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"
26 #include "evergreen_compute.h"
29 #include "sb/sb_public.h"
32 #include "pipe/p_shader_tokens.h"
33 #include "util/u_blitter.h"
34 #include "util/u_debug.h"
35 #include "util/u_format_s3tc.h"
36 #include "util/u_memory.h"
37 #include "util/u_simple_shaders.h"
38 #include "util/u_upload_mgr.h"
39 #include "util/u_math.h"
40 #include "vl/vl_decoder.h"
41 #include "vl/vl_video_buffer.h"
42 #include "radeon/radeon_uvd.h"
43 #include "os/os_time.h"
45 static const struct debug_named_value debug_options
[] = {
47 { "texdepth", DBG_TEX_DEPTH
, "Print texture depth info" },
48 { "compute", DBG_COMPUTE
, "Print compute info" },
49 { "vm", DBG_VM
, "Print virtual addresses when creating resources" },
50 { "trace_cs", DBG_TRACE_CS
, "Trace cs and write rlockup_<csid>.c file with faulty cs" },
53 { "fs", DBG_FS
, "Print fetch shaders" },
54 { "vs", DBG_VS
, "Print vertex shaders" },
55 { "gs", DBG_GS
, "Print geometry shaders" },
56 { "ps", DBG_PS
, "Print pixel shaders" },
57 { "cs", DBG_CS
, "Print compute shaders" },
60 { "nohyperz", DBG_NO_HYPERZ
, "Disable Hyper-Z" },
61 #if defined(R600_USE_LLVM)
62 { "nollvm", DBG_NO_LLVM
, "Disable the LLVM shader compiler" },
64 { "nocpdma", DBG_NO_CP_DMA
, "Disable CP DMA" },
65 { "nodma", DBG_NO_ASYNC_DMA
, "Disable asynchronous DMA" },
66 /* GL uses the word INVALIDATE, gallium uses the word DISCARD */
67 { "noinvalrange", DBG_NO_DISCARD_RANGE
, "Disable handling of INVALIDATE_RANGE map flags" },
70 { "sb", DBG_SB
, "Enable optimization of graphics shaders" },
71 { "sbcl", DBG_SB_CS
, "Enable optimization of compute shaders" },
72 { "sbdry", DBG_SB_DRY_RUN
, "Don't use optimized bytecode (just print the dumps)" },
73 { "sbstat", DBG_SB_STAT
, "Print optimization statistics for shaders" },
74 { "sbdump", DBG_SB_DUMP
, "Print IR dumps after some optimization passes" },
75 { "sbnofallback", DBG_SB_NO_FALLBACK
, "Abort on errors instead of fallback" },
76 { "sbdisasm", DBG_SB_DISASM
, "Use sb disassembler for shader dumps" },
78 DEBUG_NAMED_VALUE_END
/* must be last */
84 static struct r600_fence
*r600_create_fence(struct r600_context
*rctx
)
86 struct r600_screen
*rscreen
= rctx
->screen
;
87 struct r600_fence
*fence
= NULL
;
89 pipe_mutex_lock(rscreen
->fences
.mutex
);
91 if (!rscreen
->fences
.bo
) {
92 /* Create the shared buffer object */
93 rscreen
->fences
.bo
= (struct r600_resource
*)
94 pipe_buffer_create(&rscreen
->screen
, PIPE_BIND_CUSTOM
,
95 PIPE_USAGE_STAGING
, 4096);
96 if (!rscreen
->fences
.bo
) {
97 R600_ERR("r600: failed to create bo for fence objects\n");
100 rscreen
->fences
.data
= r600_buffer_mmap_sync_with_rings(rctx
, rscreen
->fences
.bo
, PIPE_TRANSFER_READ_WRITE
);
103 if (!LIST_IS_EMPTY(&rscreen
->fences
.pool
)) {
104 struct r600_fence
*entry
;
106 /* Try to find a freed fence that has been signalled */
107 LIST_FOR_EACH_ENTRY(entry
, &rscreen
->fences
.pool
, head
) {
108 if (rscreen
->fences
.data
[entry
->index
] != 0) {
109 LIST_DELINIT(&entry
->head
);
117 /* Allocate a new fence */
118 struct r600_fence_block
*block
;
121 if ((rscreen
->fences
.next_index
+ 1) >= 1024) {
122 R600_ERR("r600: too many concurrent fences\n");
126 index
= rscreen
->fences
.next_index
++;
128 if (!(index
% FENCE_BLOCK_SIZE
)) {
129 /* Allocate a new block */
130 block
= CALLOC_STRUCT(r600_fence_block
);
134 LIST_ADD(&block
->head
, &rscreen
->fences
.blocks
);
136 block
= LIST_ENTRY(struct r600_fence_block
, rscreen
->fences
.blocks
.next
, head
);
139 fence
= &block
->fences
[index
% FENCE_BLOCK_SIZE
];
140 fence
->index
= index
;
143 pipe_reference_init(&fence
->reference
, 1);
145 rscreen
->fences
.data
[fence
->index
] = 0;
146 r600_context_emit_fence(rctx
, rscreen
->fences
.bo
, fence
->index
, 1);
148 /* Create a dummy BO so that fence_finish without a timeout can sleep waiting for completion */
149 fence
->sleep_bo
= (struct r600_resource
*)
150 pipe_buffer_create(&rctx
->screen
->screen
, PIPE_BIND_CUSTOM
,
151 PIPE_USAGE_STAGING
, 1);
152 /* Add the fence as a dummy relocation. */
153 r600_context_bo_reloc(rctx
, &rctx
->rings
.gfx
, fence
->sleep_bo
, RADEON_USAGE_READWRITE
);
156 pipe_mutex_unlock(rscreen
->fences
.mutex
);
160 static void r600_flush(struct pipe_context
*ctx
, unsigned flags
)
162 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
163 struct pipe_query
*render_cond
= NULL
;
164 unsigned render_cond_mode
= 0;
165 boolean render_cond_cond
= FALSE
;
167 rctx
->rings
.gfx
.flushing
= true;
168 /* Disable render condition. */
169 if (rctx
->current_render_cond
) {
170 render_cond
= rctx
->current_render_cond
;
171 render_cond_cond
= rctx
->current_render_cond_cond
