From: Sanjay Patel <spatel@rotateright.com>
Date: Thu, 5 Sep 2019 16:58:18 +0000 (+0000)
Subject: [x86] add test for horizontal math bug (PR43225); NFC
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9d3619238e7564986329a741448117e1df90e0b4;p=llvm

[x86] add test for horizontal math bug (PR43225); NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@371088 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/test/CodeGen/X86/haddsub-shuf-undef-operand.ll b/test/CodeGen/X86/haddsub-shuf-undef-operand.ll
new file mode 100644
index 00000000000..3eec47a657f
--- /dev/null
+++ b/test/CodeGen/X86/haddsub-shuf-undef-operand.ll
@@ -0,0 +1,26 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-- -mattr=avx  | FileCheck %s
+
+; FIXME: Eliminating a shuffle means we have to replace an undef operand of a horizontal op.
+
+define void @PR43225(<4 x double>* %p0, <4 x double>* %p1, <4 x double> %x, <4 x double> %y, <4 x double> %z) nounwind {
+; CHECK-LABEL: PR43225:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vmovaps (%rdi), %ymm0
+; CHECK-NEXT:    vmovapd (%rsi), %ymm0
+; CHECK-NEXT:    vhsubpd %ymm0, %ymm2, %ymm0
+; CHECK-NEXT:    vmovapd %ymm0, (%rdi)
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    retq
+  %t39 = load volatile <4 x double>, <4 x double>* %p0, align 32
+  %shuffle11 = shufflevector <4 x double> %t39, <4 x double> %x, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
+  %t40 = load volatile <4 x double>, <4 x double>* %p1, align 32
+  %t41 = tail call <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double> %shuffle11, <4 x double> %t40)
+  %t42 = tail call <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double> %z, <4 x double> %t41)
+  %shuffle12 = shufflevector <4 x double> %t42, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
+  store volatile <4 x double> %shuffle12, <4 x double>* %p0, align 32
+  ret void
+}
+
+declare <4 x double> @llvm.x86.avx.hadd.pd.256(<4 x double>, <4 x double>)
+declare <4 x double> @llvm.x86.avx.hsub.pd.256(<4 x double>, <4 x double>)