[Mesa-dev] [PATCH 3/6] cso: Optimize cso_save/restore_fragment_samplers

[Michel Dänzer]

On 16/12/16 08:22 PM, Nicolai Hähnle wrote:
> On 16.12.2016 10:52, Michel Dänzer wrote:
>> From: Michel Dänzer <michel.daenzer at amd.com>
>>
>> Only copy/memset the pointers that actually need to be.
>>
>> Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
>> ---
>>  src/gallium/auxiliary/cso_cache/cso_context.c | 21 +++++++++++++++++----
>>  1 file changed, 17 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c
>> b/src/gallium/auxiliary/cso_cache/cso_context.c
>> index f52969d366..03f78eea53 100644
>> --- a/src/gallium/auxiliary/cso_cache/cso_context.c
>> +++ b/src/gallium/auxiliary/cso_cache/cso_context.c
>> @@ -1267,8 +1267,10 @@ cso_save_fragment_samplers(struct cso_context
>> *ctx)
>>     struct sampler_info *saved = &ctx->fragment_samplers_saved;
>>
>>     saved->nr_samplers = info->nr_samplers;
>> -   memcpy(saved->cso_samplers, info->cso_samplers,
>> sizeof(info->cso_samplers));
>> -   memcpy(saved->samplers, info->samplers, sizeof(info->samplers));
>> +   memcpy(saved->cso_samplers, info->cso_samplers, info->nr_samplers *
>> +          sizeof(*info->cso_samplers));
>> +   memcpy(saved->samplers, info->samplers, info->nr_samplers *
>> +          sizeof(*info->samplers));
>>  }
>>
>>
>> @@ -1277,9 +1279,20 @@ cso_restore_fragment_samplers(struct
>> cso_context *ctx)
>>  {
>>     struct sampler_info *info = &ctx->samplers[PIPE_SHADER_FRAGMENT];
>>     struct sampler_info *saved = &ctx->fragment_samplers_saved;
>> +   int delta = info->nr_samplers - saved->nr_samplers;
>> +
>> +   memcpy(info->cso_samplers, saved->cso_samplers,
>> +          saved->nr_samplers * sizeof(*info->cso_samplers));
>> +   memcpy(info->samplers, saved->samplers,
>> +          saved->nr_samplers * sizeof(*info->samplers));
>> +
>> +   if (delta > 0) {
> 
> I'd be more comfortable with an explicit info->nr_samplers >
> saved->nr_samplers given that nr_samplers is an unsigned.

How about:

   int delta = (int)info->nr_samplers - saved->nr_samplers;

This generates identical code as without the cast, whereas with the
explicit comparison the ELF text section grows by 48 bytes.


> Apart from that, patches 1-3:
> 
> Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>

Thanks!


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