summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
Diffstat (limited to 'src')
-rw-r--r--src/mesa/main/dd.h41
-rw-r--r--src/mesa/main/mtypes.h24
-rw-r--r--src/mesa/main/performance_query.c632
-rw-r--r--src/mesa/main/performance_query.h6
4 files changed, 301 insertions, 402 deletions
diff --git a/src/mesa/main/dd.h b/src/mesa/main/dd.h
index 7ebd084ca31..aba301cf22e 100644
--- a/src/mesa/main/dd.h
+++ b/src/mesa/main/dd.h
@@ -780,6 +780,47 @@ struct dd_function_table {
/*@}*/
/**
+ * \name Performance Query objects
+ */
+ /*@{*/
+ unsigned (*InitPerfQueryInfo)(struct gl_context *ctx);
+ void (*GetPerfQueryInfo)(struct gl_context *ctx,
+ unsigned queryIndex,
+ const char **name,
+ GLuint *dataSize,
+ GLuint *numCounters,
+ GLuint *numActive);
+ void (*GetPerfCounterInfo)(struct gl_context *ctx,
+ unsigned queryIndex,
+ unsigned counterIndex,
+ const char **name,
+ const char **desc,
+ GLuint *offset,
+ GLuint *data_size,
+ GLuint *type_enum,
+ GLuint *data_type_enum,
+ GLuint64 *raw_max);
+ struct gl_perf_query_object * (*NewPerfQueryObject)(struct gl_context *ctx,
+ unsigned queryIndex);
+ void (*DeletePerfQuery)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj);
+ GLboolean (*BeginPerfQuery)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj);
+ void (*EndPerfQuery)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj);
+ void (*WaitPerfQuery)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj);
+ GLboolean (*IsPerfQueryReady)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj);
+ void (*GetPerfQueryData)(struct gl_context *ctx,
+ struct gl_perf_query_object *obj,
+ GLsizei dataSize,
+ GLuint *data,
+ GLuint *bytesWritten);
+ /*@}*/
+
+
+ /**
* \name GREMEDY debug/marker functions
*/
/*@{*/
diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h
index 9f4f9613943..d07391debbb 100644
--- a/src/mesa/main/mtypes.h
+++ b/src/mesa/main/mtypes.h
@@ -1860,6 +1860,23 @@ struct gl_perf_monitor_group
/**
+ * A query object instance as described in INTEL_performance_query.
+ *
+ * NB: We want to keep this and the corresponding backend structure
+ * relatively lean considering that applications may expect to
+ * allocate enough objects to be able to query around all draw calls
+ * in a frame.
+ */
+struct gl_perf_query_object
+{
+ GLuint Id; /**< hash table ID/name */
+ unsigned Used:1; /**< has been used for 1 or more queries */
+ unsigned Active:1; /**< inside Begin/EndPerfQuery */
+ unsigned Ready:1; /**< result is ready? */
+};
+
+
+/**
* Context state for AMD_performance_monitor.
*/
struct gl_perf_monitor_state
@@ -1878,12 +1895,7 @@ struct gl_perf_monitor_state
*/
struct gl_perf_query_state
{
- /** Array of performance monitor groups (indexed by group ID) */
- const struct gl_perf_monitor_group *Groups;
- GLuint NumGroups;
-
- /** The table of all performance query objects. */
- struct _mesa_HashTable *Objects;
+ struct _mesa_HashTable *Objects; /**< The table of all performance query objects */
};
diff --git a/src/mesa/main/performance_query.c b/src/mesa/main/performance_query.c
index f0ecabe3b58..aa103516a58 100644
--- a/src/mesa/main/performance_query.c
+++ b/src/mesa/main/performance_query.c
@@ -1,5 +1,5 @@
/*
- * Copyright © 2012 Intel Corporation
+ * Copyright © 2012-2017 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -24,14 +24,6 @@
/**
* \file performance_query.c
* Core Mesa support for the INTEL_performance_query extension.
- *
- * In order to implement this extension, start by defining two enums:
- * one for Groups, and one for Counters. These will be used as indexes into
- * arrays, so they should start at 0 and increment from there.
- *
- * Counter IDs need to be globally unique. That is, you can't have counter 7
- * in group A and counter 7 in group B. A global enum of all available
- * counters is a convenient way to guarantee this.
*/
#include <stdbool.h>
@@ -42,73 +34,21 @@
#include "macros.h"
#include "mtypes.h"
#include "performance_query.h"
-#include "util/bitset.h"
#include "util/ralloc.h"
void
_mesa_init_performance_queries(struct gl_context *ctx)
{
ctx->PerfQuery.Objects = _mesa_NewHashTable();
- ctx->PerfQuery.NumGroups = 0;
- ctx->PerfQuery.Groups = NULL;
-}
-
-static inline void
-init_groups(struct gl_context *ctx)
-{
- if (unlikely(!ctx->PerfMonitor.Groups))
- ctx->Driver.InitPerfMonitorGroups(ctx);
-}
-
-static struct gl_perf_monitor_object *
-new_performance_query(struct gl_context *ctx, GLuint index)
-{
- unsigned i;
- struct gl_perf_monitor_object *m = ctx->Driver.NewPerfMonitor(ctx);
-
- if (m == NULL)
- return NULL;
-
- m->Name = index;
-
- m->Active = false;
-
- m->ActiveGroups =
- rzalloc_array(NULL, unsigned, ctx->PerfQuery.NumGroups);
-
- m->ActiveCounters =
- ralloc_array(NULL, BITSET_WORD *, ctx->PerfQuery.NumGroups);
-
- if (m->ActiveGroups == NULL || m->ActiveCounters == NULL)
- goto fail;
-
- for (i = 0; i < ctx->PerfQuery.NumGroups; i++) {
- const struct gl_perf_monitor_group *g = &ctx->PerfQuery.Groups[i];
-
- m->ActiveCounters[i] = rzalloc_array(m->ActiveCounters, BITSET_WORD,
- BITSET_WORDS(g->NumCounters));
- if (m->ActiveCounters[i] == NULL)
- goto fail;
- }
-
- return m;
-
-fail:
- ralloc_free(m->ActiveGroups);
- ralloc_free(m->ActiveCounters);
- ctx->Driver.DeletePerfMonitor(ctx, m);
- return NULL;
}
static void
free_performance_query(GLuint key, void *data, void *user)
{
- struct gl_perf_monitor_object *m = data;
+ struct gl_perf_query_object *m = data;
struct gl_context *ctx = user;
- ralloc_free(m->ActiveGroups);
- ralloc_free(m->ActiveCounters);
- ctx->Driver.DeletePerfMonitor(ctx, m);
+ ctx->Driver.DeletePerfQuery(ctx, m);
}
void
@@ -119,50 +59,39 @@ _mesa_free_performance_queries(struct gl_context *ctx)
_mesa_DeleteHashTable(ctx->PerfQuery.Objects);
}
-static inline struct gl_perf_monitor_object *
-lookup_query(struct gl_context *ctx, GLuint id)
+static inline struct gl_perf_query_object *
+lookup_object(struct gl_context *ctx, GLuint id)
{
- return (struct gl_perf_monitor_object *)
- _mesa_HashLookup(ctx->PerfQuery.Objects, id);
+ return _mesa_HashLookup(ctx->PerfQuery.Objects, id);
}
-static inline const struct gl_perf_monitor_group *
-get_group(const struct gl_context *ctx, GLuint id)
+static GLuint
+init_performance_query_info(struct gl_context *ctx)
{
- if (id >= ctx->PerfQuery.NumGroups)
- return NULL;
-
- return &ctx->PerfQuery.Groups[id];
-}
-
-static inline const struct gl_perf_monitor_counter *
-get_counter(const struct gl_perf_monitor_group *group_obj, GLuint id)
-{
- if (id >= group_obj->NumCounters)
- return NULL;
-
- return &group_obj->Counters[id];
+ if (ctx->Driver.InitPerfQueryInfo)
+ return ctx->Driver.InitPerfQueryInfo(ctx);
+ else
+ return 0;
}
-/* For INTEL_performance_query, query id 0 is reserved to be invalid. We use
- * index to Groups array + 1 as the query id. Same applies to counter id.
- */
-static inline GLuint
+/* For INTEL_performance_query, query id 0 is reserved to be invalid. */
+static inline unsigned
queryid_to_index(GLuint queryid)
{
return queryid - 1;
}
static inline GLuint
-index_to_queryid(GLuint index)
+index_to_queryid(unsigned index)
{
return index + 1;
}
static inline bool
-queryid_valid(const struct gl_context *ctx, GLuint queryid)
+queryid_valid(const struct gl_context *ctx, unsigned numQueries, GLuint queryid)
{
- return get_group(ctx, queryid_to_index(queryid)) != NULL;
+ GLuint index = queryid_to_index(queryid);
+ return index >= 0 && index < numQueries;
}
static inline GLuint
@@ -171,35 +100,33 @@ counterid_to_index(GLuint counterid)
return counterid - 1;
}
-/*****************************************************************************/
-
-/**
- * Returns how many bytes a counter's value takes up.
- */
-unsigned
-_mesa_perf_query_counter_size(const struct gl_perf_monitor_counter *c)
+static void
+output_clipped_string(GLchar *stringRet,
+ GLuint stringMaxLen,
+ const char *string)
{
- switch (c->Type) {
- case GL_FLOAT:
- case GL_PERCENTAGE_AMD:
- return sizeof(GLfloat);
- case GL_UNSIGNED_INT:
- return sizeof(GLuint);
- case GL_UNSIGNED_INT64_AMD:
- return sizeof(uint64_t);
- default:
- assert(!"Should not get here: invalid counter type");
- return 0;
- }
+ if (!stringRet)
+ return;
+
+ strncpy(stringRet, string ? string : "", stringMaxLen);
+
+ /* No specification given about whether returned strings needs
+ * to be zero-terminated. Zero-terminate the string always as we
+ * don't otherwise communicate the length of the returned
+ * string.
+ */
+ if (stringMaxLen > 0)
+ stringRet[stringMaxLen - 1] = '\0';
}
+/*****************************************************************************/
+
extern void GLAPIENTRY
_mesa_GetFirstPerfQueryIdINTEL(GLuint *queryId)
{
GET_CURRENT_CONTEXT(ctx);
- unsigned numGroups;
- init_groups(ctx);
+ unsigned numQueries;
/* The GL_INTEL_performance_query spec says:
*
@@ -211,7 +138,7 @@ _mesa_GetFirstPerfQueryIdINTEL(GLuint *queryId)
return;
}
- numGroups = ctx->PerfQuery.NumGroups;
+ numQueries = init_performance_query_info(ctx);
/* The GL_INTEL_performance_query spec says:
*
@@ -219,7 +146,7 @@ _mesa_GetFirstPerfQueryIdINTEL(GLuint *queryId)
* queries, then the value of 0 is returned and INVALID_OPERATION error
* is raised."
*/
- if (numGroups == 0) {
+ if (numQueries == 0) {
*queryId = 0;
_mesa_error(ctx, GL_INVALID_OPERATION,
"glGetFirstPerfQueryIdINTEL(no queries supported)");
@@ -233,7 +160,8 @@ extern void GLAPIENTRY
_mesa_GetNextPerfQueryIdINTEL(GLuint queryId, GLuint *nextQueryId)
{
GET_CURRENT_CONTEXT(ctx);
- init_groups(ctx);
+
+ unsigned numQueries;
/* The GL_INTEL_performance_query spec says:
*
@@ -251,29 +179,27 @@ _mesa_GetNextPerfQueryIdINTEL(GLuint queryId, GLuint *nextQueryId)
return;
}
- if (!queryid_valid(ctx, queryId)) {
- *nextQueryId = 0;
+ numQueries = init_performance_query_info(ctx);
+
+ if (!queryid_valid(ctx, numQueries, queryId)) {
_mesa_error(ctx, GL_INVALID_VALUE,
"glGetNextPerfQueryIdINTEL(invalid query)");
return;
}
- ++queryId;
-
- if (!queryid_valid(ctx, queryId)) {
- *nextQueryId = 0;
- } else {
+ if (queryid_valid(ctx, numQueries, ++queryId))
*nextQueryId = queryId;
- }
+ else
+ *nextQueryId = 0;
}
extern void GLAPIENTRY
_mesa_GetPerfQueryIdByNameINTEL(char *queryName, GLuint *queryId)
{
GET_CURRENT_CONTEXT(ctx);
- unsigned i;
- init_groups(ctx);
+ unsigned numQueries;
+ unsigned i;
/* The GL_INTEL_performance_query spec says:
*
@@ -286,15 +212,25 @@ _mesa_GetPerfQueryIdByNameINTEL(char *queryName, GLuint *queryId)
return;
}
- /* The specification does not state that this produces an error. */
+ /* The specification does not state that this produces an error but
+ * to be consistent with glGetFirstPerfQueryIdINTEL we generate an
+ * INVALID_VALUE error
+ */
if (!queryId) {
- _mesa_warning(ctx, "glGetPerfQueryIdByNameINTEL(queryId == NULL)");
+ _mesa_error(ctx, GL_INVALID_VALUE,
+ "glGetPerfQueryIdByNameINTEL(queryId == NULL)");
return;
}
- for (i = 0; i < ctx->PerfQuery.NumGroups; ++i) {
- const struct gl_perf_monitor_group *group_obj = get_group(ctx, i);
- if (strcmp(group_obj->Name, queryName) == 0) {
+ numQueries = init_performance_query_info(ctx);
+
+ for (i = 0; i < numQueries; ++i) {
+ const GLchar *name;
+ GLuint ignore;
+
+ ctx->Driver.GetPerfQueryInfo(ctx, i, &name, &ignore, &ignore, &ignore);
+
+ if (strcmp(name, queryName) == 0) {
*queryId = index_to_queryid(i);
return;
}
@@ -306,20 +242,22 @@ _mesa_GetPerfQueryIdByNameINTEL(char *queryName, GLuint *queryId)
extern void GLAPIENTRY
_mesa_GetPerfQueryInfoINTEL(GLuint queryId,
- GLuint queryNameLength, char *queryName,
- GLuint *dataSize, GLuint *noCounters,
- GLuint *noActiveInstances,
+ GLuint nameLength, GLchar *name,
+ GLuint *dataSize,
+ GLuint *numCounters,
+ GLuint *numActive,
GLuint *capsMask)
{
GET_CURRENT_CONTEXT(ctx);
- unsigned i;
-
- const struct gl_perf_monitor_group *group_obj;
- init_groups(ctx);
+ unsigned numQueries = init_performance_query_info(ctx);
+ unsigned queryIndex = queryid_to_index(queryId);
+ const char *queryName;
+ GLuint queryDataSize;
+ GLuint queryNumCounters;
+ GLuint queryNumActive;
- group_obj = get_group(ctx, queryid_to_index(queryId));
- if (group_obj == NULL) {
+ if (!queryid_valid(ctx, numQueries, queryId)) {
/* The GL_INTEL_performance_query spec says:
*
* "If queryId does not reference a valid query type, an
@@ -330,32 +268,19 @@ _mesa_GetPerfQueryInfoINTEL(GLuint queryId,
return;
}
- if (queryName) {
- strncpy(queryName, group_obj->Name, queryNameLength);
+ ctx->Driver.GetPerfQueryInfo(ctx, queryIndex,
+ &queryName,
+ &queryDataSize,
+ &queryNumCounters,
+ &queryNumActive);
- /* No specification given about whether the string needs to be
- * zero-terminated. Zero-terminate the string always as we don't
- * otherwise communicate the length of the returned string.
- */
- if (queryNameLength > 0) {
- queryName[queryNameLength - 1] = '\0';
- }
- }
+ output_clipped_string(name, nameLength, queryName);
- if (dataSize) {
- unsigned size = 0;
- for (i = 0; i < group_obj->NumCounters; ++i) {
- /* What we get from the driver is group id (uint32_t) + counter id
- * (uint32_t) + value.
- */
- size += 2 * sizeof(uint32_t) + _mesa_perf_query_counter_size(&group_obj->Counters[i]);
- }
- *dataSize = size;
- }
+ if (dataSize)
+ *dataSize = queryDataSize;
- if (noCounters) {
- *noCounters = group_obj->NumCounters;
- }
+ if (numCounters)
+ *numCounters = queryNumCounters;
/* The GL_INTEL_performance_query spec says:
*
@@ -366,130 +291,94 @@ _mesa_GetPerfQueryInfoINTEL(GLuint queryId,
* 2) Another typo in the specification, maxInstances parameter is not listed
* in the declaration of this function in the list of new functions.
*/
- if (noActiveInstances) {
- *noActiveInstances = _mesa_HashNumEntries(ctx->PerfQuery.Objects);
- }
+ if (numActive)
+ *numActive = queryNumActive;
- if (capsMask) {
- /* TODO: This information not yet available in the monitor structs. For
- * now, we hardcode SINGLE_CONTEXT, since that's what the implementation
- * currently tries very hard to do.
- */
+ /* Assume for now that all queries are per-context */
+ if (capsMask)
*capsMask = GL_PERFQUERY_SINGLE_CONTEXT_INTEL;
- }
}
extern void GLAPIENTRY
_mesa_GetPerfCounterInfoINTEL(GLuint queryId, GLuint counterId,
- GLuint counterNameLength, char *counterName,
- GLuint counterDescLength, char *counterDesc,
- GLuint *counterOffset, GLuint *counterDataSize, GLuint *counterTypeEnum,
- GLuint *counterDataTypeEnum, GLuint64 *rawCounterMaxValue)
+ GLuint nameLength, GLchar *name,
+ GLuint descLength, GLchar *desc,
+ GLuint *offset,
+ GLuint *dataSize,
+ GLuint *typeEnum,
+ GLuint *dataTypeEnum,
+ GLuint64 *rawCounterMaxValue)
{
GET_CURRENT_CONTEXT(ctx);
- const struct gl_perf_monitor_group *group_obj;
- const struct gl_perf_monitor_counter *counter_obj;
+ unsigned numQueries = init_performance_query_info(ctx);
+ unsigned queryIndex = queryid_to_index(queryId);
+ const char *queryName;
+ GLuint queryDataSize;
+ GLuint queryNumCounters;
+ GLuint queryNumActive;
unsigned counterIndex;
- unsigned i;
-
- init_groups(ctx);
-
- group_obj = get_group(ctx, queryid_to_index(queryId));
-
- /* The GL_INTEL_performance_query spec says:
- *
- * "If the pair of queryId and counterId does not reference a valid
- * counter, an INVALID_VALUE error is generated."
- */
- if (group_obj == NULL) {
+ const char *counterName;
+ const char *counterDesc;
+ GLuint counterOffset;
+ GLuint counterDataSize;
+ GLuint counterTypeEnum;
+ GLuint counterDataTypeEnum;
+ GLuint64 counterRawMax;
+
+ if (!queryid_valid(ctx, numQueries, queryId)) {
+ /* The GL_INTEL_performance_query spec says:
+ *
+ * "If the pair of queryId and counterId does not reference a valid
+ * counter, an INVALID_VALUE error is generated."
+ */
_mesa_error(ctx, GL_INVALID_VALUE,
"glGetPerfCounterInfoINTEL(invalid queryId)");
return;
}
+ ctx->Driver.GetPerfQueryInfo(ctx, queryIndex,
+ &queryName,
+ &queryDataSize,
+ &queryNumCounters,
+ &queryNumActive);
+
counterIndex = counterid_to_index(counterId);
- counter_obj = get_counter(group_obj, counterIndex);
- if (counter_obj == NULL) {
+ if (counterIndex < 0 || counterIndex >= queryNumCounters) {
_mesa_error(ctx, GL_INVALID_VALUE,
"glGetPerfCounterInfoINTEL(invalid counterId)");
return;
}
- if (counterName) {
- strncpy(counterName, counter_obj->Name, counterNameLength);
-
- /* No specification given about whether the string needs to be
- * zero-terminated. Zero-terminate the string always as we don't
- * otherwise communicate the length of the returned string.
- */
- if (counterNameLength > 0) {
- counterName[counterNameLength - 1] = '\0';
- }
- }
+ ctx->Driver.GetPerfCounterInfo(ctx, queryIndex, counterIndex,
+ &counterName,
+ &counterDesc,
+ &counterOffset,
+ &counterDataSize,
+ &counterTypeEnum,
+ &counterDataTypeEnum,
+ &counterRawMax);
- if (counterDesc) {
- /* TODO: No separate description text at the moment. We pass the name
- * again for the moment.
- */
- strncpy(counterDesc, counter_obj->Name, counterDescLength);
+ output_clipped_string(name, nameLength, counterName);
+ output_clipped_string(desc, descLength, counterDesc);
- /* No specification given about whether the string needs to be
- * zero-terminated. Zero-terminate the string always as we don't
- * otherwise communicate the length of the returned string.
- */
- if (counterDescLength > 0) {
- counterDesc[counterDescLength - 1] = '\0';
- }
- }
+ if (offset)
+ *offset = counterOffset;
- if (counterOffset) {
- unsigned offset = 0;
- for (i = 0; i < counterIndex; ++i) {
- /* What we get from the driver is group id (uint32_t) + counter id
- * (uint32_t) + value.
- */
- offset += 2 * sizeof(uint32_t) + _mesa_perf_query_counter_size(&group_obj->Counters[i]);
- }
- *counterOffset = 2 * sizeof(uint32_t) + offset;
- }
+ if (dataSize)
+ *dataSize = counterDataSize;
- if (counterDataSize) {
- *counterDataSize = _mesa_perf_query_counter_size(counter_obj);
- }
+ if (typeEnum)
+ *typeEnum = counterTypeEnum;
- if (counterTypeEnum) {
- /* TODO: Different counter types (semantic type, not data type) not
- * supported as of yet.
- */
- *counterTypeEnum = GL_PERFQUERY_COUNTER_RAW_INTEL;
- }
+ if (dataTypeEnum)
+ *dataTypeEnum = counterDataTypeEnum;
- if (counterDataTypeEnum) {
- switch (counter_obj->Type) {
- case GL_FLOAT:
- case GL_PERCENTAGE_AMD:
- *counterDataTypeEnum = GL_PERFQUERY_COUNTER_DATA_FLOAT_INTEL;
- break;
- case GL_UNSIGNED_INT:
- *counterDataTypeEnum = GL_PERFQUERY_COUNTER_DATA_UINT32_INTEL;
- break;
- case GL_UNSIGNED_INT64_AMD:
- *counterDataTypeEnum = GL_PERFQUERY_COUNTER_DATA_UINT64_INTEL;
- break;
- default:
- assert(!"Should not get here: invalid counter type");
- return;
- }
- }
+ if (rawCounterMaxValue)
+ *rawCounterMaxValue = counterRawMax;
if (rawCounterMaxValue) {
- /* This value is (implicitly) specified to be used only with
- * GL_PERFQUERY_COUNTER_RAW_INTEL counters. When semantic types for
- * counters are added, that needs to be checked.
- */
-
/* The GL_INTEL_performance_query spec says:
*
* "for some raw counters for which the maximal value is
@@ -497,10 +386,17 @@ _mesa_GetPerfCounterInfoINTEL(GLuint queryId, GLuint counterId,
* returned in the location pointed by rawCounterMaxValue, otherwise,
* the location is written with the value of 0."
*
- * The maximum value reported by the driver at the moment is not with
- * these semantics, so write 0 always to be safe.
+ * Since it's very useful to be able to report a maximum value for
+ * more that just counters using the _COUNTER_RAW_INTEL or
+ * _COUNTER_DURATION_RAW_INTEL enums (e.g. for a _THROUGHPUT tools
+ * want to be able to visualize the absolute throughput with respect
+ * to the theoretical maximum that's possible) and there doesn't seem
+ * to be any reason not to allow _THROUGHPUT counters to also be
+ * considerer "raw" here, we always leave it up to the backend to
+ * decide when it's appropriate to report a maximum counter value or 0
+ * if not.
*/
- *rawCounterMaxValue = 0;
+ *rawCounterMaxValue = counterRawMax;
}
}
@@ -508,129 +404,106 @@ extern void GLAPIENTRY
_mesa_CreatePerfQueryINTEL(GLuint queryId, GLuint *queryHandle)
{
GET_CURRENT_CONTEXT(ctx);
- GLuint first;
- GLuint group;
- const struct gl_perf_monitor_group *group_obj;
- struct gl_perf_monitor_object *m;
- unsigned i;
-
- init_groups(ctx);
-
- /* This is not specified in the extension, but is the only sane thing to
- * do.
- */
- if (queryHandle == NULL) {
- _mesa_error(ctx, GL_INVALID_VALUE,
- "glCreatePerfQueryINTEL(queryHandle == NULL)");
- return;
- }
- group = queryid_to_index(queryId);
- group_obj = get_group(ctx, group);
+ unsigned numQueries = init_performance_query_info(ctx);
+ GLuint id;
+ struct gl_perf_query_object *obj;
/* The GL_INTEL_performance_query spec says:
*
* "If queryId does not reference a valid query type, an INVALID_VALUE
* error is generated."
*/
- if (group_obj == NULL) {
+ if (!queryid_valid(ctx, numQueries, queryId)) {
_mesa_error(ctx, GL_INVALID_VALUE,
"glCreatePerfQueryINTEL(invalid queryId)");
return;
}
- /* The query object created here is the counterpart of a `monitor' in
- * AMD_performance_monitor. This call is equivalent to calling
- * GenPerfMonitorsAMD and SelectPerfMonitorCountersAMD with a list of all
- * counters in a group.
+ /* This is not specified in the extension, but is the only sane thing to
+ * do.
*/
+ if (queryHandle == NULL) {
+ _mesa_error(ctx, GL_INVALID_VALUE,
+ "glCreatePerfQueryINTEL(queryHandle == NULL)");
+ return;
+ }
- /* We keep the query ids contiguous */
- first = _mesa_HashFindFreeKeyBlock(ctx->PerfQuery.Objects, 1);
- if (!first) {
+ id = _mesa_HashFindFreeKeyBlock(ctx->PerfQuery.Objects, 1);
+ if (!id) {
/* The GL_INTEL_performance_query spec says:
*
* "If the query instance cannot be created due to exceeding the
* number of allowed instances or driver fails query creation due to
* an insufficient memory reason, an OUT_OF_MEMORY error is
* generated, and the location pointed by queryHandle returns NULL."
- */
- _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCreatePerfQueryINTEL");
+ */
+ _mesa_error_no_memory(__func__);
return;
}
- m = new_performance_query(ctx, first);
- if (m == NULL) {
+ obj = ctx->Driver.NewPerfQueryObject(ctx, queryid_to_index(queryId));
+ if (obj == NULL) {
_mesa_error_no_memory(__func__);
return;
}
- _mesa_HashInsert(ctx->PerfQuery.Objects, first, m);
- *queryHandle = first;
-
- ctx->Driver.ResetPerfMonitor(ctx, m);
+ obj->Id = id;
+ obj->Active = false;
+ obj->Ready = false;
- for (i = 0; i < group_obj->NumCounters; ++i) {
- ++m->ActiveGroups[group];
- /* Counters are a continuous range of integers, 0 to NumCounters (excl),
- * so i is the counter value to use here.
- */
- BITSET_SET(m->ActiveCounters[group], i);
- }
+ _mesa_HashInsert(ctx->PerfQuery.Objects, id, obj);
+ *queryHandle = id;
}
extern void GLAPIENTRY
_mesa_DeletePerfQueryINTEL(GLuint queryHandle)
{
GET_CURRENT_CONTEXT(ctx);
- struct gl_perf_monitor_object *m;
- /* The queryHandle is the counterpart to AMD_performance_monitor's monitor
- * id.
