2 * Copyright 2013 Advanced Micro Devices, Inc.
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * Authors: Marek Olšák <maraeo@gmail.com>
27 #include "r600_pipe_common.h"
29 #include "tgsi/tgsi_parse.h"
30 #include "util/list.h"
31 #include "util/u_draw_quad.h"
32 #include "util/u_memory.h"
33 #include "util/u_format_s3tc.h"
34 #include "util/u_upload_mgr.h"
35 #include "os/os_time.h"
36 #include "vl/vl_decoder.h"
37 #include "vl/vl_video_buffer.h"
38 #include "radeon/radeon_video.h"
40 #include <sys/utsname.h>
46 struct r600_multi_fence
{
47 struct pipe_reference reference
;
48 struct pipe_fence_handle
*gfx
;
49 struct pipe_fence_handle
*sdma
;
51 /* If the context wasn't flushed at fence creation, this is non-NULL. */
53 struct r600_common_context
*ctx
;
59 * shader binary helpers.
61 void radeon_shader_binary_init(struct radeon_shader_binary
*b
)
63 memset(b
, 0, sizeof(*b
));
66 void radeon_shader_binary_clean(struct radeon_shader_binary
*b
)
73 FREE(b
->global_symbol_offsets
);
75 FREE(b
->disasm_string
);
76 FREE(b
->llvm_ir_string
);
83 void r600_gfx_write_fence(struct r600_common_context
*ctx
,
84 uint64_t va
, uint32_t old_value
, uint32_t new_value
)
86 struct radeon_winsys_cs
*cs
= ctx
->gfx
.cs
;
88 if (ctx
->chip_class
== CIK
) {
89 /* Two EOP events are required to make all engines go idle
90 * (and optional cache flushes executed) before the timestamp
93 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE_EOP
, 4, 0));
94 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_BOTTOM_OF_PIPE_TS
) |
97 radeon_emit(cs
, (va
>> 32) | EOP_DATA_SEL(1));
98 radeon_emit(cs
, old_value
); /* immediate data */
99 radeon_emit(cs
, 0); /* unused */
102 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE_EOP
, 4, 0));
103 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_BOTTOM_OF_PIPE_TS
) |
106 radeon_emit(cs
, (va
>> 32) | EOP_DATA_SEL(1));
107 radeon_emit(cs
, new_value
); /* immediate data */
108 radeon_emit(cs
, 0); /* unused */
111 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen
*screen
)
115 if (screen
->chip_class
== CIK
)
121 void r600_gfx_wait_fence(struct r600_common_context
*ctx
,
122 uint64_t va
, uint32_t ref
, uint32_t mask
)
124 struct radeon_winsys_cs
*cs
= ctx
->gfx
.cs
;
126 radeon_emit(cs
, PKT3(PKT3_WAIT_REG_MEM
, 5, 0));
127 radeon_emit(cs
, WAIT_REG_MEM_EQUAL
| WAIT_REG_MEM_MEM_SPACE(1));
129 radeon_emit(cs
, va
>> 32);
130 radeon_emit(cs
, ref
); /* reference value */
131 radeon_emit(cs
, mask
); /* mask */
132 radeon_emit(cs
, 4); /* poll interval */
135 void r600_draw_rectangle(struct blitter_context
*blitter
,
136 int x1
, int y1
, int x2
, int y2
, float depth
,
137 enum blitter_attrib_type type
,
138 const union pipe_color_union
*attrib
)
140 struct r600_common_context
*rctx
=
141 (struct r600_common_context
*)util_blitter_get_pipe(blitter
);
142 struct pipe_viewport_state viewport
;
143 struct pipe_resource
*buf
= NULL
;
147 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
) {
148 util_blitter_draw_rectangle(blitter
, x1
, y1
, x2
, y2
, depth
, type
, attrib
);
152 /* Some operations (like color resolve on r6xx) don't work
153 * with the conventional primitive types.
154 * One that works is PT_RECTLIST, which we use here. */
157 viewport
.scale
[0] = 1.0f
;
158 viewport
.scale
[1] = 1.0f
;
159 viewport
.scale
[2] = 1.0f
;
160 viewport
.translate
[0] = 0.0f
;
161 viewport
.translate
[1] = 0.0f
;
162 viewport
.translate
[2] = 0.0f
;
163 rctx
->b
.set_viewport_states(&rctx
->b
, 0, 1, &viewport
);
165 /* Upload vertices. The hw rectangle has only 3 vertices,
166 * I guess the 4th one is derived from the first 3.
167 * The vertex specification should match u_blitter's vertex element state. */
168 u_upload_alloc(rctx
->uploader
, 0, sizeof(float) * 24, 256, &offset
, &buf
, (void**)&vb
);
188 memcpy(vb
+4, attrib
->f
, sizeof(float)*4);
189 memcpy(vb
+12, attrib
->f
, sizeof(float)*4);
190 memcpy(vb
+20, attrib
->f
, sizeof(float)*4);
194 util_draw_vertex_buffer(&rctx
->b
, NULL
, buf
, blitter
->vb_slot
, offset
,
195 R600_PRIM_RECTANGLE_LIST
, 3, 2);
196 pipe_resource_reference(&buf
, NULL
);
199 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
,
200 struct r600_resource
*dst
, struct r600_resource
*src
)
202 uint64_t vram
= 0, gtt
= 0;
205 vram
+= dst
->vram_usage
;
206 gtt
+= dst
->gart_usage
;
209 vram
+= src
->vram_usage
;
210 gtt
+= src
->gart_usage
;
213 /* Flush the GFX IB if DMA depends on it. */
214 if (radeon_emitted(ctx
->gfx
.cs
, ctx
->initial_gfx_cs_size
) &&
216 ctx
->ws
->cs_is_buffer_referenced(ctx
->gfx
.cs
, dst
->buf
,
217 RADEON_USAGE_READWRITE
)) ||
219 ctx
->ws
->cs_is_buffer_referenced(ctx
->gfx
.cs
, src
->buf
,
220 RADEON_USAGE_WRITE
))))
221 ctx
->gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
223 /* Flush if there's not enough space, or if the memory usage per IB
226 if (!ctx
->ws
->cs_check_space(ctx
->dma
.cs
, num_dw
) ||
227 !radeon_cs_memory_below_limit(ctx
->screen
, ctx
->dma
.cs
, vram
, gtt
)) {
228 ctx
->dma
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
229 assert((num_dw
+ ctx
->dma
.cs
->current
.cdw
) <= ctx
->dma
.cs
->current
.max_dw
);
232 /* If GPUVM is not supported, the CS checker needs 2 entries
233 * in the buffer list per packet, which has to be done manually.
