[PATCH] D14257: [AsmParser] Generalize matching for grammars without mnemonic-lead statements.

[Craig Topper via llvm-commits]

I didn't initially notice the binary size. I open the X86GenAsmMatcher.inc
and noticed the new classes in the table while looking at something else.
This led me to investigate the history of the change which is when I found
out it was not reviewed. And that it increased the table size on all
targets but Hexagon with no benefit to those targets.

Further review indicated that this appears to be a suboptimal algorithm
implementation for the Hexagon target anyway due to the linear scan. Do you
have any large Hexagon asm files that you could measure parsing performance
of with this linear scan vs the previous operand reordering hack you had
mentioned?



On Fri, Jan 8, 2016 at 12:03 PM, Colin LeMahieu <colinl at codeaurora.org>
wrote:

> colinl added a comment.
>
> For reference before the r256669 the binary size of all LLVM tools in
> release mode with a MinGW build was 78,331,196 and the size with the patch
> is 77,467,273, approximately 1%.  Is the binary size difference of the
> tools what spawned the investigation?
>
> Does the list have opinions on the severity of this type of issue?
>
>
> Repository:
>   rL LLVM
>
> http://reviews.llvm.org/D14257
>
>
>
>


-- 
~Craig
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160108/c98c4f32/attachment.html>