[llvm] [SLP][NFC]Add a test with the poor cost seed instructions, NFC (PR #206352)
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Sun Jun 28 10:42:56 PDT 2026
https://github.com/alexey-bataev created https://github.com/llvm/llvm-project/pull/206352
None
>From ea4a312636d840db0c754d0119cadfc9c3cf0389 Mon Sep 17 00:00:00 2001
From: Alexey Bataev <a.bataev at outlook.com>
Date: Sun, 28 Jun 2026 10:42:45 -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
---
.../X86/poor-throughput-seeds.ll | 106 ++++++++++++++++++
1 file changed, 106 insertions(+)
create mode 100644 llvm/test/Transforms/SLPVectorizer/X86/poor-throughput-seeds.ll
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/poor-throughput-seeds.ll b/llvm/test/Transforms/SLPVectorizer/X86/poor-throughput-seeds.ll
new file mode 100644
index 0000000000000..c7e91a4412191
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/poor-throughput-seeds.ll
@@ -0,0 +1,106 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=icelake-server -passes=slp-vectorizer -S | FileCheck %s
+
+declare double @llvm.sqrt.f64(double)
+
+define void @prim(double %x, double %y, double %z, double %w, ptr %p0, ptr %p1) {
+; CHECK-LABEL: define void @prim(
+; CHECK-SAME: double [[X:%.*]], double [[Y:%.*]], double [[Z:%.*]], double [[W:%.*]], ptr [[P0:%.*]], ptr [[P1:%.*]]) #[[ATTR1:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[X1:%.*]] = fsub double [[X]], [[Z]]
+; CHECK-NEXT: [[Y1:%.*]] = fadd double [[Y]], [[W]]
+; CHECK-NEXT: [[TMP8:%.*]] = fdiv double [[X1]], [[Z]]
+; CHECK-NEXT: [[TMP9:%.*]] = fdiv double [[Y1]], [[W]]
+; CHECK-NEXT: [[X3:%.*]] = fsub double [[TMP8]], [[Z]]
+; CHECK-NEXT: [[Y3:%.*]] = fadd double [[TMP9]], [[W]]
+; CHECK-NEXT: store double [[X3]], ptr [[P0]], align 8
+; CHECK-NEXT: store double [[Y3]], ptr [[P1]], align 8
+; CHECK-NEXT: ret void
+;
+entry:
+ %x1 = fsub double %x, %z
+ %y1 = fadd double %y, %w
+ %x2 = fdiv double %x1, %z
+ %y2 = fdiv double %y1, %w
+ %x3 = fsub double %x2, %z
+ %y3 = fadd double %y2, %w
+ store double %x3, ptr %p0
+ store double %y3, ptr %p1
+ ret void
+}
+
+define void @prim_sqrt(double %x, double %y, double %z, double %w, ptr %p0, ptr %p1) {
+; CHECK-LABEL: define void @prim_sqrt(
+; CHECK-SAME: double [[X:%.*]], double [[Y:%.*]], double [[Z:%.*]], double [[W:%.*]], ptr [[P0:%.*]], ptr [[P1:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[X1:%.*]] = fsub double [[X]], [[Z]]
+; CHECK-NEXT: [[Y1:%.*]] = fadd double [[Y]], [[W]]
+; CHECK-NEXT: [[TMP8:%.*]] = call double @llvm.sqrt.f64(double [[X1]])
+; CHECK-NEXT: [[TMP9:%.*]] = call double @llvm.sqrt.f64(double [[Y1]])
+; CHECK-NEXT: [[X3:%.*]] = fsub double [[TMP8]], [[Z]]
+; CHECK-NEXT: [[Y3:%.*]] = fadd double [[TMP9]], [[W]]
+; CHECK-NEXT: store double [[X3]], ptr [[P0]], align 8
+; CHECK-NEXT: store double [[Y3]], ptr [[P1]], align 8
+; CHECK-NEXT: ret void
+;
+entry:
+ %x1 = fsub double %x, %z
+ %y1 = fadd double %y, %w
+ %x2 = call double @llvm.sqrt.f64(double %x1)
+ %y2 = call double @llvm.sqrt.f64(double %y1)
+ %x3 = fsub double %x2, %z
+ %y3 = fadd double %y2, %w
+ store double %x3, ptr %p0
+ store double %y3, ptr %p1
+ ret void
+}
+
+define void @prim_v4(double %x0, double %x1, double %x2, double %x3, double %d0, double %d1, double %d2, double %d3, ptr %p0, ptr %p1, ptr %p2, ptr %p3) {
+; CHECK-LABEL: define void @prim_v4(
+; CHECK-SAME: double [[X0:%.*]], double [[X1:%.*]], double [[X2:%.*]], double [[X3:%.*]], double [[D0:%.*]], double [[D1:%.*]], double [[D2:%.*]], double [[D3:%.*]], ptr [[P0:%.*]], ptr [[P1:%.*]], ptr [[P2:%.*]], ptr [[P3:%.*]]) #[[ATTR1]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[A0:%.*]] = fsub double [[X0]], [[D0]]
+; CHECK-NEXT: [[A1:%.*]] = fadd double [[X1]], [[D1]]
+; CHECK-NEXT: [[A2:%.*]] = fsub double [[X2]], [[D2]]
+; CHECK-NEXT: [[A3:%.*]] = fadd double [[X3]], [[D3]]
+; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x double> poison, double [[A1]], i32 0
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> [[TMP0]], double [[A3]], i32 1
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[A0]], i32 2
+; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[A2]], i32 3
+; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x double> poison, double [[D1]], i32 0
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP4]], double [[D3]], i32 1
+; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x double> [[TMP5]], double [[D0]], i32 2
+; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP6]], double [[D2]], i32 3
+; CHECK-NEXT: [[TMP8:%.*]] = fdiv <4 x double> [[TMP3]], [[TMP7]]
+; CHECK-NEXT: [[TMP9:%.*]] = fadd <4 x double> [[TMP8]], [[TMP7]]
+; CHECK-NEXT: [[TMP10:%.*]] = fsub <4 x double> [[TMP8]], [[TMP7]]
+; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <4 x double> [[TMP9]], <4 x double> [[TMP10]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x double> [[TMP11]], i32 2
+; CHECK-NEXT: store double [[TMP14]], ptr [[P0]], align 8
+; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x double> [[TMP11]], i32 0
+; CHECK-NEXT: store double [[TMP15]], ptr [[P1]], align 8
+; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x double> [[TMP11]], i32 3
+; CHECK-NEXT: store double [[TMP16]], ptr [[P2]], align 8
+; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x double> [[TMP11]], i32 1
+; CHECK-NEXT: store double [[TMP17]], ptr [[P3]], align 8
+; CHECK-NEXT: ret void
+;
+entry:
+ %a0 = fsub double %x0, %d0
+ %a1 = fadd double %x1, %d1
+ %a2 = fsub double %x2, %d2
+ %a3 = fadd double %x3, %d3
+ %q0 = fdiv double %a0, %d0
+ %q1 = fdiv double %a1, %d1
+ %q2 = fdiv double %a2, %d2
+ %q3 = fdiv double %a3, %d3
+ %r0 = fsub double %q0, %d0
+ %r1 = fadd double %q1, %d1
+ %r2 = fsub double %q2, %d2
+ %r3 = fadd double %q3, %d3
+ store double %r0, ptr %p0
+ store double %r1, ptr %p1
+ store double %r2, ptr %p2
+ store double %r3, ptr %p3
+ ret void
+}
More information about the llvm-commits
mailing list