[X86] Add shuffle tests for BLEND(PERMUTE(X),PERMUTE(Y)) patterns

Some very basic tests for a case where we could fold BLEND(PERMUTE(X),PERMUTE(Y)) -> PERMUTE(BLEND(X,Y))

These assume the permute masks are the same, and "complete" (no undefs/duplicate elements) but we could relax that depending on the blend mask

Author: RKSimon

llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll

@@ -305,6 +305,37 @@ define <4 x float> @combine_vpermilvar_4f32_as_insertps(<4 x float> %a0) {
   ret <4 x float> %2
 }
 
+define <8 x i32> @combine_blend_of_permutes_v8i32(<4 x i64> %a0, <4 x i64> %a1) {
+; AVX1-LABEL: combine_blend_of_permutes_v8i32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3,4],ymm0[5,6],ymm1[7]
+; AVX1-NEXT:    ret{{[l|q]}}
+;
+; AVX2-LABEL: combine_blend_of_permutes_v8i32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,1]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm1 = ymm1[2,3,0,1]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3,4],ymm0[5,6],ymm1[7]
+; AVX2-NEXT:    ret{{[l|q]}}
+;
+; AVX512-LABEL: combine_blend_of_permutes_v8i32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512-NEXT:    vpmovsxbd {{.*#+}} ymm2 = [4,21,6,23,16,1,2,19]
+; AVX512-NEXT:    vpermt2d %zmm1, %zmm2, %zmm0
+; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
+; AVX512-NEXT:    ret{{[l|q]}}
+  %s0 = shufflevector <4 x i64> %a0, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
+  %s1 = shufflevector <4 x i64> %a1, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
+  %x0 = bitcast <4 x i64> %s0 to <8 x i32>
+  %x1 = bitcast <4 x i64> %s1 to <8 x i32>
+  %r = shufflevector <8 x i32> %x0, <8 x i32> %x1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 12, i32 5, i32 6, i32 15>
+  ret <8 x i32> %r
+}
+
 define <2 x double> @constant_fold_vpermilvar_pd() {
 ; CHECK-LABEL: constant_fold_vpermilvar_pd:
 ; CHECK:       # %bb.0:

llvm/test/CodeGen/X86/vector-shuffle-combining-avx512f.ll

@@ -973,3 +973,47 @@ define <8 x i64> @combine_vpermvar_insertion_as_broadcast_v8i64(i64 %a0) {
   %2 = call <8 x i64> @llvm.x86.avx512.permvar.di.512(<8 x i64> %1, <8 x i64> zeroinitializer)
   ret <8 x i64> %2
 }
+
+define <16 x i32> @blend_of_permutes_v16i32(<8 x i64> %a0, <8x i64> %a1) {
+; X86-AVX512F-LABEL: blend_of_permutes_v16i32:
+; X86-AVX512F:       # %bb.0:
+; X86-AVX512F-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5]
+; X86-AVX512F-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,0,1,6,7,4,5]
+; X86-AVX512F-NEXT:    movw $-25958, %ax # imm = 0x9A9A
+; X86-AVX512F-NEXT:    kmovw %eax, %k1
+; X86-AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; X86-AVX512F-NEXT:    retl
+;
+; X86-AVX512BW-LABEL: blend_of_permutes_v16i32:
+; X86-AVX512BW:       # %bb.0:
+; X86-AVX512BW-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5]
+; X86-AVX512BW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,0,1,6,7,4,5]
+; X86-AVX512BW-NEXT:    movw $-25958, %ax # imm = 0x9A9A
+; X86-AVX512BW-NEXT:    kmovd %eax, %k1
+; X86-AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; X86-AVX512BW-NEXT:    retl
+;
+; X64-AVX512F-LABEL: blend_of_permutes_v16i32:
+; X64-AVX512F:       # %bb.0:
+; X64-AVX512F-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5]
+; X64-AVX512F-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,0,1,6,7,4,5]
+; X64-AVX512F-NEXT:    movw $-25958, %ax # imm = 0x9A9A
+; X64-AVX512F-NEXT:    kmovw %eax, %k1
+; X64-AVX512F-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; X64-AVX512F-NEXT:    retq
+;
+; X64-AVX512BW-LABEL: blend_of_permutes_v16i32:
+; X64-AVX512BW:       # %bb.0:
+; X64-AVX512BW-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5]
+; X64-AVX512BW-NEXT:    vpermq {{.*#+}} zmm1 = zmm1[2,3,0,1,6,7,4,5]
+; X64-AVX512BW-NEXT:    movw $-25958, %ax # imm = 0x9A9A
+; X64-AVX512BW-NEXT:    kmovd %eax, %k1
+; X64-AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1}
+; X64-AVX512BW-NEXT:    retq
+  %s0 = shufflevector <8 x i64> %a0, <8 x i64> undef, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
+  %s1 = shufflevector <8 x i64> %a1, <8 x i64> undef, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 6, i32 7, i32 4, i32 5>
+  %x0 = bitcast <8 x i64> %s0 to <16 x i32>
+  %x1 = bitcast <8 x i64> %s1 to <16 x i32>
+  %r = shufflevector <16 x i32> %x0, <16 x i32> %x1, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 20, i32 5, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 28, i32 13, i32 14, i32 31>
+  ret <16 x i32> %r
+}

llvm/test/CodeGen/X86/vector-shuffle-combining-sse41.ll

@@ -22,6 +22,21 @@ define <16 x i8> @combine_vpshufb_as_movzx(<16 x i8> %a0) {
   ret <16 x i8> %res0
 }
 
+define <4 x i32> @combine_blend_of_permutes_v4i32(<2 x i64> %a0, <2 x i64> %a1) {
+; SSE-LABEL: combine_blend_of_permutes_v4i32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm2[0,1],xmm0[2,3],xmm2[4,5],xmm0[6,7]
+; SSE-NEXT:    retq
+  %s0 = shufflevector <2 x i64> %a0, <2 x i64> undef, <2 x i32> <i32 1, i32 0>
+  %s1 = shufflevector <2 x i64> %a1, <2 x i64> undef, <2 x i32> <i32 1, i32 0>
+  %x0 = bitcast <2 x i64> %s0 to <4 x i32>
+  %x1 = bitcast <2 x i64> %s1 to <4 x i32>
+  %r = shufflevector <4 x i32> %x0, <4 x i32> %x1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x i32> %r
+}
+
 define <16 x i8> @PR50049(ptr %p1, ptr %p2) {
 ; SSE-LABEL: PR50049:
 ; SSE:       # %bb.0: