fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-27 21:20:12 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

intel/blorp: Specify a subgroup size requirement of 16 for fast clear or repclear shaders.

Browse source 
Request a fixed subgroup size for pixel shaders that require it due to
the hardware restrictions of fast clears and repeated data clears.
This requires plumbing the "is_fast_clear" boolean across several
callers since blorp_compile_fs_brw() currently has no information
regarding whether the kernel is intended for a fast clear operation.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32664>
This commit is contained in:
Francisco Jerez 2024-09-18 14:24:21 -07:00 • committed by Marge Bot
parent 80b2355b39
commit 935f60c13c
5 changed files with 25 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/intel/blorp/blorp_blit.c
View file

@ -1525,7 +1525,7 @@ blorp_get_blit_kernel_fs(struct blorp_batch *batch,
const bool multisample_fbo = key->rt_samples > 1;
const struct blorp_program p =
blorp_compile_fs(blorp, mem_ctx, nir, multisample_fbo, false);
blorp_compile_fs(blorp, mem_ctx, nir, multisample_fbo, false, false);
bool result =
blorp->upload_shader(batch, MESA_SHADER_FRAGMENT,

3
src/intel/blorp/blorp_brw.c
View file

@ -21,6 +21,7 @@ static struct blorp_program
blorp_compile_fs_brw(struct blorp_context *blorp, void *mem_ctx,
struct nir_shader *nir,
bool multisample_fbo,
bool is_fast_clear,
bool use_repclear)
{
const struct brw_compiler *compiler = blorp->compiler->brw;
@ -33,6 +34,8 @@ blorp_compile_fs_brw(struct blorp_context *blorp, void *mem_ctx,
brw_preprocess_nir(compiler, nir, &opts);
nir_remove_dead_variables(nir, nir_var_shader_in, NULL);
nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
if (is_fast_clear || use_repclear)
nir->info.subgroup_size = SUBGROUP_SIZE_REQUIRE_16;
struct brw_wm_prog_key wm_key;
memset(&wm_key, 0, sizeof(wm_key));