;
172 render_cond_mode
= rctx
->current_render_cond_mode
;
173 ctx
->render_condition(ctx
, NULL
, FALSE
, 0);
176 r600_context_flush(rctx
, flags
);
177 rctx
->rings
.gfx
.flushing
= false;
178 r600_begin_new_cs(rctx
);
180 /* Re-enable render condition. */
182 ctx
->render_condition(ctx
, render_cond
, render_cond_cond
, render_cond_mode
);
186 static void r600_flush_from_st(struct pipe_context
*ctx
,
187 struct pipe_fence_handle
**fence
,
190 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
191 struct r600_fence
**rfence
= (struct r600_fence
**)fence
;
194 fflags
= flags
& PIPE_FLUSH_END_OF_FRAME
? RADEON_FLUSH_END_OF_FRAME
: 0;
196 *rfence
= r600_create_fence(rctx
);
198 /* flush gfx & dma ring, order does not matter as only one can be live */
199 if (rctx
->rings
.dma
.cs
) {
200 rctx
->rings
.dma
.flush(rctx
, fflags
);
202 rctx
->rings
.gfx
.flush(rctx
, fflags
);
205 static void r600_flush_gfx_ring(void *ctx
, unsigned flags
)
207 r600_flush((struct pipe_context
*)ctx
, flags
);
210 static void r600_flush_dma_ring(void *ctx
, unsigned flags
)
212 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
213 struct radeon_winsys_cs
*cs
= rctx
->rings
.dma
.cs
;
214 unsigned padding_dw
, i
;
220 /* Pad the DMA CS to a multiple of 8 dwords. */
221 padding_dw
= 8 - cs
->cdw
% 8;
222 if (padding_dw
< 8) {
223 for (i
= 0; i
< padding_dw
; i
++) {
224 cs
->buf
[cs
->cdw
++] = DMA_PACKET(DMA_PACKET_NOP
, 0, 0, 0);
228 rctx
->rings
.dma
.flushing
= true;
229 rctx
->ws
->cs_flush(cs
, flags
, 0);
230 rctx
->rings
.dma
.flushing
= false;
233 boolean
r600_rings_is_buffer_referenced(struct r600_context
*ctx
,
234 struct radeon_winsys_cs_handle
*buf
,
235 enum radeon_bo_usage usage
)
237 if (ctx
->ws
->cs_is_buffer_referenced(ctx
->rings
.gfx
.cs
, buf
, usage
)) {
240 if (ctx
->rings
.dma
.cs
) {
241 if (ctx
->ws
->cs_is_buffer_referenced(ctx
->rings
.dma
.cs
, buf
, usage
)) {
248 void *r600_buffer_mmap_sync_with_rings(struct r600_context
*ctx
,
249 struct r600_resource
*resource
,
252 enum radeon_bo_usage rusage
= RADEON_USAGE_READWRITE
;
254 bool sync_flush
= TRUE
;
256 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) {
257 return ctx
->ws
->buffer_map(resource
->cs_buf
, NULL
, usage
);
260 if (!(usage
& PIPE_TRANSFER_WRITE
)) {
261 /* have to wait for pending read */
262 rusage
= RADEON_USAGE_WRITE
;
264 if (usage
& PIPE_TRANSFER_DONTBLOCK
) {
265 flags
|= RADEON_FLUSH_ASYNC
;
268 if (ctx
->ws
->cs_is_buffer_referenced(ctx
->rings
.gfx
.cs
, resource
->cs_buf
, rusage
) && ctx
->rings
.gfx
.cs
->cdw
) {
269 ctx
->rings
.gfx
.flush(ctx
, flags
);
270 if (usage
& PIPE_TRANSFER_DONTBLOCK
) {
274 if (ctx
->rings
.dma
.cs
) {
275 if (ctx
->ws
->cs_is_buffer_referenced(ctx
->rings
.dma
.cs
, resource
->cs_buf
, rusage
) && ctx
->rings
.dma
.cs
->cdw
) {
276 ctx
->rings
.dma
.flush(ctx
, flags
);
277 if (usage
& PIPE_TRANSFER_DONTBLOCK
) {
283 if (usage
& PIPE_TRANSFER_DONTBLOCK
) {
284 if (ctx
->ws
->buffer_is_busy(resource
->buf
, rusage
)) {
289 /* Try to avoid busy-waiting in radeon_bo_wait. */
290 ctx
->ws
->cs_sync_flush(ctx
->rings
.gfx
.cs
);
291 if (ctx
->rings
.dma
.cs
) {
292 ctx
->ws
->cs_sync_flush(ctx
->rings
.dma
.cs
);
295 ctx
->ws
->buffer_wait(resource
->buf
, rusage
);
297 /* at this point everything is synchronized */
298 return ctx
->ws
->buffer_map(resource
->cs_buf
, NULL
, usage
| PIPE_TRANSFER_UNSYNCHRONIZED
);
301 static void r600_flush_from_winsys(void *ctx
, unsigned flags
)
303 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
305 rctx
->rings
.gfx
.flush(rctx
, flags
);
308 static void r600_flush_dma_from_winsys(void *ctx
, unsigned flags
)
310 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
312 rctx
->rings
.dma
.flush(rctx
, flags
);
315 static void r600_destroy_context(struct pipe_context
*context
)
317 struct r600_context
*rctx
= (struct r600_context
*)context
;
319 r600_isa_destroy(rctx
->isa
);
321 r600_sb_context_destroy(rctx
->sb_context
);
323 pipe_resource_reference((struct pipe_resource
**)&rctx
->dummy_cmask
, NULL
);
324 pipe_resource_reference((struct pipe_resource
**)&rctx
->dummy_fmask
, NULL
);
326 if (rctx
->dummy_pixel_shader
) {
327 rctx
->context
.delete_fs_state(&rctx
->context
, rctx
->dummy_pixel_shader
);
329 if (rctx
->custom_dsa_flush
) {
330 rctx
->context
.delete_depth_stencil_alpha_state(&rctx
->context
, rctx
->custom_dsa_flush
);
332 if (rctx
->custom_blend_resolve
) {
333 rctx
->context
.delete_blend_state(&rctx
->context
, rctx
->custom_blend_resolve
);
335 if (rctx
->custom_blend_decompress
) {
336 rctx
->context
.delete_blend_state(&rctx
->context
, rctx
->custom_blend_decompress
);
338 util_unreference_framebuffer_state(&rctx
->framebuffer
.state
);
341 util_blitter_destroy(rctx
->blitter
);
343 if (rctx
->uploader
) {
344 u_upload_destroy(rctx
->uploader
);
346 if (rctx
->allocator_so_filled_size
) {
347 u_suballocator_destroy(rctx
->allocator_so_filled_size
);
349 if (rctx
->allocator_fetch_shader
) {
350 u_suballocator_destroy(rctx
->allocator_fetch_shader
);
352 util_slab_destroy(&rctx
->pool_transfers
);
354 r600_release_command_buffer(&rctx
->start_cs_cmd
);
356 if (rctx
->rings
.gfx
.cs
) {
357 rctx
->ws
->cs_destroy(rctx
->rings
.gfx
.cs
);
359 if (rctx
->rings
.dma
.cs
) {
360 rctx
->ws
->cs_destroy(rctx
->rings
.dma
.cs
);
366 static struct pipe_context
*r600_create_context(struct pipe_screen
*screen
, void *priv
)
368 struct r600_context
*rctx
= CALLOC_STRUCT(r600_context
);
369 struct r600_screen
* rscreen
= (struct r600_screen
*)screen
;
374 util_slab_create(&rctx
->pool_transfers
,
375 sizeof(struct r600_transfer
), 64,
376 UTIL_SLAB_SINGLETHREADED
);
378 rctx
->context
.screen
= screen
;
379 rctx
->context
.priv
= priv
;
380 rctx
->context
.destroy
= r600_destroy_context
;
381 rctx
->context
.flush
= r600_flush_from_st
;
383 /* Easy accessing of screen/winsys. */
384 rctx
->screen
= rscreen
;
385 rctx
->ws
= rscreen
->ws
;
386 rctx
->family
= rscreen
->family
;
387 rctx
->chip_class
= rscreen
->chip_class
;
388 rctx
->keep_tiling_flags
= rscreen
->info
.drm_minor
>= 12;
390 LIST_INITHEAD(&rctx
->active_nontimer_queries
);
392 r600_init_blit_functions(rctx
);
393 r600_init_query_functions(rctx
);
394 r600_init_context_resource_functions(rctx
);
395 r600_init_surface_functions(rctx
);
397 if (rscreen
->info
.has_uvd
) {
398 rctx
->context
.create_video_decoder
= r600_uvd_create_decoder
;
399 rctx
->context
.create_video_buffer
= r600_video_buffer_create
;
401 rctx
->context
.create_video_decoder
= vl_create_decoder
;
402 rctx
->context
.create_video_buffer
= vl_video_buffer_create
;
405 r600_init_common_state_functions(rctx
);
407 switch (rctx
->chip_class
) {
410 r600_init_state_functions(rctx
);
411 r600_init_atom_start_cs(rctx
);
413 rctx
->custom_dsa_flush
= r600_create_db_flush_dsa(rctx
);
414 rctx
->custom_blend_resolve
= rctx
->chip_class
== R700
? r700_create_resolve_blend(rctx
)
415 : r600_create_resolve_blend(rctx
);
416 rctx
->custom_blend_decompress
= r600_create_decompress_blend(rctx
);
417 rctx
->has_vertex_cache
= !(rctx
->family
== CHIP_RV610
||
418 rctx
->family
== CHIP_RV620
||
419 rctx
->family
== CHIP_RS780
||
420 rctx
->family
== CHIP_RS880
||
421 rctx
->family
== CHIP_RV710
);
425 evergreen_init_state_functions(rctx
);
426 evergreen_init_atom_start_cs(rctx
);
427 evergreen_init_atom_start_compute_cs(rctx
);
429 rctx
->custom_dsa_flush
= evergreen_create_db_flush_dsa(rctx
);
430 rctx
->custom_blend_resolve
= evergreen_create_resolve_blend(rctx
);
431 rctx
->custom_blend_decompress
= evergreen_create_decompress_blend(rctx
);
432 rctx
->has_vertex_cache
= !(rctx
->family
== CHIP_CEDAR
||
433 rctx
->family
== CHIP_PALM
||
434 rctx
->family
== CHIP_SUMO
||
435 rctx
->family
== CHIP_SUMO2
||
436 rctx
->family
== CHIP_CAICOS
||
437 rctx
->family
== CHIP_CAYMAN
||
438 rctx
->family
== CHIP_ARUBA
);
441 R600_ERR("Unsupported chip class %d.\n", rctx
->chip_class
);
445 if (rscreen
->trace_bo
) {
446 rctx
->rings
.gfx
.cs
= rctx
->ws
->cs_create(rctx
->ws
, RING_GFX
, rscreen
->trace_bo
->cs_buf
);
448 rctx
->rings
.gfx
.cs
= rctx
->ws
->cs_create(rctx
->ws
, RING_GFX
, NULL
);
450 rctx
->rings
.gfx
.flush
= r600_flush_gfx_ring
;
451 rctx
->ws
->cs_set_flush_callback(rctx
->rings
.gfx
.cs
, r600_flush_from_winsys
, rctx
);
452 rctx
->rings
.gfx
.flushing
= false;
454 rctx
->rings
.dma
.cs
= NULL
;
455 if (rscreen
->info
.r600_has_dma
&& !(rscreen
->debug_flags
& DBG_NO_ASYNC_DMA
)) {
456 rctx
->rings
.dma
.cs
= rctx
->ws
->cs_create(rctx
->ws
, RING_DMA
, NULL
);
457 rctx
->rings
.dma
.flush
= r600_flush_dma_ring
;
458 rctx
->ws
->cs_set_flush_callback(rctx
->rings
.dma
.cs
, r600_flush_dma_from_winsys
, rctx
);
459 rctx
->rings
.dma
.flushing
= false;
462 rctx
->uploader
= u_upload_create(&rctx
->context
, 1024 * 1024, 256,
463 PIPE_BIND_INDEX_BUFFER
|
464 PIPE_BIND_CONSTANT_BUFFER
);
468 rctx
->allocator_fetch_shader
= u_suballocator_create(&rctx
->context
, 64 * 1024, 256,
469 0, PIPE_USAGE_STATIC
, FALSE
);
470 if (!rctx
->allocator_fetch_shader
)
473 rctx
->allocator_so_filled_size
= u_suballocator_create(&rctx
->context
, 4096, 4,
474 0, PIPE_USAGE_STATIC
, TRUE
);
475 if (!rctx
->allocator_so_filled_size
)
478 rctx
->isa
= calloc(1, sizeof(struct r600_isa
));
479 if (!rctx
->isa
|| r600_isa_init(rctx
, rctx
->isa
))
482 rctx
->blitter
= util_blitter_create(&rctx
->context
);
483 if (rctx
->blitter
== NULL
)
485 util_blitter_set_texture_multisample(rctx
->blitter
, rscreen
->has_msaa
);
486 rctx
->blitter
->draw_rectangle
= r600_draw_rectangle
;
488 r600_begin_new_cs(rctx
);
489 r600_get_backend_mask(rctx
); /* this emits commands and must be last */
491 rctx
->dummy_pixel_shader
=
492 util_make_fragment_cloneinput_shader(&rctx
->context
, 0,
493 TGSI_SEMANTIC_GENERIC
,
494 TGSI_INTERPOLATE_CONSTANT
);
495 rctx
->context
.bind_fs_state(&rctx
->context
, rctx
->dummy_pixel_shader
);
497 return &rctx
->context
;
500 r600_destroy_context(&rctx
->context
);
507 static const char* r600_get_vendor(struct pipe_screen
* pscreen
)
512 static const char *r600_get_family_name(enum radeon_family family
)
515 case CHIP_R600
: return "AMD R600";
516 case CHIP_RV610
: return "AMD RV610";
517 case CHIP_RV630
: return "AMD RV630";
518 case CHIP_RV670
: return "AMD RV670";
519 case CHIP_RV620
: return "AMD RV620";
520 case CHIP_RV635
: return "AMD RV635";
521 case CHIP_RS780
: return "AMD RS780";
522 case CHIP_RS880
: return "AMD RS880";
523 case CHIP_RV770
: return "AMD RV770";
524 case CHIP_RV730
: return "AMD RV730";
525 case CHIP_RV710
: return "AMD RV710";
526 case CHIP_RV740
: return "AMD RV740";
527 case CHIP_CEDAR
: return "AMD CEDAR";
528 case CHIP_REDWOOD
: return "AMD REDWOOD";
529 case CHIP_JUNIPER
: return "AMD JUNIPER";
530 case CHIP_CYPRESS
: return "AMD CYPRESS";
531 case CHIP_HEMLOCK
: return "AMD HEMLOCK";
532 case CHIP_PALM
: return "AMD PALM";
533 case CHIP_SUMO
: return "AMD SUMO";
534 case CHIP_SUMO2
: return "AMD SUMO2";
535 case CHIP_BARTS
: return "AMD BARTS";
536 case CHIP_TURKS
: return "AMD TURKS";
537 case CHIP_CAICOS
: return "AMD CAICOS";
538 case CHIP_CAYMAN
: return "AMD CAYMAN";
539 case CHIP_ARUBA
: return "AMD ARUBA";
540 default: return "AMD unknown";
544 static const char* r600_get_name(struct pipe_screen
* pscreen
)
546 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
548 return r600_get_family_name(rscreen
->family
);
551 static int r600_get_param(struct pipe_screen
* pscreen
, enum pipe_cap param
)
553 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
554 enum radeon_family family
= rscreen
->family
;
557 /* Supported features (boolean caps). */
558 case PIPE_CAP_NPOT_TEXTURES
:
559 case PIPE_CAP_TWO_SIDED_STENCIL
:
560 case PIPE_CAP_ANISOTROPIC_FILTER
:
561 case PIPE_CAP_POINT_SPRITE
:
562 case PIPE_CAP_OCCLUSION_QUERY
:
563 case PIPE_CAP_TEXTURE_SHADOW_MAP
:
564 case PIPE_CAP_TEXTURE_MIRROR_CLAMP
:
565 case PIPE_CAP_BLEND_EQUATION_SEPARATE
:
566 case PIPE_CAP_TEXTURE_SWIZZLE
:
567 case PIPE_CAP_DEPTH_CLIP_DISABLE
:
568 case PIPE_CAP_SHADER_STENCIL_EXPORT
:
569 case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR
:
570 case PIPE_CAP_MIXED_COLORBUFFER_FORMATS
:
571 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT
:
572 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER
:
574 case PIPE_CAP_SEAMLESS_CUBE_MAP
:
575 case PIPE_CAP_PRIMITIVE_RESTART
:
576 case PIPE_CAP_CONDITIONAL_RENDER
:
577 case PIPE_CAP_TEXTURE_BARRIER
:
578 case PIPE_CAP_VERTEX_COLOR_UNCLAMPED
:
579 case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION
:
580 case PIPE_CAP_TGSI_INSTANCEID
:
581 case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY
:
582 case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY
:
583 case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY
:
584 case PIPE_CAP_USER_INDEX_BUFFERS
:
585 case PIPE_CAP_USER_CONSTANT_BUFFERS
:
586 case PIPE_CAP_COMPUTE
:
587 case PIPE_CAP_START_INSTANCE
:
588 case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS
:
589 case PIPE_CAP_TEXTURE_BUFFER_OBJECTS
:
590 case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER
:
591 case PIPE_CAP_QUERY_PIPELINE_STATISTICS
:
592 case PIPE_CAP_TEXTURE_MULTISAMPLE
:
595 case PIPE_CAP_TGSI_TEXCOORD
:
598 case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE
:
599 return MIN2(rscreen
->info
.vram_size
, 0xFFFFFFFF);
601 case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT
:
602 return R600_MAP_BUFFER_ALIGNMENT
;
604 case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT
:
607 case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT
:
610 case PIPE_CAP_GLSL_FEATURE_LEVEL
:
613 /* Supported except the original R600. */
614 case PIPE_CAP_INDEP_BLEND_ENABLE
:
615 case PIPE_CAP_INDEP_BLEND_FUNC
:
616 /* R600 doesn't support per-MRT blends */
617 return family
== CHIP_R600
? 0 : 1;
619 /* Supported on Evergreen. */
620 case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE
:
621 case PIPE_CAP_CUBE_MAP_ARRAY
:
622 return family
>= CHIP_CEDAR
? 1 : 0;
624 /* Unsupported features. */
625 case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT
:
626 case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER
:
627 case PIPE_CAP_SCALED_RESOLVE
:
628 case PIPE_CAP_TGSI_CAN_COMPACT_VARYINGS
:
629 case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS
:
630 case PIPE_CAP_FRAGMENT_COLOR_CLAMPED
:
631 case PIPE_CAP_VERTEX_COLOR_CLAMPED
:
632 case PIPE_CAP_USER_VERTEX_BUFFERS
:
636 case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS
:
637 return rscreen
->has_streamout
? 4 : 0;
638 case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME
:
639 return rscreen
->has_streamout
? 1 : 0;
640 case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS
:
641 case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS
:
645 case PIPE_CAP_MAX_TEXTURE_2D_LEVELS
:
646 case PIPE_CAP_MAX_TEXTURE_3D_LEVELS
:
647 case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS
:
648 if (family
>= CHIP_CEDAR
)
652 case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS
:
653 return rscreen
->info
.drm_minor
>= 9 ?
654 (family
>= CHIP_CEDAR
? 16384 : 8192) : 0;
655 case PIPE_CAP_MAX_COMBINED_SAMPLERS
:
658 /* Render targets. */
659 case PIPE_CAP_MAX_RENDER_TARGETS
:
660 /* XXX some r6xx are buggy and can only do 4 */
663 /* Timer queries, present when the clock frequency is non zero. */
664 case PIPE_CAP_QUERY_TIME_ELAPSED
:
665 return rscreen
->info
.r600_clock_crystal_freq
!= 0;
666 case PIPE_CAP_QUERY_TIMESTAMP
:
667 return rscreen
->info
.drm_minor
>= 20 &&
668 rscreen
->info
.r600_clock_crystal_freq
!= 0;
670 case PIPE_CAP_MIN_TEXEL_OFFSET
:
673 case PIPE_CAP_MAX_TEXEL_OFFSET
:
676 case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK
:
677 return PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600
;
682 static float r600_get_paramf(struct pipe_screen
* pscreen
,
683 enum pipe_capf param
)
685 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
686 enum radeon_family family
= rscreen
->family
;
689 case PIPE_CAPF_MAX_LINE_WIDTH
:
690 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
691 case PIPE_CAPF_MAX_POINT_WIDTH
:
692 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
693 if (family
>= CHIP_CEDAR
)
697 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
699 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
701 case PIPE_CAPF_GUARD_BAND_LEFT
:
702 case PIPE_CAPF_GUARD_BAND_TOP
:
703 case PIPE_CAPF_GUARD_BAND_RIGHT
:
704 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
710 static int r600_get_shader_param(struct pipe_screen
* pscreen
, unsigned shader
, enum pipe_shader_cap param
)
714 case PIPE_SHADER_FRAGMENT
:
715 case PIPE_SHADER_VERTEX
:
716 case PIPE_SHADER_COMPUTE
:
718 case PIPE_SHADER_GEOMETRY
:
719 /* XXX: support and enable geometry programs */
722 /* XXX: support tessellation on Evergreen */
727 case PIPE_SHADER_CAP_MAX_INSTRUCTIONS
:
728 case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS
:
729 case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS
:
730 case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS
:
732 case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH
:
734 case PIPE_SHADER_CAP_MAX_INPUTS
:
736 case PIPE_SHADER_CAP_MAX_TEMPS
:
737 return 256; /* Max native temporaries. */
738 case PIPE_SHADER_CAP_MAX_ADDRS
:
739 /* XXX Isn't this equal to TEMPS? */
740 return 1; /* Max native address registers */
741 case PIPE_SHADER_CAP_MAX_CONSTS
:
742 return R600_MAX_CONST_BUFFER_SIZE
;
743 case PIPE_SHADER_CAP_MAX_CONST_BUFFERS
:
744 return R600_MAX_USER_CONST_BUFFERS
;
745 case PIPE_SHADER_CAP_MAX_PREDS
:
746 return 0; /* nothing uses this */
747 case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED
:
749 case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED
:
751 case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR
:
752 case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR
:
753 case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR
:
754 case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR
:
756 case PIPE_SHADER_CAP_SUBROUTINES
:
758 case PIPE_SHADER_CAP_INTEGERS
:
760 case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS
:
762 case PIPE_SHADER_CAP_PREFERRED_IR
:
763 if (shader
== PIPE_SHADER_COMPUTE
) {
764 return PIPE_SHADER_IR_LLVM
;
766 return PIPE_SHADER_IR_TGSI
;
772 static int r600_get_video_param(struct pipe_screen
*screen
,
773 enum pipe_video_profile profile
,
774 enum pipe_video_cap param
)
777 case PIPE_VIDEO_CAP_SUPPORTED
:
778 return vl_profile_supported(screen
, profile
);
779 case PIPE_VIDEO_CAP_NPOT_TEXTURES
:
781 case PIPE_VIDEO_CAP_MAX_WIDTH
:
782 case PIPE_VIDEO_CAP_MAX_HEIGHT
:
783 return vl_video_buffer_max_size(screen
);
784 case PIPE_VIDEO_CAP_PREFERED_FORMAT
:
785 return PIPE_FORMAT_NV12
