summaryrefslogtreecommitdiff
path: root/test
diff options
context:
space:
mode:
authorEvan Cheng <evan.cheng@apple.com>2011-10-07 17:21:44 +0000
committerEvan Cheng <evan.cheng@apple.com>2011-10-07 17:21:44 +0000
commit7c1780c5fe1e57bb57bf1e4bc43e159f03a8b556 (patch)
treee77f2d6805a926eabf9b93a212e3cf2709b328ff /test
parent6d2f9cec715c50bca44816d9bdea97f8b63bf2a0 (diff)
downloadllvm-7c1780c5fe1e57bb57bf1e4bc43e159f03a8b556.tar.gz
llvm-7c1780c5fe1e57bb57bf1e4bc43e159f03a8b556.tar.bz2
llvm-7c1780c5fe1e57bb57bf1e4bc43e159f03a8b556.tar.xz
High bits of movmskp{s|d} and pmovmskb are known zero. rdar://10247336
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141371 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
-rw-r--r--test/CodeGen/X86/movmsk.ll (renamed from test/CodeGen/X86/2011-05-31-movmsk.ll)31
1 files changed, 31 insertions, 0 deletions
diff --git a/test/CodeGen/X86/2011-05-31-movmsk.ll b/test/CodeGen/X86/movmsk.ll
index 2b54d5cc85..2368548bfa 100644
--- a/test/CodeGen/X86/2011-05-31-movmsk.ll
+++ b/test/CodeGen/X86/movmsk.ll
@@ -77,3 +77,34 @@ entry:
%shr.i = lshr i32 %2, 31
ret i32 %shr.i
}
+
+; rdar://10247336
+; movmskp{s|d} only set low 4/2 bits, high bits are known zero
+
+define i32 @t1(<4 x float> %x, i32* nocapture %indexTable) nounwind uwtable readonly ssp {
+entry:
+; CHECK: t1:
+; CHECK: movmskps
+; CHECK-NOT: movslq
+ %0 = tail call i32 @llvm.x86.sse.movmsk.ps(<4 x float> %x) nounwind
+ %idxprom = sext i32 %0 to i64
+ %arrayidx = getelementptr inbounds i32* %indexTable, i64 %idxprom
+ %1 = load i32* %arrayidx, align 4
+ ret i32 %1
+}
+
+define i32 @t2(<4 x float> %x, i32* nocapture %indexTable) nounwind uwtable readonly ssp {
+entry:
+; CHECK: t2:
+; CHECK: movmskpd
+; CHECK-NOT: movslq
+ %0 = bitcast <4 x float> %x to <2 x double>
+ %1 = tail call i32 @llvm.x86.sse2.movmsk.pd(<2 x double> %0) nounwind
+ %idxprom = sext i32 %1 to i64
+ %arrayidx = getelementptr inbounds i32* %indexTable, i64 %idxprom
+ %2 = load i32* %arrayidx, align 4
+ ret i32 %2
+}
+
+declare i32 @llvm.x86.sse2.movmsk.pd(<2 x double>) nounwind readnone
+declare i32 @llvm.x86.sse.movmsk.ps(<4 x float>) nounwind readnone