235 if (ctx
->screen
->info
.has_virtual_memory
) {
237 radeon_add_to_buffer_list(ctx
, &ctx
->dma
, dst
,
239 RADEON_PRIO_SDMA_BUFFER
);
241 radeon_add_to_buffer_list(ctx
, &ctx
->dma
, src
,
243 RADEON_PRIO_SDMA_BUFFER
);
247 /* This is required to prevent read-after-write hazards. */
248 void r600_dma_emit_wait_idle(struct r600_common_context
*rctx
)
250 struct radeon_winsys_cs
*cs
= rctx
->dma
.cs
;
252 /* done at the end of DMA calls, so increment this. */
253 rctx
->num_dma_calls
++;
255 /* IBs using too little memory are limited by the IB submission overhead.
256 * IBs using too much memory are limited by the kernel/TTM overhead.
257 * Too long IBs create CPU-GPU pipeline bubbles and add latency.
259 * This heuristic makes sure that DMA requests are executed
260 * very soon after the call is made and lowers memory usage.
261 * It improves texture upload performance by keeping the DMA
262 * engine busy while uploads are being submitted.
264 if (cs
->used_vram
+ cs
->used_gart
> 64 * 1024 * 1024) {
265 rctx
->dma
.flush(rctx
, RADEON_FLUSH_ASYNC
, NULL
);
269 r600_need_dma_space(rctx
, 1, NULL
, NULL
);
271 if (!radeon_emitted(cs
, 0)) /* empty queue */
274 /* NOP waits for idle on Evergreen and later. */
275 if (rctx
->chip_class
>= CIK
)
276 radeon_emit(cs
, 0x00000000); /* NOP */
277 else if (rctx
->chip_class
>= EVERGREEN
)
278 radeon_emit(cs
, 0xf0000000); /* NOP */
280 /* TODO: R600-R700 should use the FENCE packet.
281 * CS checker support is required. */
285 static void r600_memory_barrier(struct pipe_context
*ctx
, unsigned flags
)
289 void r600_preflush_suspend_features(struct r600_common_context
*ctx
)
291 /* suspend queries */
292 if (!LIST_IS_EMPTY(&ctx
->active_queries
))
293 r600_suspend_queries(ctx
);
295 ctx
->streamout
.suspended
= false;
296 if (ctx
->streamout
.begin_emitted
) {
297 r600_emit_streamout_end(ctx
);
298 ctx
->streamout
.suspended
= true;
302 void r600_postflush_resume_features(struct r600_common_context
*ctx
)
304 if (ctx
->streamout
.suspended
) {
305 ctx
->streamout
.append_bitmask
= ctx
->streamout
.enabled_mask
;
306 r600_streamout_buffers_dirty(ctx
);
310 if (!LIST_IS_EMPTY(&ctx
->active_queries
))
311 r600_resume_queries(ctx
);
314 static void r600_flush_from_st(struct pipe_context
*ctx
,
315 struct pipe_fence_handle
**fence
,
318 struct pipe_screen
*screen
= ctx
->screen
;
319 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
320 struct radeon_winsys
*ws
= rctx
->ws
;
322 struct pipe_fence_handle
*gfx_fence
= NULL
;
323 struct pipe_fence_handle
*sdma_fence
= NULL
;
324 bool deferred_fence
= false;
326 if (flags
& PIPE_FLUSH_END_OF_FRAME
)
327 rflags
|= RADEON_FLUSH_END_OF_FRAME
;
328 if (flags
& PIPE_FLUSH_DEFERRED
)
329 rflags
|= RADEON_FLUSH_ASYNC
;
332 rctx
->dma
.flush(rctx
, rflags
, fence
? &sdma_fence
: NULL
);
335 if (!radeon_emitted(rctx
->gfx
.cs
, rctx
->initial_gfx_cs_size
)) {
337 ws
->fence_reference(&gfx_fence
, rctx
->last_gfx_fence
);
338 if (!(rflags
& RADEON_FLUSH_ASYNC
))
339 ws
->cs_sync_flush(rctx
->gfx
.cs
);
341 /* Instead of flushing, create a deferred fence. Constraints:
342 * - The state tracker must allow a deferred flush.
343 * - The state tracker must request a fence.
344 * Thread safety in fence_finish must be ensured by the state tracker.
346 if (flags
& PIPE_FLUSH_DEFERRED
&& fence
) {
347 gfx_fence
= rctx
->ws
->cs_get_next_fence(rctx
->gfx
.cs
);
348 deferred_fence
= true;
350 rctx
->gfx
.flush(rctx
, rflags
, fence
? &gfx_fence
: NULL
);
354 /* Both engines can signal out of order, so we need to keep both fences. */
356 struct r600_multi_fence
*multi_fence
=
357 CALLOC_STRUCT(r600_multi_fence
);
361 multi_fence
->reference
.count
= 1;
362 /* If both fences are NULL, fence_finish will always return true. */
363 multi_fence
->gfx
= gfx_fence
;
364 multi_fence
->sdma
= sdma_fence
;
366 if (deferred_fence
) {
367 multi_fence
->gfx_unflushed
.ctx
= rctx
;
368 multi_fence
->gfx_unflushed
.ib_index
= rctx
->num_gfx_cs_flushes
;
371 screen
->fence_reference(screen
, fence
, NULL
);
372 *fence
= (struct pipe_fence_handle
*)multi_fence
;
376 static void r600_flush_dma_ring(void *ctx
, unsigned flags
,
377 struct pipe_fence_handle
**fence
)
379 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
380 struct radeon_winsys_cs
*cs
= rctx
->dma
.cs
;
381 struct radeon_saved_cs saved
;
383 (rctx
->screen
->debug_flags
& DBG_CHECK_VM
) &&
384 rctx
->check_vm_faults
;
386 if (!radeon_emitted(cs
, 0)) {
388 rctx
->ws
->fence_reference(fence
, rctx
->last_sdma_fence
);
393 radeon_save_cs(rctx
->ws
, cs
, &saved
);
395 rctx
->ws
->cs_flush(cs
, flags
, &rctx
->last_sdma_fence
);
397 rctx
->ws
->fence_reference(fence
, rctx
->last_sdma_fence
);
400 /* Use conservative timeout 800ms, after which we won't wait any
401 * longer and assume the GPU is hung.
403 rctx
->ws
->fence_wait(rctx
->ws
, rctx
->last_sdma_fence
, 800*1000*1000);
405 rctx
->check_vm_faults(rctx
, &saved
, RING_DMA
);
406 radeon_clear_saved_cs(&saved
);
411 * Store a linearized copy of all chunks of \p cs together with the buffer
414 void radeon_save_cs(struct radeon_winsys
*ws
, struct radeon_winsys_cs
*cs
,
415 struct radeon_saved_cs
*saved
)
420 /* Save the IB chunks. */
421 saved
->num_dw
= cs
->prev_dw
+ cs
->current
.cdw
;
422 saved
->ib
= MALLOC(4 * saved
->num_dw
);
427 for (i
= 0; i
< cs
->num_prev
; ++i
) {
428 memcpy(buf
, cs
->prev
[i
].buf
, cs
->prev
[i
].cdw
* 4);
429 buf
+= cs
->prev
[i
].cdw
;
431 memcpy(buf
, cs
->current
.buf
, cs
->current
.cdw
* 4);
433 /* Save the buffer list. */
434 saved
->bo_count
= ws
->cs_get_buffer_list(cs
, NULL
);
435 saved
->bo_list
= CALLOC(saved
->bo_count
,
436 sizeof(saved
->bo_list
[0]));
437 if (!saved
->bo_list
) {
441 ws
->cs_get_buffer_list(cs
, saved
->bo_list
);
446 fprintf(stderr
, "%s: out of memory\n", __func__
);
447 memset(saved
, 0, sizeof(*saved
));
450 void radeon_clear_saved_cs(struct radeon_saved_cs
*saved
)
453 FREE(saved
->bo_list
);
455 memset(saved
, 0, sizeof(*saved
));
458 static enum pipe_reset_status
r600_get_reset_status(struct pipe_context
*ctx
)
460 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
461 unsigned latest
= rctx
->ws
->query_value(rctx
->ws
,
462 RADEON_GPU_RESET_COUNTER
);
464 if (rctx
->gpu_reset_counter
== latest
)
465 return PIPE_NO_RESET
;
467 rctx
->gpu_reset_counter
= latest
;
468 return PIPE_UNKNOWN_CONTEXT_RESET
;
471 static void r600_set_debug_callback(struct pipe_context
*ctx
,
472 const struct pipe_debug_callback
*cb
)
474 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
479 memset(&rctx
->debug
, 0, sizeof(rctx
->debug
));
482 bool r600_common_context_init(struct r600_common_context
*rctx
,
483 struct r600_common_screen
*rscreen
,
484 unsigned context_flags
)
486 slab_create(&rctx
->pool_transfers
,
487 sizeof(struct r600_transfer
), 64);
489 rctx
->screen
= rscreen
;
490 rctx
->ws
= rscreen
->ws
;
491 rctx
->family
= rscreen
->family
;
492 rctx
->chip_class
= rscreen
->chip_class
;
494 if (rscreen
->chip_class
>= CIK
)
495 rctx
->max_db
= MAX2(8, rscreen
->info
.num_render_backends
);
496 else if (rscreen
->chip_class
>= EVERGREEN
)
501 rctx
->b
.invalidate_resource
= r600_invalidate_resource
;
502 rctx
->b
.transfer_map
= u_transfer_map_vtbl
;
503 rctx
->b
.transfer_flush_region
= u_transfer_flush_region_vtbl
;
504 rctx
->b
.transfer_unmap
= u_transfer_unmap_vtbl
;
505 rctx
->b
.texture_subdata
= u_default_texture_subdata
;
506 rctx
->b
.memory_barrier
= r600_memory_barrier
;
507 rctx
->b
.flush
= r600_flush_from_st
;
508 rctx
->b
.set_debug_callback
= r600_set_debug_callback
;
510 /* evergreen_compute.c has a special codepath for global buffers.
511 * Everything else can use the direct path.
