1 /**************************************************************************
3 * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
28 /* Provide additional functionality on top of bufmgr buffers:
29 * - 2d semantics and blit operations
30 * - refcounting of buffers for multiple images in a buffer.
31 * - refcounting of buffer mappings.
32 * - some logic for moving the buffers to the best memory pools for
35 * Most of this is to make it easier to implement the fixed-layout
36 * mipmap tree required by intel hardware in the face of GL's
37 * programming interface where each image can be specifed in random
38 * order and it isn't clear what layout the tree should have until the
42 #include <sys/ioctl.h>
45 #include "main/hash.h"
46 #include "intel_context.h"
47 #include "intel_regions.h"
48 #include "intel_blit.h"
49 #include "intel_buffer_objects.h"
50 #include "intel_bufmgr.h"
51 #include "intel_batchbuffer.h"
53 #define FILE_DEBUG_FLAG DEBUG_REGION
55 /* This should be set to the maximum backtrace size desired.
56 * Set it to 0 to disable backtrace debugging.
58 #define DEBUG_BACKTRACE_SIZE 0
60 #if DEBUG_BACKTRACE_SIZE == 0
61 /* Use the standard debug output */
62 #define _DBG(...) DBG(__VA_ARGS__)
64 /* Use backtracing debug output */
65 #define _DBG(...) {debug_backtrace(); DBG(__VA_ARGS__);}
67 /* Backtracing debug support */
73 void *trace
[DEBUG_BACKTRACE_SIZE
];
74 char **strings
= NULL
;
78 traceSize
= backtrace(trace
, DEBUG_BACKTRACE_SIZE
);
79 strings
= backtrace_symbols(trace
, traceSize
);
80 if (strings
== NULL
) {
85 /* Spit out all the strings with a colon separator. Ignore
86 * the first, since we don't really care about the call
87 * to debug_backtrace() itself. Skip until the final "/" in
88 * the trace to avoid really long lines.
90 for (i
= 1; i
< traceSize
; i
++) {
91 char *p
= strings
[i
], *slash
= strings
[i
];
101 /* Free up the memory, and we're done */
109 /* XXX: Thread safety?
112 intel_region_map(struct intel_context
*intel
, struct intel_region
*region
,
115 /* We have the region->map_refcount controlling mapping of the BO because
116 * in software fallbacks we may end up mapping the same buffer multiple
117 * times on Mesa's behalf, so we refcount our mappings to make sure that
118 * the pointer stays valid until the end of the unmap chain. However, we
119 * must not emit any batchbuffers between the start of mapping and the end
120 * of unmapping, or further use of the map will be incoherent with the GPU
121 * rendering done by that batchbuffer. Hence we assert in
122 * intel_batchbuffer_flush() that that doesn't happen, which means that the
123 * flush is only needed on first map of the buffer.
126 _DBG("%s %p\n", __FUNCTION__
, region
);
127 if (!region
->map_refcount
) {
128 intel_flush(&intel
->ctx
);
130 if (region
->tiling
!= I915_TILING_NONE
)
131 drm_intel_gem_bo_map_gtt(region
->bo
);
133 drm_intel_bo_map(region
->bo
, true);
135 region
->map
= region
->bo
->virtual;
138 intel
->num_mapped_regions
++;
139 region
->map_refcount
++;
146 intel_region_unmap(struct intel_context
*intel
, struct intel_region
*region
)
148 _DBG("%s %p\n", __FUNCTION__
, region
);
149 if (!--region
->map_refcount
) {
150 if (region
->tiling
!= I915_TILING_NONE
)
151 drm_intel_gem_bo_unmap_gtt(region
->bo
);
153 drm_intel_bo_unmap(region
->bo
);
156 --intel
->num_mapped_regions
;
157 assert(intel
->num_mapped_regions
>= 0);
161 static struct intel_region
*
162 intel_region_alloc_internal(struct intel_screen
*screen
,
164 GLuint width
, GLuint height
, GLuint pitch
,
165 uint32_t tiling
, drm_intel_bo
*buffer
)
167 struct intel_region
*region
;
169 region
= calloc(sizeof(*region
), 1);
174 region
->width
= width
;
175 region
->height
= height
;
176 region
->pitch
= pitch
;
177 region
->refcount
= 1;
179 region
->tiling
= tiling
;
180 region
->screen
= screen
;
182 _DBG("%s <-- %p\n", __FUNCTION__
, region
);
186 struct intel_region
*
187 intel_region_alloc(struct intel_screen
*screen
,
189 GLuint cpp
, GLuint width
, GLuint height
,
190 bool expect_accelerated_upload
)
192 drm_intel_bo
*buffer
;
193 unsigned long flags
= 0;
194 unsigned long aligned_pitch
;
195 struct intel_region
*region
;
197 if (expect_accelerated_upload
)
198 flags
|= BO_ALLOC_FOR_RENDER
;
200 buffer
= drm_intel_bo_alloc_tiled(screen
->bufmgr
, "region",
202 &tiling
, &aligned_pitch
, flags
);
206 region
= intel_region_alloc_internal(screen
, cpp
, width
, height
,
207 aligned_pitch
/ cpp
, tiling
, buffer
);
208 if (region
== NULL
) {
209 drm_intel_bo_unreference(buffer
);
217 intel_region_flink(struct intel_region
*region
, uint32_t *name
)
219 if (region
->name
== 0) {
220 if (drm_intel_bo_flink(region
->bo
, ®ion
->name
))
223 _mesa_HashInsert(region
->screen
->named_regions
,
224 region
->name
, region
);
227 *name
= region
->name
;
232 struct intel_region
*
233 intel_region_alloc_for_handle(struct intel_screen
*screen
,
235 GLuint width
, GLuint height
, GLuint pitch
,
236 GLuint handle
, const char *name
)
238 struct intel_region
*region
, *dummy
;
239 drm_intel_bo
*buffer
;
241 uint32_t bit_6_swizzle
, tiling
;
243 region
= _mesa_HashLookup(screen
->named_regions
, handle
);
244 if (region
!= NULL
) {
246 if (region
->width
!= width
|| region
->height
!= height
||
247 region
->cpp
!= cpp
|| region
->pitch
!= pitch
) {
249 "Region for name %d already exists but is not compatible\n",
253 intel_region_reference(&dummy
, region
);
257 buffer
= intel_bo_gem_create_from_name(screen
->bufmgr
, name
, handle
);
260 ret
= drm_intel_bo_get_tiling(buffer
, &tiling
, &bit_6_swizzle
);
262 fprintf(stderr
, "Couldn't get tiling of buffer %d (%s): %s\n",
263 handle
, name
, strerror(-ret
));
264 drm_intel_bo_unreference(buffer
);
268 region
= intel_region_alloc_internal(screen
, cpp
,
269 width
, height
, pitch
, tiling
, buffer
);
270 if (region
== NULL
) {
271 drm_intel_bo_unreference(buffer
);
275 region
->name
= handle
;
276 _mesa_HashInsert(screen
->named_regions
, handle
, region
);
282 intel_region_reference(struct intel_region
**dst
, struct intel_region
*src
)
284 _DBG("%s: %p(%d) -> %p(%d)\n", __FUNCTION__
,
285 *dst
, *dst
? (*dst
)->refcount
: 0, src
, src
? src
->refcount
: 0);
289 intel_region_release(dst
);
298 intel_region_release(struct intel_region
**region_handle
)
300 struct intel_region
*region
= *region_handle
;
302 if (region
== NULL
) {
303 _DBG("%s NULL\n", __FUNCTION__
);
307 _DBG("%s %p %d\n", __FUNCTION__
, region
, region
->refcount
- 1);
309 ASSERT(region
->refcount
> 0);
312 if (region
->refcount
== 0) {
313 assert(region
->map_refcount
== 0);
315 drm_intel_bo_unreference(region
->bo
);
317 if (region
->name
> 0)
318 _mesa_HashRemove(region
->screen
->named_regions
, region
->name
);
322 *region_handle
= NULL
;
326 * XXX Move this into core Mesa?
329 _mesa_copy_rect(GLubyte
* dst
,
337 GLuint src_pitch
, GLuint src_x
, GLuint src_y
)
345 dst
+= dst_y
* dst_pitch
;
346 src
+= src_y
* src_pitch
;
349 if (width
== dst_pitch
&& width
== src_pitch
)
350 memcpy(dst
, src
, height
* width
);
352 for (i
= 0; i
< height
; i
++) {
353 memcpy(dst
, src
, width
);
360 /* Copy rectangular sub-regions. Need better logic about when to
361 * push buffers into AGP - will currently do so whenever possible.
364 intel_region_copy(struct intel_context
*intel
,
365 struct intel_region
*dst
,
367 GLuint dstx
, GLuint dsty
,
368 struct intel_region
*src
,
370 GLuint srcx
, GLuint srcy
, GLuint width
, GLuint height
,
374 uint32_t src_pitch
= src
->pitch
;
376 _DBG("%s\n", __FUNCTION__
);
381 assert(src
->cpp
== dst
->cpp
);
384 src_pitch
= -src_pitch
;
386 return intelEmitCopyBlit(intel
,
388 src_pitch
, src
->bo
, src_offset
, src
->tiling
,
389 dst
->pitch
, dst
->bo
, dst_offset
, dst
->tiling
,
390 srcx
, srcy
, dstx
, dsty
, width
, height
,
395 * This function computes masks that may be used to select the bits of the X
396 * and Y coordinates that indicate the offset within a tile. If the region is
397 * untiled, the masks are set to 0.
400 intel_region_get_tile_masks(struct intel_region
*region
,
401 uint32_t *mask_x
, uint32_t *mask_y
)
403 int cpp
= region
->cpp
;
405 switch (region
->tiling
) {
408 case I915_TILING_NONE
:
409 *mask_x
= *mask_y
= 0;
412 *mask_x
= 512 / cpp
- 1;
416 *mask_x
= 128 / cpp
- 1;
423 * Compute the offset (in bytes) from the start of the region to the given x
424 * and y coordinate. For tiled regions, caller must ensure that x and y are
425 * multiples of the tile size.
428 intel_region_get_aligned_offset(struct intel_region
*region
, uint32_t x
,
431 int cpp
= region
->cpp
;
432 uint32_t pitch
= region
->pitch
* cpp
;
434 switch (region
->tiling
) {
437 case I915_TILING_NONE
:
438 return y
* pitch
+ x
* cpp
;
440 assert((x
% (512 / cpp
)) == 0);
441 assert((y
% 8) == 0);
442 return y
* pitch
+ x
/ (512 / cpp
) * 4096;
444 assert((x
% (128 / cpp
)) == 0);
445 assert((y
% 32) == 0);
446 return y
* pitch
+ x
/ (128 / cpp
) * 4096;