+void
+brw_init_shader_time(struct brw_context *brw)
+{
+ const int max_entries = 4096;
+ brw->shader_time.bo = drm_intel_bo_alloc(brw->bufmgr, "shader time",
+ max_entries * SHADER_TIME_STRIDE,
+ 4096);
+ brw->shader_time.shader_programs = rzalloc_array(brw, struct gl_shader_program *,
+ max_entries);
+ brw->shader_time.programs = rzalloc_array(brw, struct gl_program *,
+ max_entries);
+ brw->shader_time.types = rzalloc_array(brw, enum shader_time_shader_type,
+ max_entries);
+ brw->shader_time.cumulative = rzalloc_array(brw, uint64_t,
+ max_entries);
+ brw->shader_time.max_entries = max_entries;
+}
+
+static int
+compare_time(const void *a, const void *b)
+{
+ uint64_t * const *a_val = a;
+ uint64_t * const *b_val = b;
+
+ /* We don't just subtract because we're turning the value to an int. */
+ if (**a_val < **b_val)
+ return -1;
+ else if (**a_val == **b_val)
+ return 0;
+ else
+ return 1;
+}
+
+static void
+get_written_and_reset(struct brw_context *brw, int i,
+ uint64_t *written, uint64_t *reset)
+{
+ enum shader_time_shader_type type = brw->shader_time.types[i];
+ assert(type == ST_VS || type == ST_GS || type == ST_FS8 || type == ST_FS16);
+
+ /* Find where we recorded written and reset. */
+ int wi, ri;
+
+ for (wi = i; brw->shader_time.types[wi] != type + 1; wi++)
+ ;
+
+ for (ri = i; brw->shader_time.types[ri] != type + 2; ri++)
+ ;
+
+ *written = brw->shader_time.cumulative[wi];
+ *reset = brw->shader_time.cumulative[ri];
+}
+
+static void
+print_shader_time_line(const char *stage, const char *name,
+ int shader_num, uint64_t time, uint64_t total)
+{
+ fprintf(stderr, "%-6s%-18s", stage, name);
+
+ if (shader_num != -1)
+ fprintf(stderr, "%4d: ", shader_num);
+ else
+ fprintf(stderr, " : ");
+
+ fprintf(stderr, "%16lld (%7.2f Gcycles) %4.1f%%\n",
+ (long long)time,
+ (double)time / 1000000000.0,
+ (double)time / total * 100.0);
+}
+
+static void
+brw_report_shader_time(struct brw_context *brw)
+{
+ if (!brw->shader_time.bo || !brw->shader_time.num_entries)
+ return;
+
+ uint64_t scaled[brw->shader_time.num_entries];
+ uint64_t *sorted[brw->shader_time.num_entries];
+ uint64_t total_by_type[ST_FS16 + 1];
+ memset(total_by_type, 0, sizeof(total_by_type));
+ double total = 0;
+ for (int i = 0; i < brw->shader_time.num_entries; i++) {
+ uint64_t written = 0, reset = 0;
+ enum shader_time_shader_type type = brw->shader_time.types[i];
+
+ sorted[i] = &scaled[i];
+
+ switch (type) {
+ case ST_VS_WRITTEN:
+ case ST_VS_RESET:
+ case ST_GS_WRITTEN:
+ case ST_GS_RESET:
+ case ST_FS8_WRITTEN:
+ case ST_FS8_RESET:
+ case ST_FS16_WRITTEN:
+ case ST_FS16_RESET:
+ /* We'll handle these when along with the time. */
+ scaled[i] = 0;
+ continue;
+
+ case ST_VS:
+ case ST_GS:
+ case ST_FS8:
+ case ST_FS16:
+ get_written_and_reset(brw, i, &written, &reset);
+ break;
+
+ default:
+ /* I sometimes want to print things that aren't the 3 shader times.
+ * Just print the sum in that case.
+ */
+ written = 1;
+ reset = 0;
+ break;
+ }
+
+ uint64_t time = brw->shader_time.cumulative[i];
+ if (written) {
+ scaled[i] = time / written * (written + reset);
+ } else {
+ scaled[i] = time;
+ }
+
+ switch (type) {
+ case ST_VS:
+ case ST_GS:
+ case ST_FS8:
+ case ST_FS16:
+ total_by_type[type] += scaled[i];
+ break;
+ default:
+ break;
+ }
+
+ total += scaled[i];
+ }
+
+ if (total == 0) {
+ fprintf(stderr, "No shader time collected yet\n");
+ return;
+ }
+
+ qsort(sorted, brw->shader_time.num_entries, sizeof(sorted[0]), compare_time);
+
+ fprintf(stderr, "\n");
+ fprintf(stderr, "type ID cycles spent %% of total\n");
+ for (int s = 0; s < brw->shader_time.num_entries; s++) {
+ const char *shader_name;
+ const char *stage;
+ /* Work back from the sorted pointers times to a time to print. */
+ int i = sorted[s] - scaled;
+ struct gl_shader_program *prog = brw->shader_time.shader_programs[i];
+
+ if (scaled[i] == 0)
+ continue;
+
+ int shader_num = -1;
+ if (prog) {
+ shader_num = prog->Name;
+
+ /* The fixed function fragment shader generates GLSL IR with a Name
+ * of 0, and nothing else does.
