2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
23 #include "r600_pipe.h"
24 #include "r600_public.h"
27 #include "pipe/p_shader_tokens.h"
28 #include "util/u_blitter.h"
29 #include "util/u_format_s3tc.h"
30 #include "util/u_memory.h"
31 #include "util/u_simple_shaders.h"
32 #include "util/u_upload_mgr.h"
33 #include "vl/vl_decoder.h"
34 #include "vl/vl_video_buffer.h"
35 #include "os/os_time.h"
40 static struct r600_fence
*r600_create_fence(struct r600_context
*rctx
)
42 struct r600_screen
*rscreen
= rctx
->screen
;
43 struct r600_fence
*fence
= NULL
;
45 pipe_mutex_lock(rscreen
->fences
.mutex
);
47 if (!rscreen
->fences
.bo
) {
48 /* Create the shared buffer object */
49 rscreen
->fences
.bo
= (struct r600_resource
*)
50 pipe_buffer_create(&rscreen
->screen
, PIPE_BIND_CUSTOM
,
51 PIPE_USAGE_STAGING
, 4096);
52 if (!rscreen
->fences
.bo
) {
53 R600_ERR("r600: failed to create bo for fence objects\n");
56 rscreen
->fences
.data
= rctx
->ws
->buffer_map(rscreen
->fences
.bo
->cs_buf
,
58 PIPE_TRANSFER_READ_WRITE
);
61 if (!LIST_IS_EMPTY(&rscreen
->fences
.pool
)) {
62 struct r600_fence
*entry
;
64 /* Try to find a freed fence that has been signalled */
65 LIST_FOR_EACH_ENTRY(entry
, &rscreen
->fences
.pool
, head
) {
66 if (rscreen
->fences
.data
[entry
->index
] != 0) {
67 LIST_DELINIT(&entry
->head
);
75 /* Allocate a new fence */
76 struct r600_fence_block
*block
;
79 if ((rscreen
->fences
.next_index
+ 1) >= 1024) {
80 R600_ERR("r600: too many concurrent fences\n");
84 index
= rscreen
->fences
.next_index
++;
86 if (!(index
% FENCE_BLOCK_SIZE
)) {
87 /* Allocate a new block */
88 block
= CALLOC_STRUCT(r600_fence_block
);
92 LIST_ADD(&block
->head
, &rscreen
->fences
.blocks
);
94 block
= LIST_ENTRY(struct r600_fence_block
, rscreen
->fences
.blocks
.next
, head
);
97 fence
= &block
->fences
[index
% FENCE_BLOCK_SIZE
];
101 pipe_reference_init(&fence
->reference
, 1);
103 rscreen
->fences
.data
[fence
->index
] = 0;
104 r600_context_emit_fence(rctx
, rscreen
->fences
.bo
, fence
->index
, 1);
106 /* Create a dummy BO so that fence_finish without a timeout can sleep waiting for completion */
107 fence
->sleep_bo
= (struct r600_resource
*)
108 pipe_buffer_create(&rctx
->screen
->screen
, PIPE_BIND_CUSTOM
,
109 PIPE_USAGE_STAGING
, 1);
110 /* Add the fence as a dummy relocation. */
111 r600_context_bo_reloc(rctx
, fence
->sleep_bo
, RADEON_USAGE_READWRITE
);
114 pipe_mutex_unlock(rscreen
->fences
.mutex
);
119 void r600_flush(struct pipe_context
*ctx
, struct pipe_fence_handle
**fence
,
122 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
123 struct r600_fence
**rfence
= (struct r600_fence
**)fence
;
124 struct pipe_query
*render_cond
= NULL
;
125 unsigned render_cond_mode
= 0;
128 *rfence
= r600_create_fence(rctx
);
130 /* Disable render condition. */
131 if (rctx
->current_render_cond
) {
132 render_cond
= rctx
->current_render_cond
;
133 render_cond_mode
= rctx
->current_render_cond_mode
;
134 ctx
->render_condition(ctx
, NULL
, 0);
137 r600_context_flush(rctx
, flags
);
139 /* Re-enable render condition. */
141 ctx
->render_condition(ctx
, render_cond
, render_cond_mode
);
145 static void r600_flush_from_st(struct pipe_context
*ctx
,
146 struct pipe_fence_handle
**fence
)
148 r600_flush(ctx
, fence
, 0);
151 static void r600_flush_from_winsys(void *ctx
, unsigned flags
)
153 r600_flush((struct pipe_context
*)ctx
, NULL
, flags
);
156 static void r600_destroy_context(struct pipe_context
*context
)
158 struct r600_context
*rctx
= (struct r600_context
*)context
;
160 pipe_resource_reference((struct pipe_resource
**)&rctx
->dummy_cmask
, NULL
);
161 pipe_resource_reference((struct pipe_resource
**)&rctx
->dummy_fmask
, NULL
);
163 if (rctx
->dummy_pixel_shader
) {
164 rctx
->context
.delete_fs_state(&rctx
->context
, rctx
->dummy_pixel_shader
);
166 if (rctx
->custom_dsa_flush
) {
167 rctx
->context
.delete_depth_stencil_alpha_state(&rctx
->context
, rctx
->custom_dsa_flush
);
169 if (rctx
->custom_blend_resolve
) {
170 rctx
->context
.delete_blend_state(&rctx
->context
, rctx
->custom_blend_resolve
);
172 if (rctx
->custom_blend_decompress
) {
173 rctx
->context
.delete_blend_state(&rctx
->context
, rctx
->custom_blend_decompress
);
175 if (rctx
->custom_blend_fmask_decompress
) {
176 rctx
->context
.delete_blend_state(&rctx
->context
, rctx
->custom_blend_fmask_decompress
);
178 util_unreference_framebuffer_state(&rctx
->framebuffer
.state
);
180 r600_context_fini(rctx
);
183 util_blitter_destroy(rctx
->blitter
);
185 if (rctx
->uploader
) {
186 u_upload_destroy(rctx
->uploader
);
188 if (rctx
->allocator_so_filled_size
) {
189 u_suballocator_destroy(rctx
->allocator_so_filled_size
);
191 util_slab_destroy(&rctx
->pool_transfers
);
193 r600_release_command_buffer(&rctx
->start_cs_cmd
);
196 rctx
->ws
->cs_destroy(rctx
->cs
);
203 static struct pipe_context
*r600_create_context(struct pipe_screen
*screen
, void *priv
)
205 struct r600_context
*rctx
= CALLOC_STRUCT(r600_context
);
206 struct r600_screen
* rscreen
= (struct r600_screen
*)screen
;
211 util_slab_create(&rctx
->pool_transfers
,
212 sizeof(struct r600_transfer
), 64,
213 UTIL_SLAB_SINGLETHREADED
);
215 rctx
->context
.screen
= screen
;
216 rctx
->context
.priv
= priv
;
217 rctx
->context
.destroy
= r600_destroy_context
;
218 rctx
->context
.flush
= r600_flush_from_st
;
220 /* Easy accessing of screen/winsys. */
221 rctx
->screen
= rscreen
;
222 rctx
->ws
= rscreen
->ws
;
223 rctx
->family
= rscreen
->family
;
224 rctx
->chip_class
= rscreen
->chip_class
;
225 rctx
->keep_tiling_flags
= rscreen
->info
.drm_minor
>= 12;
227 LIST_INITHEAD(&rctx
->active_timer_queries
);
228 LIST_INITHEAD(&rctx
->active_nontimer_queries
);
229 LIST_INITHEAD(&rctx
->dirty
);
230 LIST_INITHEAD(&rctx
->enable_list
);
232 rctx
->range
= CALLOC(NUM_RANGES
, sizeof(struct r600_range
));
236 r600_init_blit_functions(rctx
);
237 r600_init_query_functions(rctx
);
238 r600_init_context_resource_functions(rctx
);
239 r600_init_surface_functions(rctx
);
242 rctx
->context
.create_video_decoder
= vl_create_decoder
;
243 rctx
->context
.create_video_buffer
= vl_video_buffer_create
;
245 r600_init_common_state_functions(rctx
);
247 switch (rctx
->chip_class
) {
250 r600_init_state_functions(rctx
);
251 r600_init_atom_start_cs(rctx
);
252 if (r600_context_init(rctx
))
254 rctx
->custom_dsa_flush
= r600_create_db_flush_dsa(rctx
);
255 rctx
->custom_blend_resolve
= rctx
->chip_class
== R700
? r700_create_resolve_blend(rctx
)
256 : r600_create_resolve_blend(rctx
);
257 rctx
->custom_blend_decompress
= r600_create_decompress_blend(rctx
);
258 rctx
->has_vertex_cache
= !(rctx
->family
== CHIP_RV610
||
259 rctx
->family
== CHIP_RV620
||
260 rctx
->family
== CHIP_RS780
||
261 rctx
->family
== CHIP_RS880
||
262 rctx
->family
== CHIP_RV710
);
266 evergreen_init_state_functions(rctx
);
267 evergreen_init_atom_start_cs(rctx
