#include "nvc0/nvc0_query.h"
#include "nvc0/nvc0_query_sw.h"
#include "nvc0/nvc0_query_hw.h"
+#include "nvc0/nvc0_query_hw_metric.h"
#include "nvc0/nvc0_query_hw_sm.h"
static struct pipe_query *
return q->funcs->begin_query(nvc0_context(pipe), q);
}
-static void
+static bool
nvc0_end_query(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_query *q = nvc0_query(pq);
q->funcs->end_query(nvc0_context(pipe), q);
+ return true;
}
static boolean
return q->funcs->get_query_result(nvc0_context(pipe), q, wait, result);
}
+static void
+nvc0_get_query_result_resource(struct pipe_context *pipe,
+ struct pipe_query *pq,
+ boolean wait,
+ enum pipe_query_value_type result_type,
+ int index,
+ struct pipe_resource *resource,
+ unsigned offset)
+{
+ struct nvc0_query *q = nvc0_query(pq);
+ if (!q->funcs->get_query_result_resource) {
+ assert(!"Unexpected lack of get_query_result_resource");
+ return;
+ }
+ q->funcs->get_query_result_resource(nvc0_context(pipe), q, wait, result_type,
+ index, resource, offset);
+}
+
static void
nvc0_render_condition(struct pipe_context *pipe,
struct pipe_query *pq,
- boolean condition, uint mode)
+ boolean condition, enum pipe_render_cond_flag mode)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
/* NOTE: comparison of 2 queries only works if both have completed */
switch (q->type) {
case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
+ case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
cond = condition ? NVC0_3D_COND_MODE_EQUAL :
- NVC0_3D_COND_MODE_NOT_EQUAL;
+ NVC0_3D_COND_MODE_NOT_EQUAL;
wait = true;
break;
case PIPE_QUERY_OCCLUSION_COUNTER:
case PIPE_QUERY_OCCLUSION_PREDICATE:
+ case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
if (likely(!condition)) {
if (unlikely(hq->nesting))
cond = wait ? NVC0_3D_COND_MODE_NOT_EQUAL :
nvc0->cond_mode = mode;
if (!pq) {
- PUSH_SPACE(push, 1);
+ PUSH_SPACE(push, 2);
IMMED_NVC0(push, NVC0_3D(COND_MODE), cond);
+ if (nvc0->screen->compute)
+ IMMED_NVC0(push, NVC0_CP(COND_MODE), cond);
return;
}
if (wait)
- nvc0_hw_query_fifo_wait(push, q);
+ nvc0_hw_query_fifo_wait(nvc0, q);
- PUSH_SPACE(push, 7);
+ PUSH_SPACE(push, 10);
PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NVC0(push, NVC0_3D(COND_ADDRESS_HIGH), 3);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
BEGIN_NVC0(push, NVC0_2D(COND_ADDRESS_HIGH), 2);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
PUSH_DATA (push, hq->bo->offset + hq->offset);
+ if (nvc0->screen->compute) {
+ BEGIN_NVC0(push, NVC0_CP(COND_ADDRESS_HIGH), 3);
+ PUSH_DATAh(push, hq->bo->offset + hq->offset);
+ PUSH_DATA (push, hq->bo->offset + hq->offset);
+ PUSH_DATA (push, cond);
+ }
}
-/* === DRIVER STATISTICS === */
-
-#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
-
-static const char *nvc0_sw_query_drv_stat_names[] =
-{
- "drv-tex_obj_current_count",
- "drv-tex_obj_current_bytes",
- "drv-buf_obj_current_count",
- "drv-buf_obj_current_bytes_vid",
- "drv-buf_obj_current_bytes_sys",
- "drv-tex_transfers_rd",
- "drv-tex_transfers_wr",
- "drv-tex_copy_count",
- "drv-tex_blit_count",
- "drv-tex_cache_flush_count",
- "drv-buf_transfers_rd",
- "drv-buf_transfers_wr",
- "drv-buf_read_bytes_staging_vid",
- "drv-buf_write_bytes_direct",
- "drv-buf_write_bytes_staging_vid",
- "drv-buf_write_bytes_staging_sys",
- "drv-buf_copy_bytes",
- "drv-buf_non_kernel_fence_sync_count",
- "drv-any_non_kernel_fence_sync_count",
- "drv-query_sync_count",
- "drv-gpu_serialize_count",
- "drv-draw_calls_array",
- "drv-draw_calls_indexed",
- "drv-draw_calls_fallback_count",
- "drv-user_buffer_upload_bytes",
- "drv-constbuf_upload_count",
- "drv-constbuf_upload_bytes",