- */
- m = lookup_query(ctx, queryHandle);
+ struct gl_perf_query_object *obj = lookup_object(ctx, queryHandle);
/* The GL_INTEL_performance_query spec says:
*
* "If a query handle doesn't reference a previously created performance
* query instance, an INVALID_VALUE error is generated."
*/
- if (m == NULL) {
+ if (obj == NULL) {
_mesa_error(ctx, GL_INVALID_VALUE,
"glDeletePerfQueryINTEL(invalid queryHandle)");
return;
}
- /* Let the driver stop the query if it's active. */
- if (m->Active) {
- ctx->Driver.ResetPerfMonitor(ctx, m);
- m->Ended = false;
+ /* To avoid complications in the backend we never ask the backend to
+ * delete an active query or a query object while we are still
+ * waiting for data.
+ */
+
+ if (obj->Active)
+ _mesa_EndPerfQueryINTEL(queryHandle);
+
+ if (obj->Used && !obj->Ready) {
+ ctx->Driver.WaitPerfQuery(ctx, obj);
+ obj->Ready = true;
}
_mesa_HashRemove(ctx->PerfQuery.Objects, queryHandle);
- ralloc_free(m->ActiveGroups);
- ralloc_free(m->ActiveCounters);
- ctx->Driver.DeletePerfMonitor(ctx, m);
+ ctx->Driver.DeletePerfQuery(ctx, obj);
}
extern void GLAPIENTRY
_mesa_BeginPerfQueryINTEL(GLuint queryHandle)
{
GET_CURRENT_CONTEXT(ctx);
- struct gl_perf_monitor_object *m;
-
- /* The queryHandle is the counterpart to AMD_performance_monitor's monitor
- * id.
- */
- m = lookup_query(ctx, queryHandle);
+ struct gl_perf_query_object *obj = lookup_object(ctx, queryHandle);
/* The GL_INTEL_performance_query spec says:
*
* "If a query handle doesn't reference a previously created performance
* query instance, an INVALID_VALUE error is generated."
*/
- if (m == NULL) {
+ if (obj == NULL) {
_mesa_error(ctx, GL_INVALID_VALUE,
"glBeginPerfQueryINTEL(invalid queryHandle)");
return;
@@ -646,15 +519,25 @@ _mesa_BeginPerfQueryINTEL(GLuint queryHandle)
* We also generate an INVALID_OPERATION error if the driver can't begin
* a query for its own reasons, and for nesting the same query.
*/
- if (m->Active) {
+ if (obj->Active) {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glBeginPerfQueryINTEL(already active)");
return;
}
- if (ctx->Driver.BeginPerfMonitor(ctx, m)) {
- m->Active = true;
- m->Ended = false;
+ /* To avoid complications in the backend we never ask the backend to
+ * reuse a query object and begin a new query while we are still
+ * waiting for data on that object.
+ */
+ if (obj->Used && !obj->Ready) {
+ ctx->Driver.WaitPerfQuery(ctx, obj);
+ obj->Ready = true;
+ }
+
+ if (ctx->Driver.BeginPerfQuery(ctx, obj)) {
+ obj->Used = true;
+ obj->Active = true;
+ obj->Ready = false;
} else {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glBeginPerfQueryINTEL(driver unable to begin query)");
@@ -665,39 +548,32 @@ extern void GLAPIENTRY
_mesa_EndPerfQueryINTEL(GLuint queryHandle)
{
GET_CURRENT_CONTEXT(ctx);
- struct gl_perf_monitor_object *m;
- /* The queryHandle is the counterpart to AMD_performance_monitor's monitor
- * id.
- */
+ struct gl_perf_query_object *obj = lookup_object(ctx, queryHandle);
- m = lookup_query(ctx, queryHandle);
+ /* Not explicitly covered in the spec, but for consistency... */
+ if (obj == NULL) {
+ _mesa_error(ctx, GL_INVALID_VALUE,
+ "glEndPerfQueryINTEL(invalid queryHandle)");
+ return;
+ }
/* The GL_INTEL_performance_query spec says:
*
* "If a performance query is not currently started, an
* INVALID_OPERATION error will be generated."
- *
- * The specification doesn't state that an invalid handle would be an
- * INVALID_VALUE error. Regardless, query for such a handle will not be
- * started, so we generate an INVALID_OPERATION in that case too.
