incorrect floating point results

[llvmbot]

Bugzilla Link	17988
Version	unspecified
OS	MacOS X
Attachments	Extract and type ./compile to compile the code ... Creates executable testBug
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@sunfishcode

Extended Description

Attaching a piece of C++ code which gives wrong floating point result with clang++. But gives correct results in the following compilers (tested)

1. g++ >= 4.2 and 2) icpc > 11.0

Expected output

Flux = 0.57735 0.989578 0.412228 0.412228 1.73147

Actual output from clang

Flux = 0.508632 0.797979 0.383204 0.383204 1.20769

This bug is reproducible in clang shipped with Xcode 5.0 on OSX Maverics and clang 3.1-svn. valgrind shows no errors and using -O0 does not help either.

[llvmbot]

Sorry just now confirmed this bug is visible only with the -O2 flag. So it has to do something with FP optimization.

Created attachment 11566 [details]
Extract and type ./compile to compile the code ... Creates executable testBug

Attaching a piece of C++ code which gives wrong floating point result with
clang++. But gives correct results in the following compilers (tested)

1. g++ >= 4.2 and 2) icpc > 11.0

Expected output

Flux = 0.57735 0.989578 0.412228 0.412228 1.73147

Actual output from clang

Flux = 0.508632 0.797979 0.383204 0.383204 1.20769

This bug is reproducible in clang shipped with Xcode 5.0 on OSX Maverics and
clang 3.1-svn. valgrind shows no errors and using -O0 does not help either.