2 * Copyright © 2012 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
25 * \file performance_monitor.c
26 * Core Mesa support for the AMD_performance_monitor extension.
28 * In order to implement this extension, start by defining two enums:
29 * one for Groups, and one for Counters. These will be used as indexes into
30 * arrays, so they should start at 0 and increment from there.
32 * Counter IDs need to be globally unique. That is, you can't have counter 7
33 * in group A and counter 7 in group B. A global enum of all available
34 * counters is a convenient way to guarantee this.
44 #include "performance_monitor.h"
45 #include "util/bitset.h"
46 #include "util/ralloc.h"
49 _mesa_init_performance_monitors(struct gl_context
*ctx
)
51 ctx
->PerfMonitor
.Monitors
= _mesa_NewHashTable();
52 ctx
->PerfMonitor
.NumGroups
= 0;
53 ctx
->PerfMonitor
.Groups
= NULL
;
57 init_groups(struct gl_context
*ctx
)
59 if (unlikely(!ctx
->PerfMonitor
.Groups
))
60 ctx
->Driver
.InitPerfMonitorGroups(ctx
);
63 static struct gl_perf_monitor_object
*
64 new_performance_monitor(struct gl_context
*ctx
, GLuint index
)
67 struct gl_perf_monitor_object
*m
= ctx
->Driver
.NewPerfMonitor(ctx
);
77 rzalloc_array(NULL
, unsigned, ctx
->PerfMonitor
.NumGroups
);
80 ralloc_array(NULL
, BITSET_WORD
*, ctx
->PerfMonitor
.NumGroups
);
82 if (m
->ActiveGroups
== NULL
|| m
->ActiveCounters
== NULL
)
85 for (i
= 0; i
< ctx
->PerfMonitor
.NumGroups
; i
++) {
86 const struct gl_perf_monitor_group
*g
= &ctx
->PerfMonitor
.Groups
[i
];
88 m
->ActiveCounters
[i
] = rzalloc_array(m
->ActiveCounters
, BITSET_WORD
,
89 BITSET_WORDS(g
->NumCounters
));
90 if (m
->ActiveCounters
[i
] == NULL
)
97 ralloc_free(m
->ActiveGroups
);
98 ralloc_free(m
->ActiveCounters
);
99 ctx
->Driver
.DeletePerfMonitor(ctx
, m
);
104 free_performance_monitor(GLuint key
, void *data
, void *user
)
106 struct gl_perf_monitor_object
*m
= data
;
107 struct gl_context
*ctx
= user
;
109 ralloc_free(m
->ActiveGroups
);
110 ralloc_free(m
->ActiveCounters
);
111 ctx
->Driver
.DeletePerfMonitor(ctx
, m
);
115 _mesa_free_performance_monitors(struct gl_context
*ctx
)
117 _mesa_HashDeleteAll(ctx
->PerfMonitor
.Monitors
,
118 free_performance_monitor
, ctx
);
119 _mesa_DeleteHashTable(ctx
->PerfMonitor
.Monitors
);
122 static inline struct gl_perf_monitor_object
*
123 lookup_monitor(struct gl_context
*ctx
, GLuint id
)
125 return (struct gl_perf_monitor_object
*)
126 _mesa_HashLookup(ctx
->PerfMonitor
.Monitors
, id
);
129 static inline const struct gl_perf_monitor_group
*
130 get_group(const struct gl_context
*ctx
, GLuint id
)
132 if (id
>= ctx
->PerfMonitor
.NumGroups
)
135 return &ctx
->PerfMonitor
.Groups
[id
];
138 static inline const struct gl_perf_monitor_counter
*
139 get_counter(const struct gl_perf_monitor_group
*group_obj
, GLuint id
)
141 if (id
>= group_obj
->NumCounters
)
144 return &group_obj
->Counters
[id
];
147 /* For INTEL_performance_query, query id 0 is reserved to be invalid. We use
148 * index to Groups array + 1 as the query id. Same applies to counter id.
