[PATCH] X86: lower MO_GOT_ABSOLUTE_ADDRESS in LowerSymbolOperand like all the other MO_ types.
lhames at gmail.com
Mon Jun 9 18:24:51 PDT 2014
Thanks for taking a look at this. The patch looks good in general, but I'm
concerned about the overhead of scanning all operands for all instructions
when looking for a MO_GOT_ABSOLUTE_ADDRESS. Is it possible to move the scan
in to X86MCInstLower::Lower and only do it once we see an MO_GlobalAddress?
On Mon, Jun 2, 2014 at 12:35 PM, Larry D'Anna <larry at elder-gods.org> wrote:
> I tested this by self-compiling a i686 llvm and running the regression
> * applied patch
> * compiled llvm+clang with native compiler
> * used the llvm I just built to build a i686-linux-gnu llvm
> * ran the tests on that
> I also scanned the resulting object files for the optimization that was
> by the FIXMEs that I removed. It should look like a add32ri instruction
> with a
> GOTPC reloc on the immediate operand, which has been folded into a add32mi
> instruction. This did occur in ARMDisassembler.cpp.
> Sorry for the mis-formatted attempt at submitting the patch.
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the llvm-commits