http://llvm.org/bugs/show_bug.cgi?id=20912
Bug ID: 20912
Summary: -freciprocal-math flag doesn't generate 'arcp' in LLVM
IR
Product: clang
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: LLVM Codegen
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
I'm filing this purely as a clang bug to generate the 'arcp' fast math flag
when -freciprocal-math is specified as a parameter. More changes are needed to
produce the correct machine code.
Reference: The addition of fast-math flags to LLVM IR happened in 2012:
http://lists.cs.uiuc.edu/pipermail/llvmdev/2012-October/054999.html
...but there appears to have been very little follow-up in making those flags
actually do anything (see also bug 20870).
$ ./clang -v
clang version 3.6.0 (217587)
Target: x86_64-apple-darwin13.3.0
Thread model: posix
$ cat reciprocal.c
float reciprocal(float x) {
return 1.0f / x;
}
$ ./clang -O1 -freciprocal-math reciprocal.c -emit-llvm -S -o -
...
%div = fdiv float 1.000000e+00, %x ; "arcp" ???
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
