[llvm] [SLP][NFC]Add extra test for reductions vectorization, NFC (PR #226618)
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 25 18:26:53 PDT 2026
https://github.com/alexey-bataev created https://github.com/llvm/llvm-project/pull/226618
None
>From 193f16a8acacbf392055646a93fd07b956c11fc1 Mon Sep 17 00:00:00 2001
From: Alexey Bataev <a.bataev at outlook.com>
Date: Fri, 25 Sep 2026 18:26:41 -0700
Subject: [PATCH] =?UTF-8?q?[=F0=9D=98=80=F0=9D=97=BD=F0=9D=97=BF]=20initia?=
=?UTF-8?q?l=20version?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Created using spr 1.3.7
---
.../AArch64/ordered-reduction-of-loads.ll | 48 ++++++++++++
.../SLPVectorizer/RISCV/ordered-reduction.ll | 77 +++++++++++++++++++
2 files changed, 125 insertions(+)
create mode 100644 llvm/test/Transforms/SLPVectorizer/AArch64/ordered-reduction-of-loads.ll
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/ordered-reduction-of-loads.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/ordered-reduction-of-loads.ll
new file mode 100644
index 00000000000000..f97a28867d7666
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/ordered-reduction-of-loads.ll
@@ -0,0 +1,48 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=slp-vectorizer -S -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
+
+define void @ordered_rdx_of_loads(ptr %p, ptr %q, i64 %n) {
+; CHECK-LABEL: define void @ordered_rdx_of_loads(
+; CHECK-SAME: ptr [[P:%.*]], ptr [[Q:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: br label %[[LOOP:.*]]
+; CHECK: [[LOOP]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 1, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds double, ptr [[P]], i64 [[IV]]
+; CHECK-NEXT: [[GEP_PREV:%.*]] = getelementptr inbounds i8, ptr [[GEP]], i64 -8
+; CHECK-NEXT: [[L0:%.*]] = load double, ptr [[GEP_PREV]], align 8
+; CHECK-NEXT: [[L1:%.*]] = load double, ptr [[GEP]], align 8
+; CHECK-NEXT: [[ADD0:%.*]] = fadd double [[L0]], [[L1]]
+; CHECK-NEXT: [[GEP_NEXT:%.*]] = getelementptr inbounds double, ptr [[P]], i64 [[IV_NEXT]]
+; CHECK-NEXT: [[L2:%.*]] = load double, ptr [[GEP_NEXT]], align 8
+; CHECK-NEXT: [[ADD1:%.*]] = fadd double [[ADD0]], [[L2]]
+; CHECK-NEXT: [[GEP_Q:%.*]] = getelementptr inbounds double, ptr [[Q]], i64 [[IV]]
+; CHECK-NEXT: store double [[ADD1]], ptr [[GEP_Q]], align 8
+; CHECK-NEXT: [[CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop
+
+loop:
+ %iv = phi i64 [ 1, %entry ], [ %iv.next, %loop ]
+ %iv.next = add nuw nsw i64 %iv, 1
+ %gep = getelementptr inbounds double, ptr %p, i64 %iv
+ %gep.prev = getelementptr inbounds i8, ptr %gep, i64 -8
+ %l0 = load double, ptr %gep.prev, align 8
+ %l1 = load double, ptr %gep, align 8
+ %add0 = fadd double %l0, %l1
+ %gep.next = getelementptr inbounds double, ptr %p, i64 %iv.next
+ %l2 = load double, ptr %gep.next, align 8
+ %add1 = fadd double %add0, %l2
+ %gep.q = getelementptr inbounds double, ptr %q, i64 %iv
+ store double %add1, ptr %gep.q, align 8
+ %cmp = icmp eq i64 %iv.next, %n
+ br i1 %cmp, label %exit, label %loop
+
+exit:
+ ret void
+}
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/ordered-reduction.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/ordered-reduction.ll
index 55abb29ddbdd43..a70c673aaff051 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/ordered-reduction.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/ordered-reduction.ll
@@ -74,3 +74,80 @@ for.body:
for.end:
ret void
}
+
+define void @ordered_rdx_non_pow2_fallback(ptr %p, double %x) {
+; CHECK-LABEL: define void @ordered_rdx_non_pow2_fallback(
+; CHECK-SAME: ptr [[P:%.*]], double [[X:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x double> poison, double [[X]], i64 0
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x double> [[TMP0]], <4 x double> poison, <4 x i32> <i32 poison, i32 poison, i32 0, i32 0>
+; CHECK-NEXT: br label %[[LOOP:.*]]
+; CHECK: [[LOOP]]:
+; CHECK-NEXT: [[ACC:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[SUM5:%.*]], %[[LOOP]] ]
+; CHECK-NEXT: [[TMP2:%.*]] = load <2 x double>, ptr [[P]], align 8
+; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <2 x double> [[TMP2]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x double> [[TMP1]], <4 x double> [[TMP3]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], splat (double 1.000000e+00)
+; CHECK-NEXT: [[TMP6:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP5]], <4 x double> splat (double 2.000000e+00), <4 x double> splat (double 3.000000e+00))
+; CHECK-NEXT: [[TMP7:%.*]] = fadd <4 x double> [[TMP6]], splat (double 1.000000e+00)
+; CHECK-NEXT: [[TMP8:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP7]], <4 x double> splat (double 2.000000e+00), <4 x double> splat (double 3.000000e+00))
+; CHECK-NEXT: [[TMP9:%.*]] = fadd <4 x double> [[TMP8]], splat (double 1.000000e+00)
+; CHECK-NEXT: [[TMP10:%.*]] = call <4 x double> @llvm.fmuladd.v4f64(<4 x double> [[TMP9]], <4 x double> splat (double 2.000000e+00), <4 x double> splat (double 3.000000e+00))
+; CHECK-NEXT: [[R4:%.*]] = call double @llvm.fmuladd.f64(double [[X]], double 2.000000e+00, double 3.000000e+00)
+; CHECK-NEXT: [[R5:%.*]] = call double @llvm.fmuladd.f64(double [[X]], double 2.000000e+00, double 3.000000e+00)
+; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x double> [[TMP10]], i64 0
+; CHECK-NEXT: [[SUM0:%.*]] = fadd double [[ACC]], [[TMP11]]
+; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x double> [[TMP10]], i64 1
+; CHECK-NEXT: [[SUM1:%.*]] = fadd double [[SUM0]], [[TMP12]]
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x double> [[TMP10]], i64 2
+; CHECK-NEXT: [[SUM2:%.*]] = fadd double [[SUM1]], [[TMP13]]
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x double> [[TMP10]], i64 3
+; CHECK-NEXT: [[SUM3:%.*]] = fadd double [[SUM2]], [[TMP14]]
+; CHECK-NEXT: [[SUM4:%.*]] = fadd double [[SUM3]], [[R4]]
+; CHECK-NEXT: [[SUM5]] = fadd double [[SUM4]], [[R5]]
+; CHECK-NEXT: br label %[[LOOP]]
+;
+entry:
+ br label %loop
+
+loop:
+ %acc = phi double [ 0.000000e+00, %entry ], [ %sum5, %loop ]
+ %l0 = load double, ptr %p, align 8
+ %a0 = fadd double %l0, 1.000000e+00
+ %m0 = call double @llvm.fmuladd.f64(double %a0, double 2.000000e+00, double 3.000000e+00)
+ %b0 = fadd double %m0, 1.000000e+00
+ %n0 = call double @llvm.fmuladd.f64(double %b0, double 2.000000e+00, double 3.000000e+00)
+ %c0 = fadd double %n0, 1.000000e+00
+ %r0 = call double @llvm.fmuladd.f64(double %c0, double 2.000000e+00, double 3.000000e+00)
+ %gep1 = getelementptr i8, ptr %p, i64 8
+ %l1 = load double, ptr %gep1, align 8
+ %a1 = fadd double %l1, 1.000000e+00
+ %m1 = call double @llvm.fmuladd.f64(double %a1, double 2.000000e+00, double 3.000000e+00)
+ %b1 = fadd double %m1, 1.000000e+00
+ %n1 = call double @llvm.fmuladd.f64(double %b1, double 2.000000e+00, double 3.000000e+00)
+ %c1 = fadd double %n1, 1.000000e+00
+ %r1 = call double @llvm.fmuladd.f64(double %c1, double 2.000000e+00, double 3.000000e+00)
+ %a2 = fadd double %x, 1.000000e+00
+ %m2 = call double @llvm.fmuladd.f64(double %a2, double 2.000000e+00, double 3.000000e+00)
+ %b2 = fadd double %m2, 1.000000e+00
+ %n2 = call double @llvm.fmuladd.f64(double %b2, double 2.000000e+00, double 3.000000e+00)
+ %c2 = fadd double %n2, 1.000000e+00
+ %r2 = call double @llvm.fmuladd.f64(double %c2, double 2.000000e+00, double 3.000000e+00)
+ %a3 = fadd double %x, 1.000000e+00
+ %m3 = call double @llvm.fmuladd.f64(double %a3, double 2.000000e+00, double 3.000000e+00)
+ %b3 = fadd double %m3, 1.000000e+00
+ %n3 = call double @llvm.fmuladd.f64(double %b3, double 2.000000e+00, double 3.000000e+00)
+ %c3 = fadd double %n3, 1.000000e+00
+ %r3 = call double @llvm.fmuladd.f64(double %c3, double 2.000000e+00, double 3.000000e+00)
+ %r4 = call double @llvm.fmuladd.f64(double %x, double 2.000000e+00, double 3.000000e+00)
+ %r5 = call double @llvm.fmuladd.f64(double %x, double 2.000000e+00, double 3.000000e+00)
+ %sum0 = fadd double %acc, %r0
+ %sum1 = fadd double %sum0, %r1
+ %sum2 = fadd double %sum1, %r2
+ %sum3 = fadd double %sum2, %r3
+ %sum4 = fadd double %sum3, %r4
+ %sum5 = fadd double %sum4, %r5
+ br label %loop
+}
+
+declare double @llvm.fmuladd.f64(double, double, double)
More information about the llvm-commits
mailing list