static bool
brw_codegen_cs_prog(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_compute_program *cp,
struct brw_cs_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
const GLuint *program;
void *mem_ctx = ralloc_context(NULL);
GLuint program_size;
struct brw_cs_prog_data prog_data;
bool start_busy = false;
double start_time = 0;
struct brw_shader *cs =
(struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
assert (cs);
memset(&prog_data, 0, sizeof(prog_data));
if (prog->Comp.SharedSize > 64 * 1024) {
prog->LinkStatus = false;
const char *error_str =
"Compute shader used more than 64KB of shared variables";
ralloc_strcat(&prog->InfoLog, error_str);
_mesa_problem(NULL, "Failed to link compute shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
} else {
prog_data.base.total_shared = prog->Comp.SharedSize;
}
assign_cs_binding_table_offsets(brw->intelScreen->devinfo, prog,
&cp->program.Base, &prog_data);
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*/
int param_count = cp->program.Base.nir->num_uniforms;
/* The backend also sometimes adds params for texture size. */
param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
prog_data.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.image_param =
rzalloc_array(NULL, struct brw_image_param, cs->base.NumImages);
prog_data.base.nr_params = param_count;
prog_data.base.nr_image_params = cs->base.NumImages;
brw_nir_setup_glsl_uniforms(cp->program.Base.nir, prog, &cp->program.Base,
&prog_data.base, true);
if (unlikely(brw->perf_debug)) {
start_busy = (brw->batch.last_bo &&
drm_intel_bo_busy(brw->batch.last_bo));
start_time = get_time();
}
if (unlikely(INTEL_DEBUG & DEBUG_CS))
brw_dump_ir("compute", prog, &cs->base, &cp->program.Base);
int st_index = -1;
if (INTEL_DEBUG & DEBUG_SHADER_TIME)
st_index = brw_get_shader_time_index(brw, prog, &cp->program.Base, ST_CS);
char *error_str;
program = brw_compile_cs(brw->intelScreen->compiler, brw, mem_ctx,
key, &prog_data, cp->program.Base.nir,
st_index, &program_size, &error_str);
if (program == NULL) {
prog->LinkStatus = false;
ralloc_strcat(&prog->InfoLog, error_str);
_mesa_problem(NULL, "Failed to compile compute shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
}
if (unlikely(brw->perf_debug) && cs) {
if (cs->compiled_once) {
_mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
}
cs->compiled_once = true;
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("CS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
}
if (prog_data.base.total_scratch) {
brw_get_scratch_bo(brw, &brw->cs.base.scratch_bo,
prog_data.base.total_scratch * brw->max_cs_threads);
}
if (unlikely(INTEL_DEBUG & DEBUG_CS))
fprintf(stderr, "\n");
brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
key, sizeof(*key),
program, program_size,
&prog_data, sizeof(prog_data),
&brw->cs.base.prog_offset, &brw->cs.prog_data);
ralloc_free(mem_ctx);
return true;
}
static bool
brw_codegen_cs_prog(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_program *cp,
struct brw_cs_prog_key *key)
{
const struct gen_device_info *devinfo = &brw->screen->devinfo;
struct gl_context *ctx = &brw->ctx;
const GLuint *program;
void *mem_ctx = ralloc_context(NULL);
GLuint program_size;
struct brw_cs_prog_data prog_data;
bool start_busy = false;
double start_time = 0;
struct brw_shader *cs =
(struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
assert (cs);
memset(&prog_data, 0, sizeof(prog_data));
if (prog->Comp.SharedSize > 64 * 1024) {
prog->LinkStatus = false;
const char *error_str =
"Compute shader used more than 64KB of shared variables";
ralloc_strcat(&prog->InfoLog, error_str);
_mesa_problem(NULL, "Failed to link compute shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
} else {
prog_data.base.total_shared = prog->Comp.SharedSize;
}
assign_cs_binding_table_offsets(devinfo, prog, &cp->program, &prog_data);
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*/
int param_count = cp->program.nir->num_uniforms / 4;
/* The backend also sometimes add a param for the thread local id. */
prog_data.thread_local_id_index = param_count++;
/* The backend also sometimes adds params for texture size. */
param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
prog_data.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.image_param =
rzalloc_array(NULL, struct brw_image_param, cs->base.NumImages);
prog_data.base.nr_params = param_count;
prog_data.base.nr_image_params = cs->base.NumImages;
brw_nir_setup_glsl_uniforms(cp->program.nir, prog, &cp->program,
&prog_data.base, true);
if (unlikely(brw->perf_debug)) {
start_busy = (brw->batch.last_bo &&
drm_intel_bo_busy(brw->batch.last_bo));
start_time = get_time();
}
if (unlikely(INTEL_DEBUG & DEBUG_CS))
brw_dump_ir("compute", prog, &cs->base, &cp->program);
int st_index = -1;
if (INTEL_DEBUG & DEBUG_SHADER_TIME)
st_index = brw_get_shader_time_index(brw, prog, &cp->program, ST_CS);
char *error_str;
program = brw_compile_cs(brw->screen->compiler, brw, mem_ctx, key,
&prog_data, cp->program.nir, st_index,
&program_size, &error_str);
if (program == NULL) {
prog->LinkStatus = false;
ralloc_strcat(&prog->InfoLog, error_str);
_mesa_problem(NULL, "Failed to compile compute shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
}
if (unlikely(brw->perf_debug) && cs) {
if (cs->compiled_once) {
_mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
}
cs->compiled_once = true;
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("CS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
}
const unsigned subslices = MAX2(brw->screen->subslice_total, 1);
/* WaCSScratchSize:hsw
*
* Haswell's scratch space address calculation appears to be sparse
* rather than tightly packed. The Thread ID has bits indicating
* which subslice, EU within a subslice, and thread within an EU
* it is. There's a maximum of two slices and two subslices, so these
* can be stored with a single bit. Even though there are only 10 EUs
* per subslice, this is stored in 4 bits, so there's an effective
* maximum value of 16 EUs. Similarly, although there are only 7
* threads per EU, this is stored in a 3 bit number, giving an effective
* maximum value of 8 threads per EU.
*
* This means that we need to use 16 * 8 instead of 10 * 7 for the
* number of threads per subslice.
*/
const unsigned scratch_ids_per_subslice =
brw->is_haswell ? 16 * 8 : devinfo->max_cs_threads;
brw_alloc_stage_scratch(brw, &brw->cs.base,
prog_data.base.total_scratch,
scratch_ids_per_subslice * subslices);
if (unlikely(INTEL_DEBUG & DEBUG_CS))
fprintf(stderr, "\n");
brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
key, sizeof(*key),
program, program_size,
&prog_data, sizeof(prog_data),
&brw->cs.base.prog_offset, &brw->cs.base.prog_data);
ralloc_free(mem_ctx);
return true;
}