- "drv-pushbuf_count",
- "drv-resource_validate_count"
-};
-
-#endif /* NOUVEAU_ENABLE_DRIVER_STATISTICS */
-
-/* === PERFORMANCE MONITORING COUNTERS for NVE4+ === */
-
-/* NOTE: intentionally using the same names as NV */
-static const char *nve4_hw_sm_query_names[] =
-{
- /* MP counters */
- "active_cycles",
- "active_warps",
- "atom_count",
- "branch",
- "divergent_branch",
- "gld_request",
- "global_ld_mem_divergence_replays",
- "global_store_transaction",
- "global_st_mem_divergence_replays",
- "gred_count",
- "gst_request",
- "inst_executed",
- "inst_issued",
- "inst_issued1",
- "inst_issued2",
- "l1_global_load_hit",
- "l1_global_load_miss",
- "l1_local_load_hit",
- "l1_local_load_miss",
- "l1_local_store_hit",
- "l1_local_store_miss",
- "l1_shared_load_transactions",
- "l1_shared_store_transactions",
- "local_load",
- "local_load_transactions",
- "local_store",
- "local_store_transactions",
- "prof_trigger_00",
- "prof_trigger_01",
- "prof_trigger_02",
- "prof_trigger_03",
- "prof_trigger_04",
- "prof_trigger_05",
- "prof_trigger_06",
- "prof_trigger_07",
- "shared_load",
- "shared_load_replay",
- "shared_store",
- "shared_store_replay",
- "sm_cta_launched",
- "threads_launched",
- "uncached_global_load_transaction",
- "warps_launched",
- /* metrics, i.e. functions of the MP counters */
- "metric-ipc", /* inst_executed, clock */
- "metric-ipac", /* inst_executed, active_cycles */
- "metric-ipec", /* inst_executed, (bool)inst_executed */
- "metric-achieved_occupancy", /* active_warps, active_cycles */
- "metric-sm_efficiency", /* active_cycles, clock */
- "metric-inst_replay_overhead" /* inst_issued, inst_executed */
-};
-
-/* === PERFORMANCE MONITORING COUNTERS for NVC0:NVE4 === */
-static const char *nvc0_hw_sm_query_names[] =
-{
- /* MP counters */
- "active_cycles",
- "active_warps",
- "atom_count",
- "branch",
- "divergent_branch",
- "gld_request",
- "gred_count",
- "gst_request",
- "inst_executed",
- "inst_issued1_0",
- "inst_issued1_1",
- "inst_issued2_0",
- "inst_issued2_1",
- "local_load",
- "local_store",
- "prof_trigger_00",
- "prof_trigger_01",
- "prof_trigger_02",
- "prof_trigger_03",
- "prof_trigger_04",
- "prof_trigger_05",
- "prof_trigger_06",
- "prof_trigger_07",
- "shared_load",
- "shared_store",
- "threads_launched",
- "thread_inst_executed_0",
- "thread_inst_executed_1",
- "thread_inst_executed_2",
- "thread_inst_executed_3",
- "warps_launched",
-};
-
int
nvc0_screen_get_driver_query_info(struct pipe_screen *pscreen,
unsigned id,
struct pipe_driver_query_info *info)
{
struct nvc0_screen *screen = nvc0_screen(pscreen);
- int count = 0;
-
- count += NVC0_SW_QUERY_DRV_STAT_COUNT;
+ int num_sw_queries = 0, num_hw_queries = 0;
- if (screen->base.device->drm_version >= 0x01000101) {
- if (screen->compute) {
- if (screen->base.class_3d == NVE4_3D_CLASS) {
- count += NVE4_HW_SM_QUERY_COUNT;
- } else
- if (screen->base.class_3d < NVE4_3D_CLASS) {
- count += NVC0_HW_SM_QUERY_COUNT;
- }
- }
- }
+ num_sw_queries = nvc0_sw_get_driver_query_info(screen, 0, NULL);
+ num_hw_queries = nvc0_hw_get_driver_query_info(screen, 0, NULL);
if (!info)
- return count;
+ return num_sw_queries + num_hw_queries;
/* Init default values. */
info->name = "this_is_not_the_query_you_are_looking_for";
info->max_value.u64 = 0;
info->type = PIPE_DRIVER_QUERY_TYPE_UINT64;
info->group_id = -1;
+ info->flags = 0;
#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
- if (id < NVC0_SW_QUERY_DRV_STAT_COUNT) {
- info->name = nvc0_sw_query_drv_stat_names[id];
- info->query_type = NVC0_SW_QUERY_DRV_STAT(id);
- info->max_value.u64 = 0;
- if (strstr(info->name, "bytes"))
- info->type = PIPE_DRIVER_QUERY_TYPE_BYTES;
