[llvm-commits] CVS: llvm/test/Regression/CodeGen/X86/2006-10-07-ScalarSSEMiscompile.ll

[Chris Lattner]

Changes in directory llvm/test/Regression/CodeGen/X86:

2006-10-07-ScalarSSEMiscompile.ll added (r1.1)
---
Log message:

new testcase we miscompile


---
Diffs of the changes:  (+17 -0)

 2006-10-07-ScalarSSEMiscompile.ll |   17 +++++++++++++++++
 1 files changed, 17 insertions(+)


Index: llvm/test/Regression/CodeGen/X86/2006-10-07-ScalarSSEMiscompile.ll
diff -c /dev/null llvm/test/Regression/CodeGen/X86/2006-10-07-ScalarSSEMiscompile.ll:1.1
*** /dev/null	Sat Oct  7 16:54:18 2006
--- llvm/test/Regression/CodeGen/X86/2006-10-07-ScalarSSEMiscompile.ll	Sat Oct  7 16:54:08 2006
***************
*** 0 ****
--- 1,17 ----
+ ; RUN: llvm-as < %s | llc -march=x86 -mattr=sse | grep movaps
+ ; Test that the load is NOT folded into the intrinsic, which would zero the top
+ ; elts of the loaded vector.
+ 
+ target endian = little
+ target pointersize = 32
+ target triple = "i686-apple-darwin8.7.2"
+ 
+ implementation   ; Functions:
+ 
+ <4 x float> %test(<4 x float> %A, <4 x float>* %B) {
+ 	%BV = load <4 x float>* %B
+ 	%tmp28 = tail call <4 x float> %llvm.x86.sse.sub.ss( <4 x float> %A, <4 x float> %BV)
+ 	ret <4 x float> %tmp28
+ }
+ 
+ declare <4 x float> %llvm.x86.sse.sub.ss(<4 x float>, <4 x float>)