);
268 evergreen_init_atom_start_compute_cs(rctx
);
269 if (evergreen_context_init(rctx
))
271 rctx
->custom_dsa_flush
= evergreen_create_db_flush_dsa(rctx
);
272 rctx
->custom_blend_resolve
= evergreen_create_resolve_blend(rctx
);
273 rctx
->custom_blend_decompress
= evergreen_create_decompress_blend(rctx
);
274 rctx
->custom_blend_fmask_decompress
= evergreen_create_fmask_decompress_blend(rctx
);
275 rctx
->has_vertex_cache
= !(rctx
->family
== CHIP_CEDAR
||
276 rctx
->family
== CHIP_PALM
||
277 rctx
->family
== CHIP_SUMO
||
278 rctx
->family
== CHIP_SUMO2
||
279 rctx
->family
== CHIP_CAICOS
||
280 rctx
->family
== CHIP_CAYMAN
||
281 rctx
->family
== CHIP_ARUBA
);
284 R600_ERR("Unsupported chip class %d.\n", rctx
->chip_class
);
288 rctx
->cs
= rctx
->ws
->cs_create(rctx
->ws
);
289 rctx
->ws
->cs_set_flush_callback(rctx
->cs
, r600_flush_from_winsys
, rctx
);
291 rctx
->uploader
= u_upload_create(&rctx
->context
, 1024 * 1024, 256,
292 PIPE_BIND_INDEX_BUFFER
|
293 PIPE_BIND_CONSTANT_BUFFER
);
297 rctx
->allocator_so_filled_size
= u_suballocator_create(&rctx
->context
, 4096, 4,
298 0, PIPE_USAGE_STATIC
, TRUE
);
299 if (!rctx
->allocator_so_filled_size
)
302 rctx
->blitter
= util_blitter_create(&rctx
->context
);
303 if (rctx
->blitter
== NULL
)
305 util_blitter_set_texture_multisample(rctx
->blitter
, rscreen
->has_msaa
);
306 rctx
->blitter
->draw_rectangle
= r600_draw_rectangle
;
308 r600_begin_new_cs(rctx
);
309 r600_get_backend_mask(rctx
); /* this emits commands and must be last */
311 rctx
->dummy_pixel_shader
=
312 util_make_fragment_cloneinput_shader(&rctx
->context
, 0,
313 TGSI_SEMANTIC_GENERIC
,
314 TGSI_INTERPOLATE_CONSTANT
);
315 rctx
->context
.bind_fs_state(&rctx
->context
, rctx
->dummy_pixel_shader
);
317 return &rctx
->context
;
320 r600_destroy_context(&rctx
->context
);
327 static const char* r600_get_vendor(struct pipe_screen
* pscreen
)
332 static const char *r600_get_family_name(enum radeon_family family
)
335 case CHIP_R600
: return "AMD R600";
336 case CHIP_RV610
: return "AMD RV610";
337 case CHIP_RV630
: return "AMD RV630";
338 case CHIP_RV670
: return "AMD RV670";
339 case CHIP_RV620
: return "AMD RV620";
340 case CHIP_RV635
: return "AMD RV635";
341 case CHIP_RS780
: return "AMD RS780";
342 case CHIP_RS880
: return "AMD RS880";
343 case CHIP_RV770
: return "AMD RV770";
344 case CHIP_RV730
: return "AMD RV730";
345 case CHIP_RV710
: return "AMD RV710";
346 case CHIP_RV740
: return "AMD RV740";
347 case CHIP_CEDAR
: return "AMD CEDAR";
348 case CHIP_REDWOOD
: return "AMD REDWOOD";
349 case CHIP_JUNIPER
: return "AMD JUNIPER";
350 case CHIP_CYPRESS
: return "AMD CYPRESS";
351 case CHIP_HEMLOCK
: return "AMD HEMLOCK";
352 case CHIP_PALM
: return "AMD PALM";
353 case CHIP_SUMO
: return "AMD SUMO";
354 case CHIP_SUMO2
: return "AMD SUMO2";
355 case CHIP_BARTS
: return "AMD BARTS";
356 case CHIP_TURKS
: return "AMD TURKS";
357 case CHIP_CAICOS
: return "AMD CAICOS";
358 case CHIP_CAYMAN
: return "AMD CAYMAN";
359 case CHIP_ARUBA
: return "AMD ARUBA";
360 default: return "AMD unknown";
364 static const char* r600_get_name(struct pipe_screen
* pscreen
)
366 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
368 return r600_get_family_name(rscreen
->family
);
371 static int r600_get_param(struct pipe_screen
* pscreen
, enum pipe_cap param
)
373 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
374 enum radeon_family family
= rscreen
