gallium/radeon: add a HUD query for monitoring the CS thread activity
[mesa.git] / src / gallium / drivers / radeon / radeon_winsys.h
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
23
24 #ifndef RADEON_WINSYS_H
25 #define RADEON_WINSYS_H
26
27 /* The public winsys interface header for the radeon driver. */
28
29 #include "pipebuffer/pb_buffer.h"
30
31 #include "amd/common/amd_family.h"
32
33 #define RADEON_FLUSH_ASYNC (1 << 0)
34 #define RADEON_FLUSH_END_OF_FRAME (1 << 1)
35
36 /* Tiling flags. */
37 enum radeon_bo_layout {
38 RADEON_LAYOUT_LINEAR = 0,
39 RADEON_LAYOUT_TILED,
40 RADEON_LAYOUT_SQUARETILED,
41
42 RADEON_LAYOUT_UNKNOWN
43 };
44
45 enum radeon_bo_domain { /* bitfield */
46 RADEON_DOMAIN_GTT = 2,
47 RADEON_DOMAIN_VRAM = 4,
48 RADEON_DOMAIN_VRAM_GTT = RADEON_DOMAIN_VRAM | RADEON_DOMAIN_GTT
49 };
50
51 enum radeon_bo_flag { /* bitfield */
52 RADEON_FLAG_GTT_WC = (1 << 0),
53 RADEON_FLAG_CPU_ACCESS = (1 << 1),
54 RADEON_FLAG_NO_CPU_ACCESS = (1 << 2),
55 RADEON_FLAG_HANDLE = (1 << 3), /* the buffer most not be suballocated */
56 };
57
58 enum radeon_bo_usage { /* bitfield */
59 RADEON_USAGE_READ = 2,
60 RADEON_USAGE_WRITE = 4,
61 RADEON_USAGE_READWRITE = RADEON_USAGE_READ | RADEON_USAGE_WRITE,
62
63 /* The winsys ensures that the CS submission will be scheduled after
64 * previously flushed CSs referencing this BO in a conflicting way.
65 */
66 RADEON_USAGE_SYNCHRONIZED = 8
67 };
68
69 enum ring_type {
70 RING_GFX = 0,
71 RING_COMPUTE,
72 RING_DMA,
73 RING_UVD,
74 RING_VCE,
75 RING_LAST,
76 };
77
78 enum radeon_value_id {
79 RADEON_REQUESTED_VRAM_MEMORY,
80 RADEON_REQUESTED_GTT_MEMORY,
81 RADEON_MAPPED_VRAM,
82 RADEON_MAPPED_GTT,
83 RADEON_BUFFER_WAIT_TIME_NS,
84 RADEON_NUM_MAPPED_BUFFERS,
85 RADEON_TIMESTAMP,
86 RADEON_NUM_GFX_IBS,
87 RADEON_NUM_SDMA_IBS,
88 RADEON_NUM_BYTES_MOVED,
89 RADEON_NUM_EVICTIONS,
90 RADEON_VRAM_USAGE,
91 RADEON_VRAM_VIS_USAGE,
92 RADEON_GTT_USAGE,
93 RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
94 RADEON_CURRENT_SCLK,
95 RADEON_CURRENT_MCLK,
96 RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
97 RADEON_CS_THREAD_TIME,
98 };
99
100 /* Each group of four has the same priority. */
101 enum radeon_bo_priority {
102 RADEON_PRIO_FENCE = 0,
103 RADEON_PRIO_TRACE,
104 RADEON_PRIO_SO_FILLED_SIZE,
105 RADEON_PRIO_QUERY,
106
107 RADEON_PRIO_IB1 = 4, /* main IB submitted to the kernel */
108 RADEON_PRIO_IB2, /* IB executed with INDIRECT_BUFFER */
109 RADEON_PRIO_DRAW_INDIRECT,
110 RADEON_PRIO_INDEX_BUFFER,
111
112 RADEON_PRIO_VCE = 8,
113 RADEON_PRIO_UVD,
114 RADEON_PRIO_SDMA_BUFFER,
115 RADEON_PRIO_SDMA_TEXTURE,
116
117 RADEON_PRIO_CP_DMA = 12,
118
119 RADEON_PRIO_CONST_BUFFER = 16,
120 RADEON_PRIO_DESCRIPTORS,
121 RADEON_PRIO_BORDER_COLORS,
122
123 RADEON_PRIO_SAMPLER_BUFFER = 20,
124 RADEON_PRIO_VERTEX_BUFFER,
125
126 RADEON_PRIO_SHADER_RW_BUFFER = 24,
127 RADEON_PRIO_COMPUTE_GLOBAL,
128
129 RADEON_PRIO_SAMPLER_TEXTURE = 28,
130 RADEON_PRIO_SHADER_RW_IMAGE,
131
132 RADEON_PRIO_SAMPLER_TEXTURE_MSAA = 32,
133
134 RADEON_PRIO_COLOR_BUFFER = 36,
135
136 RADEON_PRIO_DEPTH_BUFFER = 40,
137
138 RADEON_PRIO_COLOR_BUFFER_MSAA = 44,
139
140 RADEON_PRIO_DEPTH_BUFFER_MSAA = 48,
141
142 RADEON_PRIO_CMASK = 52,
143 RADEON_PRIO_DCC,
144 RADEON_PRIO_HTILE,
145 RADEON_PRIO_SHADER_BINARY, /* the hw can't hide instruction cache misses */
146
147 RADEON_PRIO_SHADER_RINGS = 56,
148
149 RADEON_PRIO_SCRATCH_BUFFER = 60,
150 /* 63 is the maximum value */
151 };
152
153 struct winsys_handle;
154 struct radeon_winsys_ctx;
155
156 struct radeon_winsys_cs_chunk {
157 unsigned cdw; /* Number of used dwords. */
158 unsigned max_dw; /* Maximum number of dwords. */
159 uint32_t *buf; /* The base pointer of the chunk. */
160 };
161
162 struct radeon_winsys_cs {
163 struct radeon_winsys_cs_chunk current;
164 struct radeon_winsys_cs_chunk *prev;
165 unsigned num_prev; /* Number of previous chunks. */
166 unsigned max_prev; /* Space in array pointed to by prev. */
167 unsigned prev_dw; /* Total number of dwords in previous chunks. */
168
169 /* Memory usage of the buffer list. These are always 0 for CE and preamble
170 * IBs. */
171 uint64_t used_vram;
172 uint64_t used_gart;
173 };
174
175 struct radeon_info {
176 /* PCI info: domain:bus:dev:func */
177 uint32_t pci_domain;
178 uint32_t pci_bus;
179 uint32_t pci_dev;
180 uint32_t pci_func;
181
182 /* Device info. */
183 uint32_t pci_id;
184 enum radeon_family family;
185 enum chip_class chip_class;
186 uint32_t gart_page_size;
187 uint64_t gart_size;
188 uint64_t vram_size;
189 uint64_t vram_vis_size;
190 uint64_t max_alloc_size;
191 uint32_t min_alloc_size;
192 bool has_dedicated_vram;
193 bool has_virtual_memory;
194 bool gfx_ib_pad_with_type2;
195 bool has_sdma;
196 bool has_uvd;
197 uint32_t uvd_fw_version;
198 uint32_t vce_fw_version;
199 uint32_t me_fw_version;
200 uint32_t pfp_fw_version;
201 uint32_t ce_fw_version;
202 uint32_t vce_harvest_config;
203 uint32_t clock_crystal_freq;
204
205 /* Kernel info. */
206 uint32_t drm_major; /* version */
207 uint32_t drm_minor;
208 uint32_t drm_patchlevel;
209 bool has_userptr;
210
211 /* Shader cores. */
212 uint32_t r600_max_quad_pipes; /* wave size / 16 */
213 uint32_t max_shader_clock;
214 uint32_t num_good_compute_units;
215 uint32_t max_se; /* shader engines */
216 uint32_t max_sh_per_se; /* shader arrays per shader engine */
217
218 /* Render backends (color + depth blocks). */
219 uint32_t r300_num_gb_pipes;
220 uint32_t r300_num_z_pipes;
221 uint32_t r600_gb_backend_map; /* R600 harvest config */
222 bool r600_gb_backend_map_valid;
223 uint32_t r600_num_banks;
224 uint32_t num_render_backends;
225 uint32_t num_tile_pipes; /* pipe count from PIPE_CONFIG */
226 uint32_t pipe_interleave_bytes;
227 uint32_t enabled_rb_mask; /* GCN harvest config */
228
229 /* Tile modes. */
230 uint32_t si_tile_mode_array[32];
231 uint32_t cik_macrotile_mode_array[16];
232 };
233
234 /* Tiling info for display code, DRI sharing, and other data. */
235 struct radeon_bo_metadata {
236 /* Tiling flags describing the texture layout for display code
237 * and DRI sharing.
238 */
239 enum radeon_bo_layout microtile;
240 enum radeon_bo_layout macrotile;
241 unsigned pipe_config;
242 unsigned bankw;
243 unsigned bankh;
244 unsigned tile_split;
245 unsigned mtilea;
246 unsigned num_banks;
247 unsigned stride;
248 bool scanout;
249
250 /* Additional metadata associated with the buffer, in bytes.
251 * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
252 * Supported by amdgpu only.
253 */
254 uint32_t size_metadata;
255 uint32_t metadata[64];
256 };
257
258 enum radeon_feature_id {
259 RADEON_FID_R300_HYPERZ_ACCESS, /* ZMask + HiZ */
260 RADEON_FID_R300_CMASK_ACCESS,
261 };
262
263 #define RADEON_SURF_MAX_LEVELS 15
264
265 enum radeon_surf_mode {
266 RADEON_SURF_MODE_LINEAR_ALIGNED = 1,
267 RADEON_SURF_MODE_1D = 2,
268 RADEON_SURF_MODE_2D = 3,
269 };
270
271 /* These are defined exactly like GB_TILE_MODEn.MICRO_TILE_MODE_NEW. */
272 enum radeon_micro_mode {
273 RADEON_MICRO_MODE_DISPLAY = 0,
274 RADEON_MICRO_MODE_THIN = 1,
275 RADEON_MICRO_MODE_DEPTH = 2,
276 RADEON_MICRO_MODE_ROTATED = 3,
277 };
278
279 /* the first 16 bits are reserved for libdrm_radeon, don't use them */
280 #define RADEON_SURF_SCANOUT (1 << 16)
281 #define RADEON_SURF_ZBUFFER (1 << 17)
282 #define RADEON_SURF_SBUFFER (1 << 18)
283 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
284 /* bits 19 and 20 are reserved for libdrm_radeon, don't use them */
285 #define RADEON_SURF_FMASK (1 << 21)
286 #define RADEON_SURF_DISABLE_DCC (1 << 22)
287 #define RADEON_SURF_TC_COMPATIBLE_HTILE (1 << 23)
288 #define RADEON_SURF_IMPORTED (1 << 24)
289 #define RADEON_SURF_OPTIMIZE_FOR_SPACE (1 << 25)
290
291 struct radeon_surf_level {
292 uint64_t offset;
293 uint64_t slice_size;
294 uint64_t dcc_offset;
295 uint64_t dcc_fast_clear_size;
296 uint16_t nblk_x;
297 uint16_t nblk_y;
298 enum radeon_surf_mode mode;
299 };
300
301 struct radeon_surf {
302 /* Format properties. */
303 unsigned blk_w:4;
304 unsigned blk_h:4;
305 unsigned bpe:5;
306 /* Number of mipmap levels where DCC is enabled starting from level 0.
307 * Non-zero levels may be disabled due to alignment constraints, but not
308 * the first level.
309 */
310 unsigned num_dcc_levels:4;
311 unsigned is_linear:1;
312 uint32_t flags;
313
314 /* These are return values. Some of them can be set by the caller, but
315 * they will be treated as hints (e.g. bankw, bankh) and might be
316 * changed by the calculator.
317 */
318 uint64_t surf_size;
319 uint64_t dcc_size;
320 uint64_t htile_size;
321
322 uint32_t surf_alignment;
323 uint32_t dcc_alignment;
324 uint32_t htile_alignment;
325
326 /* This applies to EG and later. */
327 unsigned bankw:4; /* max 8 */
328 unsigned bankh:4; /* max 8 */
329 unsigned mtilea:4; /* max 8 */
330 unsigned tile_split:13; /* max 4K */
331 unsigned stencil_tile_split:13; /* max 4K */
332 unsigned pipe_config:5; /* max 17 */
333 unsigned num_banks:5; /* max 16 */
334 unsigned macro_tile_index:4; /* max 15 */
335 unsigned micro_tile_mode:3; /* displayable, thin, depth, rotated */
336
337 /* Whether the depth miptree or stencil miptree as used by the DB are
338 * adjusted from their TC compatible form to ensure depth/stencil
339 * compatibility. If either is true, the corresponding plane cannot be
340 * sampled from.
341 */
342 unsigned depth_adjusted:1;
343 unsigned stencil_adjusted:1;
344
345 struct radeon_surf_level level[RADEON_SURF_MAX_LEVELS];
346 struct radeon_surf_level stencil_level[RADEON_SURF_MAX_LEVELS];
347 uint8_t tiling_index[RADEON_SURF_MAX_LEVELS];
348 uint8_t stencil_tiling_index[RADEON_SURF_MAX_LEVELS];
349 };
350
351 struct radeon_bo_list_item {
352 uint64_t bo_size;
353 uint64_t vm_address;
354 uint64_t priority_usage; /* mask of (1 << RADEON_PRIO_*) */
355 };
356
357 struct radeon_winsys {
358 /**
359 * The screen object this winsys was created for
360 */
361 struct pipe_screen *screen;
362
363 /**
364 * Decrement the winsys reference count.
365 *
366 * \param ws The winsys this function is called for.
367 * \return True if the winsys and screen should be destroyed.
368 */
369 bool (*unref)(struct radeon_winsys *ws);
370
371 /**
372 * Destroy this winsys.
373 *
374 * \param ws The winsys this function is called from.
375 */
376 void (*destroy)(struct radeon_winsys *ws);
377
378 /**
379 * Query an info structure from winsys.
380 *
381 * \param ws The winsys this function is called from.
382 * \param info Return structure
383 */
384 void (*query_info)(struct radeon_winsys *ws,
385 struct radeon_info *info);
386
387 /**************************************************************************
388 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
389 *
390 * Remember that gallium gets to choose the interface it needs, and the
391 * window systems must then implement that interface (rather than the
392 * other way around...).
393 *************************************************************************/
394
395 /**
396 * Create a buffer object.
397 *
398 * \param ws The winsys this function is called from.
399 * \param size The size to allocate.
400 * \param alignment An alignment of the buffer in memory.
401 * \param use_reusable_pool Whether the cache buffer manager should be used.
402 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
403 * \return The created buffer object.
404 */
405 struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
406 uint64_t size,
407 unsigned alignment,
408 enum radeon_bo_domain domain,
409 enum radeon_bo_flag flags);
410
411 /**
412 * Map the entire data store of a buffer object into the client's address
413 * space.
414 *
415 * \param buf A winsys buffer object to map.
416 * \param cs A command stream to flush if the buffer is referenced by it.
417 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
418 * \return The pointer at the beginning of the buffer.
419 */
420 void *(*buffer_map)(struct pb_buffer *buf,
421 struct radeon_winsys_cs *cs,
422 enum pipe_transfer_usage usage);
423
424 /**
425 * Unmap a buffer object from the client's address space.
426 *
427 * \param buf A winsys buffer object to unmap.
428 */
429 void (*buffer_unmap)(struct pb_buffer *buf);
430
431 /**
432 * Wait for the buffer and return true if the buffer is not used
433 * by the device.
434 *
435 * The timeout of 0 will only return the status.
436 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
437 * is idle.
438 */
439 bool (*buffer_wait)(struct pb_buffer *buf, uint64_t timeout,
440 enum radeon_bo_usage usage);
441
442 /**
443 * Return buffer metadata.
444 * (tiling info for display code, DRI sharing, and other data)
445 *
446 * \param buf A winsys buffer object to get the flags from.
447 * \param md Metadata
448 */
449 void (*buffer_get_metadata)(struct pb_buffer *buf,
450 struct radeon_bo_metadata *md);
451
452 /**
453 * Set buffer metadata.
454 * (tiling info for display code, DRI sharing, and other data)
455 *
456 * \param buf A winsys buffer object to set the flags for.
457 * \param md Metadata
458 */
459 void (*buffer_set_metadata)(struct pb_buffer *buf,
460 struct radeon_bo_metadata *md);
461
462 /**
463 * Get a winsys buffer from a winsys handle. The internal structure
464 * of the handle is platform-specific and only a winsys should access it.
465 *
466 * \param ws The winsys this function is called from.
467 * \param whandle A winsys handle pointer as was received from a state
468 * tracker.
469 * \param stride The returned buffer stride in bytes.
470 */
471 struct pb_buffer *(*buffer_from_handle)(struct radeon_winsys *ws,
472 struct winsys_handle *whandle,
473 unsigned *stride, unsigned *offset);
474
475 /**
476 * Get a winsys buffer from a user pointer. The resulting buffer can't
477 * be exported. Both pointer and size must be page aligned.
478 *
479 * \param ws The winsys this function is called from.
480 * \param pointer User pointer to turn into a buffer object.
481 * \param Size Size in bytes for the new buffer.
482 */
483 struct pb_buffer *(*buffer_from_ptr)(struct radeon_winsys *ws,
484 void *pointer, uint64_t size);
485
486 /**
487 * Whether the buffer was created from a user pointer.
488 *
489 * \param buf A winsys buffer object
490 * \return whether \p buf was created via buffer_from_ptr
491 */
492 bool (*buffer_is_user_ptr)(struct pb_buffer *buf);
493
494 /**
495 * Get a winsys handle from a winsys buffer. The internal structure
496 * of the handle is platform-specific and only a winsys should access it.
497 *
498 * \param buf A winsys buffer object to get the handle from.
499 * \param whandle A winsys handle pointer.
500 * \param stride A stride of the buffer in bytes, for texturing.
501 * \return true on success.
502 */
503 bool (*buffer_get_handle)(struct pb_buffer *buf,
504 unsigned stride, unsigned offset,
505 unsigned slice_size,
506 struct winsys_handle *whandle);
507
508 /**
509 * Return the virtual address of a buffer.
510 *
511 * When virtual memory is not in use, this is the offset relative to the
512 * relocation base (non-zero for sub-allocated buffers).
513 *
514 * \param buf A winsys buffer object
515 * \return virtual address
516 */
517 uint64_t (*buffer_get_virtual_address)(struct pb_buffer *buf);
518
519 /**
520 * Return the offset of this buffer relative to the relocation base.
521 * This is only non-zero for sub-allocated buffers.
522 *
523 * This is only supported in the radeon winsys, since amdgpu uses virtual
524 * addresses in submissions even for the video engines.
525 *
526 * \param buf A winsys buffer object
527 * \return the offset for relocations
528 */
529 unsigned (*buffer_get_reloc_offset)(struct pb_buffer *buf);
530
531 /**
532 * Query the initial placement of the buffer from the kernel driver.
533 */
534 enum radeon_bo_domain (*buffer_get_initial_domain)(struct pb_buffer *buf);
535
536 /**************************************************************************
537 * Command submission.
538 *
539 * Each pipe context should create its own command stream and submit
540 * commands independently of other contexts.
541 *************************************************************************/
542
543 /**
544 * Create a command submission context.
545 * Various command streams can be submitted to the same context.
546 */
547 struct radeon_winsys_ctx *(*ctx_create)(struct radeon_winsys *ws);
548
549 /**
550 * Destroy a context.
551 */
552 void (*ctx_destroy)(struct radeon_winsys_ctx *ctx);
553
554 /**
555 * Query a GPU reset status.
556 */
557 enum pipe_reset_status (*ctx_query_reset_status)(struct radeon_winsys_ctx *ctx);
558
559 /**
560 * Create a command stream.
561 *
562 * \param ctx The submission context
563 * \param ring_type The ring type (GFX, DMA, UVD)
564 * \param flush Flush callback function associated with the command stream.
565 * \param user User pointer that will be passed to the flush callback.
566 */
567 struct radeon_winsys_cs *(*cs_create)(struct radeon_winsys_ctx *ctx,
568 enum ring_type ring_type,
569 void (*flush)(void *ctx, unsigned flags,
570 struct pipe_fence_handle **fence),
571 void *flush_ctx);
572
573 /**
574 * Add a constant engine IB to a graphics CS. This makes the graphics CS
575 * from "cs_create" a group of two IBs that share a buffer list and are
576 * flushed together.
577 *
578 * The returned constant CS is only a stream for writing packets to the new
579 * IB. Calling other winsys functions with it is not allowed, not even
580 * "cs_destroy".
581 *
582 * In order to add buffers and check memory usage, use the graphics CS.
583 * In order to flush it, use the graphics CS, which will flush both IBs.
584 * Destroying the graphics CS will destroy both of them.
585 *
586 * \param cs The graphics CS from "cs_create" that will hold the buffer
587 * list and will be used for flushing.
588 */
589 struct radeon_winsys_cs *(*cs_add_const_ib)(struct radeon_winsys_cs *cs);
590
591 /**
592 * Add a constant engine preamble IB to a graphics CS. This add an extra IB
593 * in similar manner to cs_add_const_ib. This should always be called after
594 * cs_add_const_ib.
595 *
596 * The returned IB is a constant engine IB that only gets flushed if the
597 * context changed.
598 *
599 * \param cs The graphics CS from "cs_create" that will hold the buffer
600 * list and will be used for flushing.
601 */
602 struct radeon_winsys_cs *(*cs_add_const_preamble_ib)(struct radeon_winsys_cs *cs);
603 /**
604 * Destroy a command stream.
605 *
606 * \param cs A command stream to destroy.
607 */
608 void (*cs_destroy)(struct radeon_winsys_cs *cs);
609
610 /**
611 * Add a buffer. Each buffer used by a CS must be added using this function.
612 *
613 * \param cs Command stream
614 * \param buf Buffer
615 * \param usage Whether the buffer is used for read and/or write.
616 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
617 * \param priority A higher number means a greater chance of being
618 * placed in the requested domain. 15 is the maximum.
619 * \return Buffer index.
620 */
621 unsigned (*cs_add_buffer)(struct radeon_winsys_cs *cs,
622 struct pb_buffer *buf,
623 enum radeon_bo_usage usage,
624 enum radeon_bo_domain domain,
625 enum radeon_bo_priority priority);
626
627 /**
628 * Return the index of an already-added buffer.
629 *
630 * Not supported on amdgpu. Drivers with GPUVM should not care about
631 * buffer indices.
632 *
633 * \param cs Command stream
634 * \param buf Buffer
635 * \return The buffer index, or -1 if the buffer has not been added.
636 */
637 int (*cs_lookup_buffer)(struct radeon_winsys_cs *cs,
638 struct pb_buffer *buf);
639
640 /**
641 * Return true if there is enough memory in VRAM and GTT for the buffers
642 * added so far. If the validation fails, all buffers which have
643 * been added since the last call of cs_validate will be removed and
644 * the CS will be flushed (provided there are still any buffers).
645 *
646 * \param cs A command stream to validate.
647 */
648 bool (*cs_validate)(struct radeon_winsys_cs *cs);
649
650 /**
651 * Check whether the given number of dwords is available in the IB.
652 * Optionally chain a new chunk of the IB if necessary and supported.
653 *
654 * \param cs A command stream.
655 * \param dw Number of CS dwords requested by the caller.
656 */
657 bool (*cs_check_space)(struct radeon_winsys_cs *cs, unsigned dw);
658
659 /**
660 * Return the buffer list.
