/*
* Copyright 2015 Advanced Micro Devices, Inc.
+ * All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
- *
- * Authors:
- * Nicolai Hähnle <nicolai.haehnle@amd.com>
- *
*/
#include "util/u_memory.h"
-#include "r600_query.h"
-#include "r600_pipe_common.h"
-#include "r600d_common.h"
+#include "radeonsi/si_query.h"
+#include "radeonsi/si_pipe.h"
+#include "amd/common/sid.h"
/* Max counters per HW block */
-#define R600_QUERY_MAX_COUNTERS 16
+#define SI_QUERY_MAX_COUNTERS 16
-static const char * const r600_pc_shader_suffix[] = {
- "", "_PS", "_VS", "_GS", "_ES", "_HS", "_LS", "_CS"
-};
-
-static struct r600_perfcounter_block *
-lookup_counter(struct r600_perfcounters *pc, unsigned index,
+static struct si_perfcounter_block *
+lookup_counter(struct si_perfcounters *pc, unsigned index,
unsigned *base_gid, unsigned *sub_index)
{
- struct r600_perfcounter_block *block = pc->blocks;
+ struct si_perfcounter_block *block = pc->blocks;
unsigned bid;
*base_gid = 0;
return NULL;
}
-static struct r600_perfcounter_block *
-lookup_group(struct r600_perfcounters *pc, unsigned *index)
+static struct si_perfcounter_block *
+lookup_group(struct si_perfcounters *pc, unsigned *index)
{
unsigned bid;
- struct r600_perfcounter_block *block = pc->blocks;
+ struct si_perfcounter_block *block = pc->blocks;
for (bid = 0; bid < pc->num_blocks; ++bid, ++block) {
if (*index < block->num_groups)
return NULL;
}
-struct r600_pc_group {
- struct r600_pc_group *next;
- struct r600_perfcounter_block *block;
+struct si_pc_group {
+ struct si_pc_group *next;
+ struct si_perfcounter_block *block;
unsigned sub_gid; /* only used during init */
unsigned result_base; /* only used during init */
int se;
int instance;
unsigned num_counters;
- unsigned selectors[R600_QUERY_MAX_COUNTERS];
+ unsigned selectors[SI_QUERY_MAX_COUNTERS];
};
-struct r600_pc_counter {
+struct si_pc_counter {
unsigned base;
- unsigned dwords;
- unsigned stride;
+ unsigned qwords;
+ unsigned stride; /* in uint64s */
};
-struct r600_query_pc {
- struct r600_query_hw b;
+#define SI_PC_SHADERS_WINDOWING (1 << 31)
+
+struct si_query_pc {
+ struct si_query_hw b;
unsigned shaders;
unsigned num_counters;
- struct r600_pc_counter *counters;
- struct r600_pc_group *groups;
+ struct si_pc_counter *counters;
+ struct si_pc_group *groups;
};
-static void r600_pc_query_destroy(struct r600_common_context *ctx,
- struct r600_query *rquery)
+static void si_pc_query_destroy(struct si_screen *sscreen,
+ struct si_query *rquery)
{
- struct r600_query_pc *query = (struct r600_query_pc *)rquery;
+ struct si_query_pc *query = (struct si_query_pc *)rquery;
while (query->groups) {
- struct r600_pc_group *group = query->groups;
+ struct si_pc_group *group = query->groups;
query->groups = group->next;
FREE(group);
}
FREE(query->counters);
- r600_query_hw_destroy(ctx, rquery);
+ si_query_hw_destroy(sscreen, rquery);
}
-static void r600_pc_query_emit_start(struct r600_common_context *ctx,
- struct r600_query_hw *hwquery,
- struct r600_resource *buffer, uint64_t va)
+static bool si_pc_query_prepare_buffer(struct si_screen *screen,
+ struct si_query_hw *hwquery,
+ struct r600_resource *buffer)
{
- struct r600_perfcounters *pc = ctx->screen->perfcounters;
- struct r600_query_pc *query = (struct r600_query_pc *)hwquery;
- struct r600_pc_group *group;
+ /* no-op */
+ return true;
+}
+
+static void si_pc_query_emit_start(struct si_context *sctx,
+ struct si_query_hw *hwquery,
+ struct r600_resource *buffer, uint64_t va)
+{
+ struct si_perfcounters *pc = sctx->screen->perfcounters;
+ struct si_query_pc *query = (struct si_query_pc *)hwquery;
+ struct si_pc_group *group;
int current_se = -1;
int current_instance = -1;
if (query->shaders)
- pc->emit_shaders(ctx, query->shaders);
+ pc->emit_shaders(sctx, query->shaders);
for (group = query->groups; group; group = group->next) {
- struct r600_perfcounter_block *block = group->block;
+ struct si_perfcounter_block *block = group->block;
if (group->se != current_se || group->instance != current_instance) {
current_se = group->se;
current_instance = group->instance;
- pc->emit_instance(ctx, group->se, group->instance);
+ pc->emit_instance(sctx, group->se, group->instance);
}
- pc->emit_select(ctx, block, group->num_counters, group->selectors);
+ pc->emit_select(sctx, block, group->num_counters, group->selectors);
}
if (current_se != -1 || current_instance != -1)
- pc->emit_instance(ctx, -1, -1);
+ pc->emit_instance(sctx, -1, -1);
- pc->emit_start(ctx, buffer, va);
+ pc->emit_start(sctx, buffer, va);
}
-static void r600_pc_query_emit_stop(struct r600_common_context *ctx,
- struct r600_query_hw *hwquery,
- struct r600_resource *buffer, uint64_t va)
+static void si_pc_query_emit_stop(struct si_context *sctx,
+ struct si_query_hw *hwquery,
+ struct r600_resource *buffer, uint64_t va)
{
- struct r600_perfcounters *pc = ctx->screen->perfcounters;
- struct r600_query_pc *query = (struct r600_query_pc *)hwquery;
- struct r600_pc_group *group;
+ struct si_perfcounters *pc = sctx->screen->perfcounters;
+ struct si_query_pc *query = (struct si_query_pc *)hwquery;
+ struct si_pc_group *group;
- pc->emit_stop(ctx, buffer, va);
+ pc->emit_stop(sctx, buffer, va);
for (group = query->groups; group; group = group->next) {
- struct r600_perfcounter_block *block = group->block;
+ struct si_perfcounter_block *block = group->block;
unsigned se = group->se >= 0 ? group->se : 0;
unsigned se_end = se + 1;
- if ((block->flags & R600_PC_BLOCK_SE) && (group->se < 0))
- se_end = ctx->screen->info.max_se;
+ if ((block->flags & SI_PC_BLOCK_SE) && (group->se < 0))
+ se_end = sctx->screen->info.max_se;
do {
unsigned instance = group->instance >= 0 ? group->instance : 0;
do {
- pc->emit_instance(ctx, se, instance);
- pc->emit_read(ctx, block,
+ pc->emit_instance(sctx, se, instance);
+ pc->emit_read(sctx, block,
group->num_counters, group->selectors,
buffer, va);
- va += 4 * group->num_counters;
+ va += sizeof(uint64_t) * group->num_counters;
} while (group->instance < 0 && ++instance < block->num_instances);
} while (++se < se_end);
}
- pc->emit_instance(ctx, -1, -1);
+ pc->emit_instance(sctx, -1, -1);
}
-static void r600_pc_query_clear_result(struct r600_query_hw *hwquery,
- union pipe_query_result *result)
+static void si_pc_query_clear_result(struct si_query_hw *hwquery,
+ union pipe_query_result *result)
{
- struct r600_query_pc *query = (struct r600_query_pc *)hwquery;
+ struct si_query_pc *query = (struct si_query_pc *)hwquery;
memset(result, 0, sizeof(result->batch[0]) * query->num_counters);
}
-static void r600_pc_query_add_result(struct r600_common_context *ctx,
- struct r600_query_hw *hwquery,
- void *buffer,
- union pipe_query_result *result)
+static void si_pc_query_add_result(struct si_screen *sscreen,
+ struct si_query_hw *hwquery,
+ void *buffer,
+ union pipe_query_result *result)
{
- struct r600_query_pc *query = (struct r600_query_pc *)hwquery;
- uint32_t *results = buffer;
+ struct si_query_pc *query = (struct si_query_pc *)hwquery;
+ uint64_t *results = buffer;
unsigned i, j;
for (i = 0; i < query->num_counters; ++i) {
- struct r600_pc_counter *counter = &query->counters[i];
-
- if (counter->base == ~0)
- continue;
+ struct si_pc_counter *counter = &query->counters[i];
- for (j = 0; j < counter->dwords; ++j) {
+ for (j = 0; j < counter->qwords; ++j) {
uint32_t value = results[counter->base + j * counter->stride];
- result->batch[i].u32 += value;
+ result->batch[i].u64 += value;
}
}
}
-static struct r600_query_ops batch_query_ops = {
- .destroy = r600_pc_query_destroy,
- .begin = r600_query_hw_begin,
- .end = r600_query_hw_end,
- .get_result = r600_query_hw_get_result
+static struct si_query_ops batch_query_ops = {
+ .destroy = si_pc_query_destroy,
+ .begin = si_query_hw_begin,
+ .end = si_query_hw_end,
+ .get_result = si_query_hw_get_result
};
-static struct r600_query_hw_ops batch_query_hw_ops = {
- .emit_start = r600_pc_query_emit_start,
- .emit_stop = r600_pc_query_emit_stop,
- .clear_result = r600_pc_query_clear_result,
- .add_result = r600_pc_query_add_result,
+static struct si_query_hw_ops batch_query_hw_ops = {
+ .prepare_buffer = si_pc_query_prepare_buffer,
+ .emit_start = si_pc_query_emit_start,
+ .emit_stop = si_pc_query_emit_stop,
+ .clear_result = si_pc_query_clear_result,
+ .add_result = si_pc_query_add_result,
};
-static struct r600_pc_group *get_group_state(struct r600_common_screen *screen,
- struct r600_query_pc *query,
- struct r600_perfcounter_block *block,
+static struct si_pc_group *get_group_state(struct si_screen *screen,
+ struct si_query_pc *query,
+ struct si_perfcounter_block *block,
unsigned sub_gid)
{
- struct r600_pc_group *group = query->groups;
+ struct si_pc_group *group = query->groups;
while (group) {
if (group->block == block && group->sub_gid == sub_gid)
group = group->next;
}
- group = CALLOC_STRUCT(r600_pc_group);
+ group = CALLOC_STRUCT(si_pc_group);
if (!group)
return NULL;
group->block = block;
group->sub_gid = sub_gid;
- if (block->flags & R600_PC_BLOCK_SHADER) {
+ if (block->flags & SI_PC_BLOCK_SHADER) {
unsigned sub_gids = block->num_instances;
unsigned shader_id;
- unsigned shader_mask;
- unsigned query_shader_mask;
+ unsigned shaders;
+ unsigned query_shaders;
- if (block->flags & R600_PC_BLOCK_SE_GROUPS)
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS)
sub_gids = sub_gids * screen->info.max_se;
shader_id = sub_gid / sub_gids;
sub_gid = sub_gid % sub_gids;
- if (shader_id == 0)
- shader_mask = R600_PC_SHADER_ALL;
- else
- shader_mask = 1 << (shader_id - 1);
+ shaders = screen->perfcounters->shader_type_bits[shader_id];
- query_shader_mask = query->shaders & R600_PC_SHADER_ALL;
- if (query_shader_mask && query_shader_mask != shader_mask) {
- fprintf(stderr, "r600_perfcounter: incompatible shader groups\n");
+ query_shaders = query->shaders & ~SI_PC_SHADERS_WINDOWING;
+ if (query_shaders && query_shaders != shaders) {
+ fprintf(stderr, "si_perfcounter: incompatible shader groups\n");
FREE(group);
return NULL;
}
- query->shaders |= shader_mask;
+ query->shaders = shaders;
}
- if (block->flags & R600_PC_BLOCK_SHADER_WINDOWED) {
+ if (block->flags & SI_PC_BLOCK_SHADER_WINDOWED && !query->shaders) {
// A non-zero value in query->shaders ensures that the shader
// masking is reset unless the user explicitly requests one.
