[Mesa-dev] [PATCH 02/19] intel/fs: Use groups for SIMD16 LINTERP on gen11+

Jason Ekstrand jason at jlekstrand.net
Mon May 21 14:04:39 UTC 2018


On May 20, 2018 23:45:27 Kenneth Graunke <kenneth at whitecape.org> wrote:

> On Friday, May 18, 2018 4:35:15 PM PDT Jason Ekstrand wrote:
>> This is better than compression control because it naturally extends to
>> SIMD32.
>> ---
>> src/intel/compiler/brw_fs_generator.cpp | 6 ++----
>> 1 file changed, 2 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/intel/compiler/brw_fs_generator.cpp 
>> b/src/intel/compiler/brw_fs_generator.cpp
>> index 0c050a7..f435f97 100644
>> --- a/src/intel/compiler/brw_fs_generator.cpp
>> +++ b/src/intel/compiler/brw_fs_generator.cpp
>> @@ -795,16 +795,14 @@ fs_generator::generate_linterp(fs_inst *inst,
>>   */
>>  brw_inst_set_saturate(p->devinfo, i[0], false);
>> } else {
>> -         brw_set_default_compression_control(p, BRW_COMPRESSION_NONE);
>> +         brw_set_default_group(p, inst->group);
>>  i[0] = brw_MAD(p,            acc, dwR, offset(delta_x, 0), dwP);
>>  i[1] = brw_MAD(p, offset(dst, 0), acc, offset(delta_x, 1), dwQ);
>>
>> -         brw_set_default_compression_control(p, BRW_COMPRESSION_2NDHALF);
>> +         brw_set_default_group(p, inst->group + 8);
>>  i[2] = brw_MAD(p,            acc, dwR, offset(delta_y, 0), dwP);
>>  i[3] = brw_MAD(p, offset(dst, 1), acc, offset(delta_y, 1), dwQ);
>>
>> -         brw_set_default_compression_control(p, BRW_COMPRESSION_COMPRESSED);
>> -
>>  brw_inst_set_cond_modifier(p->devinfo, i[1], inst->conditional_mod);
>>  brw_inst_set_cond_modifier(p->devinfo, i[3], inst->conditional_mod);
>
> The last hunk you dropped is supposed to restore the original state, so
> subsequent things continue working...I think you either want to change
> it to brw_set_default_group(inst->group)...or push/pop the state around
> this whole section...

It's not needed because no more instructions get emitted and we push/pop 
around the giant generator switch anyway.  Still, I guess it doesn't hurt 
to push/pop over extra time.

> I like the change, though.





More information about the mesa-dev mailing list