[Mesa-dev] [PATCH 2/2] radeonsi: Consider input SGPR count for compute shader SGPR count.

[Marek Olšák]

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

It looks like the key change here is the use of si_shader_create,
which calls si_fix_num_sgprs, which is what the title describes. This
is something that should be mentioned in the commit message, because
it's not immediately obvious from the patch.

Marek

On Tue, Apr 19, 2016 at 2:51 PM, Bas Nieuwenhuizen
<bas at basnieuwenhuizen.nl> wrote:
> Signed-off-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
> ---
>  src/gallium/drivers/radeonsi/si_compute.c | 16 +++++++++++-----
>  src/gallium/drivers/radeonsi/si_shader.c  |  3 ++-
>  2 files changed, 13 insertions(+), 6 deletions(-)
>
> diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
> index 557e892..905c169 100644
> --- a/src/gallium/drivers/radeonsi/si_compute.c
> +++ b/src/gallium/drivers/radeonsi/si_compute.c
> @@ -81,14 +81,20 @@ static void *si_create_compute_state(
>
>                 program->shader.selector = &sel;
>
> -               if (si_compile_tgsi_shader(sscreen, sctx->tm, &program->shader,
> -                                          true, &sctx->b.debug)) {
> +               if (si_shader_create(sscreen, sctx->tm, &program->shader,
> +                                    &sctx->b.debug)) {
>                         FREE(sel.tokens);
>                         return NULL;
>                 }
>
>                 scratch_enabled = shader->config.scratch_bytes_per_wave > 0;
>
> +               shader->config.rsrc1 =
> +                          S_00B848_VGPRS((shader->config.num_vgprs - 1) / 4) |
> +                          S_00B848_SGPRS((shader->config.num_sgprs - 1) / 8) |
> +                          S_00B848_DX10_CLAMP(1) |
> +                          S_00B848_FLOAT_MODE(shader->config.float_mode);
> +
>                 shader->config.rsrc2 = S_00B84C_USER_SGPR(SI_CS_NUM_USER_SGPR) |
>                            S_00B84C_SCRATCH_EN(scratch_enabled) |
>                            S_00B84C_TGID_X_EN(1) | S_00B84C_TGID_Y_EN(1) |
> @@ -105,10 +111,10 @@ static void *si_create_compute_state(
>                 radeon_elf_read(code, header->num_bytes, &program->shader.binary);
>                 si_shader_binary_read_config(&program->shader.binary,
>                              &program->shader.config, 0);
> +               si_shader_dump(sctx->screen, &program->shader, &sctx->b.debug,
> +                              PIPE_SHADER_COMPUTE, stderr);
> +               si_shader_binary_upload(sctx->screen, &program->shader);
>         }
> -       si_shader_dump(sctx->screen, &program->shader, &sctx->b.debug,
> -                      TGSI_PROCESSOR_COMPUTE, stderr);
> -       si_shader_binary_upload(sctx->screen, &program->shader);
>
>         return program;
>  }
> diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
> index 605b964..3bf68eb 100644
> --- a/src/gallium/drivers/radeonsi/si_shader.c
> +++ b/src/gallium/drivers/radeonsi/si_shader.c
> @@ -7022,7 +7022,8 @@ int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
>              (shader->key.vs.as_es != mainp->key.vs.as_es ||
>               shader->key.vs.as_ls != mainp->key.vs.as_ls)) ||
>             (shader->selector->type == PIPE_SHADER_TESS_EVAL &&
> -            shader->key.tes.as_es != mainp->key.tes.as_es)) {
> +            shader->key.tes.as_es != mainp->key.tes.as_es) ||
> +           shader->selector->type == PIPE_SHADER_COMPUTE) {
>                 /* Monolithic shader (compiled as a whole, has many variants,
>                  * may take a long time to compile).
>                  */
> --
> 2.8.0
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev