radeon_queryobj.c revision cefee4e327c92daa2f01b6de650a43eddd348063 
1/*
2 * Copyright © 2008-2009 Maciej Cencora <m.cencora@gmail.com>
3 *
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:
10 *
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
13 * Software.
14 *
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 DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 *    Maciej Cencora <m.cencora@gmail.com>
25 *
26 */
27#include "radeon_common.h"
28#include "radeon_queryobj.h"
29#include "radeon_debug.h"
30
31#include "main/imports.h"
32#include "main/simple_list.h"
33
34static void radeonQueryGetResult(GLcontext *ctx, struct gl_query_object *q)
35{
36	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
37	struct radeon_query_object *query = (struct radeon_query_object *)q;
38        uint32_t *result;
39	int i;
40
41	radeon_print(RADEON_STATE, RADEON_VERBOSE,
42			"%s: query id %d, result %d\n",
43			__FUNCTION__, query->Base.Id, (int) query->Base.Result);
44
45	radeon_bo_map(query->bo, GL_FALSE);
46        result = query->bo->ptr;
47
48	query->Base.Result = 0;
49	if (IS_R600_CLASS(radeon->radeonScreen)) {
50		/* ZPASS EVENT writes alternating qwords
51		 * At query start we set the start offset to 0 and
52		 * hw writes zpass start counts to qwords 0, 2, 4, 6.
53		 * At query end we set the start offset to 8 and
54		 * hw writes zpass end counts to qwords 1, 3, 5, 7.
55		 * then we substract. MSB is the valid bit.
56		 */
57		for (i = 0; i < 16; i += 4) {
58			uint64_t start = (uint64_t)LE32_TO_CPU(result[i]) |
59					 (uint64_t)LE32_TO_CPU(result[i + 1]) << 32;
60			uint64_t end = (uint64_t)LE32_TO_CPU(result[i + 2]) |
61				       (uint64_t)LE32_TO_CPU(result[i + 3]) << 32;
62			if ((start & 0x8000000000000000) && (end & 0x8000000000000000)) {
63				uint64_t query_count = end - start;
64				query->Base.Result += query_count;
65
66			}
67			radeon_print(RADEON_STATE, RADEON_TRACE,
68				     "%d start: %lx, end: %lx %ld\n", i, start, end, end - start);
69		}
70	} else {
71		for (i = 0; i < query->curr_offset/sizeof(uint32_t); ++i) {
72			query->Base.Result += LE32_TO_CPU(result[i]);
73			radeon_print(RADEON_STATE, RADEON_TRACE, "result[%d] = %d\n", i, LE32_TO_CPU(result[i]));
74		}
75	}
76
77	radeon_bo_unmap(query->bo);
78}
79
80static struct gl_query_object * radeonNewQueryObject(GLcontext *ctx, GLuint id)
81{
82	struct radeon_query_object *query;
83
84	query = _mesa_calloc(sizeof(struct radeon_query_object));
85
86	query->Base.Id = id;
87	query->Base.Result = 0;
88	query->Base.Active = GL_FALSE;
89	query->Base.Ready = GL_TRUE;
90
91	radeon_print(RADEON_STATE, RADEON_VERBOSE,"%s: query id %d\n", __FUNCTION__, query->Base.Id);
92
93	return &query->Base;
94}
95
96static void radeonDeleteQuery(GLcontext *ctx, struct gl_query_object *q)
97{
98	struct radeon_query_object *query = (struct radeon_query_object *)q;
99
100	radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id);
101
102	if (query->bo) {
103		radeon_bo_unref(query->bo);
104	}
105
106	_mesa_free(query);
107}
108
109static void radeonWaitQuery(GLcontext *ctx, struct gl_query_object *q)
110{
111	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
112	struct radeon_query_object *query = (struct radeon_query_object *)q;
113
114	/* If the cmdbuf with packets for this query hasn't been flushed yet, do it now */
115	if (radeon_bo_is_referenced_by_cs(query->bo, radeon->cmdbuf.cs))
116		ctx->Driver.Flush(ctx);
117
118	radeon_print(RADEON_STATE, RADEON_VERBOSE, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, q->Id, query->bo, query->curr_offset);
119
120	radeonQueryGetResult(ctx, q);
121
122	query->Base.Ready = GL_TRUE;
123}
124
125
126static void radeonBeginQuery(GLcontext *ctx, struct gl_query_object *q)
127{
128	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
129	struct radeon_query_object *query = (struct radeon_query_object *)q;
130
131	radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id);
132
133	assert(radeon->query.current == NULL);
134
135	if (radeon->dma.flush)
136		radeon->dma.flush(radeon->glCtx);
137
138	if (!query->bo) {
139		query->bo = radeon_bo_open(radeon->radeonScreen->bom, 0, RADEON_QUERY_PAGE_SIZE, RADEON_QUERY_PAGE_SIZE, RADEON_GEM_DOMAIN_GTT, 0);
140	}
141	query->curr_offset = 0;
142
143	radeon->query.current = query;
144
145	radeon->query.queryobj.dirty = GL_TRUE;
146	radeon->hw.is_dirty = GL_TRUE;
147}
148
149void radeonEmitQueryEnd(GLcontext *ctx)
150{
151	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
152	struct radeon_query_object *query = radeon->query.current;
153
154	if (!query)
155		return;
156
157	if (query->emitted_begin == GL_FALSE)
158		return;
159
160	radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d, bo %p, offset %d\n", __FUNCTION__, query->Base.Id, query->bo, query->curr_offset);
161
162	radeon_cs_space_check_with_bo(radeon->cmdbuf.cs,
163				      query->bo,
164				      0, RADEON_GEM_DOMAIN_GTT);
165
166	radeon->vtbl.emit_query_finish(radeon);
167}
168
169static void radeonEndQuery(GLcontext *ctx, struct gl_query_object *q)
170{
171	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
172
173	radeon_print(RADEON_STATE, RADEON_NORMAL, "%s: query id %d\n", __FUNCTION__, q->Id);
174
175	if (radeon->dma.flush)
176		radeon->dma.flush(radeon->glCtx);
177	radeonEmitQueryEnd(ctx);
178
179	radeon->query.current = NULL;
180}
181
182static void radeonCheckQuery(GLcontext *ctx, struct gl_query_object *q)
183{
184	radeon_print(RADEON_STATE, RADEON_TRACE, "%s: query id %d\n", __FUNCTION__, q->Id);
185
186#ifdef DRM_RADEON_GEM_BUSY
187	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
188
189	if (radeon->radeonScreen->kernel_mm) {
190		struct radeon_query_object *query = (struct radeon_query_object *)q;
191		uint32_t domain;
192
193		/* Need to perform a flush, as per ARB_occlusion_query spec */
194		if (radeon_bo_is_referenced_by_cs(query->bo, radeon->cmdbuf.cs)) {
195			ctx->Driver.Flush(ctx);
196		}
197
198		if (radeon_bo_is_busy(query->bo, &domain) == 0) {
199			radeonQueryGetResult(ctx, q);
200			query->Base.Ready = GL_TRUE;
201		}
202	} else {
203		radeonWaitQuery(ctx, q);
204	}
205#else
206	radeonWaitQuery(ctx, q);
207#endif
208}
209
210void radeonInitQueryObjFunctions(struct dd_function_table *functions)
211{
212	functions->NewQueryObject = radeonNewQueryObject;
213	functions->DeleteQuery = radeonDeleteQuery;
214	functions->BeginQuery = radeonBeginQuery;
215	functions->EndQuery = radeonEndQuery;
216	functions->CheckQuery = radeonCheckQuery;
217	functions->WaitQuery = radeonWaitQuery;
218}
219
220int radeon_check_query_active(GLcontext *ctx, struct radeon_state_atom *atom)
221{
222	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
223	struct radeon_query_object *query = radeon->query.current;
224
225	if (!query || query->emitted_begin)
226		return 0;
227	return atom->cmd_size;
228}
229
230void radeon_emit_queryobj(GLcontext *ctx, struct radeon_state_atom *atom)
231{
232	radeonContextPtr radeon = RADEON_CONTEXT(ctx);
233	BATCH_LOCALS(radeon);
234	int dwords;
235
236	dwords = (*atom->check) (ctx, atom);
237
238	BEGIN_BATCH_NO_AUTOSTATE(dwords);
239	OUT_BATCH_TABLE(atom->cmd, dwords);
240	END_BATCH();
241
242	radeon->query.current->emitted_begin = GL_TRUE;
243}
244