[llvm] 3c5e24a - [SLP] Add tests showing over-eager SLP when scalar fma can be used.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 29 10:59:23 PDT 2022
Author: Florian Hahn
Date: 2022-08-29T18:58:56+01:00
New Revision: 3c5e24a51ce072fd0083396dcf0ea107b1858d11
URL: https://github.com/llvm/llvm-project/commit/3c5e24a51ce072fd0083396dcf0ea107b1858d11
DIFF: https://github.com/llvm/llvm-project/commit/3c5e24a51ce072fd0083396dcf0ea107b1858d11.diff
LOG: [SLP] Add tests showing over-eager SLP when scalar fma can be used.
Add test cases for AArch64 that show over-eager SLP vectorization on
AArch64, where keeping the things scalar allows efficient lowering using
scalar fmas.
Added:
llvm/test/Transforms/SLPVectorizer/AArch64/slp-fma-loss.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/slp-fma-loss.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-fma-loss.ll
new file mode 100644
index 0000000000000..5f8c5c16800b9
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-fma-loss.ll
@@ -0,0 +1,134 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=slp-vectorizer -mtriple=arm64-apple-ios -S %s | FileCheck %s
+
+; Test case where not vectorizing is more profitable because multiple
+; fmul/{fadd,fsub} pairs can be lowered to fma instructions.
+define void @slp_not_profitable(ptr %A, ptr %B) {
+; CHECK-LABEL: @slp_not_profitable(
+; CHECK-NEXT: [[GEP_B_1:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 1
+; CHECK-NEXT: [[A_0:%.*]] = load float, ptr [[A:%.*]], align 4
+; CHECK-NEXT: [[B_0:%.*]] = load float, ptr [[B]], align 4
+; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[GEP_B_1]], align 4
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x float> poison, float [[B_0]], i32 0
+; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x float> [[TMP2]], float [[B_0]], i32 1
+; CHECK-NEXT: [[TMP4:%.*]] = fmul fast <2 x float> [[TMP3]], [[TMP1]]
+; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[A_0]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[A_0]], i32 1
+; CHECK-NEXT: [[TMP7:%.*]] = fmul fast <2 x float> [[TMP1]], [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = fsub fast <2 x float> [[TMP7]], [[SHUFFLE]]
+; CHECK-NEXT: [[TMP9:%.*]] = fadd fast <2 x float> [[TMP7]], [[SHUFFLE]]
+; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> [[TMP9]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: store <2 x float> [[TMP10]], ptr [[A]], align 4
+; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x float> [[TMP1]], i32 1
+; CHECK-NEXT: store float [[TMP11]], ptr [[B]], align 4
+; CHECK-NEXT: ret void
+;
+ %gep.B.1 = getelementptr inbounds float, ptr %B, i64 1
+ %A.0 = load float, ptr %A, align 4
+ %B.1 = load float, ptr %gep.B.1, align 4
+ %mul.0 = fmul fast float %B.1, %A.0
+ %B.0 = load float, ptr %B, align 4
+ %gep.B.2 = getelementptr inbounds float, ptr %B, i64 2
+ %B.2 = load float, ptr %gep.B.2, align 4
+ %mul.1 = fmul fast float %B.2, %B.0
+ %sub = fsub fast float %mul.0, %mul.1
+ %mul.2 = fmul fast float %B.0, %B.1
+ %mul.3 = fmul fast float %B.2, %A.0
+ %add = fadd fast float %mul.3, %mul.2
+ store float %sub, ptr %A, align 4
+ %gep.A.1 = getelementptr inbounds float, ptr %A, i64 1
+ store float %add, ptr %gep.A.1, align 4
+ store float %B.2, ptr %B, align 4
+ ret void
+}
+
+; Test case where not vectorizing is more profitable because multiple
+; fmul/{fadd,fsub} pairs can be lowered to fma instructions.