+ */
+ if (prog->Label) {
+ shader_name = prog->Label;
+ } else if (shader_num == 0 &&
+ (brw->shader_time.types[i] == ST_FS8 ||
+ brw->shader_time.types[i] == ST_FS16)) {
+ shader_name = "ff";
+ shader_num = -1;
+ } else {
+ shader_name = "glsl";
+ }
+ } else if (brw->shader_time.programs[i]) {
+ shader_num = brw->shader_time.programs[i]->Id;
+ if (shader_num == 0) {
+ shader_name = "ff";
+ shader_num = -1;
+ } else {
+ shader_name = "prog";
+ }
+ } else {
+ shader_name = "other";
+ }
+
+ switch (brw->shader_time.types[i]) {
+ case ST_VS:
+ stage = "vs";
+ break;
+ case ST_GS:
+ stage = "gs";
+ break;
+ case ST_FS8:
+ stage = "fs8";
+ break;
+ case ST_FS16:
+ stage = "fs16";
+ break;
+ default:
+ stage = "other";
+ break;
+ }
+
+ print_shader_time_line(stage, shader_name, shader_num,
+ scaled[i], total);
+ }
+
+ fprintf(stderr, "\n");
+ print_shader_time_line("total", "vs", -1, total_by_type[ST_VS], total);
+ print_shader_time_line("total", "gs", -1, total_by_type[ST_GS], total);
+ print_shader_time_line("total", "fs8", -1, total_by_type[ST_FS8], total);
+ print_shader_time_line("total", "fs16", -1, total_by_type[ST_FS16], total);
+}
+
+static void
+brw_collect_shader_time(struct brw_context *brw)
+{
+ if (!brw->shader_time.bo)
+ return;
+
+ /* This probably stalls on the last rendering. We could fix that by
+ * delaying reading the reports, but it doesn't look like it's a big
+ * overhead compared to the cost of tracking the time in the first place.
+ */
+ drm_intel_bo_map(brw->shader_time.bo, true);
+
+ uint32_t *times = brw->shader_time.bo->virtual;
+
+ for (int i = 0; i < brw->shader_time.num_entries; i++) {
+ brw->shader_time.cumulative[i] += times[i * SHADER_TIME_STRIDE / 4];
+ }
+
+ /* Zero the BO out to clear it out for our next collection.
+ */
+ memset(times, 0, brw->shader_time.bo->size);
+ drm_intel_bo_unmap(brw->shader_time.bo);
+}
+
+void
+brw_collect_and_report_shader_time(struct brw_context *brw)
+{
+ brw_collect_shader_time(brw);
+
+ if (brw->shader_time.report_time == 0 ||
+ get_time() - brw->shader_time.report_time >= 1.0) {
+ brw_report_shader_time(brw);
+ brw->shader_time.report_time = get_time();
+ }
+}
+
+/**
+ * Chooses an index in the shader_time buffer and sets up tracking information
+ * for our printouts.
+ *
+ * Note that this holds on to references to the underlying programs, which may
+ * change their lifetimes compared to normal operation.
+ */
+int
+brw_get_shader_time_index(struct brw_context *brw,
+ struct gl_shader_program *shader_prog,
+ struct gl_program *prog,
+ enum shader_time_shader_type type)
+{
+ struct gl_context *ctx = &brw->ctx;
+
+ int shader_time_index = brw->shader_time.num_entries++;
+ assert(shader_time_index < brw->shader_time.max_entries);
+ brw->shader_time.types[shader_time_index] = type;
+
+ _mesa_reference_shader_program(ctx,
+ &brw->shader_time.shader_programs[shader_time_index],
+ shader_prog);
+
+ _mesa_reference_program(ctx,
+ &brw->shader_time.programs[shader_time_index],
+ prog);
+
+ return shader_time_index;
+}
+
+void
+brw_destroy_shader_time(struct brw_context *brw)
+{
+ drm_intel_bo_unreference(brw->shader_time.bo);
+ brw->shader_time.bo = NULL;
+}
+
+void
+brw_mark_surface_used(struct brw_stage_prog_data *prog_data,
+ unsigned surf_index)
+{
+ assert(surf_index < BRW_MAX_SURFACES);
+
+ prog_data->binding_table.size_bytes =
+ MAX2(prog_data->binding_table.size_bytes, (surf_index + 1) * 4);
+}
+
+bool
+brw_stage_prog_data_compare(const struct brw_stage_prog_data *a,
+ const struct brw_stage_prog_data *b)
+{
+ /* Compare all the struct up to the pointers. */
+ if (memcmp(a, b, offsetof(struct brw_stage_prog_data, param)))
+ return false;
+
+ if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
+ return false;
+
+ if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
+ return false;
+
+ return true;
+}
+
+void
+brw_stage_prog_data_free(const void *p)
+{
+ struct brw_stage_prog_data *prog_data = (struct brw_stage_prog_data *)p;
+
+ ralloc_free(prog_data->param);
+ ralloc_free(prog_data->pull_param);
+}
+
+void
+brw_dump_ir(struct brw_context *brw, const char *stage,
+ struct gl_shader_program *shader_prog,
+ struct gl_shader *shader, struct gl_program *prog)
+{
+ if (shader_prog) {
+ fprintf(stderr,
+ "GLSL IR for native %s shader %d:\n", stage, shader_prog->Name);
+ _mesa_print_ir(stderr, shader->ir, NULL);
+ fprintf(stderr, "\n\n");
+ } else {
+ fprintf(stderr, "ARB_%s_program %d ir for native %s shader\n",
+ stage, prog->Id, stage);
+ _mesa_print_program(prog);
+ }
+}