151 queryid_to_index(GLuint queryid
)
157 index_to_queryid(GLuint index
)
163 queryid_valid(const struct gl_context
*ctx
, GLuint queryid
)
165 return get_group(ctx
, queryid_to_index(queryid
)) != NULL
;
169 counterid_to_index(GLuint counterid
)
171 return counterid
- 1;
174 /*****************************************************************************/
177 _mesa_GetPerfMonitorGroupsAMD(GLint
*numGroups
, GLsizei groupsSize
,
180 GET_CURRENT_CONTEXT(ctx
);
183 if (numGroups
!= NULL
)
184 *numGroups
= ctx
->PerfMonitor
.NumGroups
;
186 if (groupsSize
> 0 && groups
!= NULL
) {
188 unsigned n
= MIN2((GLuint
) groupsSize
, ctx
->PerfMonitor
.NumGroups
);
190 /* We just use the index in the Groups array as the ID. */
191 for (i
= 0; i
< n
; i
++)
197 _mesa_GetPerfMonitorCountersAMD(GLuint group
, GLint
*numCounters
,
198 GLint
*maxActiveCounters
,
199 GLsizei countersSize
, GLuint
*counters
)
201 GET_CURRENT_CONTEXT(ctx
);
202 const struct gl_perf_monitor_group
*group_obj
;
206 group_obj
= get_group(ctx
, group
);
207 if (group_obj
== NULL
) {
208 _mesa_error(ctx
, GL_INVALID_VALUE
,
209 "glGetPerfMonitorCountersAMD(invalid group)");
213 if (maxActiveCounters
!= NULL
)
214 *maxActiveCounters
= group_obj
->MaxActiveCounters
;
216 if (numCounters
!= NULL
)
217 *numCounters
= group_obj
->NumCounters
;
219 if (counters
!= NULL
) {
221 unsigned n
= MIN2(group_obj
->NumCounters
, (GLuint
) countersSize
);
222 for (i
= 0; i
< n
; i
++) {
223 /* We just use the index in the Counters array as the ID. */
230 _mesa_GetPerfMonitorGroupStringAMD(GLuint group
, GLsizei bufSize
,
231 GLsizei
*length
, GLchar
*groupString
)
233 GET_CURRENT_CONTEXT(ctx
);
234 const struct gl_perf_monitor_group
*group_obj
;
238 group_obj
= get_group(ctx
, group
);
239 if (group_obj
== NULL
) {
240 _mesa_error(ctx
, GL_INVALID_VALUE
, "glGetPerfMonitorGroupStringAMD");
245 /* Return the number of characters that would be required to hold the
246 * group string, excluding the null terminator.
249 *length
= strlen(group_obj
->Name
);
252 *length
= MIN2(strlen(group_obj
->Name
), bufSize
);
253 if (groupString
!= NULL
)
254 strncpy(groupString
, group_obj
->Name
, bufSize
);
259 _mesa_GetPerfMonitorCounterStringAMD(GLuint group
, GLuint counter
,
260 GLsizei bufSize
, GLsizei
*length
,
261 GLchar
*counterString
)
263 GET_CURRENT_CONTEXT(ctx
);
265 const struct gl_perf_monitor_group
*group_obj
;
266 const struct gl_perf_monitor_counter
*counter_obj
;
270 group_obj
= get_group(ctx
, group
);
272 if (group_obj
== NULL
) {
273 _mesa_error(ctx
, GL_INVALID_VALUE
,
274 "glGetPerfMonitorCounterStringAMD(invalid group)");
278 counter_obj
= get_counter(group_obj
, counter
);
280 if (counter_obj
== NULL
) {
281 _mesa_error(ctx
, GL_INVALID_VALUE
,
282 "glGetPerfMonitorCounterStringAMD(invalid counter)");
287 /* Return the number of characters that would be required to hold the
288 * counter string, excluding the null terminator.
291 *length
= strlen(counter_obj
->Name
);
294 *length
= MIN2(strlen(counter_obj
->Name
), bufSize
);
295 if (counterString
!= NULL
)
296 strncpy(counterString
, counter_obj
->Name
, bufSize
);
301 _mesa_GetPerfMonitorCounterInfoAMD(GLuint group
, GLuint counter
, GLenum pname
,
304 GET_CURRENT_CONTEXT(ctx
);
306 const struct gl_perf_monitor_group
*group_obj
;
307 const struct gl_perf_monitor_counter
*counter_obj
;
311 group_obj
= get_group(ctx
, group
);
313 if (group_obj
== NULL
) {
314 _mesa_error(ctx
, GL_INVALID_VALUE
,
315 "glGetPerfMonitorCounterInfoAMD(invalid group)");
319 counter_obj
= get_counter(group_obj
, counter
);
321 if (counter_obj
== NULL
) {
322 _mesa_error(ctx
, GL_INVALID_VALUE
,
323 "glGetPerfMonitorCounterInfoAMD(invalid counter)");
328 case GL_COUNTER_TYPE_AMD
:
329 *((GLenum
*) data
) = counter_obj
->Type
;
332 case GL_COUNTER_RANGE_AMD
:
333 switch (counter_obj
->Type
) {
335 case GL_PERCENTAGE_AMD
: {
336 float *f_data
= data
;
337 f_data
[0] = counter_obj
->Minimum
.f
;
338 f_data
[1] = counter_obj
->Maximum
.f
;
341 case GL_UNSIGNED_INT
: {
342 uint32_t *u32_data
= data
;
343 u32_data
[0] = counter_obj
->Minimum
.u32
;
344 u32_data
[1] = counter_obj
->Maximum
.u32
;
347 case GL_UNSIGNED_INT64_AMD
: {
348 uint64_t *u64_data
= data
;
349 u64_data
[0] = counter_obj
->Minimum
.u64
;
350 u64_data
[1] = counter_obj
->Maximum
.u64
;
354 assert(!"Should not get here: invalid counter type");
359 _mesa_error(ctx
, GL_INVALID_ENUM
,
360 "glGetPerfMonitorCounterInfoAMD(pname)");
366 _mesa_GenPerfMonitorsAMD(GLsizei n
, GLuint
*monitors
)
369 GET_CURRENT_CONTEXT(ctx
);
371 if (MESA_VERBOSE
& VERBOSE_API
)
372 _mesa_debug(ctx
, "glGenPerfMonitorsAMD(%d)\n", n
);
377 _mesa_error(ctx
, GL_INVALID_VALUE
, "glGenPerfMonitorsAMD(n < 0)");
381 if (monitors
== NULL
)
384 /* We don't actually need them to be contiguous, but this is what
385 * the rest of Mesa does, so we may as well.
387 first
= _mesa_HashFindFreeKeyBlock(ctx
->PerfMonitor
.Monitors
, n
);
390 for (i
= 0; i
< n
; i
++) {
391 struct gl_perf_monitor_object
*m
=
392 new_performance_monitor(ctx
, first
+ i
);
394 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "glGenPerfMonitorsAMD");
397 monitors
[i
] = first
+ i
;
398 _mesa_HashInsert(ctx
->PerfMonitor
.Monitors
, first
+ i
, m
);
401 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "glGenPerfMonitorsAMD");
407 _mesa_DeletePerfMonitorsAMD(GLsizei n
, GLuint
*monitors
)
410 GET_CURRENT_CONTEXT(ctx
);
412 if (MESA_VERBOSE
& VERBOSE_API
)
413 _mesa_debug(ctx
, "glDeletePerfMonitorsAMD(%d)\n", n
);
416 _mesa_error(ctx
, GL_INVALID_VALUE
, "glDeletePerfMonitorsAMD(n < 0)");
420 if (monitors
== NULL
)
423 for (i
= 0; i
< n
; i
++) {
424 struct gl_perf_monitor_object
*m
= lookup_monitor(ctx
, monitors
[i
]);
427 /* Give the driver a chance to stop the monitor if it's active. */
429 ctx
->Driver
.ResetPerfMonitor(ctx
, m
);
433 _mesa_HashRemove(ctx
->PerfMonitor
.Monitors
, monitors
[i
]);
434 ralloc_free(m
->ActiveGroups
);
435 ralloc_free(m
->ActiveCounters
);
436 ctx
->Driver
.DeletePerfMonitor(ctx
, m
);
438 /* "INVALID_VALUE error will be generated if any of the monitor IDs
439 * in the <monitors> parameter to DeletePerfMonitorsAMD do not
440 * reference a valid generated monitor ID."
442 _mesa_error(ctx
, GL_INVALID_VALUE
,
443 "glDeletePerfMonitorsAMD(invalid monitor)");
449 _mesa_SelectPerfMonitorCountersAMD(GLuint monitor
, GLboolean enable
,
450 GLuint group
, GLint numCounters
,
453 GET_CURRENT_CONTEXT(ctx
);
455 struct gl_perf_monitor_object
*m
;
456 const struct gl_perf_monitor_group
*group_obj
;
458 m
= lookup_monitor(ctx
, monitor
);
460 /* "INVALID_VALUE error will be generated if the <monitor> parameter to
461 * SelectPerfMonitorCountersAMD does not reference a monitor created by
462 * GenPerfMonitorsAMD."
465 _mesa_error(ctx
, GL_INVALID_VALUE
,
466 "glSelectPerfMonitorCountersAMD(invalid monitor)");
470 group_obj
= get_group(ctx
, group
);
472 /* "INVALID_VALUE error will be generated if the <group> parameter to
473 * GetPerfMonitorCountersAMD, GetPerfMonitorCounterStringAMD,
474 * GetPerfMonitorCounterStringAMD, GetPerfMonitorCounterInfoAMD, or
475 * SelectPerfMonitorCountersAMD does not reference a valid group ID."
477 if (group_obj
== NULL
) {
478 _mesa_error(ctx
, GL_INVALID_VALUE
,
479 "glSelectPerfMonitorCountersAMD(invalid group)");
483 /* "INVALID_VALUE error will be generated if the <numCounters> parameter to
484 * SelectPerfMonitorCountersAMD is less than 0."
486 if (numCounters
< 0) {
487 _mesa_error(ctx
, GL_INVALID_VALUE
,
488 "glSelectPerfMonitorCountersAMD(numCounters < 0)");
492 /* "When SelectPerfMonitorCountersAMD is called on a monitor, any outstanding
493 * results for that monitor become invalidated and the result queries
494 * PERFMON_RESULT_SIZE_AMD and PERFMON_RESULT_AVAILABLE_AMD are reset to 0."
496 ctx
->Driver
.ResetPerfMonitor(ctx
, m
);
498 /* Sanity check the counter ID list. */
499 for (i
= 0; i
< numCounters
; i
++) {
500 if (counterList
[i
] >= group_obj
->NumCounters
) {
501 _mesa_error(ctx
, GL_INVALID_VALUE
,
502 "glSelectPerfMonitorCountersAMD(invalid counter ID)");
508 /* Enable the counters */
509 for (i
= 0; i
< numCounters
; i
++) {
510 ++m
->ActiveGroups
[group
];
511 BITSET_SET(m
->ActiveCounters
[group
], counterList
[i
]);
514 /* Disable the counters */
515 for (i
= 0; i
< numCounters
; i
++) {
516 --m
->ActiveGroups
[group
];
517 BITSET_CLEAR(m
->ActiveCounters
[group
], counterList
[i
]);
523 _mesa_BeginPerfMonitorAMD(GLuint monitor
)
525 GET_CURRENT_CONTEXT(ctx
);
527 struct gl_perf_monitor_object
*m
= lookup_monitor(ctx
, monitor
);
530 _mesa_error(ctx
, GL_INVALID_VALUE
,
531 "glBeginPerfMonitorAMD(invalid monitor)");
535 /* "INVALID_OPERATION error will be generated if BeginPerfMonitorAMD is
536 * called when a performance monitor is already active."
539 _mesa_error(ctx
, GL_INVALID_OPERATION
,
540 "glBeginPerfMonitor(already active)");
544 /* The driver is free to return false if it can't begin monitoring for
545 * any reason. This translates into an INVALID_OPERATION error.
