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