*/
- if (m == NULL) {
- _mesa_error(ctx, GL_INVALID_OPERATION,
- "glEndPerfQueryINTEL(invalid queryHandle)");
- return;
- }
- if (!m->Active) {
+ if (!obj->Active) {
_mesa_error(ctx, GL_INVALID_OPERATION,
"glEndPerfQueryINTEL(not active)");
return;
}
- ctx->Driver.EndPerfMonitor(ctx, m);
+ ctx->Driver.EndPerfQuery(ctx, obj);
- m->Active = false;
- m->Ended = true;
+ obj->Active = false;
+ obj->Ready = false;
}
extern void GLAPIENTRY
@@ -705,8 +581,15 @@ _mesa_GetPerfQueryDataINTEL(GLuint queryHandle, GLuint flags,
GLsizei dataSize, void *data, GLuint *bytesWritten)
{
GET_CURRENT_CONTEXT(ctx);
- struct gl_perf_monitor_object *m;
- bool result_available;
+
+ struct gl_perf_query_object *obj = lookup_object(ctx, queryHandle);
+
+ /* Not explicitly covered in the spec, but for consistency... */
+ if (obj == NULL) {
+ _mesa_error(ctx, GL_INVALID_VALUE,
+ "glEndPerfQueryINTEL(invalid queryHandle)");
+ return;
+ }
/* The GL_INTEL_performance_query spec says:
*
@@ -719,66 +602,33 @@ _mesa_GetPerfQueryDataINTEL(GLuint queryHandle, GLuint flags,
return;
}
- /* The queryHandle is the counterpart to AMD_performance_monitor's monitor
- * id.
- */
-
- m = lookup_query(ctx, queryHandle);
-
- /* The specification doesn't state that an invalid handle generates an
- * error. We could interpret that to mean the case should be handled as
- * "measurement not ready for this query", but what should be done if
- * `flags' equals PERFQUERY_WAIT_INTEL?
- *
- * To resolve this, we just generate an INVALID_VALUE from an invalid query
- * handle.
+ /* Just for good measure in case a lazy application is only
+ * checking this and not checking for errors...
*/
- if (m == NULL) {
- _mesa_error(ctx, GL_INVALID_VALUE,
- "glGetPerfQueryDataINTEL(invalid queryHandle)");
- return;
- }
+ *bytesWritten = 0;
- /* We need at least enough room for a single value. */
- if (dataSize < sizeof(GLuint)) {
- *bytesWritten = 0;
- return;
- }
-
- /* The GL_INTEL_performance_query spec says:
- *
- * "The call may end without returning any data if they are not ready
- * for reading as the measurement session is still pending (the
- * EndPerfQueryINTEL() command processing is not finished by
- * hardware). In this case location pointed by the bytesWritten
- * parameter will be set to 0."
- *
- * If EndPerfQueryINTEL() is not called at all, we follow this.
+ /* Not explicitly covered in the spec but to be consistent with
+ * EndPerfQuery which validates that an application only ends an
+ * active query we also validate that an application doesn't try
+ * and get the data for a still active query...
*/
- if (!m->Ended) {
- *bytesWritten = 0;
+ if (obj->Active) {
+ _mesa_error(ctx, GL_INVALID_OPERATION,
+ "glGetPerfQueryDataINTEL(query still active)");
return;
}
- result_available = ctx->Driver.IsPerfMonitorResultAvailable(ctx, m);
+ obj->Ready = ctx->Driver.IsPerfQueryReady(ctx, obj);
- if (!result_available) {
+ if (!obj->Ready) {
if (flags == GL_PERFQUERY_FLUSH_INTEL) {
ctx->Driver.Flush(ctx);
} else if (flags == GL_PERFQUERY_WAIT_INTEL) {
- /* Assume Finish() is both enough and not too much to wait for
- * results. If results are still not available after Finish(), the
- * later code automatically bails out with 0 for bytesWritten.
- */
- ctx->Driver.Finish(ctx);
- result_available =
- ctx->Driver.IsPerfMonitorResultAvailable(ctx, m);
+ ctx->Driver.WaitPerfQuery(ctx, obj);
+ obj->Ready = true;
}
}
- if (result_available) {
- ctx->Driver.GetPerfMonitorResult(ctx, m, dataSize, data, (GLint*)bytesWritten);
- } else {
- *bytesWritten = 0;
- }
+ if (obj->Ready)
+ ctx->Driver.GetPerfQueryData(ctx, obj, dataSize, data, bytesWritten);
}
diff --git a/src/mesa/main/performance_query.h b/src/mesa/main/performance_query.h
index 3fed5eae1b8..8268f0ef194 100644
--- a/src/mesa/main/performance_query.h
+++ b/src/mesa/main/performance_query.h
@@ -1,5 +1,5 @@
/*
- * Copyright © 2012 Intel Corporation
+ * Copyright © 2012,2015 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
@@ -38,10 +38,6 @@ _mesa_init_performance_queries(struct gl_context *ctx);
extern void
_mesa_free_performance_queries(struct gl_context *ctx);
-unsigned
-_mesa_perf_query_counter_size(const struct gl_perf_monitor_counter *);
-
-
extern void GLAPIENTRY
_mesa_GetFirstPerfQueryIdINTEL(GLuint *queryId);
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-30 11:31:36 +0000