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/u_draw_quad.h"
31 #include "util/u_memory.h"
32 #include "util/u_format_s3tc.h"
33 #include "util/u_upload_mgr.h"
34 #include "vl/vl_decoder.h"
35 #include "vl/vl_video_buffer.h"
36 #include "radeon/radeon_video.h"
47 void r600_draw_rectangle(struct blitter_context
*blitter
,
48 int x1
, int y1
, int x2
, int y2
, float depth
,
49 enum blitter_attrib_type type
,
50 const union pipe_color_union
*attrib
)
52 struct r600_common_context
*rctx
=
53 (struct r600_common_context
*)util_blitter_get_pipe(blitter
);
54 struct pipe_viewport_state viewport
;
55 struct pipe_resource
*buf
= NULL
;
59 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
) {
60 util_blitter_draw_rectangle(blitter
, x1
, y1
, x2
, y2
, depth
, type
, attrib
);
64 /* Some operations (like color resolve on r6xx) don't work
65 * with the conventional primitive types.
66 * One that works is PT_RECTLIST, which we use here. */
69 viewport
.scale
[0] = 1.0f
;
70 viewport
.scale
[1] = 1.0f
;
71 viewport
.scale
[2] = 1.0f
;
72 viewport
.translate
[0] = 0.0f
;
73 viewport
.translate
[1] = 0.0f
;
74 viewport
.translate
[2] = 0.0f
;
75 rctx
->b
.set_viewport_states(&rctx
->b
, 0, 1, &viewport
);
77 /* Upload vertices. The hw rectangle has only 3 vertices,
78 * I guess the 4th one is derived from the first 3.
79 * The vertex specification should match u_blitter's vertex element state. */
80 u_upload_alloc(rctx
->uploader
, 0, sizeof(float) * 24, &offset
, &buf
, (void**)&vb
);
97 memcpy(vb
+4, attrib
->f
, sizeof(float)*4);
98 memcpy(vb
+12, attrib
->f
, sizeof(float)*4);
99 memcpy(vb
+20, attrib
->f
, sizeof(float)*4);
103 util_draw_vertex_buffer(&rctx
->b
, NULL
, buf
, blitter
->vb_slot
, offset
,
104 R600_PRIM_RECTANGLE_LIST
, 3, 2);
105 pipe_resource_reference(&buf
, NULL
);
108 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
)
110 /* Flush if there's not enough space. */
111 if ((num_dw
+ ctx
->rings
.dma
.cs
->cdw
) > ctx
->rings
.dma
.cs
->max_dw
) {
112 ctx
->rings
.dma
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
113 assert((num_dw
+ ctx
->rings
.dma
.cs
->cdw
) <= ctx
->rings
.dma
.cs
->max_dw
);
117 static void r600_memory_barrier(struct pipe_context
*ctx
, unsigned flags
)
121 void r600_preflush_suspend_features(struct r600_common_context
*ctx
)
123 /* Disable render condition. */
124 ctx
->saved_render_cond
= NULL
;
125 ctx
->saved_render_cond_cond
= FALSE
;
126 ctx
->saved_render_cond_mode
= 0;
127 if (ctx
->current_render_cond
) {
128 ctx
->saved_render_cond
= ctx
->current_render_cond
;
129 ctx
->saved_render_cond_cond
= ctx
->current_render_cond_cond
;
130 ctx
->saved_render_cond_mode
= ctx
->current_render_cond_mode
;
131 ctx
->b
.render_condition(&ctx
->b
, NULL
, FALSE
, 0);
134 /* suspend queries */
135 ctx
->queries_suspended_for_flush
= false;
136 if (ctx
->num_cs_dw_nontimer_queries_suspend
) {
137 r600_suspend_nontimer_queries(ctx
);
138 r600_suspend_timer_queries(ctx
);
139 ctx
->queries_suspended_for_flush
= true;
142 ctx
->streamout
.suspended
= false;
143 if (ctx
->streamout
.begin_emitted
) {
144 r600_emit_streamout_end(ctx
);
145 ctx
->streamout
.suspended
= true;
149 void r600_postflush_resume_features(struct r600_common_context
*ctx
)
151 if (ctx
->streamout
.suspended
) {
152 ctx
->streamout
.append_bitmask
= ctx
->streamout
.enabled_mask
;
153 r600_streamout_buffers_dirty(ctx
);
157 if (ctx
->queries_suspended_for_flush
) {
158 r600_resume_nontimer_queries(ctx
);
159 r600_resume_timer_queries(ctx
);
162 /* Re-enable render condition. */
163 if (ctx
->saved_render_cond
) {
164 ctx
->b
.render_condition(&ctx
->b
, ctx
->saved_render_cond
,
165 ctx
->saved_render_cond_cond
,
166 ctx
->saved_render_cond_mode
);
170 static void r600_flush_from_st(struct pipe_context
