[llvm-commits] [llvm-gcc-4.2] r53113 - /llvm-gcc-4.2/trunk/gcc/config/darwin.h

[Devang Patel]

On Jul 7, 2008, at 2:59 PM, Eric Christopher wrote:

>> I do not prefer separate new -mllvm option for the linker.
>>
>
> Well, if it's not an option then don't pass it :)


I meant separate option, say -mllvm-lto, for this feature.

$ llvm-gcc -mllvm "-print-isel-input -print-emitted-asm" /tmp/a.c -c

prints info useful to llvm codegen folks.

$ llvm-gcc -mllvm "-print-isel-input -print-emitted-asm" /tmp/a.c -O4

prints info useful to  llvm codegen folks when LTO is used. If the  
linker does not support -mllvm then the user wants to know that,  
instead of no output.

I do not prefer

$ llvm-gcc -mllvm-lto "-print-isel-input -print-emitted-asm" /tmp/a.c - 
O4

because, if I want to print isel-input and set loop-unroll threshold  
then I should be able to say

$ llvm-gcc -mllvm "-print-isel-input -unroll-threshold=1000" /tmp/a.c - 
O4

instead of

$ llvm-gcc -mllvm " -unroll-threshold=1000" -mllvm-lto "-print-isel- 
input" /tmp/a.c -O4

This is unnecessary and it may get confusing when LTO does loop  
unrolling also at linker level. This is why I prefer -mllvm "blah  
blah" for compiler as well as the linker.

However, if someone feels strongly, feel free to apply appropriate  
patch to do the right thing!
-
Devang