llvm-gcc build failure on x86 linux (@llvm.x86.sse.comile.ss)

[llvmbot]

Bugzilla Link	2282
Resolution	FIXED
Resolved on	May 04, 2008 16:28
Version	unspecified
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@asl

Extended Description

The LLVM build fails on x86-32 and x86-64 linux:

cc1: llvm/lib/VMCore/Instructions.cpp:301: void llvm::CallInst::init(llvm::Value , llvm::Value): Assertion `(FTy->getNumParams() == 1 || (FTy->isVarArg() && FTy->getNumParams() == 0)) && "Calling a function with bad signature"' failed.
gcc/config/i386/crtfastmath.c: In function ‘set_fast_math’:
gcc/config/i386/crtfastmath.c:48: internal compiler error: Aborted

The problem is that this code (64 bit case):

unsigned int mxcsr = __builtin_ia32_stmxcsr ();
mxcsr |= MXCSR_DAZ | MXCSR_FTZ;
__builtin_ia32_ldmxcsr (mxcsr);

is generating a call to

declare i32 @​llvm.x86.sse.comile.ss(<4 x float>, <4 x float>) nounwind readnone

but is only passing one parameter to it:
%tmp6 = bitcast i32* %memtmp5 to i8* ; <i8*> [#uses=1]

[lattner]

That is really strange. It sounds like some enum (generated from intrinsics.td) get changed but not all of llvm-gcc got properly rebuilt. Does this occur with a clean build?

[llvmbot]

This was a clean build.

[asl]

Confirmed, same here. There wasn't any problem ~5 days ago

[asl]

Fixed here:

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20080428/061936.html

[asl]

Erm, really fixed :)