*ctx
,
171 struct pipe_fence_handle
**fence
,
174 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
177 if (flags
& PIPE_FLUSH_END_OF_FRAME
)
178 rflags
|= RADEON_FLUSH_END_OF_FRAME
;
180 if (rctx
->rings
.dma
.cs
) {
181 rctx
->rings
.dma
.flush(rctx
, rflags
, NULL
);
183 rctx
->rings
.gfx
.flush(rctx
, rflags
, fence
);
186 static void r600_flush_dma_ring(void *ctx
, unsigned flags
,
187 struct pipe_fence_handle
**fence
)
189 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
190 struct radeon_winsys_cs
*cs
= rctx
->rings
.dma
.cs
;
196 rctx
->rings
.dma
.flushing
= true;
197 rctx
->ws
->cs_flush(cs
, flags
, fence
, 0);
198 rctx
->rings
.dma
.flushing
= false;
201 static enum pipe_reset_status
r600_get_reset_status(struct pipe_context
*ctx
)
203 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
204 unsigned latest
= rctx
->ws
->query_value(rctx
->ws
,
205 RADEON_GPU_RESET_COUNTER
);
207 if (rctx
->gpu_reset_counter
== latest
)
208 return PIPE_NO_RESET
;
210 rctx
->gpu_reset_counter
= latest
;
211 return PIPE_UNKNOWN_CONTEXT_RESET
;
214 bool r600_common_context_init(struct r600_common_context
*rctx
,
215 struct r600_common_screen
*rscreen
)
217 util_slab_create(&rctx
->pool_transfers
,
218 sizeof(struct r600_transfer
), 64,
219 UTIL_SLAB_SINGLETHREADED
);
221 rctx
->screen
= rscreen
;
222 rctx
->ws
= rscreen
->ws
;
223 rctx
->family
= rscreen
->family
;
224 rctx
->chip_class
= rscreen
->chip_class
;
226 if (rscreen
->family
== CHIP_HAWAII
)
228 else if (rscreen
->chip_class
>= EVERGREEN
)
233 rctx
->b
.transfer_map
= u_transfer_map_vtbl
;
234 rctx
->b
.transfer_flush_region
= u_default_transfer_flush_region
;
235 rctx
->b
.transfer_unmap
= u_transfer_unmap_vtbl
;
236 rctx
->b
.transfer_inline_write
= u_default_transfer_inline_write
;
237 rctx
->b
.memory_barrier
= r600_memory_barrier
;
238 rctx
->b
.flush
= r600_flush_from_st
;
240 if (rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
>= 43) {
241 rctx
->b
.get_device_reset_status
= r600_get_reset_status
;
242 rctx
->gpu_reset_counter
=
243 rctx
->ws
->query_value(rctx
->ws
,
244 RADEON_GPU_RESET_COUNTER
);
247 LIST_INITHEAD(&rctx
->texture_buffers
);
249 r600_init_context_texture_functions(rctx
);
250 r600_streamout_init(rctx
);
251 r600_query_init(rctx
);
252 cayman_init_msaa(&rctx
->b
);
254 rctx
->allocator_so_filled_size
= u_suballocator_create(&rctx
->b
, 4096, 4,
255 0, PIPE_USAGE_DEFAULT
, TRUE
);
256 if (!rctx
->allocator_so_filled_size
)
259 rctx
->uploader
= u_upload_create(&rctx
->b
, 1024 * 1024, 256,
260 PIPE_BIND_INDEX_BUFFER
|
261 PIPE_BIND_CONSTANT_BUFFER
);
265 rctx
->ctx
= rctx
->ws
->ctx_create(rctx
->ws
);
269 if (rscreen
->info
.r600_has_dma
&& !(rscreen
->debug_flags
& DBG_NO_ASYNC_DMA
)) {
270 rctx
->rings
.dma
.cs
= rctx
->ws
->cs_create(rctx
->ctx
, RING_DMA
,
273 rctx
->rings
.dma
.flush
= r600_flush_dma_ring
;
279 void r600_common_context_cleanup(struct r600_common_context
*rctx
)
281 if (rctx
->rings
.gfx
.cs
)
282 rctx
->ws
->cs_destroy(rctx
->rings
.gfx
.cs
);
283 if (rctx
->rings
.dma
.cs
)
284 rctx
->ws
->cs_destroy(rctx
->rings
.dma
.cs
);
286 rctx
->ws
->ctx_destroy(rctx
->ctx
);
288 if (rctx
->uploader
) {
289 u_upload_destroy(rctx
->uploader
);
292 util_slab_destroy(&rctx
->pool_transfers
);
294 if (rctx
->allocator_so_filled_size
) {
295 u_suballocator_destroy(rctx
->allocator_so_filled_size
);
299 void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
301 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
302 struct r600_resource
*rr
= (struct r600_resource
*)r
;
309 * The idea is to compute a gross estimate of memory requirement of
310 * each draw call. After each draw call, memory will be precisely
311 * accounted. So the uncertainty is only on the current draw call.
312 * In practice this gave very good estimate (+/- 10% of the target
315 if (rr
->domains
& RADEON_DOMAIN_GTT
) {
316 rctx
->gtt
+= rr
->buf
->size
;
318 if (rr
->domains
& RADEON_DOMAIN_VRAM
) {
319 rctx
->vram
+= rr
->buf
->size
;
327 static const struct debug_named_value common_debug_options
[] = {
329 { "tex", DBG_TEX
, "Print texture info" },
330 { "texmip", DBG_TEXMIP
, "Print texture info (mipmapped only)" },
331 { "compute", DBG_COMPUTE
, "Print compute info" },
332 { "vm", DBG_VM
, "Print virtual addresses when creating resources" },
333 { "trace_cs", DBG_TRACE_CS
, "Trace cs and write rlockup_<csid>.c file with faulty cs" },
334 { "info", DBG_INFO
, "Print driver information" },
337 { "fs", DBG_FS
, "Print fetch shaders" },
338 { "vs", DBG_VS
, "Print vertex shaders" },
339 { "gs", DBG_GS
, "Print geometry shaders" },
340 { "ps", DBG_PS
, "Print pixel shaders" },
341 { "cs", DBG_CS
, "Print compute shaders" },
342 { "tcs", DBG_TCS
, "Print tessellation control shaders" },
343 { "tes", DBG_TES
, "Print tessellation evaluation shaders" },
344 { "noir", DBG_NO_IR
, "Don't print the LLVM IR"},
345 { "notgsi", DBG_NO_TGSI
, "Don't print the TGSI"},
346 { "noasm", DBG_NO_ASM
, "Don't print disassembled shaders"},
349 { "nodma", DBG_NO_ASYNC_DMA
, "Disable asynchronous DMA" },
350 { "nohyperz", DBG_NO_HYPERZ
, "Disable Hyper-Z" },
351 /* GL uses the word INVALIDATE, gallium uses the word DISCARD */
352 { "noinvalrange", DBG_NO_DISCARD_RANGE
, "Disable handling of INVALIDATE_RANGE map flags" },
353 { "no2d", DBG_NO_2D_TILING
, "Disable 2D tiling" },
354 { "notiling", DBG_NO_TILING
, "Disable tiling" },
355 { "switch_on_eop", DBG_SWITCH_ON_EOP
, "Program WD/IA to switch on end-of-packet." },
356 { "forcedma", DBG_FORCE_DMA
, "Use asynchronous DMA for all operations when possible." },
357 { "precompile", DBG_PRECOMPILE
, "Compile one shader variant at shader creation." },
359 DEBUG_NAMED_VALUE_END
/* must be last */
362 static const char* r600_get_vendor(struct pipe_screen
* pscreen
)
367 static const char* r600_get_device_vendor(struct pipe_screen
* pscreen
)
372 static const char* r600_get_chip_name(struct r600_common_screen
*rscreen
)
374 switch (rscreen
->info
.family
) {
375 case CHIP_R600
: return "AMD R600";
376 case CHIP_RV610
: return "AMD RV610";
377 case CHIP_RV630
: return "AMD RV630";
378 case CHIP_RV670
: return "AMD RV670";
379 case CHIP_RV620
: return "AMD RV620";
380 case CHIP_RV635
: return "AMD RV635";
381 case CHIP_RS780
: return "AMD RS780";
382 case CHIP_RS880
: return "AMD RS880";
383 case CHIP_RV770
: return "AMD RV770";
384 case CHIP_RV730
: return "AMD RV730";
385 case CHIP_RV710
: return "AMD RV710";
386 case CHIP_RV740
: return "AMD RV740";
387 case CHIP_CEDAR
: return "AMD CEDAR";
388 case CHIP_REDWOOD
: return "AMD REDWOOD";
389 case CHIP_JUNIPER
: return "AMD JUNIPER";
390 case CHIP_CYPRESS
: return "AMD CYPRESS";
391 case CHIP_HEMLOCK
: return "AMD HEMLOCK";
392 case CHIP_PALM
: return "AMD PALM";
393 case CHIP_SUMO
: return "AMD SUMO";
394 case CHIP_SUMO2
: return "AMD SUMO2";
395 case CHIP_BARTS
: return "AMD BARTS";
396 case CHIP_TURKS