- query->shaders |= R600_PC_SHADER_WINDOWING;
+ query->shaders = SI_PC_SHADERS_WINDOWING;
}
- if (block->flags & R600_PC_BLOCK_SE_GROUPS) {
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS) {
group->se = sub_gid / block->num_instances;
sub_gid = sub_gid % block->num_instances;
} else {
group->se = -1;
}
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS) {
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS) {
group->instance = sub_gid;
} else {
group->instance = -1;
return group;
}
-struct pipe_query *r600_create_batch_query(struct pipe_context *ctx,
- unsigned num_queries,
- unsigned *query_types)
+struct pipe_query *si_create_batch_query(struct pipe_context *ctx,
+ unsigned num_queries,
+ unsigned *query_types)
{
- struct r600_common_context *rctx = (struct r600_common_context *)ctx;
- struct r600_common_screen *screen = rctx->screen;
- struct r600_perfcounters *pc = screen->perfcounters;
- struct r600_perfcounter_block *block;
- struct r600_pc_group *group;
- struct r600_query_pc *query;
+ struct si_screen *screen =
+ (struct si_screen *)ctx->screen;
+ struct si_perfcounters *pc = screen->perfcounters;
+ struct si_perfcounter_block *block;
+ struct si_pc_group *group;
+ struct si_query_pc *query;
unsigned base_gid, sub_gid, sub_index;
unsigned i, j;
if (!pc)
return NULL;
- query = CALLOC_STRUCT(r600_query_pc);
+ query = CALLOC_STRUCT(si_query_pc);
if (!query)
return NULL;
query->b.b.ops = &batch_query_ops;
query->b.ops = &batch_query_hw_ops;
- query->b.flags = R600_QUERY_HW_FLAG_TIMER;
query->num_counters = num_queries;
for (i = 0; i < num_queries; ++i) {
unsigned sub_gid;
- if (query_types[i] < R600_QUERY_FIRST_PERFCOUNTER)
+ if (query_types[i] < SI_QUERY_FIRST_PERFCOUNTER)
goto error;
- block = lookup_counter(pc, query_types[i] - R600_QUERY_FIRST_PERFCOUNTER,
+ block = lookup_counter(pc, query_types[i] - SI_QUERY_FIRST_PERFCOUNTER,
&base_gid, &sub_index);
if (!block)
goto error;
}
/* Compute result bases and CS size per group */
- query->b.num_cs_dw_begin = pc->num_start_cs_dwords;
query->b.num_cs_dw_end = pc->num_stop_cs_dwords;
-
- query->b.num_cs_dw_begin += pc->num_instance_cs_dwords; /* conservative */
query->b.num_cs_dw_end += pc->num_instance_cs_dwords;
i = 0;
for (group = query->groups; group; group = group->next) {
- struct r600_perfcounter_block *block = group->block;
- unsigned select_dw, read_dw;
+ struct si_perfcounter_block *block = group->block;
+ unsigned read_dw;
unsigned instances = 1;
- if ((block->flags & R600_PC_BLOCK_SE) && group->se < 0)
- instances = rctx->screen->info.max_se;
+ if ((block->flags & SI_PC_BLOCK_SE) && group->se < 0)
+ instances = screen->info.max_se;
if (group->instance < 0)
instances *= block->num_instances;
group->result_base = i;
- query->b.result_size += 4 * instances * group->num_counters;
+ query->b.result_size += sizeof(uint64_t) * instances * group->num_counters;
i += instances * group->num_counters;
- pc->get_size(block, group->num_counters, group->selectors,
- &select_dw, &read_dw);
- query->b.num_cs_dw_begin += select_dw;
+ read_dw = 6 * group->num_counters;
query->b.num_cs_dw_end += instances * read_dw;
- query->b.num_cs_dw_begin += pc->num_instance_cs_dwords; /* conservative */
query->b.num_cs_dw_end += instances * pc->num_instance_cs_dwords;
}
if (query->shaders) {
- if ((query->shaders & R600_PC_SHADER_ALL) == 0)
- query->shaders |= R600_PC_SHADER_ALL;
- query->b.num_cs_dw_begin += pc->num_shaders_cs_dwords;
+ if (query->shaders == SI_PC_SHADERS_WINDOWING)
+ query->shaders = 0xffffffff;
}
/* Map user-supplied query array to result indices */
query->counters = CALLOC(num_queries, sizeof(*query->counters));
for (i = 0; i < num_queries; ++i) {
- struct r600_pc_counter *counter = &query->counters[i];
- struct r600_perfcounter_block *block;
+ struct si_pc_counter *counter = &query->counters[i];
+ struct si_perfcounter_block *block;
- block = lookup_counter(pc, query_types[i] - R600_QUERY_FIRST_PERFCOUNTER,
+ block = lookup_counter(pc, query_types[i] - SI_QUERY_FIRST_PERFCOUNTER,
&base_gid, &sub_index);
sub_gid = sub_index / block->num_selectors;
counter->base = group->result_base + j;
counter->stride = group->num_counters;
- counter->dwords = 1;
- if ((block->flags & R600_PC_BLOCK_SE) && group->se < 0)
- counter->dwords = screen->info.max_se;
+ counter->qwords = 1;
+ if ((block->flags & SI_PC_BLOCK_SE) && group->se < 0)
+ counter->qwords = screen->info.max_se;
if (group->instance < 0)
- counter->dwords *= block->num_instances;
+ counter->qwords *= block->num_instances;
}
- if (!r600_query_hw_init(rctx, &query->b))
+ if (!si_query_hw_init(screen, &query->b))
goto error;
return (struct pipe_query *)query;
error:
- r600_pc_query_destroy(rctx, &query->b.b);
+ si_pc_query_destroy(screen, &query->b.b);
return NULL;
}
-int r600_get_perfcounter_info(struct r600_common_screen *screen,
- unsigned index,
- struct pipe_driver_query_info *info)
+static bool si_init_block_names(struct si_screen *screen,
+ struct si_perfcounter_block *block)
{
- struct r600_perfcounters *pc = screen->perfcounters;
- struct r600_perfcounter_block *block;
+ unsigned i, j, k;
+ unsigned groups_shader = 1, groups_se = 1, groups_instance = 1;
+ unsigned namelen;
+ char *groupname;
+ char *p;
+
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS)
+ groups_instance = block->num_instances;
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS)
+ groups_se = screen->info.max_se;
+ if (block->flags & SI_PC_BLOCK_SHADER)
+ groups_shader = screen->perfcounters->num_shader_types;
+
+ namelen = strlen(block->basename);
+ block->group_name_stride = namelen + 1;
+ if (block->flags & SI_PC_BLOCK_SHADER)
+ block->group_name_stride += 3;
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS) {
+ assert(groups_se <= 10);
+ block->group_name_stride += 1;
+
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS)
+ block->group_name_stride += 1;
+ }
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS) {
+ assert(groups_instance <= 100);
+ block->group_name_stride += 2;
+ }
+
+ block->group_names = MALLOC(block->num_groups * block->group_name_stride);
+ if (!block->group_names)
+ return false;
+
+ groupname = block->group_names;
+ for (i = 0; i < groups_shader; ++i) {
+ const char *shader_suffix = screen->perfcounters->shader_type_suffixes[i];
+ unsigned shaderlen = strlen(shader_suffix);
+ for (j = 0; j < groups_se; ++j) {
+ for (k = 0; k < groups_instance; ++k) {
+ strcpy(groupname, block->basename);
+ p = groupname + namelen;
+
+ if (block->flags & SI_PC_BLOCK_SHADER) {
+ strcpy(p, shader_suffix);
+ p += shaderlen;
+ }
+
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS) {
+ p += sprintf(p, "%d", j);
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS)
+ *p++ = '_';
+ }
+
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS)
+ p += sprintf(p, "%d", k);
+
+ groupname += block->group_name_stride;
+ }
+ }
+ }
+
+ assert(block->num_selectors <= 1000);
+ block->selector_name_stride = block->group_name_stride + 4;
+ block->selector_names = MALLOC(block->num_groups * block->num_selectors *
+ block->selector_name_stride);
+ if (!block->selector_names)
+ return false;
+
