[llvm] 7657e50 - [DAGCombiner] Fold avg(x, x) --> x
Jun Zhang via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 3 01:58:07 PDT 2023
Author: Jun Zhang
Date: 2023-04-03T16:57:50+08:00
New Revision: 7657e50fefe0af9151f8bef0437cd1343f0a3220
URL: https://github.com/llvm/llvm-project/commit/7657e50fefe0af9151f8bef0437cd1343f0a3220
DIFF: https://github.com/llvm/llvm-project/commit/7657e50fefe0af9151f8bef0437cd1343f0a3220.diff
LOG: [DAGCombiner] Fold avg(x, x) --> x
Signed-off-by: Jun Zhang <jun at junz.org>
Differential Revision: https://reviews.llvm.org/D147404
Added:
Modified:
llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/test/CodeGen/AArch64/hadd-combine.ll
llvm/test/CodeGen/X86/avg.ll
llvm/test/CodeGen/X86/combine-pavg.ll
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 31b584ee9f8d..f79a9a1b74d3 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -5085,6 +5085,10 @@ SDValue DAGCombiner::visitAVG(SDNode *N) {
if (N1.isUndef())
return N0;
+ // Fold (avg x, x) --> x
+ if (N0 == N1 && Level >= AfterLegalizeTypes)
+ return N0;
+
// TODO If we use avg for scalars anywhere, we can add (avgfl x, 0) -> x >> 1
return SDValue();
diff --git a/llvm/test/CodeGen/AArch64/hadd-combine.ll b/llvm/test/CodeGen/AArch64/hadd-combine.ll
index 62ccf2e8d99f..73e3a9f1d569 100644
--- a/llvm/test/CodeGen/AArch64/hadd-combine.ll
+++ b/llvm/test/CodeGen/AArch64/hadd-combine.ll
@@ -671,7 +671,6 @@ define <8 x i16> @rhadds_i_undef(<8 x i16> %t, <8 x i16> %src1) {
define <8 x i8> @shadd_v8i8(<8 x i8> %x) {
; CHECK-LABEL: shadd_v8i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.8b, v0.8b, v0.8b
; CHECK-NEXT: ret
%r = tail call <8 x i8> @llvm.aarch64.neon.shadd.v8i8(<8 x i8> %x, <8 x i8> %x)
ret <8 x i8> %r
@@ -680,7 +679,6 @@ define <8 x i8> @shadd_v8i8(<8 x i8> %x) {
define <4 x i16> @shadd_v4i16(<4 x i16> %x) {
; CHECK-LABEL: shadd_v4i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.4h, v0.4h, v0.4h
; CHECK-NEXT: ret
%r = tail call <4 x i16> @llvm.aarch64.neon.shadd.v4i16(<4 x i16> %x, <4 x i16> %x)
ret <4 x i16> %r
@@ -689,7 +687,6 @@ define <4 x i16> @shadd_v4i16(<4 x i16> %x) {
define <2 x i32> @shadd_v2i32(<2 x i32> %x) {
; CHECK-LABEL: shadd_v2i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.2s, v0.2s, v0.2s
; CHECK-NEXT: ret
%r = tail call <2 x i32> @llvm.aarch64.neon.shadd.v2i32(<2 x i32> %x, <2 x i32> %x)
ret <2 x i32> %r
@@ -698,7 +695,6 @@ define <2 x i32> @shadd_v2i32(<2 x i32> %x) {
define <16 x i8> @shadd_v16i8(<16 x i8> %x) {
; CHECK-LABEL: shadd_v16i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.16b, v0.16b, v0.16b
; CHECK-NEXT: ret
%r = tail call <16 x i8> @llvm.aarch64.neon.shadd.v16i8(<16 x i8> %x, <16 x i8> %x)
ret <16 x i8> %r
@@ -707,7 +703,6 @@ define <16 x i8> @shadd_v16i8(<16 x i8> %x) {
define <8 x i16> @shadd_v8i16(<8 x i16> %x) {
; CHECK-LABEL: shadd_v8i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.8h, v0.8h, v0.8h
; CHECK-NEXT: ret
%r = tail call <8 x i16> @llvm.aarch64.neon.shadd.v8i16(<8 x i16> %x, <8 x i16> %x)
ret <8 x i16> %r
@@ -716,7 +711,6 @@ define <8 x i16> @shadd_v8i16(<8 x i16> %x) {
define <4 x i32> @shadd_v4i32(<4 x i32> %x) {
; CHECK-LABEL: shadd_v4i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: shadd v0.4s, v0.4s, v0.4s
; CHECK-NEXT: ret
%r = tail call <4 x i32> @llvm.aarch64.neon.shadd.v4i32(<4 x i32> %x, <4 x i32> %x)
ret <4 x i32> %r
@@ -725,7 +719,6 @@ define <4 x i32> @shadd_v4i32(<4 x i32> %x) {
