Example of before:

define void %testVaCopyArg(i8* %fmt, ...) {
entry:
%ap = alloca i8*, align 4 ; <i8**> [#uses=3]
%aq = alloca i8*, align 4 ; <i8**> [#uses=3]
%memtmp = alloca i8* ; <i8**> [#uses=2]
call void %llvm.va_start( i8** %ap )
%tmp = load i8** %ap ; <i8*> [#uses=1]
store i8* %tmp, i8** %memtmp
call void %llvm.va_copy( i8** %aq, i8** %memtmp )
%tmp1 = load i8** %aq ; <i8*> [#uses=1]
call void %test( i8* %fmt, i8* %tmp1 )
call void %llvm.va_end( i8** %aq )
call void %llvm.va_end( i8** %ap )
ret void
}

after:

define void %testVaCopyArg(i8* %fmt, ...) {
entry:
%ap = alloca i8*, align 4 ; <i8**> [#uses=2]
%aq = alloca i8*, align 4 ; <i8**> [#uses=2]
%memtmp = alloca i8* ; <i8**> [#uses=2]
%ap = bitcast i8** %ap to i8* ; <i8*> [#uses=2]
call void %llvm.va_start( i8* %ap )
%tmp = load i8** %ap ; <i8*> [#uses=1]
store i8* %tmp, i8** %memtmp
%aq = bitcast i8** %aq to i8* ; <i8*> [#uses=2]
%memtmp = bitcast i8** %memtmp to i8* ; <i8*> [#uses=1]
call void %llvm.va_copy( i8* %aq, i8* %memtmp )
%tmp1 = load i8** %aq ; <i8*> [#uses=1]
call void %test( i8* %fmt, i8* %tmp1 )
call void %llvm.va_end( i8* %aq )
call void %llvm.va_end( i8* %ap )
ret void
}

the difference is that the intrinsics all take i8* as the valist pointer.

Reid, please make llvm-upgrade handle this. Thanks!

-Chris

The CFE and LangRef have been updated, the only piece left is llvm-upgrade.

What exactly is the transform you want on llvm-upgrade? I'm seeing it as:

for any intrinsic starting with llvm.va_, bitcast any pointer argument to i8*
before the instruction.

Is that correct?

yep, sounds right!

Fixed with these patches:

http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070108/042600.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070108/042601.html
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070108/042608.html