- info->group_id = NVC0_SW_QUERY_DRV_STAT_GROUP;
- return 1;
- } else
+ if (id < num_sw_queries)
+ return nvc0_sw_get_driver_query_info(screen, id, info);
#endif
- if (id < count) {
- if (screen->compute) {
- if (screen->base.class_3d == NVE4_3D_CLASS) {
- info->name = nve4_hw_sm_query_names[id - NVC0_SW_QUERY_DRV_STAT_COUNT];
- info->query_type = NVE4_HW_SM_QUERY(id - NVC0_SW_QUERY_DRV_STAT_COUNT);
- info->max_value.u64 =
- (id < NVE4_HW_SM_QUERY_METRIC_MP_OCCUPANCY) ? 0 : 100;
- info->group_id = NVC0_HW_SM_QUERY_GROUP;
- return 1;
- } else
- if (screen->base.class_3d < NVE4_3D_CLASS) {
- info->name = nvc0_hw_sm_query_names[id - NVC0_SW_QUERY_DRV_STAT_COUNT];
- info->query_type = NVC0_HW_SM_QUERY(id - NVC0_SW_QUERY_DRV_STAT_COUNT);
- info->group_id = NVC0_HW_SM_QUERY_GROUP;
- return 1;
- }
- }
- }
- /* user asked for info about non-existing query */
- return 0;
+
+ return nvc0_hw_get_driver_query_info(screen, id - num_sw_queries, info);
}
int
count++;
#endif
- if (screen->base.device->drm_version >= 0x01000101) {
+ if (screen->base.drm->version >= 0x01000101) {
if (screen->compute) {
- if (screen->base.class_3d == NVE4_3D_CLASS) {
- count++;
- } else
- if (screen->base.class_3d < NVE4_3D_CLASS) {
- count++;
+ if (screen->base.class_3d <= GM200_3D_CLASS) {
+ count += 2;
}
}
}
if (id == NVC0_HW_SM_QUERY_GROUP) {
if (screen->compute) {
info->name = "MP counters";
- info->type = PIPE_DRIVER_QUERY_GROUP_TYPE_GPU;
-
- /* Because we can't expose the number of hardware counters needed for
- * each different query, we don't want to allow more than one active
- * query simultaneously to avoid failure when the maximum number of
- * counters is reached. Note that these groups of GPU counters are
- * currently only used by AMD_performance_monitor.
- */
- info->max_active_queries = 1;
-
- if (screen->base.class_3d == NVE4_3D_CLASS) {
- info->num_queries = NVE4_HW_SM_QUERY_COUNT;
- return 1;
- } else
- if (screen->base.class_3d < NVE4_3D_CLASS) {
- info->num_queries = NVC0_HW_SM_QUERY_COUNT;
+
+ /* Expose the maximum number of hardware counters available, although
+ * some queries use more than one counter. Expect failures in that
+ * case but as performance counters are for developers, this should
+ * not have a real impact. */
+ info->max_active_queries = 8;
+ info->num_queries = nvc0_hw_sm_get_num_queries(screen);
+ return 1;
+ }
+ } else
+ if (id == NVC0_HW_METRIC_QUERY_GROUP) {
+ if (screen->compute) {
+ if (screen->base.class_3d <= GM200_3D_CLASS) {
+ info->name = "Performance metrics";
+ info->max_active_queries = 4; /* A metric uses at least 2 queries */
+ info->num_queries = nvc0_hw_metric_get_num_queries(screen);
return 1;
}
}
#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
else if (id == NVC0_SW_QUERY_DRV_STAT_GROUP) {
info->name = "Driver statistics";
- info->type = PIPE_DRIVER_QUERY_GROUP_TYPE_CPU;
info->max_active_queries = NVC0_SW_QUERY_DRV_STAT_COUNT;
info->num_queries = NVC0_SW_QUERY_DRV_STAT_COUNT;
return 1;
info->name = "this_is_not_the_query_group_you_are_looking_for";
info->max_active_queries = 0;
info->num_queries = 0;
- info->type = 0;
return 0;
}
+static void
+nvc0_set_active_query_state(struct pipe_context *pipe, boolean enable)
+{
+}
+
void
nvc0_init_query_functions(struct nvc0_context *nvc0)
{
pipe->begin_query = nvc0_begin_query;
pipe->end_query = nvc0_end_query;
pipe->get_query_result = nvc0_get_query_result;
+ pipe->get_query_result_resource = nvc0_get_query_result_resource;
+ pipe->set_active_query_state = nvc0_set_active_query_state;
pipe->render_condition = nvc0_render_condition;
+ nvc0->cond_condmode = NVC0_3D_COND_MODE_ALWAYS;
}