[X86] lowerV8F32Shuffle - use lowerShufflePairAsUNPCKAndPermute on AVX1 targets (#140881)

If we're not going to split the v8f32 shuffle anyway, attempt to match with lowerShufflePairAsUNPCKAndPermute

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -16705,10 +16705,11 @@ static SDValue lowerV8F32Shuffle(const SDLoc &DL, ArrayRef<int> Mask,
 
   // Try to match an interleave of two v8f32s and lower them as unpck and
   // permutes using ymms. This needs to go before we try to split the vectors.
-  //
-  // TODO: Expand this to AVX1. Currently v8i32 is casted to v8f32 and hits
-  // this path inadvertently.
-  if (Subtarget.hasAVX2() && !Subtarget.hasAVX512())
+  // Don't attempt on AVX1 if we're likely to split vectors anyway.
+  if ((Subtarget.hasAVX2() ||
+       !(isFreeToSplitVector(peekThroughBitcasts(V1), DAG) ||
+         isFreeToSplitVector(peekThroughBitcasts(V2), DAG))) &&
+      !Subtarget.hasAVX512())
     if (SDValue V = lowerShufflePairAsUNPCKAndPermute(DL, MVT::v8f32, V1, V2,
                                                       Mask, DAG))
       return V;

llvm/test/CodeGen/X86/vector-interleave.ll

@@ -265,15 +265,10 @@ define <16 x i32> @interleave2x8i32(<8 x i32> %a, <8 x i32> %b) {
 ;
 ; AVX1-LABEL: interleave2x8i32:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vunpckhps {{.*#+}} xmm2 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vunpcklps {{.*#+}} xmm3 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm3, %ymm2
-; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
-; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vunpckhps {{.*#+}} xmm3 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; AVX1-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm1
-; AVX1-NEXT:    vmovaps %ymm2, %ymm0
+; AVX1-NEXT:    vunpckhps {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; AVX1-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm1 = ymm0[2,3],ymm2[2,3]
+; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: interleave2x8i32: