#include "brw_state.h"
#include "brw_ff_gs.h"
#include "brw_nir.h"
+#include "brw_program.h"
+#include "compiler/glsl/ir_uniform.h"
static void
-assign_gs_binding_table_offsets(const struct brw_device_info *devinfo,
+brw_gs_debug_recompile(struct brw_context *brw,
+ struct gl_shader_program *shader_prog,
+ const struct brw_gs_prog_key *key)
+{
+ struct brw_cache_item *c = NULL;
+ const struct brw_gs_prog_key *old_key = NULL;
+ bool found = false;
+
+ perf_debug("Recompiling geometry shader for program %d\n",
+ shader_prog->Name);
+
+ for (unsigned int i = 0; i < brw->cache.size; i++) {
+ for (c = brw->cache.items[i]; c; c = c->next) {
+ if (c->cache_id == BRW_CACHE_GS_PROG) {
+ old_key = c->key;
+
+ if (old_key->program_string_id == key->program_string_id)
+ break;
+ }
+ }
+ if (c)
+ break;
+ }
+
+ if (!c) {
+ perf_debug(" Didn't find previous compile in the shader cache for "
+ "debug\n");
+ return;
+ }
+
+ found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
+
+ if (!found) {
+ perf_debug(" Something else\n");
+ }
+}
+
+static void
+assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
const struct gl_shader_program *shader_prog,
const struct gl_program *prog,
struct brw_gs_prog_data *prog_data)
struct brw_geometry_program *gp,
struct brw_gs_prog_key *key)
{
- struct brw_compiler *compiler = brw->intelScreen->compiler;
- struct gl_shader *shader = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
+ struct brw_compiler *compiler = brw->screen->compiler;
+ const struct gen_device_info *devinfo = &brw->screen->devinfo;
struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_gs_prog_data prog_data;
+ bool start_busy = false;
+ double start_time = 0;
+
memset(&prog_data, 0, sizeof(prog_data));
- assign_gs_binding_table_offsets(brw->intelScreen->devinfo, prog,
+ assign_gs_binding_table_offsets(devinfo, prog,
&gp->program.Base, &prog_data);
/* Allocate the references to the uniforms that will end up in the
* padding around uniform values below vec4 size, so the worst case is that
* every uniform is a float which gets padded to the size of a vec4.
*/
- struct gl_shader *gs = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
- int param_count = gp->program.Base.nir->num_uniforms * 4;
+ struct gl_linked_shader *gs = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
+ struct brw_shader *bgs = (struct brw_shader *) gs;
+ int param_count = gp->program.Base.nir->num_uniforms / 4;
prog_data.base.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.base.nr_image_params = gs->NumImages;
brw_nir_setup_glsl_uniforms(gp->program.Base.nir, prog, &gp->program.Base,
- &prog_data.base.base, compiler->scalar_gs);
+ &prog_data.base.base,
+ compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
GLbitfield64 outputs_written = gp->program.Base.OutputsWritten;
- brw_compute_vue_map(brw->intelScreen->devinfo,
+ prog_data.base.cull_distance_mask =
+ ((1 << gp->program.Base.CullDistanceArraySize) - 1) <<
+ gp->program.Base.ClipDistanceArraySize;
+
+ brw_compute_vue_map(devinfo,
&prog_data.base.vue_map, outputs_written,
- prog ? prog->SeparateShader : false);
+ prog->SeparateShader);
if (unlikely(INTEL_DEBUG & DEBUG_GS))
brw_dump_ir("geometry", prog, gs, NULL);
if (INTEL_DEBUG & DEBUG_SHADER_TIME)
st_index = brw_get_shader_time_index(brw, prog, NULL, ST_GS);
+ if (unlikely(brw->perf_debug)) {
+ start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
+ start_time = get_time();
+ }
+
void *mem_ctx = ralloc_context(NULL);
unsigned program_size;
char *error_str;
const unsigned *program =
- brw_compile_gs(brw->intelScreen->compiler, brw, mem_ctx, key,
- &prog_data, shader->Program->nir, prog,
+ brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
+ &prog_data, gs->Program->nir, prog,
st_index, &program_size, &error_str);
if (program == NULL) {
+ ralloc_strcat(&prog->InfoLog, error_str);
+ _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
+
ralloc_free(mem_ctx);
return false;
}
- /* Scratch space is used for register spilling */
- if (prog_data.base.base.total_scratch) {
- brw_get_scratch_bo(brw, &stage_state->scratch_bo,
- prog_data.base.base.total_scratch *
- brw->max_gs_threads);
+ if (unlikely(brw->perf_debug)) {
+ if (bgs->compiled_once) {
+ brw_gs_debug_recompile(brw, prog, key);
+ }
+ if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
+ perf_debug("GS compile took %.03f ms and stalled the GPU\n",
+ (get_time() - start_time) * 1000);
+ }
+ bgs->compiled_once = true;
}
+ /* Scratch space is used for register spilling */
+ brw_alloc_stage_scratch(brw, stage_state,
+ prog_data.base.base.total_scratch,
+ devinfo->max_gs_threads);
+
brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
key, sizeof(*key),
program, program_size,
}
static bool
-brw_gs_state_dirty(struct brw_context *brw)
+brw_gs_state_dirty(const struct brw_context *brw)
{
return brw_state_dirty(brw,
_NEW_TEXTURE,
struct brw_gs_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
- struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_geometry_program *gp =
(struct brw_geometry_program *) brw->geometry_program;
struct gl_program *prog = &gp->program.Base;
key->program_string_id = gp->id;
/* _NEW_TEXTURE */
- brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
- &key->tex);
+ brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
}
void