[X86] lowerShuffleAsVTRUNC - ensure we peek through bitcasts when looking for freely-concatable subvectors

Fixes #111611

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -10356,7 +10356,7 @@ static SDValue lowerShuffleAsVTRUNC(const SDLoc &DL, MVT VT, SDValue V1,
           }
           return false;
         };
-        if (!IsCheapConcat(V1, V2))
+        if (!IsCheapConcat(peekThroughBitcasts(V1), peekThroughBitcasts(V2)))
           continue;
       }
 

llvm/test/CodeGen/X86/shuffle-vs-trunc-128.ll

@@ -797,69 +797,12 @@ define <16 x i8> @oddelts_v32i16_shuffle_v16i16_to_v16i8(<32 x i16> %n2) nounwin
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512F-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
-; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512F-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512F-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2,6,10,14,2,6,10,14,2,6,10,14,2,6,10,14]
-; AVX512F-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512F-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX512F-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; AVX512F-NEXT:    vpsrld $16, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512F-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512F-NEXT:    vzeroupper
-; AVX512F-NEXT:    retq
-;
-; AVX512VL-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512VL-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512VL-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2,6,10,14,2,6,10,14,2,6,10,14,2,6,10,14]
-; AVX512VL-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512VL-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX512VL-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; AVX512VL-NEXT:    vpsrld $16, %ymm0, %ymm0
-; AVX512VL-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512VL-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512VL-NEXT:    vzeroupper
-; AVX512VL-NEXT:    retq
-;
-; AVX512BW-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BW-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2,6,10,14,2,6,10,14,2,6,10,14,2,6,10,14]
-; AVX512BW-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512BW-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX512BW-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; AVX512BW-NEXT:    vpsrld $16, %ymm0, %ymm0
-; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512BW-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512BW-NEXT:    vzeroupper
-; AVX512BW-NEXT:    retq
-;
-; AVX512BWVL-ONLY-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
-; AVX512BWVL-ONLY:       # %bb.0:
-; AVX512BWVL-ONLY-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BWVL-ONLY-NEXT:    vextracti128 $1, %ymm1, %xmm2
-; AVX512BWVL-ONLY-NEXT:    vpbroadcastd {{.*#+}} xmm3 = [2,6,10,14,2,6,10,14,2,6,10,14,2,6,10,14]
-; AVX512BWVL-ONLY-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX512BWVL-ONLY-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
-; AVX512BWVL-ONLY-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; AVX512BWVL-ONLY-NEXT:    vpsrld $16, %ymm0, %ymm0
-; AVX512BWVL-ONLY-NEXT:    vpmovdb %ymm0, %xmm0
-; AVX512BWVL-ONLY-NEXT:    vpblendd {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3]
-; AVX512BWVL-ONLY-NEXT:    vzeroupper
-; AVX512BWVL-ONLY-NEXT:    retq
-;
-; AVX512VBMI-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
-; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm1 = [2,6,10,14,18,22,26,30,34,38,42,46,50,54,58,62]
-; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm0
-; AVX512VBMI-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
-; AVX512VBMI-NEXT:    vzeroupper
-; AVX512VBMI-NEXT:    retq
+; AVX512-LABEL: oddelts_v32i16_shuffle_v16i16_to_v16i8:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpsrld $16, %zmm0, %zmm0
+; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512-NEXT:    vzeroupper
+; AVX512-NEXT:    retq
   %n0 = bitcast <32 x i16> %n2 to <64 x i8>
   %p = shufflevector <64 x i8> %n0, <64 x i8> poison, <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>
   ret <16 x i8> %p
@@ -1548,3 +1491,5 @@ define <16 x i8> @oddelts_v32i16_trunc_v16i16_to_v16i8(<32 x i16> %n2) nounwind
   %n1 = trunc <16 x i16> %n0 to <16 x i8>
   ret <16 x i8> %n1
 }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; AVX512VBMI: {{.*}}

llvm/test/CodeGen/X86/shuffle-vs-trunc-512.ll

@@ -389,57 +389,28 @@ define <32 x i8> @trunc_shuffle_v32i16_v32i8_ofs1(<32 x i16> %a0) {
 ; AVX512VL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
 ; AVX512VL-FAST-PERLANE-NEXT:    retq
 ;
-; AVX512BW-FAST-ALL-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
-; AVX512BW-FAST-ALL:       # %bb.0:
-; AVX512BW-FAST-ALL-NEXT:    vpmovsxbq {{.*#+}} ymm1 = [0,2,9,11]
-; AVX512BW-FAST-ALL-NEXT:    vextracti64x4 $1, %zmm0, %ymm2
-; AVX512BW-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[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]
-; AVX512BW-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]
-; AVX512BW-FAST-ALL-NEXT:    vpermt2q %zmm2, %zmm1, %zmm0
-; AVX512BW-FAST-ALL-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
-; AVX512BW-FAST-ALL-NEXT:    retq
-;
-; AVX512BW-FAST-PERLANE-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
-; AVX512BW-FAST-PERLANE:       # %bb.0:
-; AVX512BW-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BW-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]
-; AVX512BW-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
-; AVX512BW-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
-; AVX512BW-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX512BW-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX512BW-FAST-PERLANE-NEXT:    retq
-;
-; AVX512BWVL-FAST-ALL-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
-; AVX512BWVL-FAST-ALL:       # %bb.0:
-; AVX512BWVL-FAST-ALL-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BWVL-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]
-; AVX512BWVL-FAST-ALL-NEXT:    vpshufb {{.*#+}} ymm2 = 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]
-; AVX512BWVL-FAST-ALL-NEXT:    vpmovsxbq {{.*#+}} ymm0 = [0,2,5,7]
-; AVX512BWVL-FAST-ALL-NEXT:    vpermi2q %ymm1, %ymm2, %ymm0
-; AVX512BWVL-FAST-ALL-NEXT:    retq
+; AVX512BW-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BW-NEXT:    retq
 ;
-; AVX512BWVL-FAST-PERLANE-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
-; AVX512BWVL-FAST-PERLANE:       # %bb.0:
-; AVX512BWVL-FAST-PERLANE-NEXT:    vextracti64x4 $1, %zmm0, %ymm1
-; AVX512BWVL-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]
-; AVX512BWVL-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm1, %ymm1
-; AVX512BWVL-FAST-PERLANE-NEXT:    vpshufb %ymm2, %ymm0, %ymm0
-; AVX512BWVL-FAST-PERLANE-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7]
-; AVX512BWVL-FAST-PERLANE-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX512BWVL-FAST-PERLANE-NEXT:    retq
+; AVX512BWVL-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
+; AVX512BWVL:       # %bb.0:
+; AVX512BWVL-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512BWVL-NEXT:    vpmovwb %zmm0, %ymm0
+; AVX512BWVL-NEXT:    retq
 ;
 ; AVX512VBMI-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,5,7,9,11,13,15,97,99,101,103,105,107,109,111,17,19,21,23,25,27,29,31,113,115,117,119,121,123,125,127]
-; AVX512VBMI-NEXT:    vpermi2b %zmm0, %zmm0, %zmm1
-; AVX512VBMI-NEXT:    vpermq {{.*#+}} ymm0 = ymm1[0,2,1,3]
+; AVX512VBMI-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VBMI-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMI-NEXT:    retq
 ;
 ; AVX512VBMIVL-LABEL: trunc_shuffle_v32i16_v32i8_ofs1:
 ; AVX512VBMIVL:       # %bb.0:
-; AVX512VBMIVL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,45,47,49,51,53,55,57,59,61,63]
-; AVX512VBMIVL-NEXT:    vpermb %zmm0, %zmm1, %zmm0
-; AVX512VBMIVL-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512VBMIVL-NEXT:    vpsrlw $8, %zmm0, %zmm0
+; AVX512VBMIVL-NEXT:    vpmovwb %zmm0, %ymm0
 ; AVX512VBMIVL-NEXT:    retq
   %bc = bitcast <32 x i16> %a0 to <64 x i8>
   %res = shufflevector <64 x i8> %bc, <64 x i8> poison, <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>
@@ -523,3 +494,8 @@ define <16 x i8> @trunc_v8i64_to_v8i8_return_v16i8(<8 x i64> %vec) nounwind {
   ret <16 x i8> %result
 }
 
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; AVX512BW-FAST-ALL: {{.*}}
+; AVX512BW-FAST-PERLANE: {{.*}}
+; AVX512BWVL-FAST-ALL: {{.*}}
+; AVX512BWVL-FAST-PERLANE: {{.*}}

llvm/test/CodeGen/X86/x86-interleaved-access.ll

@@ -405,14 +405,13 @@ define <8 x i8> @interleaved_load_vf8_i8_stride4(ptr %ptr) nounwind {
 ; AVX512-NEXT:    vmovdqu (%rdi), %ymm0
 ; AVX512-NEXT:    vmovdqa 16(%rdi), %xmm1
 ; AVX512-NEXT:    vpmovsxbw {{.*#+}} xmm2 = [3,7,11,15,7,15,6,7]
-; AVX512-NEXT:    vpshufb %xmm2, %xmm1, %xmm3
+; AVX512-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX512-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
-; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm2 = xmm2[0],xmm3[0]
-; AVX512-NEXT:    vpsrld $16, %xmm1, %xmm1
-; AVX512-NEXT:    vpsrld $16, %xmm0, %xmm3
-; AVX512-NEXT:    vpackusdw %xmm1, %xmm3, %xmm1
-; AVX512-NEXT:    vpaddb %xmm1, %xmm2, %xmm1
+; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
 ; AVX512-NEXT:    vmovdqu (%rdi), %ymm2
+; AVX512-NEXT:    vpsrld $16, %ymm2, %ymm3
+; AVX512-NEXT:    vpmovdw %zmm3, %ymm3
+; AVX512-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovdw %zmm2, %ymm2
 ; AVX512-NEXT:    vpsrlw $8, %ymm0, %ymm0
 ; AVX512-NEXT:    vpmovwb %zmm0, %ymm0