[Mesa-dev] [PATCH 2/3] radeonsi: implement PK2H and UP2H opcodes

[Michel Dänzer]

On 03.02.2016 05:15, Marek Olšák wrote:
> On Sat, Jan 30, 2016 at 12:46 AM, Marek Olšák <maraeo at gmail.com> wrote:
>> From: Marek Olšák <marek.olsak at amd.com>
>>
>> Based on a gallivm patch by Ilia Mirkin.
>>
>> +8 piglit regressions due to precision issues

You're saying this patch causes 8 piglit tests to fail? What are the
benefits we get in exchange for that?


>> diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
>> index 05f30bd..4dacdd1 100644
>> --- a/src/gallium/drivers/radeonsi/si_pipe.c
>> +++ b/src/gallium/drivers/radeonsi/si_pipe.c
>> @@ -308,6 +308,7 @@ static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
>>         case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
>>         case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
>>         case PIPE_CAP_INVALIDATE_BUFFER:
>> +       case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
>>                 return 1;
> 
> FYI, I'm going to change it to this:
> return HAVE_LLVM >= 0x0306
> 
> because Roland said LLVM < 3.6 doesn't support fptrunc 32->16.
> 
> Alternatively, we can axe the LLVM 3.5 support code, which would clean
> up our compute code paths, but not graphics code paths.

FWIW, I think removing support for LLVM 3.5 from radeonsi would be
reasonable at this point, even more so after 11.2 branches.


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer