From df4af41b9b734e9fdf0453ab092013199b4ad4e0 Mon Sep 17 00:00:00 2001
From: Benjamin Kramer <benny.kra@googlemail.com>
Date: Tue, 25 Dec 2012 12:54:19 +0000
Subject: X86: Custom lower <2 x i64> eq and ne when SSE41 is not available.

pcmpeqd, pshufd, pshufd, pand is cheaper than unpack + cmpq, sbbq, cmpq, sbbq + pack.
Small speedup on loop-vectorized viterbi (-march=core2).

llvm-svn: 171063
---
 llvm/test/CodeGen/X86/vec_compare.ll | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

(limited to 'llvm/test/CodeGen/X86/vec_compare.ll')

diff --git a/llvm/test/CodeGen/X86/vec_compare.ll b/llvm/test/CodeGen/X86/vec_compare.ll
index 367dd27f307..cf86c737c73 100644
--- a/llvm/test/CodeGen/X86/vec_compare.ll
+++ b/llvm/test/CodeGen/X86/vec_compare.ll
@@ -41,3 +41,29 @@ define <4 x i32> @test4(<4 x i32> %A, <4 x i32> %B) nounwind {
         %D = sext <4 x i1> %C to <4 x i32>
 	ret <4 x i32> %D
 }
+
+define <2 x i64> @test5(<2 x i64> %A, <2 x i64> %B) nounwind {
+; CHECK: test5:
+; CHECK: pcmpeqd
+; CHECK: pshufd $-11
+; CHECK: pshufd $-96
+; CHECK: pand
+; CHECK: ret
+	%C = icmp eq <2 x i64> %A, %B
+	%D = sext <2 x i1> %C to <2 x i64>
+	ret <2 x i64> %D
+}
+
+define <2 x i64> @test6(<2 x i64> %A, <2 x i64> %B) nounwind {
+; CHECK: test6:
+; CHECK: pcmpeqd
+; CHECK: pshufd $-11
+; CHECK: pshufd $-96
+; CHECK: pand
+; CHECK: pcmpeqd
+; CHECK: pxor
+; CHECK: ret
+	%C = icmp ne <2 x i64> %A, %B
+	%D = sext <2 x i1> %C to <2 x i64>
+	ret <2 x i64> %D
+}
-- 
cgit v1.2.3