+define float @slp_not_profitable_in_loop(float %x, ptr %A) {
+; CHECK-LABEL: @slp_not_profitable_in_loop(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[GEP_A_1:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 1
+; CHECK-NEXT: [[L_0:%.*]] = load float, ptr [[GEP_A_1]], align 4
+; CHECK-NEXT: [[GEP_A_2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 2
+; CHECK-NEXT: [[L_1:%.*]] = load float, ptr [[GEP_A_2]], align 4
+; CHECK-NEXT: [[L_2:%.*]] = load float, ptr [[A]], align 4
+; CHECK-NEXT: [[L_3:%.*]] = load float, ptr [[A]], align 4
+; CHECK-NEXT: br label [[LOOP:%.*]]
+; CHECK: loop:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[RED:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[RED_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[MUL11:%.*]] = fmul fast float 3.000000e+00, [[L_0]]
+; CHECK-NEXT: [[MUL12:%.*]] = fmul fast float 3.000000e+00, [[L_1]]
+; CHECK-NEXT: [[MUL14:%.*]] = fmul fast float [[X:%.*]], [[L_2]]
+; CHECK-NEXT: [[MUL16:%.*]] = fmul fast float 3.000000e+00, [[L_3]]
+; CHECK-NEXT: [[ADD:%.*]] = fadd fast float [[MUL12]], [[MUL11]]
+; CHECK-NEXT: [[ADD13:%.*]] = fadd fast float [[ADD]], [[MUL14]]
+; CHECK-NEXT: [[RED_NEXT]] = fadd fast float [[ADD13]], [[MUL16]]
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV]], 10
+; CHECK-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK: exit:
+; CHECK-NEXT: ret float [[RED_NEXT]]
+;
+entry:
+ %gep.A.1 = getelementptr inbounds float, ptr %A, i64 1
+ %l.0 = load float, ptr %gep.A.1, align 4
+ %gep.A.2 = getelementptr inbounds float, ptr %A, i64 2
+ %l.1 = load float, ptr %gep.A.2, align 4
+ %l.2 = load float, ptr %A, align 4
+ %l.3 = load float, ptr %A, align 4
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %red = phi float [ 0.000000e+00, %entry ], [ %red.next, %loop ]
+ %mul11 = fmul fast float 3.000000e+00, %l.0
+ %mul12 = fmul fast float 3.000000e+00, %l.1
+ %mul14 = fmul fast float %x, %l.2
+ %mul16 = fmul fast float 3.000000e+00, %l.3
+ %add = fadd fast float %mul12, %mul11
+ %add13 = fadd fast float %add, %mul14
+ %red.next = fadd fast float %add13, %mul16
+ %iv.next = add nuw nsw i64 %iv, 1
+ %cmp = icmp eq i64 %iv, 10
+ br i1 %cmp, label %exit, label %loop
+
+exit:
+ ret float %red.next
+}
+
+define void @slp_profitable(ptr %A, ptr %B, float %0) {
+; CHECK-LABEL: @slp_profitable(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[SUB_I1096:%.*]] = fsub fast float 1.000000e+00, [[TMP0:%.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[A:%.*]], align 4
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x float> poison, float [[TMP0]], i32 0
+; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x float> [[TMP2]], float [[TMP0]], i32 1
+; CHECK-NEXT: [[TMP4:%.*]] = fmul fast <2 x float> [[TMP1]], [[TMP3]]
+; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x float> [[TMP4]], <2 x float> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[SUB_I1096]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[SUB_I1096]], i32 1
+; CHECK-NEXT: [[TMP7:%.*]] = fmul fast <2 x float> [[TMP1]], [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = fadd fast <2 x float> [[SHUFFLE]], [[TMP7]]
+; CHECK-NEXT: [[TMP9:%.*]] = fsub fast <2 x float> [[SHUFFLE]], [[TMP7]]
+; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x float> [[TMP8]], <2 x float> [[TMP9]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: store <2 x float> [[TMP10]], ptr [[B:%.*]], align 4
+; CHECK-NEXT: ret void
+;
+entry:
+ %gep.A.1 = getelementptr inbounds float, ptr %A, i64 1
+ %sub.i1096 = fsub fast float 1.000000e+00, %0
+ %1 = load float, ptr %A, align 4
+ %mul.i1100 = fmul fast float %1, %sub.i1096
+ %2 = load float, ptr %gep.A.1, align 4
+ %mul7.i1101 = fmul fast float %2, %0
+ %add.i1102 = fadd fast float %mul7.i1101, %mul.i1100
+ %mul14.i = fmul fast float %1, %0
+ %3 = fmul fast float %2, %sub.i1096
+ %add15.i = fsub fast float %mul14.i, %3
+ store float %add.i1102, ptr %B, align 4
+ %gep.B.1 = getelementptr inbounds float, ptr %B, i64 1
+ store float %add15.i, ptr %gep.B.1, align 4
+ ret void
+}
More information about the llvm-commits
mailing list