->family
;
377 /* Supported features (boolean caps). */
378 case PIPE_CAP_NPOT_TEXTURES
:
379 case PIPE_CAP_TWO_SIDED_STENCIL
:
380 case PIPE_CAP_ANISOTROPIC_FILTER
:
381 case PIPE_CAP_POINT_SPRITE
:
382 case PIPE_CAP_OCCLUSION_QUERY
:
383 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
384 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
385 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
386 case PIPE_CAP_TEXTURE_SWIZZLE
:
387 case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE
:
388 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
389 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
390 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
391 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
392 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
393 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
395 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
396 case PIPE_CAP_PRIMITIVE_RESTART
:
397 case PIPE_CAP_CONDITIONAL_RENDER
:
398 case PIPE_CAP_TEXTURE_BARRIER
:
399 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
400 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
401 case PIPE_CAP_TGSI_INSTANCEID
:
402 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
403 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
404 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
405 case PIPE_CAP_USER_INDEX_BUFFERS
:
406 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
407 case PIPE_CAP_COMPUTE
:
408 case PIPE_CAP_START_INSTANCE
:
409 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
412 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
413 return R600_MAP_BUFFER_ALIGNMENT
;
415 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
418 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
421 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
422 return rscreen
->msaa_texture_support
!= MSAA_TEXTURE_SAMPLE_ZERO
;
424 /* Supported except the original R600. */
425 case PIPE_CAP_INDEP_BLEND_ENABLE
:
426 case PIPE_CAP_INDEP_BLEND_FUNC
:
427 /* R600 doesn't support per-MRT blends */
428 return family
== CHIP_R600
? 0 : 1;
430 /* Supported on Evergreen. */
431 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
432 case PIPE_CAP_CUBE_MAP_ARRAY
:
433 return family
>= CHIP_CEDAR
? 1 : 0;
435 /* Unsupported features. */
436 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
437 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
438 case PIPE_CAP_SCALED_RESOLVE
:
439 case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS
:
440 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
441 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
442 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
443 case PIPE_CAP_USER_VERTEX_BUFFERS
:
444 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
448 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
449 return rscreen
->has_streamout
? 4 : 0;
450 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
451 return rscreen
->has_streamout
? 1 : 0;
452 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
453 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
457 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
458 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
459 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
460 if (family
>= CHIP_CEDAR
)
464 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
465 return rscreen
->info
.drm_minor
>= 9 ?
