[SLP]Support minbitwidth analisys for buildvector nodes.

alexey-bataev · web-flow · commit 37ae4ad0eef3 · 2024-04-29T09:57:37.000-04:00
Metric: size..text Program size..text exp ref diff test-suite :: MultiSource/Benchmarks/mediabench/gsm/toast/toast.test 42906.00 42986.00 0.2% test-suite :: MultiSource/Benchmarks/MiBench/telecomm-gsm/telecomm-gsm.test 42909.00 42989.00 0.2% test-suite :: External/SPEC/CINT2017rate/525.x264_r/525.x264_r.test 664581.00 664661.00 0.0% test-suite :: External/SPEC/CINT2017speed/625.x264_s/625.x264_s.test 664581.00 664661.00 0.0% Less is better. Replaces `buildvector <p x in> + trunc <p x in> to <p x im>` sequences to `buildvector <p x im> of { trunc in to im }` scalars, which is free in most cases, results in better code. Reviewers: RKSimon Reviewed By: RKSimon Pull Request: #88504
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/gather-buildvector-with-minbitwidth-user.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/gather-buildvector-with-minbitwidth-user.ll
@@ -5,12 +5,7 @@ define void @h() {
 ; CHECK-LABEL: define void @h() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr i8, ptr null, i64 16
-; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <8 x i32> <i32 undef, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>, i32 0, i32 0
-; CHECK-NEXT:    [[TMP1:%.*]] = trunc <8 x i32> [[TMP0]] to <8 x i1>
-; CHECK-NEXT:    [[TMP2:%.*]] = or <8 x i1> zeroinitializer, [[TMP1]]
-; CHECK-NEXT:    [[TMP4:%.*]] = or <8 x i1> [[TMP2]], zeroinitializer
-; CHECK-NEXT:    [[TMP3:%.*]] = zext <8 x i1> [[TMP4]] to <8 x i16>
-; CHECK-NEXT:    store <8 x i16> [[TMP3]], ptr [[ARRAYIDX2]], align 2
+; CHECK-NEXT:    store <8 x i16> zeroinitializer, ptr [[ARRAYIDX2]], align 2
 ; CHECK-NEXT:    ret void
 ;
 entry:
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/gather-with-minbith-user.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/gather-with-minbith-user.ll
@@ -5,7 +5,8 @@ define void @h() {
 ; CHECK-LABEL: define void @h() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr i8, ptr null, i64 16
-; CHECK-NEXT:    [[TMP0:%.*]] = trunc <8 x i32> zeroinitializer to <8 x i1>
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc i32 0 to i1
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <8 x i1> <i1 false, i1 false, i1 false, i1 false, i1 poison, i1 false, i1 false, i1 false>, i1 [[TMP6]], i32 4
 ; CHECK-NEXT:    [[TMP1:%.*]] = sub <8 x i1> [[TMP0]], zeroinitializer
 ; CHECK-NEXT:    [[TMP2:%.*]] = add <8 x i1> [[TMP0]], zeroinitializer
 ; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP1]], <8 x i1> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/user-node-not-in-bitwidths.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/user-node-not-in-bitwidths.ll
@@ -5,7 +5,12 @@ define void @h() {
 ; CHECK-LABEL: define void @h() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr i8, ptr null, i64 16
-; CHECK-NEXT:    store <8 x i16> zeroinitializer, ptr [[ARRAYIDX2]], align 2
+; CHECK-NEXT:    [[TMP0:%.*]] = trunc i32 0 to i1
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <8 x i1> <i1 false, i1 false, i1 false, i1 false, i1 poison, i1 false, i1 false, i1 false>, i1 [[TMP0]], i32 4
+; CHECK-NEXT:    [[TMP2:%.*]] = or <8 x i1> zeroinitializer, [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = or <8 x i1> zeroinitializer, [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = zext <8 x i1> [[TMP3]] to <8 x i16>
+; CHECK-NEXT:    store <8 x i16> [[TMP4]], ptr [[ARRAYIDX2]], align 2
 ; CHECK-NEXT:    ret void
 ;
 entry:
diff --git a/llvm/test/Transforms/SLPVectorizer/SystemZ/minbitwidth-root-trunc.ll b/llvm/test/Transforms/SLPVectorizer/SystemZ/minbitwidth-root-trunc.ll
@@ -7,9 +7,9 @@ define void @test(ptr %a, i8 %0, i16 %b.promoted.i) {
 ; CHECK-NEXT:    [[TMP2:%.*]] = zext i8 [[TMP0]] to i128
 ; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x i16> poison, i16 [[B_PROMOTED_I]], i32 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i16> [[TMP3]], <4 x i16> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x i128> poison, i128 [[TMP2]], i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <4 x i128> [[TMP5]], <4 x i128> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP7:%.*]] = trunc <4 x i128> [[TMP6]] to <4 x i16>
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i128 [[TMP2]] to i16
+; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <4 x i16> poison, i16 [[TMP5]], i32 0
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <4 x i16> [[TMP6]], <4 x i16> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP8:%.*]] = or <4 x i16> [[TMP4]], [[TMP7]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = call i16 @llvm.vector.reduce.and.v4i16(<4 x i16> [[TMP8]])
 ; CHECK-NEXT:    [[TMP11:%.*]] = zext i16 [[TMP9]] to i64
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-node-with-multi-users.ll b/llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-node-with-multi-users.ll
@@ -16,8 +16,7 @@ define void @test() {
 ; CHECK-NEXT:    [[TMP9:%.*]] = trunc <4 x i8> [[TMP8]] to <4 x i1>
 ; CHECK-NEXT:    [[TMP10:%.*]] = or <4 x i1> zeroinitializer, [[TMP15]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq <4 x i1> [[TMP9]], [[TMP10]]
-; CHECK-NEXT:    [[TMP16:%.*]] = shufflevector <4 x i1> [[TMP15]], <4 x i1> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[TMP6:%.*]] = zext <4 x i1> [[TMP16]] to <4 x i32>
+; CHECK-NEXT:    [[TMP6:%.*]] = zext <4 x i1> [[TMP15]] to <4 x i32>
 ; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> <i32 0, i32 0, i32 poison, i32 0>, <4 x i32> <i32 4, i32 5, i32 2, i32 7>
 ; CHECK-NEXT:    [[TMP13:%.*]] = select <4 x i1> [[TMP11]], <4 x i32> [[TMP12]], <4 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP14:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP13]])