define <8 x i8> @uhadd_v8i8(<8 x i8> %x) {
; CHECK-LABEL: uhadd_v8i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.8b, v0.8b, v0.8b
; CHECK-NEXT: ret
%r = tail call <8 x i8> @llvm.aarch64.neon.uhadd.v8i8(<8 x i8> %x, <8 x i8> %x)
ret <8 x i8> %r
@@ -734,7 +727,6 @@ define <8 x i8> @uhadd_v8i8(<8 x i8> %x) {
define <4 x i16> @uhadd_v4i16(<4 x i16> %x) {
; CHECK-LABEL: uhadd_v4i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.4h, v0.4h, v0.4h
; CHECK-NEXT: ret
%r = tail call <4 x i16> @llvm.aarch64.neon.uhadd.v4i16(<4 x i16> %x, <4 x i16> %x)
ret <4 x i16> %r
@@ -743,7 +735,6 @@ define <4 x i16> @uhadd_v4i16(<4 x i16> %x) {
define <2 x i32> @uhadd_v2i32(<2 x i32> %x) {
; CHECK-LABEL: uhadd_v2i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.2s, v0.2s, v0.2s
; CHECK-NEXT: ret
%r = tail call <2 x i32> @llvm.aarch64.neon.uhadd.v2i32(<2 x i32> %x, <2 x i32> %x)
ret <2 x i32> %r
@@ -752,7 +743,6 @@ define <2 x i32> @uhadd_v2i32(<2 x i32> %x) {
define <16 x i8> @uhadd_v16i8(<16 x i8> %x) {
; CHECK-LABEL: uhadd_v16i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.16b, v0.16b, v0.16b
; CHECK-NEXT: ret
%r = tail call <16 x i8> @llvm.aarch64.neon.uhadd.v16i8(<16 x i8> %x, <16 x i8> %x)
ret <16 x i8> %r
@@ -761,7 +751,6 @@ define <16 x i8> @uhadd_v16i8(<16 x i8> %x) {
define <8 x i16> @uhadd_v8i16(<8 x i16> %x) {
; CHECK-LABEL: uhadd_v8i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.8h, v0.8h, v0.8h
; CHECK-NEXT: ret
%r = tail call <8 x i16> @llvm.aarch64.neon.uhadd.v8i16(<8 x i16> %x, <8 x i16> %x)
ret <8 x i16> %r
@@ -770,7 +759,6 @@ define <8 x i16> @uhadd_v8i16(<8 x i16> %x) {
define <4 x i32> @uhadd_v4i32(<4 x i32> %x) {
; CHECK-LABEL: uhadd_v4i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: uhadd v0.4s, v0.4s, v0.4s
; CHECK-NEXT: ret
%r = tail call <4 x i32> @llvm.aarch64.neon.uhadd.v4i32(<4 x i32> %x, <4 x i32> %x)
ret <4 x i32> %r
@@ -778,7 +766,6 @@ define <4 x i32> @uhadd_v4i32(<4 x i32> %x) {
define <8 x i8> @srhadd_v8i8(<8 x i8> %x) {
; CHECK-LABEL: srhadd_v8i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.8b, v0.8b, v0.8b
; CHECK-NEXT: ret
%r = tail call <8 x i8> @llvm.aarch64.neon.srhadd.v8i8(<8 x i8> %x, <8 x i8> %x)
ret <8 x i8> %r
@@ -787,7 +774,6 @@ define <8 x i8> @srhadd_v8i8(<8 x i8> %x) {
define <4 x i16> @srhadd_v4i16(<4 x i16> %x) {
; CHECK-LABEL: srhadd_v4i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.4h, v0.4h, v0.4h
; CHECK-NEXT: ret
%r = tail call <4 x i16> @llvm.aarch64.neon.srhadd.v4i16(<4 x i16> %x, <4 x i16> %x)
ret <4 x i16> %r
@@ -796,7 +782,6 @@ define <4 x i16> @srhadd_v4i16(<4 x i16> %x) {
define <2 x i32> @srhadd_v2i32(<2 x i32> %x) {
; CHECK-LABEL: srhadd_v2i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.2s, v0.2s, v0.2s
; CHECK-NEXT: ret
%r = tail call <2 x i32> @llvm.aarch64.neon.srhadd.v2i32(<2 x i32> %x, <2 x i32> %x)
ret <2 x i32> %r
@@ -805,7 +790,6 @@ define <2 x i32> @srhadd_v2i32(<2 x i32> %x) {
define <16 x i8> @srhadd_v16i8(<16 x i8> %x) {
; CHECK-LABEL: srhadd_v16i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.16b, v0.16b, v0.16b
; CHECK-NEXT: ret
%r = tail call <16 x i8> @llvm.aarch64.neon.srhadd.v16i8(<16 x i8> %x, <16 x i8> %x)
ret <16 x i8> %r
@@ -814,7 +798,6 @@ define <16 x i8> @srhadd_v16i8(<16 x i8> %x) {
define <8 x i16> @srhadd_v8i16(<8 x i16> %x) {
; CHECK-LABEL: srhadd_v8i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.8h, v0.8h, v0.8h
; CHECK-NEXT: ret
%r = tail call <8 x i16> @llvm.aarch64.neon.srhadd.v8i16(<8 x i16> %x, <8 x i16> %x)
ret <8 x i16> %r
@@ -823,7 +806,6 @@ define <8 x i16> @srhadd_v8i16(<8 x i16> %x) {
define <4 x i32> @srhadd_v4i32(<4 x i32> %x) {
; CHECK-LABEL: srhadd_v4i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: srhadd v0.4s, v0.4s, v0.4s
; CHECK-NEXT: ret
%r = tail call <4 x i32> @llvm.aarch64.neon.srhadd.v4i32(<4 x i32> %x, <4 x i32> %x)
ret <4 x i32> %r
@@ -832,7 +814,6 @@ define <4 x i32> @srhadd_v4i32(<4 x i32> %x) {
define <8 x i8> @urhadd_v8i8(<8 x i8> %x) {
; CHECK-LABEL: urhadd_v8i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.8b, v0.8b, v0.8b
; CHECK-NEXT: ret
%r = tail call <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8> %x, <8 x i8> %x)
ret <8 x i8> %r
@@ -841,7 +822,6 @@ define <8 x i8> @urhadd_v8i8(<8 x i8> %x) {
define <4 x i16> @urhadd_v4i16(<4 x i16> %x) {
; CHECK-LABEL: urhadd_v4i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.4h, v0.4h, v0.4h
; CHECK-NEXT: ret
%r = tail call <4 x i16> @llvm.aarch64.neon.urhadd.v4i16(<4 x i16> %x, <4 x i16> %x)
ret <4 x i16> %r
@@ -850,7 +830,6 @@ define <4 x i16> @urhadd_v4i16(<4 x i16> %x) {
define <2 x i32> @urhadd_v2i32(<2 x i32> %x) {
; CHECK-LABEL: urhadd_v2i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.2s, v0.2s, v0.2s
; CHECK-NEXT: ret
%r = tail call <2 x i32> @llvm.aarch64.neon.urhadd.v2i32(<2 x i32> %x, <2 x i32> %x)
ret <2 x i32> %r
@@ -859,7 +838,6 @@ define <2 x i32> @urhadd_v2i32(<2 x i32> %x) {
define <16 x i8> @urhadd_v16i8(<16 x i8> %x) {
; CHECK-LABEL: urhadd_v16i8:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.16b, v0.16b, v0.16b
; CHECK-NEXT: ret
%r = tail call <16 x i8> @llvm.aarch64.neon.urhadd.v16i8(<16 x i8> %x, <16 x i8> %x)
ret <16 x i8> %r
@@ -868,7 +846,6 @@ define <16 x i8> @urhadd_v16i8(<16 x i8> %x) {
define <8 x i16> @urhadd_v8i16(<8 x i16> %x) {
; CHECK-LABEL: urhadd_v8i16:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.8h, v0.8h, v0.8h
; CHECK-NEXT: ret
%r = tail call <8 x i16> @llvm.aarch64.neon.urhadd.v8i16(<8 x i16> %x, <8 x i16> %x)
ret <8 x i16> %r
@@ -877,7 +854,6 @@ define <8 x i16> @urhadd_v8i16(<8 x i16> %x) {
define <4 x i32> @urhadd_v4i32(<4 x i32> %x) {
; CHECK-LABEL: urhadd_v4i32:
; CHECK: // %bb.0:
-; CHECK-NEXT: urhadd v0.4s, v0.4s, v0.4s
; CHECK-NEXT: ret
%r = tail call <4 x i32> @llvm.aarch64.neon.urhadd.v4i32(<4 x i32> %x, <4 x i32> %x)
ret <4 x i32> %r
diff --git a/llvm/test/CodeGen/X86/avg.ll b/llvm/test/CodeGen/X86/avg.ll
index 72c9948c2f59..516d3d3312a2 100644
--- a/llvm/test/CodeGen/X86/avg.ll
+++ b/llvm/test/CodeGen/X86/avg.ll
@@ -728,65 +728,40 @@ define void @avg_v32i8_2(ptr %a, ptr %b) nounwind {
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: 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: vmovdqu64 %zmm0, (%rax)
-; AVX512BW-NEXT: vzeroupper
-; AVX512BW-NEXT: retq
+; AVX512-LABEL: avg_v64i8_2:
+; AVX512: # %bb.0:
+; AVX512-NEXT: vmovaps (%rsi), %zmm0
+; AVX512-NEXT: vmovups %zmm0, (%rax)
+; AVX512-NEXT: vzeroupper
+; AVX512-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>
diff --git a/llvm/test/CodeGen/X86/combine-pavg.ll b/llvm/test/CodeGen/X86/combine-pavg.ll
index 493a1af9a591..a8923210427e 100644
--- a/llvm/test/CodeGen/X86/combine-pavg.ll
+++ b/llvm/test/CodeGen/X86/combine-pavg.ll
@@ -6,16 +6,13 @@
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
More information about the llvm-commits
mailing list