+ groupname = block->group_names;
+ p = block->selector_names;
+ for (i = 0; i < block->num_groups; ++i) {
+ for (j = 0; j < block->num_selectors; ++j) {
+ sprintf(p, "%s_%03d", groupname, j);
+ p += block->selector_name_stride;
+ }
+ groupname += block->group_name_stride;
+ }
+
+ return true;
+}
+
+int si_get_perfcounter_info(struct si_screen *screen,
+ unsigned index,
+ struct pipe_driver_query_info *info)
+{
+ struct si_perfcounters *pc = screen->perfcounters;
+ struct si_perfcounter_block *block;
unsigned base_gid, sub;
if (!pc)
if (!block)
return 0;
+ if (!block->selector_names) {
+ if (!si_init_block_names(screen, block))
+ return 0;
+ }
info->name = block->selector_names + sub * block->selector_name_stride;
- info->query_type = R600_QUERY_FIRST_PERFCOUNTER + index;
+ info->query_type = SI_QUERY_FIRST_PERFCOUNTER + index;
info->max_value.u64 = 0;
- info->type = PIPE_DRIVER_QUERY_TYPE_UINT;
- info->result_type = PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE;
+ info->type = PIPE_DRIVER_QUERY_TYPE_UINT64;
+ info->result_type = PIPE_DRIVER_QUERY_RESULT_TYPE_AVERAGE;
info->group_id = base_gid + sub / block->num_selectors;
info->flags = PIPE_DRIVER_QUERY_FLAG_BATCH;
+ if (sub > 0 && sub + 1 < block->num_selectors * block->num_groups)
+ info->flags |= PIPE_DRIVER_QUERY_FLAG_DONT_LIST;
return 1;
}
-int r600_get_perfcounter_group_info(struct r600_common_screen *screen,
- unsigned index,
- struct pipe_driver_query_group_info *info)
+int si_get_perfcounter_group_info(struct si_screen *screen,
+ unsigned index,
+ struct pipe_driver_query_group_info *info)
{
- struct r600_perfcounters *pc = screen->perfcounters;
- struct r600_perfcounter_block *block;
+ struct si_perfcounters *pc = screen->perfcounters;
+ struct si_perfcounter_block *block;
if (!pc)
return 0;
block = lookup_group(pc, &index);
if (!block)
return 0;
+
+ if (!block->group_names) {
+ if (!si_init_block_names(screen, block))
+ return 0;
+ }
info->name = block->group_names + index * block->group_name_stride;
info->num_queries = block->num_selectors;
info->max_active_queries = block->num_counters;
return 1;
}
-void r600_perfcounters_destroy(struct r600_common_screen *rscreen)
+void si_perfcounters_destroy(struct si_screen *sscreen)
{
- if (rscreen->perfcounters)
- rscreen->perfcounters->cleanup(rscreen);
+ if (sscreen->perfcounters)
+ sscreen->perfcounters->cleanup(sscreen);
}
-boolean r600_perfcounters_init(struct r600_perfcounters *pc,
- unsigned num_blocks)
+bool si_perfcounters_init(struct si_perfcounters *pc,
+ unsigned num_blocks)
{
- pc->blocks = CALLOC(num_blocks, sizeof(struct r600_perfcounter_block));
+ pc->blocks = CALLOC(num_blocks, sizeof(struct si_perfcounter_block));
if (!pc->blocks)
- return FALSE;
+ return false;
- pc->separate_se = debug_get_bool_option("RADEON_PC_SEPARATE_SE", FALSE);
- pc->separate_instance = debug_get_bool_option("RADEON_PC_SEPARATE_INSTANCE", FALSE);
+ pc->separate_se = debug_get_bool_option("RADEON_PC_SEPARATE_SE", false);
+ pc->separate_instance = debug_get_bool_option("RADEON_PC_SEPARATE_INSTANCE", false);
- return TRUE;
+ return true;
}
-boolean r600_perfcounters_add_block(struct r600_common_screen *rscreen,
- struct r600_perfcounters *pc,
- const char *name, unsigned flags,
- unsigned counters, unsigned selectors,
- unsigned instances, void *data)
+void si_perfcounters_add_block(struct si_screen *sscreen,
+ struct si_perfcounters *pc,
+ const char *name, unsigned flags,
+ unsigned counters, unsigned selectors,
+ unsigned instances, void *data)
{
- struct r600_perfcounter_block *block = &pc->blocks[pc->num_blocks];
