[DAGCombiner] Fold avg(x, x) --> x

[junaire]

Summary

Signed-off-by: Jun Zhang <[e-mail]>

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

SDValue DAGCombiner::visitAVG(SDNode *N) {
   }
 
   // fold (avg x, undef) -> x
   if (N0.isUndef())
     return N1;
   if (N1.isUndef())
     return N0;
 
+  // Fold (avg x, x) --> x
+  if (N0 == N1)
+    return N0;
+
   // TODO If we use avg for scalars anywhere, we can add (avgfl x, 0) -> x >> 1
 
   return SDValue();
 }
 
 SDValue DAGCombiner::visitABD(SDNode *N) {
   unsigned Opcode = N->getOpcode();
   SDValue N0 = N->getOperand(0);

llvm/test/CodeGen/X86/avg.ll

; AVX512-NEXT:    retq
   %8 = trunc <32 x i32> %7 to <32 x i8>
   store <32 x i8> %8, ptr undef, align 4
   ret void
 }
 
 define void @avg_v64i8_2(ptr %a, ptr %b) nounwind {
 ; SSE2-LABEL: avg_v64i8_2:
 ; SSE2:       # %bb.0:
-; SSE2-NEXT:    movdqa (%rsi), %xmm0
-; SSE2-NEXT:    movdqa 16(%rsi), %xmm1
-; SSE2-NEXT:    movdqa 32(%rsi), %xmm2
-; SSE2-NEXT:    movdqa 48(%rsi), %xmm3
-; SSE2-NEXT:    pavgb %xmm0, %xmm0
-; SSE2-NEXT:    pavgb %xmm1, %xmm1
-; SSE2-NEXT:    pavgb %xmm2, %xmm2
-; SSE2-NEXT:    pavgb %xmm3, %xmm3
-; SSE2-NEXT:    movdqu %xmm3, (%rax)
-; SSE2-NEXT:    movdqu %xmm2, (%rax)
-; SSE2-NEXT:    movdqu %xmm1, (%rax)
-; SSE2-NEXT:    movdqu %xmm0, (%rax)
+; SSE2-NEXT:    movaps (%rsi), %xmm0
+; SSE2-NEXT:    movaps 16(%rsi), %xmm1
+; SSE2-NEXT:    movaps 32(%rsi), %xmm2
+; SSE2-NEXT:    movaps 48(%rsi), %xmm3
+; SSE2-NEXT:    movups %xmm3, (%rax)
+; SSE2-NEXT:    movups %xmm2, (%rax)
+; SSE2-NEXT:    movups %xmm1, (%rax)
+; SSE2-NEXT:    movups %xmm0, (%rax)
 ; SSE2-NEXT:    retq
 ;
 ; AVX1-LABEL: avg_v64i8_2:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vmovdqa (%rsi), %xmm0
-; AVX1-NEXT:    vmovdqa 16(%rsi), %xmm1
-; AVX1-NEXT:    vmovdqa 32(%rsi), %xmm2
-; AVX1-NEXT:    vmovdqa 48(%rsi), %xmm3
-; AVX1-NEXT:    vpavgb %xmm0, %xmm0, %xmm0
-; AVX1-NEXT:    vpavgb %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpavgb %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpavgb %xmm3, %xmm3, %xmm3
-; AVX1-NEXT:    vmovdqu %xmm3, (%rax)
-; AVX1-NEXT:    vmovdqu %xmm2, (%rax)
-; AVX1-NEXT:    vmovdqu %xmm1, (%rax)
-; AVX1-NEXT:    vmovdqu %xmm0, (%rax)
+; AVX1-NEXT:    vmovaps (%rsi), %ymm0
+; AVX1-NEXT:    vmovaps 32(%rsi), %ymm1
+; AVX1-NEXT:    vmovups %ymm1, (%rax)
+; AVX1-NEXT:    vmovups %ymm0, (%rax)
+; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
 ;
 ; AVX2-LABEL: avg_v64i8_2:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vmovdqa (%rsi), %ymm0
-; AVX2-NEXT:    vmovdqa 32(%rsi), %ymm1
-; AVX2-NEXT:    vpavgb %ymm0, %ymm0, %ymm0
-; AVX2-NEXT:    vpavgb %ymm1, %ymm1, %ymm1
-; AVX2-NEXT:    vmovdqu %ymm1, (%rax)
-; AVX2-NEXT:    vmovdqu %ymm0, (%rax)
+; AVX2-NEXT:    vmovaps (%rsi), %ymm0
+; AVX2-NEXT:    vmovaps 32(%rsi), %ymm1
+; AVX2-NEXT:    vmovups %ymm1, (%rax)
+; AVX2-NEXT:    vmovups %ymm0, (%rax)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