: return "AMD TURKS";
397 case CHIP_CAICOS
: return "AMD CAICOS";
398 case CHIP_CAYMAN
: return "AMD CAYMAN";
399 case CHIP_ARUBA
: return "AMD ARUBA";
400 case CHIP_TAHITI
: return "AMD TAHITI";
401 case CHIP_PITCAIRN
: return "AMD PITCAIRN";
402 case CHIP_VERDE
: return "AMD CAPE VERDE";
403 case CHIP_OLAND
: return "AMD OLAND";
404 case CHIP_HAINAN
: return "AMD HAINAN";
405 case CHIP_BONAIRE
: return "AMD BONAIRE";
406 case CHIP_KAVERI
: return "AMD KAVERI";
407 case CHIP_KABINI
: return "AMD KABINI";
408 case CHIP_HAWAII
: return "AMD HAWAII";
409 case CHIP_MULLINS
: return "AMD MULLINS";
410 default: return "AMD unknown";
414 static const char* r600_get_name(struct pipe_screen
* pscreen
)
416 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)pscreen
;
418 return rscreen
->renderer_string
;
421 static float r600_get_paramf(struct pipe_screen
* pscreen
,
422 enum pipe_capf param
)
424 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)pscreen
;
427 case PIPE_CAPF_MAX_LINE_WIDTH
:
428 case PIPE_CAPF_MAX_LINE_WIDTH_AA
:
429 case PIPE_CAPF_MAX_POINT_WIDTH
:
430 case PIPE_CAPF_MAX_POINT_WIDTH_AA
:
431 if (rscreen
->family
>= CHIP_CEDAR
)
435 case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY
:
437 case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS
:
439 case PIPE_CAPF_GUARD_BAND_LEFT
:
440 case PIPE_CAPF_GUARD_BAND_TOP
:
441 case PIPE_CAPF_GUARD_BAND_RIGHT
:
442 case PIPE_CAPF_GUARD_BAND_BOTTOM
:
448 static int r600_get_video_param(struct pipe_screen
*screen
,
449 enum pipe_video_profile profile
,
450 enum pipe_video_entrypoint entrypoint
,
451 enum pipe_video_cap param
)
454 case PIPE_VIDEO_CAP_SUPPORTED
:
455 return vl_profile_supported(screen
, profile
, entrypoint
);
456 case PIPE_VIDEO_CAP_NPOT_TEXTURES
:
458 case PIPE_VIDEO_CAP_MAX_WIDTH
:
459 case PIPE_VIDEO_CAP_MAX_HEIGHT
:
460 return vl_video_buffer_max_size(screen
);
461 case PIPE_VIDEO_CAP_PREFERED_FORMAT
:
462 return PIPE_FORMAT_NV12
;
463 case PIPE_VIDEO_CAP_PREFERS_INTERLACED
:
465 case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED
:
467 case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE
:
469 case PIPE_VIDEO_CAP_MAX_LEVEL
:
470 return vl_level_supported(screen
, profile
);
476 const char *r600_get_llvm_processor_name(enum radeon_family family
)
519 case CHIP_TAHITI
: return "tahiti";
520 case CHIP_PITCAIRN
: return "pitcairn";
521 case CHIP_VERDE
: return "verde";
522 case CHIP_OLAND
: return "oland";
523 case CHIP_HAINAN
: return "hainan";
524 case CHIP_BONAIRE
: return "bonaire";
525 case CHIP_KABINI
: return "kabini";
526 case CHIP_KAVERI
: return "kaveri";
527 case CHIP_HAWAII
: return "hawaii";
529 #if HAVE_LLVM >= 0x0305
538 static int r600_get_compute_param(struct pipe_screen
*screen
,
539 enum pipe_compute_cap param
,
542 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
544 //TODO: select these params by asic
546 case PIPE_COMPUTE_CAP_IR_TARGET
: {
549 if (rscreen
->family
<= CHIP_ARUBA
|| HAVE_LLVM
< 0x0306) {
554 switch(rscreen
->family
) {
555 /* Clang < 3.6 is missing Hainan in its list of
556 * GPUs, so we need to use the name of a similar GPU.
558 #if HAVE_LLVM < 0x0306
564 gpu
= r600_get_llvm_processor_name(rscreen
->family
);
568 sprintf(ret
, "%s-%s", gpu
, triple
);
570 /* +2 for dash and terminating NIL byte */
571 return (strlen(triple
) + strlen(gpu
) + 2) * sizeof(char);
573 case PIPE_COMPUTE_CAP_GRID_DIMENSION
:
575 uint64_t *grid_dimension
= ret
;
576 grid_dimension
[0] = 3;
578 return 1 * sizeof(uint64_t);
580 case PIPE_COMPUTE_CAP_MAX_GRID_SIZE
:
582 uint64_t *grid_size
= ret
;
583 grid_size
[0] = 65535;
584 grid_size
[1] = 65535;
587 return 3 * sizeof(uint64_t) ;
589 case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE
:
591 uint64_t *block_size
= ret
;
596 return 3 * sizeof(uint64_t);
598 case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK
:
600 uint64_t *max_threads_per_block
= ret
;
601 *max_threads_per_block
= 256;
603 return sizeof(uint64_t);
605 case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE
:
607 uint64_t *max_global_size
= ret
;
608 uint64_t max_mem_alloc_size
;
610 r600_get_compute_param(screen
,
611 PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
,
612 &max_mem_alloc_size
);
614 /* In OpenCL, the MAX_MEM_ALLOC_SIZE must be at least
615 * 1/4 of the MAX_GLOBAL_SIZE. Since the
616 * MAX_MEM_ALLOC_SIZE is fixed for older kernels,
617 * make sure we never report more than
618 * 4 * MAX_MEM_ALLOC_SIZE.
620 *max_global_size
= MIN2(4 * max_mem_alloc_size
,
621 rscreen
->info
.gart_size
+
622 rscreen
->info
.vram_size
);
624 return sizeof(uint64_t);
626 case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE
:
628 uint64_t *max_local_size
= ret
;
629 /* Value reported by the closed source driver. */
630 *max_local_size
= 32768;
632 return sizeof(uint64_t);
634 case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE
:
636 uint64_t *max_input_size
= ret
;
637 /* Value reported by the closed source driver. */
638 *max_input_size
= 1024;
640 return sizeof(uint64_t);
642 case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE
:
644 uint64_t *max_mem_alloc_size
= ret
;
646 /* XXX: The limit in older kernels is 256 MB. We
647 * should add a query here for newer kernels.
649 *max_mem_alloc_size
= 256 * 1024 * 1024;
651 return sizeof(uint64_t);
653 case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY
:
655 uint32_t *max_clock_frequency
= ret
;
656 *max_clock_frequency
= rscreen
->info
.max_sclk
;
658 return sizeof(uint32_t);
660 case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS
:
662 uint32_t *max_compute_units
= ret
;
663 *max_compute_units
= rscreen
->info
.max_compute_units
;
665 return sizeof(uint32_t);
667 case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED
:
669 uint32_t *images_supported
= ret
;
670 *images_supported
= 0;
672 return sizeof(uint32_t);
673 case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE
:
675 case PIPE_COMPUTE_CAP_SUBGROUP_SIZE
:
677 uint32_t *subgroup_size
= ret
;
678 *subgroup_size
= r600_wavefront_size(rscreen
->family
);
680 return sizeof(uint32_t);
683 fprintf(stderr
, "unknown PIPE_COMPUTE_CAP %d\n", param
);
687 static uint64_t r600_get_timestamp(struct pipe_screen
*screen
)
689 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
691 return 1000000 * rscreen
->ws
->query_value(rscreen
->ws
, RADEON_TIMESTAMP
) /
692 rscreen
->info
.r600_clock_crystal_freq
;
