABEL: shuffle_v16i32_to_v8i32_1:117; AVX512BW-FAST-PERLANE: # %bb.0:118; AVX512BW-FAST-PERLANE-NEXT: vmovaps (%rdi), %ymm0119; AVX512BW-FAST-PERLANE-NEXT: vshufps {{.*#+}} ymm0 = ymm0[1,3],mem[1,3],ymm0[5,7],mem[5,7]120; AVX512BW-FAST-PERLANE-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]121; AVX512BW-FAST-PERLANE-NEXT: vmovaps %ymm0, (%rsi)122; AVX512BW-FAST-PERLANE-NEXT: vzeroupper123; AVX512BW-FAST-PERLANE-NEXT: retq124;125; AVX512BWVL-FAST-ALL-LABEL: shuffle_v16i32_to_v8i32_1:126; AVX512BWVL-FAST-ALL: # %bb.0:127; AVX512BWVL-FAST-ALL-NEXT: vpmovsxbd {{.*#+}} ymm0 = [1,3,5,7,9,11,13,15]128; AVX512BWVL-FAST-ALL-NEXT: vpermps (%rdi), %zmm0, %zmm0129; AVX512BWVL-FAST-ALL-NEXT: vmovaps %ymm0, (%rsi)130; AVX512BWVL-FAST-ALL-NEXT: vzeroupper131; AVX512BWVL-FAST-ALL-NEXT: retq132;133; AVX512BWVL-FAST-PERLANE-LABEL: shuffle_v16i32_to_v8i32_1:134; AVX512BWVL-FAST-PERLANE: # %bb.0:135; AVX512BWVL-FAST-PERLANE-NEXT: vmovaps (%rdi), %ymm0136; AVX512BWVL-FAST-PERLANE-NEXT: vshufps {{.*#+}} ymm0 = ymm0[1,3],mem[1,3],ymm0[5,7],mem[5,7]137; AVX512BWVL-FAST-PERLANE-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]138; AVX512BWVL-FAST-PERLANE-NEXT: vmovaps %ymm0, (%rsi)139; AVX512BWVL-FAST-PERLANE-NEXT: vzeroupper140; AVX512BWVL-FAST-PERLANE-NEXT: retq141 %vec = load <16 x i32>, ptr %L142 %strided.vec = shufflevector <16 x i32> %vec, <16 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>143 store <8 x i32> %strided.vec, ptr %S144 ret void145}146 147define void @shuffle_v64i8_to_v16i8_1(ptr %L, ptr %S) nounwind {148; AVX512-LABEL: shuffle_v64i8_to_v16i8_1:149; AVX512: # %bb.0:150; AVX512-NEXT: vpsrld $8, (%rdi), %zmm0151; AVX512-NEXT: vpmovdb %zmm0, (%rsi)152; AVX512-NEXT: vzeroupper153; AVX512-NEXT: retq154 %vec = load <64 x i8>, ptr %L155 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 33, i32 37, i32 41, i32 45, i32 49, i32 53, i32 57, i32 61>156 store <16 x i8> %strided.vec, ptr %S157 ret void158}159 160define void @shuffle_v64i8_to_v16i8_2(ptr %L, ptr %S) nounwind {161; AVX512-LABEL: shuffle_v64i8_to_v16i8_2:162; AVX512: # %bb.0:163; AVX512-NEXT: vpsrld $16, (%rdi), %zmm0164; AVX512-NEXT: vpmovdb %zmm0, (%rsi)165; AVX512-NEXT: vzeroupper166; AVX512-NEXT: retq167 %vec = load <64 x i8>, ptr %L168 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 34, i32 38, i32 42, i32 46, i32 50, i32 54, i32 58, i32 62>169 store <16 x i8> %strided.vec, ptr %S170 ret void171}172 173define void @shuffle_v64i8_to_v16i8_3(ptr %L, ptr %S) nounwind {174; AVX512-LABEL: shuffle_v64i8_to_v16i8_3:175; AVX512: # %bb.0:176; AVX512-NEXT: vpsrld $24, (%rdi), %zmm0177; AVX512-NEXT: vpmovdb %zmm0, (%rsi)178; AVX512-NEXT: vzeroupper179; AVX512-NEXT: retq180 %vec = load <64 x i8>, ptr %L181 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <16 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31, i32 35, i32 39, i32 43, i32 47, i32 51, i32 55, i32 59, i32 63>182 store <16 x i8> %strided.vec, ptr %S183 ret void184}185 186define void @shuffle_v32i16_to_v8i16_1(ptr %L, ptr %S) nounwind {187; AVX512-LABEL: shuffle_v32i16_to_v8i16_1:188; AVX512: # %bb.0:189; AVX512-NEXT: vpsrlq $16, (%rdi), %zmm0190; AVX512-NEXT: vpmovqw %zmm0, (%rsi)191; AVX512-NEXT: vzeroupper192; AVX512-NEXT: retq193 %vec = load <32 x i16>, ptr %L194 %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29>195 store <8 x i16> %strided.vec, ptr %S196 ret void197}198 199define void @shuffle_v32i16_to_v8i16_2(ptr %L, ptr %S) nounwind {200; AVX512-LABEL: shuffle_v32i16_to_v8i16_2:201; AVX512: # %bb.0:202; AVX512-NEXT: vpsrlq $32, (%rdi), %zmm0203; AVX512-NEXT: vpmovqw %zmm0, (%rsi)204; AVX512-NEXT: vzeroupper205; AVX512-NEXT: retq206 %vec = load <32 x i16>, ptr %L207 %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>208 store <8 x i16> %strided.vec, ptr %S209 ret void210}211 212define void @shuffle_v32i16_to_v8i16_3(ptr %L, ptr %S) nounwind {213; AVX512-LABEL: shuffle_v32i16_to_v8i16_3:214; AVX512: # %bb.0:215; AVX512-NEXT: vpsrlq $48, (%rdi), %zmm0216; AVX512-NEXT: vpmovqw %zmm0, (%rsi)217; AVX512-NEXT: vzeroupper218; AVX512-NEXT: retq219 %vec = load <32 x i16>, ptr %L220 %strided.vec = shufflevector <32 x i16> %vec, <32 x i16> undef, <8 x i32> <i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>221 store <8 x i16> %strided.vec, ptr %S222 ret void223}224 225define void @shuffle_v64i8_to_v8i8_1(ptr %L, ptr %S) nounwind {226; AVX512-LABEL: shuffle_v64i8_to_v8i8_1:227; AVX512: # %bb.0:228; AVX512-NEXT: vpsrlq $8, (%rdi), %zmm0229; AVX512-NEXT: vpmovqb %zmm0, (%rsi)230; AVX512-NEXT: vzeroupper231; AVX512-NEXT: retq232 %vec = load <64 x i8>, ptr %L233 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 1, i32 9, i32 17, i32 25, i32 33, i32 41, i32 49, i32 57>234 store <8 x i8> %strided.vec, ptr %S235 ret void236}237 238define void @shuffle_v64i8_to_v8i8_2(ptr %L, ptr %S) nounwind {239; AVX512-LABEL: shuffle_v64i8_to_v8i8_2:240; AVX512: # %bb.0:241; AVX512-NEXT: vpsrlq $16, (%rdi), %zmm0242; AVX512-NEXT: vpmovqb %zmm0, (%rsi)243; AVX512-NEXT: vzeroupper244; AVX512-NEXT: retq245 %vec = load <64 x i8>, ptr %L246 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 2, i32 10, i32 18, i32 26, i32 34, i32 42, i32 50, i32 58>247 store <8 x i8> %strided.vec, ptr %S248 ret void249}250 251define void @shuffle_v64i8_to_v8i8_3(ptr %L, ptr %S) nounwind {252; AVX512-LABEL: shuffle_v64i8_to_v8i8_3:253; AVX512: # %bb.0:254; AVX512-NEXT: vpsrlq $24, (%rdi), %zmm0255; AVX512-NEXT: vpmovqb %zmm0, (%rsi)256; AVX512-NEXT: vzeroupper257; AVX512-NEXT: retq258 %vec = load <64 x i8>, ptr %L259 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 3, i32 11, i32 19, i32 27, i32 35, i32 43, i32 51, i32 59>260 store <8 x i8> %strided.vec, ptr %S261 ret void262}263 264define void @shuffle_v64i8_to_v8i8_4(ptr %L, ptr %S) nounwind {265; AVX512-LABEL: shuffle_v64i8_to_v8i8_4:266; AVX512: # %bb.0:267; AVX512-NEXT: vpsrlq $32, (%rdi), %zmm0268; AVX512-NEXT: vpmovqb %zmm0, (%rsi)269; AVX512-NEXT: vzeroupper270; AVX512-NEXT: retq271 %vec = load <64 x i8>, ptr %L272 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 4, i32 12, i32 20, i32 28, i32 36, i32 44, i32 52, i32 60>273 store <8 x i8> %strided.vec, ptr %S274 ret void275}276 277define void @shuffle_v64i8_to_v8i8_5(ptr %L, ptr %S) nounwind {278; AVX512-LABEL: shuffle_v64i8_to_v8i8_5:279; AVX512: # %bb.0:280; AVX512-NEXT: vpsrlq $40, (%rdi), %zmm0281; AVX512-NEXT: vpmovqb %zmm0, (%rsi)282; AVX512-NEXT: vzeroupper283; AVX512-NEXT: retq284 %vec = load <64 x i8>, ptr %L285 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 5, i32 13, i32 21, i32 29, i32 37, i32 45, i32 53, i32 61>286 store <8 x i8> %strided.vec, ptr %S287 ret void288}289 290define void @shuffle_v64i8_to_v8i8_6(ptr %L, ptr %S) nounwind {291; AVX512-LABEL: shuffle_v64i8_to_v8i8_6:292; AVX512: # %bb.0:293; AVX512-NEXT: vpsrlq $48, (%rdi), %zmm0294; AVX512-NEXT: vpmovqb %zmm0, (%rsi)295; AVX512-NEXT: vzeroupper296; AVX512-NEXT: retq297 %vec = load <64 x i8>, ptr %L298 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 6, i32 14, i32 22, i32 30, i32 38, i32 46, i32 54, i32 62>299 store <8 x i8> %strided.vec, ptr %S300 ret void301}302 303define void @shuffle_v64i8_to_v8i8_7(ptr %L, ptr %S) nounwind {304; AVX512-LABEL: shuffle_v64i8_to_v8i8_7:305; AVX512: # %bb.0:306; AVX512-NEXT: vpsrlq $56, (%rdi), %zmm0307; AVX512-NEXT: vpmovqb %zmm0, (%rsi)308; AVX512-NEXT: vzeroupper309; AVX512-NEXT: retq310 %vec = load <64 x i8>, ptr %L311 %strided.vec = shufflevector <64 x i8> %vec, <64 x i8> undef, <8 x i32> <i32 7, i32 15, i32 23, i32 31, i32 39, i32 47, i32 55, i32 63>312 store <8 x i8> %strided.vec, ptr %S313 ret void314}315 316;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:317; AVX512VL: {{.*}}318