[Mesa-dev] [PATCH 1/6] i965/fs: Lower integer multiplication after optimizations.
jason at jlekstrand.net
Sat May 16 12:40:15 PDT 2015
On Sat, May 16, 2015 at 11:25 AM, Matt Turner <mattst88 at gmail.com> wrote:
> On Sat, May 16, 2015 at 8:37 AM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>> On Fri, May 15, 2015 at 2:02 PM, Matt Turner <mattst88 at gmail.com> wrote:
>>> 32-bit x 32-bit integer multiplication requires multiple instructions
>>> until Broadwell. This patch just lets us treat the MUL instruction in
>>> the FS backend like it operates on Broadwell, and after optimizations
>>> we lower it into a sequence of instructions on older platforms.
>> Any reason why you chose to re-use the MUL instruction with D types
>> rather than making a logical IMUL instruction? I think I have a minor
>> preference for IMUL but I also understand that it can be a pain to
>> plumb through.
> Yes, basically I'm just changing the integer multiply IR from
> pre-Broadwell to Broadwell semantics, and the lowering pass sorts out
> the pre-Broadwell differences. Since you still have to have MUL, this
> seems cleaner than adding an additional virtual operation (which
> wouldn't be necessary on Broadwell).
Fine by me.
More information about the mesa-dev