695 static int r600_get_driver_query_info(struct pipe_screen
*screen
,
697 struct pipe_driver_query_info
*info
)
699 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
700 struct pipe_driver_query_info list
[] = {
701 {"num-compilations", R600_QUERY_NUM_COMPILATIONS
, {0}, PIPE_DRIVER_QUERY_TYPE_UINT64
,
702 PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE
},
703 {"num-shaders-created", R600_QUERY_NUM_SHADERS_CREATED
, {0}, PIPE_DRIVER_QUERY_TYPE_UINT64
,
704 PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE
},
705 {"draw-calls", R600_QUERY_DRAW_CALLS
, {0}},
706 {"requested-VRAM", R600_QUERY_REQUESTED_VRAM
, {rscreen
->info
.vram_size
}, PIPE_DRIVER_QUERY_TYPE_BYTES
},
707 {"requested-GTT", R600_QUERY_REQUESTED_GTT
, {rscreen
->info
.gart_size
}, PIPE_DRIVER_QUERY_TYPE_BYTES
},
708 {"buffer-wait-time", R600_QUERY_BUFFER_WAIT_TIME
, {0}, PIPE_DRIVER_QUERY_TYPE_MICROSECONDS
,
709 PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE
},
710 {"num-cs-flushes", R600_QUERY_NUM_CS_FLUSHES
, {0}},
711 {"num-bytes-moved", R600_QUERY_NUM_BYTES_MOVED
, {0}, PIPE_DRIVER_QUERY_TYPE_BYTES
,
712 PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE
},
713 {"VRAM-usage", R600_QUERY_VRAM_USAGE
, {rscreen
->info
.vram_size
}, PIPE_DRIVER_QUERY_TYPE_BYTES
},
714 {"GTT-usage", R600_QUERY_GTT_USAGE
, {rscreen
->info
.gart_size
}, PIPE_DRIVER_QUERY_TYPE_BYTES
},
715 {"temperature", R600_QUERY_GPU_TEMPERATURE
, {100}},
716 {"shader-clock", R600_QUERY_CURRENT_GPU_SCLK
, {0}, PIPE_DRIVER_QUERY_TYPE_HZ
},
717 {"memory-clock", R600_QUERY_CURRENT_GPU_MCLK
, {0}, PIPE_DRIVER_QUERY_TYPE_HZ
},
718 {"GPU-load", R600_QUERY_GPU_LOAD
, {100}}
720 unsigned num_queries
;
722 if (rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
>= 42)
723 num_queries
= Elements(list
);
730 if (index
>= num_queries
)
737 static void r600_fence_reference(struct pipe_screen
*screen
,
738 struct pipe_fence_handle
**ptr
,
739 struct pipe_fence_handle
*fence
)
741 struct radeon_winsys
*rws
= ((struct r600_common_screen
*)screen
)->ws
;
743 rws
->fence_reference(ptr
, fence
);
746 static boolean
r600_fence_finish(struct pipe_screen
*screen
,
747 struct pipe_fence_handle
*fence
,
750 struct radeon_winsys
*rws
= ((struct r600_common_screen
*)screen
)->ws
;
752 return rws
->fence_wait(rws
, fence
, timeout
);
755 static bool r600_interpret_tiling(struct r600_common_screen
*rscreen
,
756 uint32_t tiling_config
)
758 switch ((tiling_config
& 0xe) >> 1) {
760 rscreen
->tiling_info
.num_channels
= 1;
763 rscreen
->tiling_info
.num_channels
= 2;
766 rscreen
->tiling_info
.num_channels
= 4;
769 rscreen
->tiling_info
.num_channels
= 8;
775 switch ((tiling_config
& 0x30) >> 4) {
777 rscreen
->tiling_info
.num_banks
= 4;
780 rscreen
->tiling_info
.num_banks
= 8;
786 switch ((tiling_config
& 0xc0) >> 6) {
788 rscreen
->tiling_info
.group_bytes
= 256;
791 rscreen
->tiling_info
.group_bytes
= 512;
799 static bool evergreen_interpret_tiling(struct r600_common_screen
*rscreen
,
800 uint32_t tiling_config
)
802 switch (tiling_config
& 0xf) {
804 rscreen
->tiling_info
.num_channels
= 1;
807 rscreen
->tiling_info
.num_channels
= 2;
810 rscreen
->tiling_info
.num_channels
= 4;
813 rscreen
->tiling_info
.num_channels
= 8;
819 switch ((tiling_config
& 0xf0) >> 4) {
821 rscreen
->tiling_info
.num_banks
= 4;
824 rscreen