 ; AVX512F-LABEL: avg_v64i8_2:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vmovdqa (%rsi), %ymm0
-; AVX512F-NEXT:    vmovdqa 32(%rsi), %ymm1
-; AVX512F-NEXT:    vpavgb %ymm0, %ymm0, %ymm0
-; AVX512F-NEXT:    vpavgb %ymm1, %ymm1, %ymm1
-; AVX512F-NEXT:    vmovdqu %ymm1, (%rax)
-; AVX512F-NEXT:    vmovdqu %ymm0, (%rax)
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm2 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovzxbd {{.*#+}} zmm3 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512F-NEXT:    vpmovdb %zmm3, (%rax)
+; AVX512F-NEXT:    vpmovdb %zmm2, (%rax)
+; AVX512F-NEXT:    vpmovdb %zmm1, (%rax)
+; AVX512F-NEXT:    vpmovdb %zmm0, (%rax)
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: avg_v64i8_2:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovdqa64 (%rsi), %zmm0
-; AVX512BW-NEXT:    vpavgb %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm1 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm2 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512BW-NEXT:    vpmovzxbd {{.*#+}} zmm3 = mem[0],zero,zero,zero,mem[1],zero,zero,zero,mem[2],zero,zero,zero,mem[3],zero,zero,zero,mem[4],zero,zero,zero,mem[5],zero,zero,zero,mem[6],zero,zero,zero,mem[7],zero,zero,zero,mem[8],zero,zero,zero,mem[9],zero,zero,zero,mem[10],zero,zero,zero,mem[11],zero,zero,zero,mem[12],zero,zero,zero,mem[13],zero,zero,zero,mem[14],zero,zero,zero,mem[15],zero,zero,zero
+; AVX512BW-NEXT:    vpmovdb %zmm3, %xmm3
+; AVX512BW-NEXT:    vpmovdb %zmm2, %xmm2
+; AVX512BW-NEXT:    vinserti128 $1, %xmm2, %ymm3, %ymm2
+; AVX512BW-NEXT:    vpmovdb %zmm1, %xmm1
+; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512BW-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
+; AVX512BW-NEXT:    vinserti64x4 $1, %ymm2, %zmm0, %zmm0
 ; AVX512BW-NEXT:    vmovdqu64 %zmm0, (%rax)
 ; AVX512BW-NEXT:    vzeroupper
 ; AVX512BW-NEXT:    retq
   %1 = load <64 x i8>, ptr %a
   %2 = load <64 x i8>, ptr %b
   %3 = zext <64 x i8> %1 to <64 x i32>
   %4 = zext <64 x i8> %2 to <64 x i32>
   %5 = add nuw nsw <64 x i32> %4, %4

llvm/test/CodeGen/X86/combine-pavg.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX
 
 declare <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8>, <16 x i8>) nounwind readnone
 declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone
 
 ; TODO: AVG(X,X) -> X
 define <16 x i8> @combine_pavgb_self(<16 x i8> %a0) {
 ; SSE-LABEL: combine_pavgb_self:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    pavgb %xmm0, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: combine_pavgb_self:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpavgb %xmm0, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %1 = call <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8> %a0, <16 x i8> %a0)
   ret <16 x i8> %1
 }
 
 define <16 x i8> @combine_pavgw_knownbits(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {
 ; SSE-LABEL: combine_pavgw_knownbits:
 ; SSE:       # %bb.0: