[Mesa-dev] [PATCH 8/9] r600: add support for hw atomic counters. (v2)

[Dave Airlie]

On 3 November 2017 at 23:59, Gert Wollny <gw.fossdev at gmail.com> wrote:
> Am Freitag, den 03.11.2017, 17:24 +1000 schrieb Dave Airlie:
>>
>> diff --git a/src/gallium/drivers/r600/r600_pipe.c
>> b/src/gallium/drivers/r600/r600_pipe.c
>> index d67a22b..434596b 100644
>> --- a/src/gallium/drivers/r600/r600_pipe.c
>> +++ b/src/gallium/drivers/r600/r600_pipe.c
>> @@ -74,6 +74,8 @@ static void r600_destroy_context(struct
>> pipe_context *context)
>>       r600_resource_reference(&rctx->dummy_cmask, NULL);
>>       r600_resource_reference(&rctx->dummy_fmask, NULL);
>>
>> +     if (rctx->append_fence)
>> +             pipe_resource_reference((struct
>> pipe_resource**)&rctx->append_fence, NULL);
>>       for (sh = 0; sh < PIPE_SHADER_TYPES; sh++) {
>>               rctx->b.b.set_constant_buffer(&rctx->b.b, sh,
>> R600_BUFFER_INFO_CONST_BUFFER, NULL);
>>               free(rctx->driver_consts[sh].constants);
>> @@ -186,6 +188,9 @@ static struct pipe_context
>> *r600_create_context(struct pipe_screen *screen,
>>                                          rctx->b.family ==
>> CHIP_CAICOS ||
>>                                          rctx->b.family ==
>> CHIP_CAYMAN ||
>>                                          rctx->b.family ==
>> CHIP_ARUBA);
>> +
>> +             rctx->append_fence = pipe_buffer_create(rctx-
>> >b.b.screen, PIPE_BIND_CUSTOM,
>> +                                                      PIPE_USAGE_
>> DEFAULT, 32);
>>               break;
>>       default:
>>               R600_ERR("Unsupported chip class %d.\n", rctx-
>> >b.chip_class);
>> @@ -605,7 +610,15 @@ static int r600_get_shader_param(struct
>> pipe_screen* pscreen,
>>       case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
>>       case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
>>       case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
>> +             if (rscreen->b.family >= CHIP_CEDAR && rscreen-
>> >has_atomics)
>> +                     return 8;
>> +             return 0;
>
> I don't think the other cases should fall through. It makes quite a few
> piglits fail on BARTS because the number of registers is too high if no
> merging is applied.

Oops cut-n-paste/rebase error,

Thanks,
Dave.