->tiling_info
.num_banks
= 8;
827 rscreen
->tiling_info
.num_banks
= 16;
833 switch ((tiling_config
& 0xf00) >> 8) {
835 rscreen
->tiling_info
.group_bytes
= 256;
838 rscreen
->tiling_info
.group_bytes
= 512;
846 static bool r600_init_tiling(struct r600_common_screen
*rscreen
)
848 uint32_t tiling_config
= rscreen
->info
.r600_tiling_config
;
850 /* set default group bytes, overridden by tiling info ioctl */
851 if (rscreen
->chip_class
<= R700
) {
852 rscreen
->tiling_info
.group_bytes
= 256;
854 rscreen
->tiling_info
.group_bytes
= 512;
860 if (rscreen
->chip_class
<= R700
) {
861 return r600_interpret_tiling(rscreen
, tiling_config
);
863 return evergreen_interpret_tiling(rscreen
, tiling_config
);
867 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
868 const struct pipe_resource
*templ
)
870 if (templ
->target
== PIPE_BUFFER
) {
871 return r600_buffer_create(screen
, templ
, 4096);
873 return r600_texture_create(screen
, templ
);
877 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
878 struct radeon_winsys
*ws
)
880 char llvm_string
[32] = {};
882 ws
->query_info(ws
, &rscreen
->info
);
885 snprintf(llvm_string
, sizeof(llvm_string
),
886 ", LLVM %i.%i.%i", (HAVE_LLVM
>> 8) & 0xff,
887 HAVE_LLVM
& 0xff, MESA_LLVM_VERSION_PATCH
);
889 snprintf(rscreen
->renderer_string
, sizeof(rscreen
->renderer_string
),
890 "%s (DRM %i.%i.%i%s)",
891 r600_get_chip_name(rscreen
), rscreen
->info
.drm_major
,
892 rscreen
->info
.drm_minor
, rscreen
->info
.drm_patchlevel
,
895 rscreen
->b
.get_name
= r600_get_name
;
896 rscreen
->b
.get_vendor
= r600_get_vendor
;
897 rscreen
->b
.get_device_vendor
= r600_get_device_vendor
;
898 rscreen
->b
.get_compute_param
= r600_get_compute_param
;
899 rscreen
->b
.get_paramf
= r600_get_paramf
;
900 rscreen
->b
.get_driver_query_info
= r600_get_driver_query_info
;
901 rscreen
->b
.get_timestamp
= r600_get_timestamp
;
902 rscreen
->b
.fence_finish
= r600_fence_finish
;
903 rscreen
->b
.fence_reference
= r600_fence_reference
;
904 rscreen
->b
.resource_destroy
= u_resource_destroy_vtbl
;
905 rscreen
->b
.resource_from_user_memory
= r600_buffer_from_user_memory
;
907 if (rscreen
->info
.has_uvd
) {
908 rscreen
->b
.get_video_param
= rvid_get_video_param
;
909 rscreen
->b
.is_video_format_supported
= rvid_is_format_supported
;
911 rscreen
->b
.get_video_param
= r600_get_video_param
;
912 rscreen
->b
.is_video_format_supported
= vl_video_buffer_is_format_supported
;
915 r600_init_screen_texture_functions(rscreen
);
918 rscreen
->family
= rscreen
->info
.family
;
919 rscreen
->chip_class
= rscreen
->info
.chip_class
;
920 rscreen
->debug_flags
= debug_get_flags_option("R600_DEBUG", common_debug_options
, 0);
922 if (!r600_init_tiling(rscreen
)) {
925 util_format_s3tc_init();
926 pipe_mutex_init(rscreen
->aux_context_lock
);
927 pipe_mutex_init(rscreen
->gpu_load_mutex
);
929 if (rscreen
->info
.drm_minor
>= 28 && (rscreen
->debug_flags
& DBG_TRACE_CS
)) {
930 rscreen
->trace_bo
= (struct r600_resource
*)pipe_buffer_create(&rscreen
->b
,
934 if (rscreen
->trace_bo
) {
935 rscreen
->trace_ptr
= rscreen
->ws
->buffer_map(rscreen
->trace_bo
->cs_buf
, NULL
,
936 PIPE_TRANSFER_UNSYNCHRONIZED
);
940 if (rscreen
->debug_flags
& DBG_INFO
) {
941 printf("pci_id = 0x%x\n", rscreen
->info
.pci_id
);
942 printf("family = %i\n", rscreen
