[llvm-commits] [LLVMdev] [PATCH] BasicBlock Autovectorization Pass

[Duncan Sands]

Hi James,

>> this would make LLVM's IR optimizers slower than GCC's tree optimizers
>> according to my benchmarking of their relative speeds.
>
> Purely OOI, by how much, roughly?

actually it looks like I was wrong.  I just recomputed an estimate from some
benchmark numbers I have comparing gcc-4.6 and LLVM on darwin.  Currently the
GCC tree level optimizers take more or less 60% longer than LLVM's at -O3.  So
if LLVM's slow down by 23% that still leaves GCC's as 30% slower.  That said,
I'm pretty sure that the last time I compared these times on Linux (for a small
set of huge programs) the difference was much smaller.  This may be due to me
remembering wrong, to the different set of programs, or to LLVM being optimized
for darwin (eg: thanks to all the work to reduce the malloc traffic - it seems
that malloc is particularly slow on darwin).

Ciao, Duncan.

>
> Cheers,
>
> James
>
> -----Original Message-----
> From: llvm-commits-bounces at cs.uiuc.edu
> [mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Duncan Sands
> Sent: 01 December 2011 08:27
> To: llvm-commits at cs.uiuc.edu
> Subject: Re: [llvm-commits] [LLVMdev] [PATCH] BasicBlock Autovectorization
> Pass
>
> Hi Hal,
>
>> Over the entire test suite:
>>
>> default vs. -vectorize:
>>
>> The mean compile-time slowdown was 23%,
>
> this would make LLVM's IR optimizers slower than GCC's tree optimizers
> according to my benchmarking of their relative speeds.
>
> Ciao, Duncan.
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
>