547 if (ctx
->Driver
.BeginPerfMonitor(ctx
, m
)) {
551 _mesa_error(ctx
, GL_INVALID_OPERATION
,
552 "glBeginPerfMonitor(driver unable to begin monitoring)");
557 _mesa_EndPerfMonitorAMD(GLuint monitor
)
559 GET_CURRENT_CONTEXT(ctx
);
561 struct gl_perf_monitor_object
*m
= lookup_monitor(ctx
, monitor
);
564 _mesa_error(ctx
, GL_INVALID_VALUE
, "glEndPerfMonitorAMD(invalid monitor)");
568 /* "INVALID_OPERATION error will be generated if EndPerfMonitorAMD is called
569 * when a performance monitor is not currently started."
572 _mesa_error(ctx
, GL_INVALID_OPERATION
, "glBeginPerfMonitor(not active)");
576 ctx
->Driver
.EndPerfMonitor(ctx
, m
);
583 * Return the number of bytes needed to store a monitor's result.
586 perf_monitor_result_size(const struct gl_context
*ctx
,
587 const struct gl_perf_monitor_object
*m
)
589 unsigned group
, counter
;
592 for (group
= 0; group
< ctx
->PerfMonitor
.NumGroups
; group
++) {
593 const struct gl_perf_monitor_group
*g
= &ctx
->PerfMonitor
.Groups
[group
];
596 BITSET_FOREACH_SET(counter
, tmp
, m
->ActiveCounters
[group
], g
->NumCounters
) {
597 const struct gl_perf_monitor_counter
*c
= &g
->Counters
[counter
];
599 size
+= sizeof(uint32_t); /* Group ID */
600 size
+= sizeof(uint32_t); /* Counter ID */
601 size
+= _mesa_perf_monitor_counter_size(c
);
608 _mesa_GetPerfMonitorCounterDataAMD(GLuint monitor
, GLenum pname
,
609 GLsizei dataSize
, GLuint
*data
,
612 GET_CURRENT_CONTEXT(ctx
);
614 struct gl_perf_monitor_object
*m
= lookup_monitor(ctx
, monitor
);
615 bool result_available
;
618 _mesa_error(ctx
, GL_INVALID_VALUE
,
619 "glGetPerfMonitorCounterDataAMD(invalid monitor)");
623 /* "It is an INVALID_OPERATION error for <data> to be NULL." */
625 _mesa_error(ctx
, GL_INVALID_OPERATION
,
626 "glGetPerfMonitorCounterDataAMD(data == NULL)");
630 /* We need at least enough room for a single value. */
631 if (dataSize
< sizeof(GLuint
)) {
632 if (bytesWritten
!= NULL
)
637 /* If the monitor has never ended, there is no result. */
638 result_available
= m
->Ended
&&
639 ctx
->Driver
.IsPerfMonitorResultAvailable(ctx
, m
);
641 /* AMD appears to return 0 for all queries unless a result is available. */
642 if (!result_available
) {
644 if (bytesWritten
!= NULL
)
645 *bytesWritten
= sizeof(GLuint
);
650 case GL_PERFMON_RESULT_AVAILABLE_AMD
:
652 if (bytesWritten
!= NULL
)
653 *bytesWritten
= sizeof(GLuint
);
655 case GL_PERFMON_RESULT_SIZE_AMD
:
656 *data
= perf_monitor_result_size(ctx
, m
);
657 if (bytesWritten
!= NULL
)
658 *bytesWritten
= sizeof(GLuint
);
660 case GL_PERFMON_RESULT_AMD
:
661 ctx
->Driver
.GetPerfMonitorResult(ctx
, m
, dataSize
, data
, bytesWritten
);
664 _mesa_error(ctx
, GL_INVALID_ENUM
,
665 "glGetPerfMonitorCounterDataAMD(pname)");
670 * Returns how many bytes a counter's value takes up.
673 _mesa_perf_monitor_counter_size(const struct gl_perf_monitor_counter
*c
)
677 case GL_PERCENTAGE_AMD
:
678 return sizeof(GLfloat
);
679 case GL_UNSIGNED_INT
:
680 return sizeof(GLuint
);
681 case GL_UNSIGNED_INT64_AMD
:
682 return sizeof(uint64_t);
684 assert(!"Should not get here: invalid counter type");