513 if ((rscreen
->chip_class
== EVERGREEN
|| rscreen
->chip_class
== CAYMAN
) &&
514 (context_flags
& PIPE_CONTEXT_COMPUTE_ONLY
))
515 rctx
->b
.buffer_subdata
= u_default_buffer_subdata
;
517 rctx
->b
.buffer_subdata
= r600_buffer_subdata
;
519 if (rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
>= 43) {
520 rctx
->b
.get_device_reset_status
= r600_get_reset_status
;
521 rctx
->gpu_reset_counter
=
522 rctx
->ws
->query_value(rctx
->ws
,
523 RADEON_GPU_RESET_COUNTER
);
526 LIST_INITHEAD(&rctx
->texture_buffers
);
528 r600_init_context_texture_functions(rctx
);
529 r600_init_viewport_functions(rctx
);
530 r600_streamout_init(rctx
);
531 r600_query_init(rctx
);
532 cayman_init_msaa(&rctx
->b
);
534 rctx
->allocator_zeroed_memory
=
535 u_suballocator_create(&rctx
->b
, rscreen
->info
.gart_page_size
,
536 0, PIPE_USAGE_DEFAULT
, true);
537 if (!rctx
->allocator_zeroed_memory
)
540 rctx
->uploader
= u_upload_create(&rctx
->b
, 1024 * 1024,
541 PIPE_BIND_INDEX_BUFFER
|
542 PIPE_BIND_CONSTANT_BUFFER
, PIPE_USAGE_STREAM
);
546 rctx
->ctx
= rctx
->ws
->ctx_create(rctx
->ws
);
550 if (rscreen
->info
.has_sdma
&& !(rscreen
->debug_flags
& DBG_NO_ASYNC_DMA
)) {
551 rctx
->dma
.cs
= rctx
->ws
->cs_create(rctx
->ctx
, RING_DMA
,
554 rctx
->dma
.flush
= r600_flush_dma_ring
;
560 void r600_common_context_cleanup(struct r600_common_context
*rctx
)
564 /* Release DCC stats. */
565 for (i
= 0; i
< ARRAY_SIZE(rctx
->dcc_stats
); i
++) {
566 assert(!rctx
->dcc_stats
[i
].query_active
);
568 for (j
= 0; j
< ARRAY_SIZE(rctx
->dcc_stats
[i
].ps_stats
); j
++)
569 if (rctx
->dcc_stats
[i
].ps_stats
[j
])
570 rctx
->b
.destroy_query(&rctx
->b
,
571 rctx
->dcc_stats
[i
].ps_stats
[j
]);
573 r600_texture_reference(&rctx
->dcc_stats
[i
].tex
, NULL
);
576 if (rctx
->query_result_shader
)
577 rctx
->b
.delete_compute_state(&rctx
->b
, rctx
->query_result_shader
);
580 rctx
->ws
->cs_destroy(rctx
->gfx
.cs
);
582 rctx
->ws
->cs_destroy(rctx
->dma
.cs
);
584 rctx
->ws
->ctx_destroy(rctx
->ctx
);
586 if (rctx
->uploader
) {
587 u_upload_destroy(rctx
->uploader
);
590 slab_destroy(&rctx
->pool_transfers
);
592 if (rctx
->allocator_zeroed_memory
) {
593 u_suballocator_destroy(rctx
->allocator_zeroed_memory
);
595 rctx
->ws
->fence_reference(&rctx
->last_gfx_fence
, NULL
);
596 rctx
->ws
->fence_reference(&rctx
->last_sdma_fence
, NULL
);
599 void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
601 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
602 struct r600_resource
*rr
= (struct r600_resource
*)r
;
609 * The idea is to compute a gross estimate of memory requirement of
610 * each draw call. After each draw call, memory will be precisely
611 * accounted. So the uncertainty is only on the current draw call.
612 * In practice this gave very good estimate (+/- 10% of the target
615 rctx
->vram
+= rr
->vram_usage
;
616 rctx
->gtt
+= rr
->gart_usage
;
623 static const struct debug_named_value common_debug_options
[] = {
625 { "tex", DBG_TEX
, "Print texture info" },
626 { "compute", DBG_COMPUTE
, "Print compute info" },
627 { "vm", DBG_VM
, "Print virtual addresses when creating resources" },
628 { "info", DBG_INFO
, "Print driver information" },
631 { "fs", DBG_FS
, "Print fetch shaders" },
632 { "vs", DBG_VS
, "Print vertex shaders" },
633 { "gs", DBG_GS
, "Print geometry shaders" },
634 { "ps", DBG_PS
, "Print pixel shaders" },
635 { "cs", DBG_CS
, "Print compute shaders" },
636 { "tcs", DBG_TCS
, "Print tessellation control shaders" },
637 { "tes", DBG_TES
, "Print tessellation evaluation shaders" },
638 { "noir", DBG_NO_IR
, "Don't print the LLVM IR"},
639 { "notgsi", DBG_NO_TGSI
, "Don't print the TGSI"},
640 { "noasm", DBG_NO_ASM
, "Don't print disassembled shaders"},
641 { "preoptir", DBG_PREOPT_IR
, "Print the LLVM IR before initial optimizations" },
643 { "testdma", DBG_TEST_DMA
, "Invoke SDMA tests and exit." },
646 { "nodma", DBG_NO_ASYNC_DMA
, "Disable asynchronous DMA" },
647 { "nohyperz", DBG_NO_HYPERZ
, "Disable Hyper-Z" },
648 /* GL uses the word INVALIDATE, gallium uses the word DISCARD */
649 { "noinvalrange", DBG_NO_DISCARD_RANGE
, "Disable handling of INVALIDATE_RANGE map flags" },
650 { "no2d", DBG_NO_2D_TILING
, "Disable 2D tiling" },
651 { "notiling", DBG_NO_TILING
, "Disable tiling" },
652 { "switch_on_eop", DBG_SWITCH_ON_EOP
, "Program WD/IA to switch on end-of-packet." },
653 { "forcedma", DBG_FORCE_DMA
, "Use asynchronous DMA for all operations when possible." },
654 { "precompile", DBG_PRECOMPILE
, "Compile one shader variant at shader creation." },
655 { "nowc", DBG_NO_WC
, "Disable GTT write combining" },
656 { "check_vm", DBG_CHECK_VM
, "Check VM faults and dump debug info." },
657 { "nodcc", DBG_NO_DCC
, "Disable DCC." },
658 { "nodccclear", DBG_NO_DCC_CLEAR
, "Disable DCC fast clear." },
659 { "norbplus", DBG_NO_RB_PLUS
, "Disable RB+ on Stoney." },
660 { "sisched", DBG_SI_SCHED
, "Enable LLVM SI Machine Instruction Scheduler." },
661 { "mono", DBG_MONOLITHIC_SHADERS
, "Use old-style monolithic shaders compiled on demand" },
662 { "noce", DBG_NO_CE
, "Disable the constant engine"},
663 { "unsafemath", DBG_UNSAFE_MATH
, "Enable unsafe math shader optimizations" },
664 { "nodccfb", DBG_NO_DCC_FB
, "Disable separate DCC on the main framebuffer" },
666 DEBUG_NAMED_VALUE_END
/* must be last */
669 static const char* r600_get_vendor(struct pipe_screen
* pscreen
)
674 static const char* r600_get_device_vendor(struct pipe_screen
* pscreen
)
679 static const char* r600_get_chip_name(struct r600_common_screen
*rscreen
)
681 switch (rscreen
->info
.family
) {
682 case CHIP_R600
: return "AMD R600";
683 case CHIP_RV610
: return "AMD RV610";
684 case CHIP_RV630
: return "AMD RV630";
685 case CHIP_RV670
: return "AMD RV670";
686 case CHIP_RV620
: return "AMD RV620";
687 case CHIP_RV635
: return "AMD RV635";
688 case CHIP_RS780
: return "AMD RS780";
689 case CHIP_RS880
: return "AMD RS880";
690 case CHIP_RV770
: return "AMD RV770";
691 case CHIP_RV730
: return "AMD RV730";
692 case CHIP_RV710
: return "AMD RV710";
693 case CHIP_RV740
: return "AMD RV740";
694 case CHIP_CEDAR
: return "AMD CEDAR";
695 case CHIP_REDWOOD
: return "AMD REDWOOD";
696 case CHIP_JUNIPER
: return "AMD JUNIPER";
697 case CHIP_CYPRESS
: return "AMD CYPRESS";
698 case CHIP_HEMLOCK
: return "AMD HEMLOCK";
699 case CHIP_PALM
: return "AMD PALM";
700 case CHIP_SUMO
: return "AMD SUMO";
701 case CHIP_SUMO2
: return "AMD SUMO2";
702 case CHIP_BARTS
: return "AMD BARTS";
703 case CHIP_TURKS
: return "AMD TURKS";
704 case CHIP_CAICOS
: return "AMD CAICOS";
705 case CHIP_CAYMAN
: return "AMD CAYMAN";
706 case CHIP_ARUBA
: return "AMD ARUBA";
707 case CHIP_TAHITI
: return "AMD TAHITI";
708 case CHIP_PITCAIRN
: return "AMD PITCAIRN";
709 case CHIP_VERDE
: return "AMD CAPE VERDE";
710 case CHIP_OLAND
: return "AMD OLAND";
711 case CHIP_HAINAN
: return "AMD HAINAN";
712 case CHIP_BONAIRE
: return "AMD BONAIRE";
713 case CHIP_KAVERI
: return "AMD KAVERI";
714 case CHIP_KABINI
: return "AMD KABINI";
715 case CHIP_HAWAII
: return "AMD HAWAII";
716 case CHIP_MULLINS
: return "AMD MULLINS";
717 case CHIP_TONGA
: return "AMD TONGA";
718 case CHIP_ICELAND
: return "AMD ICELAND";
719 case CHIP_CARRIZO
: return "AMD CARRIZO";
720 case CHIP_FIJI
: return "AMD FIJI";
721 case CHIP_POLARIS10
: return "AMD POLARIS10";
722 case CHIP_POLARIS11
: return "AMD POLARIS11";
723 case CHIP_STONEY
: return "AMD STONEY";
724 default: return "AMD unknown";
728 static const char* r600_get_name(struct pipe_screen
* pscreen
)
730 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)pscreen
;
732 return rscreen
->renderer_string
;
735 static float r600_get_paramf(struct pipe_screen
* pscreen
,
736 enum pipe_capf param
)
738 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)pscreen
;
741 case PIPE_CAPF_MAX_LINE_WIDTH
:
742 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
743 case PIPE_CAPF_MAX_POINT_WIDTH
:
744 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
745 if (rscreen
->family
>= CHIP_CEDAR
)
749 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
751 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
753 case PIPE_CAPF_GUARD_BAND_LEFT
:
754 case PIPE_CAPF_GUARD_BAND_TOP
:
755 case PIPE_CAPF_GUARD_BAND_RIGHT
:
756 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
762 static int r600_get_video_param(struct pipe_screen
*screen
,
763 enum pipe_video_profile profile
,
764 enum pipe_video_entrypoint entrypoint
,
765 enum pipe_video_cap param
)
768 case PIPE_VIDEO_CAP_SUPPORTED
:
769 return vl_profile_supported(screen
, profile
, entrypoint
);
770 case PIPE_VIDEO_CAP_NPOT_TEXTURES
:
772 case PIPE_VIDEO_CAP_MAX_WIDTH
:
773 case PIPE_VIDEO_CAP_MAX_HEIGHT
:
774 return vl_video_buffer_max_size(screen
);
775 case PIPE_VIDEO_CAP_PREFERED_FORMAT
:
776 return PIPE_FORMAT_NV12
;
777 case PIPE_VIDEO_CAP_PREFERS_INTERLACED
:
779 case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED
:
781 case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE
:
783 case PIPE_VIDEO_CAP_MAX_LEVEL
:
784 return vl_level_supported(screen
, profile
);
790 const char *r600_get_llvm_processor_name(enum radeon_family family
)
833 case CHIP_TAHITI
: return "tahiti";
834 case CHIP_PITCAIRN
: return "pitcairn";
835 case CHIP_VERDE
: return "verde";
836 case CHIP_OLAND
: return "oland";
837 case CHIP_HAINAN
: return "hainan";
838 case CHIP_BONAIRE
: return "bonaire";
839 case CHIP_KABINI
: return "kabini";
840 case CHIP_KAVERI
: return "kaveri";
841 case CHIP_HAWAII
: return "hawaii";
844 case CHIP_TONGA
: return "tonga";
845 case CHIP_ICELAND
: return "iceland";
846 case CHIP_CARRIZO
: return "carrizo";
847 #if HAVE_LLVM <= 0x0307
848 case CHIP_FIJI
: return "tonga";
849 case CHIP_STONEY
: return "carrizo";
851 case CHIP_FIJI
: return "fiji";
852 case CHIP_STONEY
: return "stoney";
854 #if HAVE_LLVM <= 0x0308
855 case CHIP_POLARIS10
: return "tonga";
856 case CHIP_POLARIS11
: return "tonga";
858 case CHIP_POLARIS10
: return "polaris10";
859 case CHIP_POLARIS11
: return "polaris11";
865 static int r600_get_compute_param(struct pipe_screen
*screen
,
866 enum pipe_shader_ir ir_type
,
867 enum pipe_compute_cap param
,
870 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
872 //TODO: select these params by asic
874 case PIPE_COMPUTE_CAP_IR_TARGET
: {
877 if (rscreen
->family
<= CHIP_ARUBA
) {
880 if (HAVE_LLVM
< 0x0400) {
883 triple
= "amdgcn-mesa-mesa3d";
886 switch(rscreen
->family
) {
887 /* Clang < 3.6 is missing Hainan in its list of
888 * GPUs, so we need to use the name of a similar GPU.
891 gpu
= r600_get_llvm_processor_name(rscreen
->family
);
895 sprintf(ret
, "%s-%s", gpu
, triple
);
897 /* +2 for dash and terminating NIL byte */
898 return (strlen(triple
) + strlen(gpu
) + 2) * sizeof(char);
900 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
902 uint64_t *grid_dimension
= ret
;
903 grid_dimension
[0] = 3;
905 return 1 * sizeof(uint64_t);
907 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
909 uint64_t *grid_size
= ret
;
910 grid_size
[0] = 65535;
911 grid_size
[1] = 65535;
912 grid_size
[2] = 65535;
914 return 3 * sizeof(uint64_t) ;
916 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
918 uint64_t *block_size
= ret
;
919 if (rscreen
->chip_class
>= SI
&& HAVE_LLVM
>= 0x309 &&
920 ir_type
== PIPE_SHADER_IR_TGSI
) {
921 block_size
[0] = 2048;
922 block_size
[1] = 2048;
923 block_size
[2] = 2048;
930 return 3 * sizeof(uint64_t);
932 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
934 uint64_t *max_threads_per_block
= ret
;
935 if (rscreen
->chip_class
>= SI
&& HAVE_LLVM
>= 0x309 &&
936 ir_type
== PIPE_SHADER_IR_TGSI
)
937 *max_threads_per_block
= 2048;
939 *max_threads_per_block
= 256;
941 return sizeof(uint64_t);
942 case PIPE_COMPUTE_CAP_ADDRESS_BITS
:
944 uint32_t *address_bits
= ret
;
945 address_bits
[0] = 32;
946 if (rscreen
->chip_class
>= SI
)
947 address_bits
[0] = 64;
949 return 1 * sizeof(uint32_t);
951 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
:
953 uint64_t *max_global_size
= ret
;
954 uint64_t max_mem_alloc_size
;
956 r600_get_compute_param(screen
, ir_type
,
957 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
,
958 &max_mem_alloc_size
);
960 /* In OpenCL, the MAX_MEM_ALLOC_SIZE must be at least
961 * 1/4 of the MAX_GLOBAL_SIZE. Since the
962 * MAX_MEM_ALLOC_SIZE is fixed for older kernels,
963 * make sure we never report more than
964 * 4 * MAX_MEM_ALLOC_SIZE.
966 *max_global_size
= MIN2(4 * max_mem_alloc_size
,
967 MAX2(rscreen
->info
.gart_size
,
968 rscreen
->info
.vram_size
));
970 return sizeof(uint64_t);
972 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
:
974 uint64_t *max_local_size
= ret
;
975 /* Value reported by the closed source driver. */
976 *max_local_size
= 32768;
978 return sizeof(uint64_t);
980 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
:
982 uint64_t *max_input_size
= ret
;
983 /* Value reported by the closed source driver. */
984 *max_input_size
= 1024;
986 return sizeof(uint64_t);
988 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
990 uint64_t *max_mem_alloc_size
= ret
;
992 *max_mem_alloc_size
= rscreen
->info
.max_alloc_size
;
994 return sizeof(uint64_t);
996 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
998 uint32_t *max_clock_frequency
= ret
;
999 *max_clock_frequency
= rscreen
->info
.max_shader_clock
;
1001 return sizeof(uint32_t);
1003 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
1005 uint32_t *max_compute_units
= ret
;
1006 *max_compute_units
= rscreen
->info
.num_good_compute_units
;
1008 return sizeof(uint32_t);
1010 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
1012 uint32_t *images_supported
= ret
;
1013 *images_supported
= 0;
1015 return sizeof(uint32_t);
1016 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
:
1018 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
1020 uint32_t *subgroup_size
= ret
;
1021 *subgroup_size
= r600_wavefront_size(rscreen
->family
);
1023 return sizeof(uint32_t);
1026 fprintf(stderr
, "unknown PIPE_COMPUTE_CAP %d\n", param
);
1030 static uint64_t r600_get_timestamp(struct pipe_screen
*screen
)
1032 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
1034 return 1000000 * rscreen
->ws
->query_value(rscreen
->ws
, RADEON_TIMESTAMP
) /
1035 rscreen
->info
.clock_crystal_freq
;
1038 static void r600_fence_reference(struct pipe_screen
*screen
,
1039 struct pipe_fence_handle
**dst
,
1040 struct pipe_fence_handle
*src
)
1042 struct radeon_winsys
*ws
= ((struct r600_common_screen
*)screen
)->ws
;
1043 struct r600_multi_fence
**rdst
= (struct r600_multi_fence
**)dst
;
1044 struct r600_multi_fence
*rsrc
= (struct r600_multi_fence
*)src
;
1046 if (pipe_reference(&(*rdst
)->reference
, &rsrc
->reference
)) {
1047 ws
->fence_reference(&(*rdst
)->gfx
, NULL
);
1048 ws
->fence_reference(&(*rdst
)->sdma
, NULL
);
1054 static boolean
r600_fence_finish(struct pipe_screen
*screen
,
1055 struct pipe_context
*ctx
,
1056 struct pipe_fence_handle
*fence
,
1059 struct radeon_winsys
*rws
= ((struct r600_common_screen
*)screen
)->ws
;
1060 struct r600_multi_fence
*rfence
= (struct r600_multi_fence
*)fence
;
1061 struct r600_common_context
*rctx
=
1062 ctx
? (struct r600_common_context
*)ctx
: NULL
;
1063 int64_t abs_timeout
= os_time_get_absolute_timeout(timeout
);
1066 if (!rws
->fence_wait(rws
, rfence
->sdma
, timeout
))
1069 /* Recompute the timeout after waiting. */
1070 if (timeout
&& timeout
!= PIPE_TIMEOUT_INFINITE
) {
1071 int64_t time
= os_time_get_nano();
1072 timeout
= abs_timeout
> time
? abs_timeout
- time
: 0;
1079 /* Flush the gfx IB if it hasn't been flushed yet. */
1081 rfence
->gfx_unflushed
.ctx
== rctx
&&
1082 rfence
->gfx_unflushed
.ib_index
== rctx
->num_gfx_cs_flushes
) {
1083 rctx
->gfx
.flush(rctx
, timeout
? 0 : RADEON_FLUSH_ASYNC
, NULL
);
1084 rfence
->gfx_unflushed
.ctx
= NULL
;
1089 /* Recompute the timeout after all that. */
1090 if (timeout
&& timeout
!= PIPE_TIMEOUT_INFINITE
) {
1091 int64_t time
= os_time_get_nano();
1092 timeout
= abs_timeout
> time
? abs_timeout
- time
: 0;
1096 return rws
->fence_wait(rws
, rfence
->gfx
, timeout
);
1099 static void r600_query_memory_info(struct pipe_screen
*screen
,
1100 struct pipe_memory_info
*info
)
1102 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
1103 struct radeon_winsys
*ws
= rscreen
->ws
;
1104 unsigned vram_usage
, gtt_usage
;
1106 info
->total_device_memory
= rscreen
->info
.vram_size
/ 1024;
1107 info
->total_staging_memory
= rscreen
->info
.gart_size
/ 1024;
1109 /* The real TTM memory usage is somewhat random, because:
1111 * 1) TTM delays freeing memory, because it can only free it after
1114 * 2) The memory usage can be really low if big VRAM evictions are
1115 * taking place, but the real usage is well above the size of VRAM.
1117 * Instead, return statistics of this process.
1119 vram_usage
= ws
->query_value(ws
, RADEON_REQUESTED_VRAM_MEMORY
) / 1024;
1120 gtt_usage
= ws
->query_value(ws
, RADEON_REQUESTED_GTT_MEMORY
) / 1024;
1122 info
->avail_device_memory
=
1123 vram_usage
<= info
->total_device_memory
?
1124 info
->total_device_memory
- vram_usage
: 0;
1125 info
->avail_staging_memory
=
1126 gtt_usage
<= info
->total_staging_memory
?
1127 info
->total_staging_memory
- gtt_usage
: 0;
1129 info
->device_memory_evicted
=
1130 ws
->query_value(ws
, RADEON_NUM_BYTES_MOVED
) / 1024;
1132 if (rscreen
->info
.drm_major
== 3 && rscreen
->info
.drm_minor
>= 4)
1133 info
->nr_device_memory_evictions
=
1134 ws
->query_value(ws
, RADEON_NUM_EVICTIONS
);
1136 /* Just return the number of evicted 64KB pages. */
1137 info
->nr_device_memory_evictions
= info
->device_memory_evicted
/ 64;
1140 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
1141 const struct pipe_resource
*templ
)
1143 if (templ
->target
== PIPE_BUFFER
) {
1144 return r600_buffer_create(screen
, templ
, 256);
1146 return r600_texture_create(screen
, templ
);
1150 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
1151 struct radeon_winsys
*ws
)
1153 char llvm_string
[32] = {}, kernel_version
[128] = {};
1154 struct utsname uname_data
;
1156 ws
->query_info(ws
, &rscreen
->info
);
1158 if (uname(&uname_data
) == 0)
1159 snprintf(kernel_version
, sizeof(kernel_version
),
1160 " / %s", uname_data
.release
);
1163 snprintf(llvm_string
, sizeof(llvm_string
),
1164 ", LLVM %i.%i.%i", (HAVE_LLVM
>> 8) & 0xff,
1165 HAVE_LLVM
& 0xff, MESA_LLVM_VERSION_PATCH
);
1168 snprintf(rscreen
->renderer_string
, sizeof(rscreen
->renderer_string
),
1169 "%s (DRM %i.%i.%i%s%s)",
1170 r600_get_chip_name(rscreen
), rscreen
->info
.drm_major
,
1171 rscreen
->info
.drm_minor
, rscreen
->info
.drm_patchlevel
,
1172 kernel_version
, llvm_string
);
1174 rscreen
->b
.get_name
= r600_get_name
;
1175 rscreen
->b
.get_vendor
= r600_get_vendor
;
1176 rscreen
->b
.get_device_vendor
= r600_get_device_vendor
;
1177 rscreen
->b
.get_compute_param
= r600_get_compute_param
;
1178 rscreen
->b
.get_paramf
= r600_get_paramf
;
1179 rscreen
->b
.get_timestamp
= r600_get_timestamp
;
1180 rscreen
->b
.fence_finish
= r600_fence_finish
;
1181 rscreen
->b
.fence_reference
= r600_fence_reference
;
1182 rscreen
->b
.resource_destroy
= u_resource_destroy_vtbl
;
1183 rscreen
->b
.resource_from_user_memory
= r600_buffer_from_user_memory
;
1184 rscreen
->b
.query_memory_info
= r600_query_memory_info
;
1186 if (rscreen
->info
.has_uvd
) {
1187 rscreen
->b
.get_video_param
= rvid_get_video_param
;
1188 rscreen
->b
.is_video_format_supported
= rvid_is_format_supported
;
1190 rscreen
->b
.get_video_param
= r600_get_video_param
;
1191 rscreen
->b
.is_video_format_supported