661 *
662 * This is the buffer list as passed to the kernel, i.e. it only contains
663 * the parent buffers of sub-allocated buffers.
664 *
665 * \param cs Command stream
666 * \param list Returned buffer list. Set to NULL to query the count only.
667 * \return The buffer count.
668 */
669 unsigned (*cs_get_buffer_list)(struct radeon_winsys_cs *cs,
670 struct radeon_bo_list_item *list);
671
672 /**
673 * Flush a command stream.
674 *
675 * \param cs A command stream to flush.
676 * \param flags, RADEON_FLUSH_ASYNC or 0.
677 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
678 * after the CS and is returned through this parameter.
679 * \return Negative POSIX error code or 0 for success.
680 * Asynchronous submissions never return an error.
681 */
682 int (*cs_flush)(struct radeon_winsys_cs *cs,
683 unsigned flags,
684 struct pipe_fence_handle **fence);
685
686 /**
687 * Create a fence before the CS is flushed.
688 * The user must flush manually to complete the initializaton of the fence.
689 * The fence must not be used before the flush.
690 */
691 struct pipe_fence_handle *(*cs_get_next_fence)(struct radeon_winsys_cs *cs);
692
693 /**
694 * Return true if a buffer is referenced by a command stream.
695 *
696 * \param cs A command stream.
697 * \param buf A winsys buffer.
698 */
699 bool (*cs_is_buffer_referenced)(struct radeon_winsys_cs *cs,
700 struct pb_buffer *buf,
701 enum radeon_bo_usage usage);
702
703 /**
704 * Request access to a feature for a command stream.
705 *
706 * \param cs A command stream.
707 * \param fid Feature ID, one of RADEON_FID_*
708 * \param enable Whether to enable or disable the feature.
709 */
710 bool (*cs_request_feature)(struct radeon_winsys_cs *cs,
711 enum radeon_feature_id fid,
712 bool enable);
713 /**
714 * Make sure all asynchronous flush of the cs have completed
715 *
716 * \param cs A command stream.
717 */
718 void (*cs_sync_flush)(struct radeon_winsys_cs *cs);
719
720 /**
721 * Wait for the fence and return true if the fence has been signalled.
722 * The timeout of 0 will only return the status.
723 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
724 * is signalled.
725 */
726 bool (*fence_wait)(struct radeon_winsys *ws,
727 struct pipe_fence_handle *fence,
728 uint64_t timeout);
729
730 /**
731 * Reference counting for fences.
732 */
733 void (*fence_reference)(struct pipe_fence_handle **dst,
734 struct pipe_fence_handle *src);
735
736 /**
737 * Initialize surface
738 *
739 * \param ws The winsys this function is called from.
740 * \param tex Input texture description
741 * \param flags Bitmask of RADEON_SURF_* flags
742 * \param bpe Bytes per pixel, it can be different for Z buffers.
743 * \param mode Preferred tile mode. (linear, 1D, or 2D)
744 * \param surf Output structure
745 */
746 int (*surface_init)(struct radeon_winsys *ws,
747 const struct pipe_resource *tex,
748 unsigned flags, unsigned bpe,
749 enum radeon_surf_mode mode,
750 struct radeon_surf *surf);
751
752 uint64_t (*query_value)(struct radeon_winsys *ws,
753 enum radeon_value_id value);
754
755 bool (*read_registers)(struct radeon_winsys *ws, unsigned reg_offset,
756 unsigned num_registers, uint32_t *out);
757 };
758
759 static inline bool radeon_emitted(struct radeon_winsys_cs *cs, unsigned num_dw)
760 {
761 return cs && (cs->prev_dw + cs->current.cdw > num_dw);
762 }
763
764 static inline void radeon_emit(struct radeon_winsys_cs *cs, uint32_t value)
765 {
766 cs->current.buf[cs->current.cdw++] = value;
767 }
768
769 static inline void radeon_emit_array(struct radeon_winsys_cs *cs,
770 const uint32_t *values, unsigned count)
771 {
772 memcpy(cs->current.buf + cs->current.cdw, values, count * 4);
773 cs->current.cdw += count;
774 }
775
776 #endif