[Mesa-dev] [PATCH 1/7] draw/tgsi: implement geom shader invocation support.

[Ilia Mirkin]

On Wed, May 27, 2015 at 7:42 AM, Roland Scheidegger <sroland at vmware.com> wrote:
> Am 27.05.2015 um 09:45 schrieb Dave Airlie:
>> From: Dave Airlie <airlied at redhat.com>
>>
>> This is just for softpipe, llvmpipe won't work without
>> some changes.
>>
>> Signed-off-by: Dave Airlie <airlied at redhat.com>
>> ---
>>  src/gallium/auxiliary/draw/draw_gs.c   | 47 +++++++++++++++++++++-------------
>>  src/gallium/auxiliary/draw/draw_gs.h   |  2 ++
>>  src/gallium/auxiliary/tgsi/tgsi_scan.c |  2 ++
>>  src/gallium/auxiliary/tgsi/tgsi_scan.h |  1 +
>>  4 files changed, 34 insertions(+), 18 deletions(-)
>>
>> diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
>> index 6375d41..755e527 100644
>> --- a/src/gallium/auxiliary/draw/draw_gs.c
>> +++ b/src/gallium/auxiliary/draw/draw_gs.c
>> @@ -190,9 +190,15 @@ static void tgsi_gs_prepare(struct draw_geometry_shader *shader,
>>                              const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS])
>>  {
>>     struct tgsi_exec_machine *machine = shader->machine;
>> -
>> +   int j;
>>     tgsi_exec_set_constant_buffers(machine, PIPE_MAX_CONSTANT_BUFFERS,
>>                                    constants, constants_size);
>> +
>> +   if (shader->info.uses_invocationid) {
>> +      unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INVOCATIONID];
>> +      for (j = 0; j < TGSI_QUAD_SIZE; j++)
>> +         machine->SystemValue[i].i[j] = shader->invocation_id;
>> +   }
>>  }
>>
>>  static unsigned tgsi_gs_run(struct draw_geometry_shader *shader,
>> @@ -555,7 +561,7 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
>>      * overflown vertices into some area where they won't harm anyone */
>>     unsigned total_verts_per_buffer = shader->primitive_boundary *
>>        num_in_primitives;
>> -
>> +   unsigned invocation;
>>     //Assume at least one primitive
>>     max_out_prims = MAX2(max_out_prims, 1);
>>
>> @@ -564,7 +570,7 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
>>     output_verts->stride = output_verts->vertex_size;
>>     output_verts->verts =
>>        (struct vertex_header *)MALLOC(output_verts->vertex_size *
>> -                                     total_verts_per_buffer);
>> +                                     total_verts_per_buffer * shader->num_invocations);
> This can't get very large, right?

With ARB_gpu_shader5, it has to be at least 32. In mesa, it's always
defined as 32.

  -ilia