= vl_video_buffer_is_format_supported
;
1194 r600_init_screen_texture_functions(rscreen
);
1195 r600_init_screen_query_functions(rscreen
);
1198 rscreen
->family
= rscreen
->info
.family
;
1199 rscreen
->chip_class
= rscreen
->info
.chip_class
;
1200 rscreen
->debug_flags
= debug_get_flags_option("R600_DEBUG", common_debug_options
, 0);
1202 rscreen
->force_aniso
= MIN2(16, debug_get_num_option("R600_TEX_ANISO", -1));
1203 if (rscreen
->force_aniso
>= 0) {
1204 printf("radeon: Forcing anisotropy filter to %ix\n",
1205 /* round down to a power of two */
1206 1 << util_logbase2(rscreen
->force_aniso
));
1209 util_format_s3tc_init();
1210 pipe_mutex_init(rscreen
->aux_context_lock
);
1211 pipe_mutex_init(rscreen
->gpu_load_mutex
);
1213 if (rscreen
->debug_flags
& DBG_INFO
) {
1214 printf("pci_id = 0x%x\n", rscreen
->info
.pci_id
);
1215 printf("family = %i (%s)\n", rscreen
->info
.family
,
1216 r600_get_chip_name(rscreen
));
1217 printf("chip_class = %i\n", rscreen
->info
.chip_class
);
1218 printf("gart_size = %i MB\n", (int)DIV_ROUND_UP(rscreen
->info
.gart_size
, 1024*1024));
1219 printf("vram_size = %i MB\n", (int)DIV_ROUND_UP(rscreen
->info
.vram_size
, 1024*1024));
1220 printf("max_alloc_size = %i MB\n",
1221 (int)DIV_ROUND_UP(rscreen
->info
.max_alloc_size
, 1024*1024));
1222 printf("has_virtual_memory = %i\n", rscreen
->info
.has_virtual_memory
);
1223 printf("gfx_ib_pad_with_type2 = %i\n", rscreen
->info
.gfx_ib_pad_with_type2
);
1224 printf("has_sdma = %i\n", rscreen
->info
.has_sdma
);
1225 printf("has_uvd = %i\n", rscreen
->info
.has_uvd
);
1226 printf("me_fw_version = %i\n", rscreen
->info
.me_fw_version
);
1227 printf("pfp_fw_version = %i\n", rscreen
->info
.pfp_fw_version
);
1228 printf("ce_fw_version = %i\n", rscreen
->info
.ce_fw_version
);
1229 printf("vce_fw_version = %i\n", rscreen
->info
.vce_fw_version
);
1230 printf("vce_harvest_config = %i\n", rscreen
->info
.vce_harvest_config
);
1231 printf("clock_crystal_freq = %i\n", rscreen
->info
.clock_crystal_freq
);
1232 printf("drm = %i.%i.%i\n", rscreen
->info
.drm_major
,
1233 rscreen
->info
.drm_minor
, rscreen
->info
.drm_patchlevel
);
1234 printf("has_userptr = %i\n", rscreen
->info
.has_userptr
);
1236 printf("r600_max_quad_pipes = %i\n", rscreen
->info
.r600_max_quad_pipes
);
1237 printf("max_shader_clock = %i\n", rscreen
->info
.max_shader_clock
);
1238 printf("num_good_compute_units = %i\n", rscreen
->info
.num_good_compute_units
);
1239 printf("max_se = %i\n", rscreen
->info
.max_se
);
1240 printf("max_sh_per_se = %i\n", rscreen
->info
.max_sh_per_se
);
1242 printf("r600_gb_backend_map = %i\n", rscreen
->info
.r600_gb_backend_map
);
1243 printf("r600_gb_backend_map_valid = %i\n", rscreen
->info
.r600_gb_backend_map_valid
);
1244 printf("r600_num_banks = %i\n", rscreen
->info
.r600_num_banks
);
1245 printf("num_render_backends = %i\n", rscreen
->info
.num_render_backends
);
1246 printf("num_tile_pipes = %i\n", rscreen
->info
.num_tile_pipes
);
1247 printf("pipe_interleave_bytes = %i\n", rscreen
->info
.pipe_interleave_bytes
);
1252 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
)
1254 r600_perfcounters_destroy(rscreen
);
1255 r600_gpu_load_kill_thread(rscreen
);
1257 pipe_mutex_destroy(rscreen
->gpu_load_mutex
);
1258 pipe_mutex_destroy(rscreen
->aux_context_lock
);
1259 rscreen
->aux_context
->destroy(rscreen
->aux_context
);
1261 rscreen
->ws
->destroy(rscreen
->ws
);
1265 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
1268 switch (processor
) {
1269 case PIPE_SHADER_VERTEX
:
1270 return (rscreen
->debug_flags
& DBG_VS
) != 0;
1271 case PIPE_SHADER_TESS_CTRL
:
1272 return (rscreen
->debug_flags
& DBG_TCS
) != 0;
1273 case PIPE_SHADER_TESS_EVAL
:
1274 return (rscreen
->debug_flags
& DBG_TES
) != 0;
1275 case PIPE_SHADER_GEOMETRY
:
1276 return (rscreen
->debug_flags
& DBG_GS
) != 0;
1277 case PIPE_SHADER_FRAGMENT
:
1278 return (rscreen
->debug_flags
& DBG_PS
) != 0;
1279 case PIPE_SHADER_COMPUTE
:
1280 return (rscreen
->debug_flags
& DBG_CS
) != 0;
1286 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
1287 uint64_t offset
, uint64_t size
, unsigned value
,
1288 enum r600_coherency coher
)
1290 struct r600_common_context
*rctx
= (struct r600_common_context
*)rscreen
->aux_context
;
1292 pipe_mutex_lock(rscreen
->aux_context_lock
);
1293 rctx
->clear_buffer(&rctx
->b
, dst
, offset
, size
, value
, coher
);
1294 rscreen
->aux_context
->flush(rscreen
->aux_context
, NULL
, 0);
1295 pipe_mutex_unlock(rscreen
->aux_context_lock
);