27
src/intel/blorp/blorp_clear.c
View file

@ -40,6 +40,7 @@
struct blorp_const_color_prog_key
{
struct blorp_base_key base;
bool is_fast_clear;
bool use_simd16_replicated_data;
bool clear_rgb_as_red;
uint8_t local_y;
@ -49,6 +50,7 @@ struct blorp_const_color_prog_key
static bool
blorp_params_get_clear_kernel_fs(struct blorp_batch *batch,
struct blorp_params *params,
bool is_fast_clear,
bool want_replicated_data,
bool clear_rgb_as_red)
{
@ -59,6 +61,7 @@ blorp_params_get_clear_kernel_fs(struct blorp_batch *batch,
const struct blorp_const_color_prog_key blorp_key = {
.base = BLORP_BASE_KEY_INIT(BLORP_SHADER_TYPE_CLEAR),
.base.shader_pipeline = BLORP_SHADER_PIPELINE_RENDER,
.is_fast_clear = is_fast_clear,
.use_simd16_replicated_data = use_replicated_data,
.clear_rgb_as_red = clear_rgb_as_red,
.local_y = 0,
@ -95,7 +98,8 @@ blorp_params_get_clear_kernel_fs(struct blorp_batch *batch,
const bool multisample_fbo = false;
struct blorp_program p =
blorp_compile_fs(blorp, mem_ctx, b.shader, multisample_fbo, use_replicated_data);
blorp_compile_fs(blorp, mem_ctx, b.shader, multisample_fbo,
is_fast_clear, use_replicated_data);
bool result =
blorp->upload_shader(batch, MESA_SHADER_FRAGMENT,
@ -183,6 +187,7 @@ blorp_params_get_clear_kernel_cs(struct blorp_batch *batch,
static bool
blorp_params_get_clear_kernel(struct blorp_batch *batch,
struct blorp_params *params,
bool is_fast_clear,
bool use_replicated_data,
bool clear_rgb_as_red)
{
@ -190,7 +195,7 @@ blorp_params_get_clear_kernel(struct blorp_batch *batch,
assert(!use_replicated_data);
return blorp_params_get_clear_kernel_cs(batch, params, clear_rgb_as_red);
} else {
return blorp_params_get_clear_kernel_fs(batch, params,
return blorp_params_get_clear_kernel_fs(batch, params, is_fast_clear,
use_replicated_data,
clear_rgb_as_red);
}
@ -481,7 +486,7 @@ blorp_fast_clear(struct blorp_batch *batch,
get_fast_clear_rect(batch->blorp->isl_dev, surf->surf, surf->aux_surf,
&params.x0, &params.y0, &params.x1, &params.y1);
if (!blorp_params_get_clear_kernel(batch, &params, true, false))
if (!blorp_params_get_clear_kernel(batch, &params, true, true, false))
return;
blorp_surface_info_init(batch, &params.dst, surf, level,
@ -654,7 +659,7 @@ blorp_clear(struct blorp_batch *batch,
if (color_write_disable)
use_simd16_replicated_data = false;
if (!blorp_params_get_clear_kernel(batch, &params,
if (!blorp_params_get_clear_kernel(batch, &params, false,
use_simd16_replicated_data,
clear_rgb_as_red))
return;
@ -820,7 +825,7 @@ blorp_clear_stencil_as_rgba(struct blorp_batch *batch,
blorp_params_init(&params);
params.op = BLORP_OP_SLOW_DEPTH_CLEAR;
if (!blorp_params_get_clear_kernel(batch, &params, true, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, true, false))
return false;
memset(&params.wm_inputs.clear_color, stencil_value,
@ -913,7 +918,7 @@ blorp_clear_depth_stencil(struct blorp_batch *batch,
* we disable statistics in 3DSTATE_WM. Give it the usual clear shader
* to work around the issue.
*/
if (!blorp_params_get_clear_kernel(batch, &params, false, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, false, false))
return;
}
@ -1137,7 +1142,7 @@ blorp_clear_attachments(struct blorp_batch *batch,
* is tiled or not, we have to assume it may be linear. This means no
* SIMD16_REPDATA for us. :-(
*/
if (!blorp_params_get_clear_kernel(batch, &params, false, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, false, false))
return;
}
@ -1269,7 +1274,7 @@ blorp_ccs_resolve(struct blorp_batch *batch,
* color" message.
*/
if (!blorp_params_get_clear_kernel(batch, &params, true, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, true, false))
return;
batch->blorp->exec(batch, &params);
@ -1359,7 +1364,7 @@ blorp_params_get_mcs_partial_resolve_kernel(struct blorp_batch *batch,
const bool multisample_fbo = true;
const struct blorp_program p =
blorp_compile_fs(blorp, mem_ctx, b.shader, multisample_fbo, false);
blorp_compile_fs(blorp, mem_ctx, b.shader, multisample_fbo, false, false);
bool result =
blorp->upload_shader(batch, MESA_SHADER_FRAGMENT,
@ -1506,7 +1511,7 @@ blorp_mcs_ambiguate(struct blorp_batch *batch,
params.wm_inputs.clear_color[0] = pixel & 0xFFFFFFFF;
params.wm_inputs.clear_color[1] = pixel >> 32;
if (!blorp_params_get_clear_kernel(batch, &params, true, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, true, false))
return;
batch->blorp->exec(batch, &params);
@ -1665,7 +1670,7 @@ blorp_ccs_ambiguate(struct blorp_batch *batch,
memset(&params.wm_inputs.clear_color, 0,
sizeof(params.wm_inputs.clear_color));
if (!blorp_params_get_clear_kernel(batch, &params, true, false))
if (!blorp_params_get_clear_kernel(batch, &params, false, true, false))
return;
batch->blorp->exec(batch, &params);

1
src/intel/blorp/blorp_elk.c
View file

@ -22,6 +22,7 @@ static struct blorp_program
blorp_compile_fs_elk(struct blorp_context *blorp, void *mem_ctx,
struct nir_shader *nir,
bool multisample_fbo,
bool is_fast_clear,
bool use_repclear)
{
const struct elk_compiler *compiler = blorp->compiler->elk;

5
src/intel/blorp/blorp_priv.h
View file

@ -48,6 +48,7 @@ struct blorp_compiler {
struct blorp_program (*compile_fs)(struct blorp_context *blorp, void *mem_ctx,
struct nir_shader *nir,
bool multisample_fbo,
bool is_fast_clear,
bool use_repclear);
struct blorp_program (*compile_vs)(struct blorp_context *blorp, void *mem_ctx,
struct nir_shader *nir);
@ -443,9 +444,11 @@ static inline struct blorp_program
blorp_compile_fs(struct blorp_context *blorp, void *mem_ctx,
struct nir_shader *nir,
bool multisample_fbo,
bool is_fast_clear,
bool use_repclear)
{
return blorp->compiler->compile_fs(blorp, mem_ctx, nir, multisample_fbo, use_repclear);
return blorp->compiler->compile_fs(blorp, mem_ctx, nir, multisample_fbo,
is_fast_clear, use_repclear);
}
static inline struct blorp_program