[X86] combineINSERT_SUBVECTOR - attempt to recursively shuffle combine if both base/sub-vectors are already shuffles (#130304)

3 files changed, 45 insertions, 72 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 0602f50..f9661b0 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -58758,6 +58758,14 @@ static SDValue combineINSERT_SUBVECTOR(SDNode *N, SelectionDAG &DAG,
     }
   }
 
+  // Attempt to recursively combine to a shuffle.
+  if (isTargetShuffle(peekThroughBitcasts(Vec).getOpcode()) &&
+      isTargetShuffle(peekThroughBitcasts(SubVec).getOpcode())) {
+    SDValue Op(N, 0);
+    if (SDValue Res = combineX86ShufflesRecursively(Op, DAG, Subtarget))
+      return Res;
+  }
+
   return SDValue();
 }
 
diff --git a/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast.ll b/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast.ll
index 1102044..2dfa515 100644
--- a/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast.ll
+++ b/llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast.ll
@@ -3519,9 +3519,7 @@ define void @vec384_i16_widen_to_i32_factor2_broadcast_to_v12i32_factor12(ptr %i
 ; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} xmm1 = [0,25,0,27,0,29,0,31]
-; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-NEXT:    vzeroupper
@@ -3651,9 +3649,7 @@ define void @vec384_i16_widen_to_i48_factor3_broadcast_to_v8i48_factor8(ptr %in.
 ; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} xmm1 = [0,25,26,0,28,29,0,31]
-; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm0, %xmm0
-; AVX512BW-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-NEXT:    vzeroupper
@@ -3776,9 +3772,7 @@ define void @vec384_i16_widen_to_i64_factor4_broadcast_to_v6i64_factor6(ptr %in.
 ; AVX512BW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [0,25,26,27,0,29,30,31,0,25,26,27,0,29,30,31]
 ; AVX512BW-NEXT:    # ymm1 = mem[0,1,0,1]
-; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-NEXT:    vzeroupper
@@ -3909,9 +3903,7 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
 ; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vpmovsxbw {{.*#+}} xmm1 = [0,25,26,27,28,29,0,31]
-; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm1
-; AVX512BW-NEXT:    vpbroadcastw %xmm0, %ymm0
-; AVX512BW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
+; AVX512BW-NEXT:    vpermw %zmm0, %zmm1, %zmm0
 ; AVX512BW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-NEXT:    vzeroupper
@@ -4401,10 +4393,8 @@ define void @vec384_i32_widen_to_i96_factor3_broadcast_to_v4i96_factor4(ptr %in.
 ; AVX512BW-SLOW:       # %bb.0:
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %zmm0
 ; AVX512BW-SLOW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    vpmovsxbd {{.*#+}} xmm1 = [0,13,14,0]
-; AVX512BW-SLOW-NEXT:    vpermd %zmm0, %zmm1, %zmm1
-; AVX512BW-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,1,1]
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
+; AVX512BW-SLOW-NEXT:    vpmovsxbd {{.*#+}} zmm1 = [0,13,14,0,0,0,0,0,0,0,1,1,0,0,0,0]
+; AVX512BW-SLOW-NEXT:    vpermd %zmm0, %zmm1, %zmm0
 ; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-SLOW-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
index 409d889..35f25d3 100644
--- a/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
+++ b/llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast.ll
@@ -3652,13 +3652,11 @@ define void @vec384_i16_widen_to_i32_factor2_broadcast_to_v12i32_factor12(ptr %i
 ; AVX512BW-FAST-LABEL: vec384_i16_widen_to_i32_factor2_broadcast_to_v12i32_factor12:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} ymm1 = [0,25,0,27,0,29,0,31,0,41,0,43,0,45,0,47]
 ; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1],zero,zero,xmm0[0,1],zero,zero,xmm0[0,1],zero,zero,xmm0[0,1],zero,zero
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} zmm2 = [0,25,0,27,0,29,0,31,0,41,0,43,0,45,0,47,0,49,0,51,0,53,0,55,0,0,0,0,0,0,0,0]
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
@@ -3857,13 +3855,11 @@ define void @vec384_i16_widen_to_i48_factor3_broadcast_to_v8i48_factor8(ptr %in.
 ; AVX512BW-FAST-LABEL: vec384_i16_widen_to_i48_factor3_broadcast_to_v8i48_factor8:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} ymm1 = [0,25,26,0,28,29,0,31,40,0,42,43,0,45,46,0]
 ; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1],zero,zero,zero,zero,xmm0[0,1],zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} zmm2 = [0,25,26,0,28,29,0,31,40,0,42,43,0,45,46,0,48,49,0,51,52,0,54,55,0,0,0,0,0,0,0,0]
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
@@ -4085,13 +4081,11 @@ define void @vec384_i16_widen_to_i64_factor4_broadcast_to_v6i64_factor6(ptr %in.
 ; AVX512BW-FAST-LABEL: vec384_i16_widen_to_i64_factor4_broadcast_to_v6i64_factor6:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} ymm1 = [0,25,26,27,0,29,30,31,0,41,42,43,0,45,46,47]
 ; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1],zero,zero,zero,zero,zero,zero,xmm0[0,1],zero,zero,zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} zmm2 = [0,25,26,27,0,29,30,31,0,41,42,43,0,45,46,47,0,49,50,51,0,53,54,55,0,0,0,0,0,0,0,0]
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
@@ -4292,13 +4286,11 @@ define void @vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4(ptr %in.
 ; AVX512BW-FAST-LABEL: vec384_i16_widen_to_i96_factor6_broadcast_to_v4i96_factor4:
 ; AVX512BW-FAST:       # %bb.0:
 ; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} ymm1 = [0,25,26,27,28,29,0,31,40,41,42,43,0,45,46,47]
 ; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-FAST-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1
-; AVX512BW-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = zero,zero,zero,zero,xmm0[0,1],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
-; AVX512BW-FAST-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-FAST-NEXT:    vpmovsxbw {{.*#+}} zmm2 = [0,25,26,27,28,29,0,31,40,41,42,43,0,45,46,47,48,49,0,51,52,53,54,55,0,0,0,0,0,0,0,0]
+; AVX512BW-FAST-NEXT:    vpermi2w %zmm1, %zmm0, %zmm2
+; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-FAST-NEXT:    vzeroupper
 ; AVX512BW-FAST-NEXT:    retq
@@ -5101,32 +5093,17 @@ define void @vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3(ptr %i
 ; AVX512DQ-NEXT:    vzeroupper
 ; AVX512DQ-NEXT:    retq
 ;
-; AVX512BW-SLOW-LABEL: vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3:
-; AVX512BW-SLOW:       # %bb.0:
-; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-SLOW-NEXT:    vpmovsxbd {{.*#+}} ymm1 = [0,13,14,15,0,21,22,23]
-; AVX512BW-SLOW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-SLOW-NEXT:    vpermi2d %zmm2, %zmm0, %zmm1
-; AVX512BW-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-SLOW-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3,4,5,6,7]
-; AVX512BW-SLOW-NEXT:    vinserti64x4 $1, %ymm0, %zmm1, %zmm0
-; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, (%rcx)
-; AVX512BW-SLOW-NEXT:    vzeroupper
-; AVX512BW-SLOW-NEXT:    retq
-;
-; AVX512BW-FAST-LABEL: vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3:
-; AVX512BW-FAST:       # %bb.0:
-; AVX512BW-FAST-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-FAST-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-FAST-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX512BW-FAST-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [0,13,14,15,0,21,22,23,0,25,26,27,28,29,30,31]
-; AVX512BW-FAST-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
-; AVX512BW-FAST-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
-; AVX512BW-FAST-NEXT:    vmovdqa64 %zmm0, (%rcx)
-; AVX512BW-FAST-NEXT:    vzeroupper
-; AVX512BW-FAST-NEXT:    retq
+; AVX512BW-LABEL: vec384_i32_widen_to_i128_factor4_broadcast_to_v3i128_factor3:
+; AVX512BW:       # %bb.0:
+; AVX512BW-NEXT:    vmovdqa64 (%rdi), %zmm0
+; AVX512BW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
+; AVX512BW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-NEXT:    vpmovsxbd {{.*#+}} zmm2 = [0,13,14,15,0,21,22,23,0,25,26,27,28,29,30,31]
+; AVX512BW-NEXT:    vpermi2d %zmm1, %zmm0, %zmm2
+; AVX512BW-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 %zmm0, (%rcx)
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
   %in.vec.base = load <64 x i8>, ptr %in.vec.base.ptr, align 64
   %in.vec.bias = load <64 x i8>, ptr %in.vec.bias.ptr, align 64
   %in.vec = add <64 x i8> %in.vec.base, %in.vec.bias
@@ -5381,13 +5358,11 @@ define void @vec384_i64_widen_to_i128_factor2_broadcast_to_v3i128_factor3(ptr %i
 ; AVX512BW-SLOW-LABEL: vec384_i64_widen_to_i128_factor2_broadcast_to_v3i128_factor3:
 ; AVX512BW-SLOW:       # %bb.0:
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 (%rdi), %zmm0
-; AVX512BW-SLOW-NEXT:    vpmovsxbq {{.*#+}} ymm1 = [0,7,0,11]
 ; AVX512BW-SLOW-NEXT:    vpaddb (%rsi), %zmm0, %zmm0
-; AVX512BW-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX512BW-SLOW-NEXT:    vpermi2q %zmm2, %zmm0, %zmm1
-; AVX512BW-SLOW-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
-; AVX512BW-SLOW-NEXT:    vinserti32x4 $2, %xmm0, %zmm1, %zmm0
-; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm0, %zmm0
+; AVX512BW-SLOW-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; AVX512BW-SLOW-NEXT:    vpmovsxbq {{.*#+}} zmm2 = [0,7,0,11,0,13,0,0]
+; AVX512BW-SLOW-NEXT:    vpermi2q %zmm1, %zmm0, %zmm2
+; AVX512BW-SLOW-NEXT:    vpaddb (%rdx), %zmm2, %zmm0
 ; AVX512BW-SLOW-NEXT:    vmovdqa64 %zmm0, (%rcx)
 ; AVX512BW-SLOW-NEXT:    vzeroupper
 ; AVX512BW-SLOW-NEXT:    retq