;
786 case PIPE_VIDEO_CAP_PREFERS_INTERLACED
:
788 case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED
:
790 case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE
:
797 const char * r600_llvm_gpu_string(enum radeon_family family
)
799 const char * gpu_family
;
812 gpu_family
= "rs880";
815 gpu_family
= "rv710";
818 gpu_family
= "rv730";
822 gpu_family
= "rv770";
826 gpu_family
= "cedar";
833 gpu_family
= "redwood";
836 gpu_family
= "juniper";
840 gpu_family
= "cypress";
843 gpu_family
= "barts";
846 gpu_family
= "turks";
849 gpu_family
= "caicos";
853 gpu_family
= "cayman";
857 fprintf(stderr
, "Chip not supported by r600 llvm "
858 "backend, please file a bug at " PACKAGE_BUGREPORT
"\n");
865 static int r600_get_compute_param(struct pipe_screen
*screen
,
866 enum pipe_compute_cap param
,
869 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
870 //TODO: select these params by asic
872 case PIPE_COMPUTE_CAP_IR_TARGET
: {
873 const char *gpu
= r600_llvm_gpu_string(rscreen
->family
);
875 sprintf(ret
, "%s-r600--", gpu
);
877 return (8 + strlen(gpu
)) * sizeof(char);
879 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
881 uint64_t * grid_dimension
= ret
;
882 grid_dimension
[0] = 3;
884 return 1 * sizeof(uint64_t);
886 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
888 uint64_t * grid_size
= ret
;
889 grid_size
[0] = 65535;
890 grid_size
[1] = 65535;
893 return 3 * sizeof(uint64_t) ;
895 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
897 uint64_t * block_size
= ret
;
902 return 3 * sizeof(uint64_t);
904 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
906 uint64_t * max_threads_per_block
= ret
;
907 *max_threads_per_block
= 256;
909 return sizeof(uint64_t);
911 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
:
913 uint64_t * max_global_size
= ret
;
914 /* XXX: This is what the proprietary driver reports, we
915 * may want to use a different value. */
916 *max_global_size
= 201326592;
918 return sizeof(uint64_t);
920 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
:
922 uint64_t * max_input_size
= ret
;
923 *max_input_size
= 1024;
925 return sizeof(uint64_t);
927 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
:
929 uint64_t * max_local_size
= ret
;
930 /* XXX: This is what the proprietary driver reports, we
931 * may want to use a different value. */
932 *max_local_size
= 32768;
934 return sizeof(uint64_t);
936 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
938 uint64_t max_global_size
;
939 uint64_t * max_mem_alloc_size
= ret
;
940 r600_get_compute_param(screen
,
941 PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
,
943 /* OpenCL requres this value be at least
944 * max(MAX_GLOBAL_SIZE / 4, 128 * 1024 *1024)
945 * I'm really not sure what value to report here, but
946 * MAX_GLOBAL_SIZE / 4 seems resonable.
948 *max_mem_alloc_size
= max_global_size
/ 4;
950 return sizeof(uint64_t);
953 fprintf(stderr
, "unknown PIPE_COMPUTE_CAP %d\n", param
);
958 static void r600_destroy_screen(struct pipe_screen
* pscreen
)
960 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
965 pipe_mutex_destroy(rscreen
->aux_context_lock
);
966 rscreen
->aux_context
->destroy(rscreen
->aux_context
);
968 if (rscreen
->global_pool
) {
969 compute_memory_pool_delete(rscreen
->global_pool
);
972 if (rscreen
->fences
.bo
) {
973 struct r600_fence_block
*entry
, *tmp
;
975 LIST_FOR_EACH_ENTRY_SAFE(entry
, tmp
, &rscreen
->fences
.blocks
, head
) {
976 LIST_DEL(&entry
->head
);
980 rscreen
->ws
->buffer_unmap(rscreen
->fences
.bo
->cs_buf
);
981 pipe_resource_reference((struct pipe_resource
**)&rscreen
->fences
.bo
, NULL
);
983 if (rscreen
->trace_bo
) {
984 rscreen
->ws
->buffer_unmap(rscreen
->trace_bo
->cs_buf
);
985 pipe_resource_reference((struct pipe_resource
**)&rscreen
->trace_bo
, NULL
);
987 pipe_mutex_destroy(rscreen
->fences
.mutex
);
989 rscreen
->ws
->destroy(rscreen
->ws
);
993 static void r600_fence_reference(struct pipe_screen
*pscreen
,
994 struct pipe_fence_handle
**ptr
,
995 struct pipe_fence_handle
*fence
)
997 struct r600_fence
**oldf
= (struct r600_fence
**)ptr
;
998 struct r600_fence
*newf
= (struct r600_fence
*)fence
;
1000 if (pipe_reference(&(*oldf
)->reference
, &newf
->reference
)) {
1001 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
1002 pipe_mutex_lock(rscreen
->fences
.mutex
);
1003 pipe_resource_reference((struct pipe_resource
**)&(*oldf
)->sleep_bo
, NULL
);
1004 LIST_ADDTAIL(&(*oldf
)->head
, &rscreen
->fences
.pool
);
1005 pipe_mutex_unlock(rscreen
->fences
.mutex
);
1011 static boolean
r600_fence_signalled(struct pipe_screen
*pscreen
,
1012 struct pipe_fence_handle
*fence
)
1014 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