466 (family
>= CHIP_CEDAR
? 16384 : 8192) : 0;
467 case PIPE_CAP_MAX_COMBINED_SAMPLERS
:
470 /* Render targets. */
471 case PIPE_CAP_MAX_RENDER_TARGETS
:
472 /* XXX some r6xx are buggy and can only do 4 */
475 /* Timer queries, present when the clock frequency is non zero. */
476 case PIPE_CAP_TIMER_QUERY
:
477 return rscreen
->info
.r600_clock_crystal_freq
!= 0;
478 case PIPE_CAP_QUERY_TIMESTAMP
:
479 return rscreen
->info
.drm_minor
>= 20 &&
480 rscreen
->info
.r600_clock_crystal_freq
!= 0;
482 case PIPE_CAP_MIN_TEXEL_OFFSET
:
485 case PIPE_CAP_MAX_TEXEL_OFFSET
:
491 static float r600_get_paramf(struct pipe_screen
* pscreen
,
492 enum pipe_capf param
)
494 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
495 enum radeon_family family
= rscreen
->family
;
498 case PIPE_CAPF_MAX_LINE_WIDTH
:
499 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
500 case PIPE_CAPF_MAX_POINT_WIDTH
:
501 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
502 if (family
>= CHIP_CEDAR
)
506 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
508 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
510 case PIPE_CAPF_GUARD_BAND_LEFT
:
511 case PIPE_CAPF_GUARD_BAND_TOP
:
512 case PIPE_CAPF_GUARD_BAND_RIGHT
:
513 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
519 static int r600_get_shader_param(struct pipe_screen
* pscreen
, unsigned shader
, enum pipe_shader_cap param
)
523 case PIPE_SHADER_FRAGMENT
:
524 case PIPE_SHADER_VERTEX
:
525 case PIPE_SHADER_COMPUTE
:
527 case PIPE_SHADER_GEOMETRY
:
528 /* XXX: support and enable geometry programs */
531 /* XXX: support tessellation on Evergreen */
535 /* XXX: all these should be fixed, since r600 surely supports much more! */
537 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
538 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
539 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
540 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
542 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
544 case PIPE_SHADER_CAP_MAX_INPUTS
:
546 case PIPE_SHADER_CAP_MAX_TEMPS
:
547 return 256; /* Max native temporaries. */
548 case PIPE_SHADER_CAP_MAX_ADDRS
:
549 /* XXX Isn't this equal to TEMPS? */
550 return 1; /* Max native address registers */
551 case PIPE_SHADER_CAP_MAX_CONSTS
:
552 return R600_MAX_CONST_BUFFER_SIZE
;
553 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
554 return R600_MAX_USER_CONST_BUFFERS
;
555 case PIPE_SHADER_CAP_MAX_PREDS
:
556 return 0; /* nothing uses this */
557 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
559 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
560 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
561 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
562 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
564 case PIPE_SHADER_CAP_SUBROUTINES
:
566 case PIPE_SHADER_CAP_INTEGERS
:
568 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
570 case PIPE_SHADER_CAP_PREFERRED_IR
:
571 if (shader
== PIPE_SHADER_COMPUTE
) {
572 return PIPE_SHADER_IR_LLVM
;
574 return PIPE_SHADER_IR_TGSI
;
580 static int r600_get_video_param(struct pipe_screen
*screen
,
581 enum pipe_video_profile profile
,
582 enum pipe_video_cap param
)
585 case PIPE_VIDEO_CAP_SUPPORTED
:
586 return vl_profile_supported(screen
, profile
);
587 case PIPE_VIDEO_CAP_NPOT_TEXTURES
:
589 case PIPE_VIDEO_CAP_MAX_WIDTH
:
590 case PIPE_VIDEO_CAP_MAX_HEIGHT
:
591 return vl_video_buffer_max_size(screen
);
592 case PIPE_VIDEO_CAP_PREFERED_FORMAT
:
593 return PIPE_FORMAT_NV12
;
594 case PIPE_VIDEO_CAP_PREFERS_INTERLACED
:
596 case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED
:
598 case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE
:
605 static int r600_get_compute_param(struct pipe_screen
*screen
,
606 enum pipe_compute_cap param
,
609 //TODO: select these params by asic
611 case PIPE_COMPUTE_CAP_IR_TARGET
:
613 strcpy(ret
, "r600--");
615 return 7 * sizeof(char);
617 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
619 uint64_t * grid_dimension
= ret
;
620 grid_dimension
[0] = 3;
622 return 1 * sizeof(uint64_t);
624 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
626 uint64_t * grid_size
= ret
;
627 grid_size
[0] = 65535;
628 grid_size
[1] = 65535;
631 return 3 * sizeof(uint64_t) ;
633 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
635 uint64_t * block_size
= ret
;
640 return 3 * sizeof(uint64_t);
642 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
644 uint64_t * max_threads_per_block
= ret
;
645 *max_threads_per_block
= 256;
647 return sizeof(uint64_t);
649 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
:
651 uint64_t * max_global_size
= ret
;
652 /* XXX: This is what the proprietary driver reports, we
653 * may want to use a different value. */
654 *max_global_size
= 201326592;
656 return sizeof(uint64_t);
658 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
:
660 uint64_t * max_input_size
= ret
;
661 *max_input_size
= 1024;
663 return sizeof(uint64_t);
665 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
:
667 uint64_t * max_local_size
= ret
;
668 /* XXX: This is what the proprietary driver reports, we
669 * may want to use a different value. */
670 *max_local_size
= 32768;
672 return sizeof(uint64_t);
674 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
676 uint64_t max_global_size
;
677 uint64_t * max_mem_alloc_size
= ret
;
678 r600_get_compute_param(screen
,
679 PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
,
681 /* OpenCL requres this value be at least
682 * max(MAX_GLOBAL_SIZE / 4, 128 * 1024 *1024)
683 * I'm really not sure what value to report here, but
684 * MAX_GLOBAL_SIZE / 4 seems resonable.
686 *max_mem_alloc_size
= max_global_size
/ 4;
688 return sizeof(uint64_t);
691 fprintf(stderr
, "unknown PIPE_COMPUTE_CAP %d\n", param
);
696 static void r600_destroy_screen(struct pipe_screen
* pscreen
)
698 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
703 if (rscreen
->global_pool
) {
704 compute_memory_pool_delete(rscreen
->global_pool
);
707 if (rscreen
->fences
.bo
) {
708 struct r600_fence_block
*entry
, *tmp
;
710 LIST_FOR_EACH_ENTRY_SAFE(entry
, tmp
, &rscreen
->fences
.blocks
, head
) {
711 LIST_DEL(&entry
->head
);
715 rscreen
->ws
->buffer_unmap(rscreen
->fences
.bo
->cs_buf
);
716 pipe_resource_reference((struct pipe_resource
**)&rscreen
->fences
.bo
, NULL
);
718 pipe_mutex_destroy(rscreen
->fences
.mutex
);
720 rscreen
->ws
->destroy(rscreen
->ws
);
724 static void r600_fence_reference(struct pipe_screen
*pscreen
,
725 struct pipe_fence_handle
**ptr
,
726 struct pipe_fence_handle
*fence
)
728 struct r600_fence
**oldf
= (struct r600_fence
**)ptr
;
729 struct r600_fence
*newf
= (struct r600_fence
*)fence
;
731 if (pipe_reference(&(*oldf
)->reference
, &newf
->reference
)) {
732 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
733 pipe_mutex_lock(rscreen
->fences
.mutex
);
734 pipe_resource_reference((struct pipe_resource
**)&(*oldf
)->sleep_bo
, NULL
);
735 LIST_ADDTAIL(&(*oldf
)->head
, &rscreen
->fences
.pool
);
736 pipe_mutex_unlock(rscreen
->fences
.mutex
);
742 static boolean
r600_fence_signalled(struct pipe_screen
*pscreen
,
743 struct pipe_fence_handle
*fence
)
745 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
746 struct r600_fence
*rfence
= (struct r600_fence
*)fence
;
748 return rscreen
->fences
.data
[rfence
->index
];
751 static boolean
r600_fence_finish(struct pipe_screen
*pscreen
,
752 struct pipe_fence_handle
*fence
,
755 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
756 struct r600_fence
*rfence
= (struct r600_fence
*)fence
;
757 int64_t start_time
= 0;
760 if (timeout
!= PIPE_TIMEOUT_INFINITE
) {
761 start_time
= os_time_get();
763 /* Convert to microseconds. */
767 while (rscreen
->fences
.data
[rfence
->index
] == 0) {
768 /* Special-case infinite timeout - wait for the dummy BO to become idle */
769 if (timeout
== PIPE_TIMEOUT_INFINITE
) {
770 rscreen
->ws
->buffer_wait(rfence
->sleep_bo
->buf
, RADEON_USAGE_READWRITE
);
774 /* The dummy BO will be busy until the CS including the fence has completed, or
775 * the GPU is reset. Don't bother continuing to spin when the BO is idle. */
776 if (!rscreen
->ws
->buffer_is_busy(rfence
->sleep_bo
->buf
, RADEON_USAGE_READWRITE
))
786 if (timeout
!= PIPE_TIMEOUT_INFINITE
&&
787 os_time_get() - start_time
>= timeout
) {
792 return rscreen
->fences
.data
[rfence
->index
] != 0;
795 static int r600_interpret_tiling(struct r600_screen
*rscreen
, uint32_t tiling_config
)
797 switch ((tiling_config
& 0xe) >> 1) {
799 rscreen
->tiling_info
.num_channels
= 1;
802 rscreen
->tiling_info
.num_channels
= 2;
805 rscreen
->tiling_info
.num_channels
= 4;
808 rscreen
->tiling_info
.num_channels
= 8;
814 switch ((tiling_config
& 0x30) >> 4) {
816 rscreen
->tiling_info
.num_banks
= 4;
819 rscreen
->tiling_info
.num_banks
= 8;
825 switch ((tiling_config
& 0xc0) >> 6) {
827 rscreen
->tiling_info
.group_bytes
= 256;
830 rscreen
->tiling_info
.group_bytes
= 512;
838 static int evergreen_interpret_tiling(struct r600_screen
*rscreen
, uint32_t tiling_config
)
840 switch (tiling_config
& 0xf) {
842 rscreen
->tiling_info
.num_channels
= 1;
845 rscreen
->tiling_info
.num_channels
= 2;
848 rscreen
->tiling_info
.num_channels
= 4;
851 rscreen
->tiling_info
.num_channels
= 8;
857 switch ((tiling_config
& 0xf0) >> 4) {
859 rscreen
->tiling_info
.num_banks
= 4;
862 rscreen
->tiling_info
.num_banks
= 8;
865 rscreen
->tiling_info
.num_banks
= 16;
871 switch ((tiling_config
& 0xf00) >> 8) {
873 rscreen
->tiling_info
.group_bytes
= 256;
876 rscreen
->tiling_info
.group_bytes
= 512;
884 static int r600_init_tiling(struct r600_screen
*rscreen
)
886 uint32_t tiling_config
= rscreen
->info
.r600_tiling_config
;
888 /* set default group bytes, overridden by tiling info ioctl */
889 if (rscreen
->chip_class
<= R700
) {
890 rscreen
->tiling_info
.group_bytes
= 256;
892 rscreen
->tiling_info
.group_bytes
= 512;
898 if (rscreen
->chip_class
<= R700
) {
899 return r600_interpret_tiling(rscreen
, tiling_config
);
901 return evergreen_interpret_tiling(rscreen
, tiling_config
);
905 static unsigned radeon_family_from_device(unsigned device
)
908 #define CHIPSET(pciid, name, family) case pciid: return CHIP_##family;
909 #include "pci_ids/r600_pci_ids.h"
916 static uint64_t r600_get_timestamp(struct pipe_screen
*screen
)
918 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
920 return 1000000 * rscreen
->ws
->query_timestamp(rscreen
->ws
) /
921 rscreen
->info
.r600_clock_crystal_freq
;
924 struct pipe_screen
*r600_screen_create(struct radeon_winsys
*ws
)
926 struct r600_screen
*rscreen
= CALLOC_STRUCT(r600_screen
);
928 if (rscreen
== NULL
) {
933 ws
->query_info(ws
, &rscreen
->info
);
935 rscreen
->family
= radeon_family_from_device(rscreen
->info
.pci_id
);
936 if (rscreen
->family
== CHIP_UNKNOWN
) {
937 fprintf(stderr
, "r600: Unknown chipset 0x%04X\n", rscreen
->info
.pci_id
);
943 if (rscreen
->family
>= CHIP_CAYMAN
) {
944 rscreen
->chip_class
= CAYMAN
;
945 } else if (rscreen
->family
>= CHIP_CEDAR
) {
946 rscreen
->chip_class
= EVERGREEN
;
947 } else if (rscreen
->family
>= CHIP_RV770
) {
948 rscreen
->chip_class
= R700
;
950 rscreen
->chip_class
= R600
;
953 /* Figure out streamout kernel support. */
954 switch (rscreen
->chip_class
) {
956 if (rscreen
->family
< CHIP_RS780
) {
957 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 14;
959 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 23;
963 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 17;
967 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 14;
972 switch (rscreen
->chip_class
) {
975 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 22;
976 rscreen
->msaa_texture_support
= MSAA_TEXTURE_DECOMPRESSED
;
979 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 19;
980 rscreen
->msaa_texture_support
=
981 rscreen
->info
.drm_minor
>= 24 ? MSAA_TEXTURE_COMPRESSED
:
982 MSAA_TEXTURE_DECOMPRESSED
;
985 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 19;
986 /* We should be able to read compressed MSAA textures, but it doesn't work. */
987 rscreen
->msaa_texture_support
= MSAA_TEXTURE_SAMPLE_ZERO
;
991 if (r600_init_tiling(rscreen
)) {
996 rscreen
->screen
.destroy
= r600_destroy_screen
;
997 rscreen
->screen
.get_name
= r600_get_name
;
998 rscreen
->screen
.get_vendor
= r600_get_vendor
;
999 rscreen
->screen
.get_param
= r600_get_param
;
1000 rscreen
->screen
.get_shader_param
= r600_get_shader_param
;
1001 rscreen
->screen
.get_paramf
= r600_get_paramf
;
1002 rscreen
->screen
.get_video_param
= r600_get_video_param
;
1003 rscreen
->screen
.get_compute_param
= r600_get_compute_param
;
1004 rscreen
->screen
.get_timestamp
= r600_get_timestamp
;
1006 if (rscreen
->chip_class
>= EVERGREEN
) {
1007 rscreen
->screen
.is_format_supported
= evergreen_is_format_supported
;
1009 rscreen
->screen
.is_format_supported
= r600_is_format_supported
;
1011 rscreen
->screen
.is_video_format_supported
= vl_video_buffer_is_format_supported
;
1012 rscreen
->screen
.context_create
= r600_create_context
;
1013 rscreen
->screen
.fence_reference
= r600_fence_reference
;
1014 rscreen
->screen
.fence_signalled
= r600_fence_signalled
;
1015 rscreen
->screen
.fence_finish
= r600_fence_finish
;
1016 r600_init_screen_resource_functions(&rscreen
->screen
);
1018 util_format_s3tc_init();
1020 rscreen
->fences
.bo
= NULL
;
1021 rscreen
->fences
.data
= NULL
;
1022 rscreen
->fences
.next_index
= 0;
1023 LIST_INITHEAD(&rscreen
->fences
.pool
);
1024 LIST_INITHEAD(&rscreen
->fences
.blocks
);
1025 pipe_mutex_init(rscreen
->fences
.mutex
);
1027 rscreen
->global_pool
= compute_memory_pool_new(rscreen
);
1029 return &rscreen
->screen
;