->info
.family
);
943 printf("chip_class = %i\n", rscreen
->info
.chip_class
);
944 printf("gart_size = %i MB\n", (int)(rscreen
->info
.gart_size
>> 20));
945 printf("vram_size = %i MB\n", (int)(rscreen
->info
.vram_size
>> 20));
946 printf("max_sclk = %i\n", rscreen
->info
.max_sclk
);
947 printf("max_compute_units = %i\n", rscreen
->info
.max_compute_units
);
948 printf("max_se = %i\n", rscreen
->info
.max_se
);
949 printf("max_sh_per_se = %i\n", rscreen
->info
.max_sh_per_se
);
950 printf("drm = %i.%i.%i\n", rscreen
->info
.drm_major
,
951 rscreen
->info
.drm_minor
, rscreen
->info
.drm_patchlevel
);
952 printf("has_uvd = %i\n", rscreen
->info
.has_uvd
);
953 printf("vce_fw_version = %i\n", rscreen
->info
.vce_fw_version
);
954 printf("r600_num_backends = %i\n", rscreen
->info
.r600_num_backends
);
955 printf("r600_clock_crystal_freq = %i\n", rscreen
->info
.r600_clock_crystal_freq
);
956 printf("r600_tiling_config = 0x%x\n", rscreen
->info
.r600_tiling_config
);
957 printf("r600_num_tile_pipes = %i\n", rscreen
->info
.r600_num_tile_pipes
);
958 printf("r600_max_pipes = %i\n", rscreen
->info
.r600_max_pipes
);
959 printf("r600_virtual_address = %i\n", rscreen
->info
.r600_virtual_address
);
960 printf("r600_has_dma = %i\n", rscreen
->info
.r600_has_dma
);
961 printf("r600_backend_map = %i\n", rscreen
->info
.r600_backend_map
);
962 printf("r600_backend_map_valid = %i\n", rscreen
->info
.r600_backend_map_valid
);
963 printf("si_tile_mode_array_valid = %i\n", rscreen
->info
.si_tile_mode_array_valid
);
964 printf("cik_macrotile_mode_array_valid = %i\n", rscreen
->info
.cik_macrotile_mode_array_valid
);
969 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
)
971 r600_gpu_load_kill_thread(rscreen
);
973 pipe_mutex_destroy(rscreen
->gpu_load_mutex
);
974 pipe_mutex_destroy(rscreen
->aux_context_lock
);
975 rscreen
->aux_context
->destroy(rscreen
->aux_context
);
977 if (rscreen
->trace_bo
)
978 pipe_resource_reference((struct pipe_resource
**)&rscreen
->trace_bo
, NULL
);
980 rscreen
->ws
->destroy(rscreen
->ws
);
984 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
985 const struct tgsi_token
*tokens
)
987 /* Compute shader don't have tgsi_tokens */
989 return (rscreen
->debug_flags
& DBG_CS
) != 0;
991 switch (tgsi_get_processor_type(tokens
)) {
992 case TGSI_PROCESSOR_VERTEX
:
993 return (rscreen
->debug_flags
& DBG_VS
) != 0;
994 case TGSI_PROCESSOR_TESS_CTRL
:
995 return (rscreen
->debug_flags
& DBG_TCS
) != 0;
996 case TGSI_PROCESSOR_TESS_EVAL
:
997 return (rscreen
->debug_flags
& DBG_TES
) != 0;
998 case TGSI_PROCESSOR_GEOMETRY
:
999 return (rscreen
->debug_flags
& DBG_GS
) != 0;
1000 case TGSI_PROCESSOR_FRAGMENT
:
1001 return (rscreen
->debug_flags
& DBG_PS
) != 0;
1002 case TGSI_PROCESSOR_COMPUTE
:
1003 return (rscreen
->debug_flags
& DBG_CS
) != 0;
1009 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
1010 unsigned offset
, unsigned size
, unsigned value
,
1011 bool is_framebuffer
)
1013 struct r600_common_context
*rctx
= (struct r600_common_context
*)rscreen
->aux_context
;
1015 pipe_mutex_lock(rscreen
->aux_context_lock
);
1016 rctx
->clear_buffer(&rctx
->b
, dst
, offset
, size
, value
, is_framebuffer
);
1017 rscreen
->aux_context
->flush(rscreen
->aux_context
, NULL
, 0);
1018 pipe_mutex_unlock(rscreen
->aux_context_lock
);