1015 struct r600_fence
*rfence
= (struct r600_fence
*)fence
;
1017 return rscreen
->fences
.data
[rfence
->index
] != 0;
1020 static boolean
r600_fence_finish(struct pipe_screen
*pscreen
,
1021 struct pipe_fence_handle
*fence
,
1024 struct r600_screen
*rscreen
= (struct r600_screen
*)pscreen
;
1025 struct r600_fence
*rfence
= (struct r600_fence
*)fence
;
1026 int64_t start_time
= 0;
1029 if (timeout
!= PIPE_TIMEOUT_INFINITE
) {
1030 start_time
= os_time_get();
1032 /* Convert to microseconds. */
1036 while (rscreen
->fences
.data
[rfence
->index
] == 0) {
1037 /* Special-case infinite timeout - wait for the dummy BO to become idle */
1038 if (timeout
== PIPE_TIMEOUT_INFINITE
) {
1039 rscreen
->ws
->buffer_wait(rfence
->sleep_bo
->buf
, RADEON_USAGE_READWRITE
);
1043 /* The dummy BO will be busy until the CS including the fence has completed, or
1044 * the GPU is reset. Don't bother continuing to spin when the BO is idle. */
1045 if (!rscreen
->ws
->buffer_is_busy(rfence
->sleep_bo
->buf
, RADEON_USAGE_READWRITE
))
1055 if (timeout
!= PIPE_TIMEOUT_INFINITE
&&
1056 os_time_get() - start_time
>= timeout
) {
1061 return rscreen
->fences
.data
[rfence
->index
] != 0;
1064 static int r600_interpret_tiling(struct r600_screen
*rscreen
, uint32_t tiling_config
)
1066 switch ((tiling_config
& 0xe) >> 1) {
1068 rscreen
->tiling_info
.num_channels
= 1;
1071 rscreen
->tiling_info
.num_channels
= 2;
1074 rscreen
->tiling_info
.num_channels
= 4;
1077 rscreen
->tiling_info
.num_channels
= 8;
1083 switch ((tiling_config
& 0x30) >> 4) {
1085 rscreen
->tiling_info
.num_banks
= 4;
1088 rscreen
->tiling_info
.num_banks
= 8;
1094 switch ((tiling_config
& 0xc0) >> 6) {
1096 rscreen
->tiling_info
.group_bytes
= 256;
1099 rscreen
->tiling_info
.group_bytes
= 512;
1107 static int evergreen_interpret_tiling(struct r600_screen
*rscreen
, uint32_t tiling_config
)
1109 switch (tiling_config
& 0xf) {
1111 rscreen
->tiling_info
.num_channels
= 1;
1114 rscreen
->tiling_info
.num_channels
= 2;
1117 rscreen
->tiling_info
.num_channels
= 4;
1120 rscreen
->tiling_info
.num_channels
= 8;
1126 switch ((tiling_config
& 0xf0) >> 4) {
1128 rscreen
->tiling_info
.num_banks
= 4;
1131 rscreen
->tiling_info
.num_banks
= 8;
1134 rscreen
->tiling_info
.num_banks
= 16;
1140 switch ((tiling_config
& 0xf00) >> 8) {
1142 rscreen
->tiling_info
.group_bytes
= 256;
1145 rscreen
->tiling_info
.group_bytes
= 512;
1153 static int r600_init_tiling(struct r600_screen
*rscreen
)
1155 uint32_t tiling_config
= rscreen
->info
.r600_tiling_config
;
1157 /* set default group bytes, overridden by tiling info ioctl */
1158 if (rscreen
->chip_class
<= R700
) {
1159 rscreen
->tiling_info
.group_bytes
= 256;
1161 rscreen
->tiling_info
.group_bytes
= 512;
1167 if (rscreen
->chip_class
<= R700
) {
1168 return r600_interpret_tiling(rscreen
, tiling_config
);
1170 return evergreen_interpret_tiling(rscreen
, tiling_config
);
1174 static uint64_t r600_get_timestamp(struct pipe_screen
*screen
)
1176 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
1178 return 1000000 * rscreen
->ws
->query_value(rscreen
->ws
, RADEON_TIMESTAMP
) /
1179 rscreen
->info
.r600_clock_crystal_freq
;
1182 static int r600_get_driver_query_info(struct pipe_screen
*screen
,
1184 struct pipe_driver_query_info
*info
)
1186 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
1187 struct pipe_driver_query_info list
[] = {
1188 {"draw-calls", R600_QUERY_DRAW_CALLS
, 0},
1189 {"requested-VRAM", R600_QUERY_REQUESTED_VRAM
, rscreen
->info
.vram_size
, TRUE
},
1190 {"requested-GTT", R600_QUERY_REQUESTED_GTT
, rscreen
->info
.gart_size
, TRUE
},
1191 {"buffer-wait-time", R600_QUERY_BUFFER_WAIT_TIME
, 0, FALSE
}
1195 return Elements(list
);
1197 if (index
>= Elements(list
))
1200 *info
= list
[index
];
1204 struct pipe_screen
*r600_screen_create(struct radeon_winsys
*ws
)
1206 struct r600_screen
*rscreen
= CALLOC_STRUCT(r600_screen
);
1208 if (rscreen
== NULL
) {
1213 ws
->query_info(ws
, &rscreen
->info
);
1215 rscreen
->debug_flags
= debug_get_flags_option("R600_DEBUG", debug_options
, 0);
1216 if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE
))
1217 rscreen
->debug_flags
|= DBG_COMPUTE
;
1218 if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE
))
1219 rscreen
->debug_flags
|= DBG_FS
| DBG_VS
| DBG_GS
| DBG_PS
| DBG_CS
;
1220 if (!debug_get_bool_option("R600_HYPERZ", TRUE
))
1221 rscreen
->debug_flags
|= DBG_NO_HYPERZ
;
1222 if (!debug_get_bool_option("R600_LLVM", TRUE
))
1223 rscreen
->debug_flags
|= DBG_NO_LLVM
;
1224 if (debug_get_bool_option("R600_PRINT_TEXDEPTH", FALSE
))
1225 rscreen
->debug_flags
|= DBG_TEX_DEPTH
;
1226 rscreen
->family
= rscreen
->info
.family
;
1227 rscreen
->chip_class
= rscreen
->info
.chip_class
;
1229 if (rscreen
->family
== CHIP_UNKNOWN
) {
1230 fprintf(stderr
, "r600: Unknown chipset 0x%04X\n", rscreen
->info
.pci_id
);
1235 /* Figure out streamout kernel support. */
1236 switch (rscreen
->chip_class
) {
1238 if (rscreen
->family
< CHIP_RS780
) {
1239 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 14;
1241 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 23;
1245 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 17;
1249 rscreen
->has_streamout
= rscreen
->info
.drm_minor
>= 14;
1252 rscreen
->has_streamout
= FALSE
;
1257 switch (rscreen
->chip_class
) {
1260 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 22;
1261 rscreen
->has_compressed_msaa_texturing
= false;
1264 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 19;
1265 rscreen
->has_compressed_msaa_texturing
= rscreen
->info
.drm_minor
>= 24;
1268 rscreen
->has_msaa
= rscreen
->info
.drm_minor
>= 19;
1269 rscreen
->has_compressed_msaa_texturing
= true;
1272 rscreen
->has_msaa
= FALSE
;
1273 rscreen
->has_compressed_msaa_texturing
= false;
1276 rscreen
->has_cp_dma
= rscreen
->info
.drm_minor
>= 27 &&
1277 !(rscreen
->debug_flags
& DBG_NO_CP_DMA
);
1279 if (r600_init_tiling(rscreen
)) {
1284 rscreen
->screen
.destroy
= r600_destroy_screen
;
1285 rscreen
->screen
.get_name
= r600_get_name
;
1286 rscreen
->screen
.get_vendor
= r600_get_vendor
;
1287 rscreen
->screen
.get_param
= r600_get_param
;
1288 rscreen
->screen
.get_shader_param
= r600_get_shader_param
;
1289 rscreen
->screen
.get_paramf
= r600_get_paramf
;
1290 rscreen
->screen
.get_compute_param
= r600_get_compute_param
;
1291 rscreen
->screen
.get_timestamp
= r600_get_timestamp
;
1293 if (rscreen
->chip_class
>= EVERGREEN
) {
1294 rscreen
->screen
.is_format_supported
= evergreen_is_format_supported
;
1295 rscreen
->dma_blit
= &evergreen_dma_blit
;
1297 rscreen
->screen
.is_format_supported
= r600_is_format_supported
;
1298 rscreen
->dma_blit
= &r600_dma_blit
;
1300 rscreen
->screen
.context_create
= r600_create_context
;
1301 rscreen
->screen
.fence_reference
= r600_fence_reference
;
1302 rscreen
->screen
.fence_signalled
= r600_fence_signalled
;
1303 rscreen
->screen
.fence_finish
= r600_fence_finish
;
1304 rscreen
->screen
.get_driver_query_info
= r600_get_driver_query_info
;
1306 if (rscreen
->info
.has_uvd
) {
1307 rscreen
->screen
.get_video_param
= r600_uvd_get_video_param
;
1308 rscreen
->screen
.is_video_format_supported
= ruvd_is_format_supported
;
1310 rscreen
->screen
.get_video_param
= r600_get_video_param
;
1311 rscreen
->screen
.is_video_format_supported
= vl_video_buffer_is_format_supported
;
1314 r600_init_screen_resource_functions(&rscreen
->screen
);
1316 util_format_s3tc_init();
1318 rscreen
->fences
.bo
= NULL
;
1319 rscreen
->fences
.data
= NULL
;
1320 rscreen
->fences
.next_index
= 0;
1321 LIST_INITHEAD(&rscreen
->fences
.pool
);
1322 LIST_INITHEAD(&rscreen
->fences
.blocks
);
1323 pipe_mutex_init(rscreen
->fences
.mutex
);
1325 rscreen
->global_pool
= compute_memory_pool_new(rscreen
);
1327 rscreen
->cs_count
= 0;
1328 if (rscreen
->info
.drm_minor
>= 28 && (rscreen
->debug_flags
& DBG_TRACE_CS
)) {
1329 rscreen
->trace_bo
= (struct r600_resource
*)pipe_buffer_create(&rscreen
->screen
,
1333 if (rscreen
->trace_bo
) {
1334 rscreen
->trace_ptr
= rscreen
->ws
->buffer_map(rscreen
->trace_bo
->cs_buf
, NULL
,
1335 PIPE_TRANSFER_UNSYNCHRONIZED
);
1339 /* Create the auxiliary context. */
1340 pipe_mutex_init(rscreen
->aux_context_lock
);
1341 rscreen
->aux_context
= rscreen
->screen
.context_create(&rscreen
->screen
, NULL
);
1343 #if 0 /* This is for testing whether aux_context and buffer clearing work correctly. */
1344 struct pipe_resource templ
= {};
1347 templ
.height0
= 2048;
1349 templ
.array_size
= 1;
1350 templ
.target
= PIPE_TEXTURE_2D
;
1351 templ
.format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
1352 templ
.usage
= PIPE_USAGE_STATIC
;
1354 struct r600_resource
*res
= r600_resource(rscreen
->screen
.resource_create(&rscreen
->screen
, &templ
));
1355 unsigned char *map
= ws
->buffer_map(res
->cs_buf
, NULL
, PIPE_TRANSFER_WRITE
);
1357 memset(map
, 0, 256);
1359 r600_screen_clear_buffer(rscreen
, &res
->b
.b
, 4, 4, 0xCC);
1360 r600_screen_clear_buffer(rscreen
, &res
->b
.b
, 8, 4, 0xDD);
1361 r600_screen_clear_buffer(rscreen
, &res
->b
.b
, 12, 4, 0xEE);
1362 r600_screen_clear_buffer(rscreen
, &res
->b
.b
, 20, 4, 0xFF);
1363 r600_screen_clear_buffer(rscreen
, &res
->b
.b
, 32, 20, 0x87);
1365 ws
->buffer_wait(res
->buf
, RADEON_USAGE_WRITE
);
1368 for (i
= 0; i
< 256; i
++) {
1369 printf("%02X", map
[i
]);
1375 return &rscreen
->screen
;