- unsigned i, j, k;
- unsigned groups_shader = 1, groups_se = 1, groups_instance = 1;
- unsigned namelen;
- char *groupname;
- char *p;
+ struct si_perfcounter_block *block = &pc->blocks[pc->num_blocks];
- assert(counters <= R600_QUERY_MAX_COUNTERS);
+ assert(counters <= SI_QUERY_MAX_COUNTERS);
block->basename = name;
block->flags = flags;
block->num_instances = MAX2(instances, 1);
block->data = data;
- if (pc->separate_se && (block->flags & R600_PC_BLOCK_SE))
- block->flags |= R600_PC_BLOCK_SE_GROUPS;
+ if (pc->separate_se && (block->flags & SI_PC_BLOCK_SE))
+ block->flags |= SI_PC_BLOCK_SE_GROUPS;
if (pc->separate_instance && block->num_instances > 1)
- block->flags |= R600_PC_BLOCK_INSTANCE_GROUPS;
+ block->flags |= SI_PC_BLOCK_INSTANCE_GROUPS;
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS) {
- groups_instance = block->num_instances;
- block->num_groups = groups_instance;
+ if (block->flags & SI_PC_BLOCK_INSTANCE_GROUPS) {
+ block->num_groups = block->num_instances;
} else {
block->num_groups = 1;
}
- if (block->flags & R600_PC_BLOCK_SE_GROUPS) {
- groups_se = rscreen->info.max_se;
- block->num_groups *= groups_se;
- }
-
- if (block->flags & R600_PC_BLOCK_SHADER) {
- groups_shader = ARRAY_SIZE(r600_pc_shader_suffix);
- block->num_groups *= groups_shader;
- }
-
- namelen = strlen(name);
- block->group_name_stride = namelen + 1;
- if (block->flags & R600_PC_BLOCK_SHADER)
- block->group_name_stride += 3;
- if (block->flags & R600_PC_BLOCK_SE_GROUPS) {
- assert(groups_se <= 10);
- block->group_name_stride += 1;
-
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS)
- block->group_name_stride += 1;
- }
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS) {
- assert(groups_instance <= 100);
- block->group_name_stride += 2;
- }
-
- block->group_names = MALLOC(block->num_groups * block->group_name_stride);
- if (!block->group_names)
- goto error;
-
- groupname = block->group_names;
- for (i = 0; i < groups_shader; ++i) {
- unsigned shaderlen = strlen(r600_pc_shader_suffix[i]);
- for (j = 0; j < groups_se; ++j) {
- for (k = 0; k < groups_instance; ++k) {
- strcpy(groupname, name);
- p = groupname + namelen;
-
- if (block->flags & R600_PC_BLOCK_SHADER) {
- strcpy(p, r600_pc_shader_suffix[i]);
- p += shaderlen;
- }
-
- if (block->flags & R600_PC_BLOCK_SE_GROUPS) {
- p += sprintf(p, "%d", j);
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS)
- *p++ = '_';
- }
-
- if (block->flags & R600_PC_BLOCK_INSTANCE_GROUPS)
- p += sprintf(p, "%d", k);
-
- groupname += block->group_name_stride;
- }
- }
- }
-
- assert(selectors <= 1000);
- block->selector_name_stride = block->group_name_stride + 4;
- block->selector_names = MALLOC(block->num_groups * selectors *
- block->selector_name_stride);
- if (!block->selector_names)
- goto error_groupnames;
-
- groupname = block->group_names;
- p = block->selector_names;
- for (i = 0; i < block->num_groups; ++i) {
- for (j = 0; j < selectors; ++j) {
- sprintf(p, "%s_%03d", groupname, j);
- p += block->selector_name_stride;
- }
- groupname += block->group_name_stride;
- }
+ if (block->flags & SI_PC_BLOCK_SE_GROUPS)
+ block->num_groups *= sscreen->info.max_se;
+ if (block->flags & SI_PC_BLOCK_SHADER)
+ block->num_groups *= pc->num_shader_types;
++pc->num_blocks;
pc->num_groups += block->num_groups;
-
- return TRUE;
-
-error_groupnames:
- FREE(block->group_names);
-error:
- return FALSE;
}
-void r600_perfcounters_do_destroy(struct r600_perfcounters *pc)
+void si_perfcounters_do_destroy(struct si_perfcounters *pc)
{
unsigned i;
projects / mesa.git / blobdiff