radeonsi: implement TC-compatible HTILE
[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_TIMESTAMP,
85 RADEON_NUM_CS_FLUSHES,
86 RADEON_NUM_BYTES_MOVED,
87 RADEON_NUM_EVICTIONS,
88 RADEON_VRAM_USAGE,
89 RADEON_GTT_USAGE,
90 RADEON_GPU_TEMPERATURE, /* DRM 2.42.0 */
91 RADEON_CURRENT_SCLK,
92 RADEON_CURRENT_MCLK,
93 RADEON_GPU_RESET_COUNTER, /* DRM 2.43.0 */
94 };
95
96 /* Each group of four has the same priority. */
97 enum radeon_bo_priority {
98 RADEON_PRIO_FENCE = 0,
99 RADEON_PRIO_TRACE,
100 RADEON_PRIO_SO_FILLED_SIZE,
101 RADEON_PRIO_QUERY,
102
103 RADEON_PRIO_IB1 = 4, /* main IB submitted to the kernel */
104 RADEON_PRIO_IB2, /* IB executed with INDIRECT_BUFFER */
105 RADEON_PRIO_DRAW_INDIRECT,
106 RADEON_PRIO_INDEX_BUFFER,
107
108 RADEON_PRIO_VCE = 8,
109 RADEON_PRIO_UVD,
110 RADEON_PRIO_SDMA_BUFFER,
111 RADEON_PRIO_SDMA_TEXTURE,
112
113 RADEON_PRIO_CP_DMA = 12,
114
115 RADEON_PRIO_CONST_BUFFER = 16,
116 RADEON_PRIO_DESCRIPTORS,
117 RADEON_PRIO_BORDER_COLORS,
118
119 RADEON_PRIO_SAMPLER_BUFFER = 20,
120 RADEON_PRIO_VERTEX_BUFFER,
121
122 RADEON_PRIO_SHADER_RW_BUFFER = 24,
123 RADEON_PRIO_COMPUTE_GLOBAL,
124
125 RADEON_PRIO_SAMPLER_TEXTURE = 28,
126 RADEON_PRIO_SHADER_RW_IMAGE,
127
128 RADEON_PRIO_SAMPLER_TEXTURE_MSAA = 32,
129
130 RADEON_PRIO_COLOR_BUFFER = 36,
131
132 RADEON_PRIO_DEPTH_BUFFER = 40,
133
134 RADEON_PRIO_COLOR_BUFFER_MSAA = 44,
135
136 RADEON_PRIO_DEPTH_BUFFER_MSAA = 48,
137
138 RADEON_PRIO_CMASK = 52,
139 RADEON_PRIO_DCC,
140 RADEON_PRIO_HTILE,
141 RADEON_PRIO_SHADER_BINARY, /* the hw can't hide instruction cache misses */
142
143 RADEON_PRIO_SHADER_RINGS = 56,
144
145 RADEON_PRIO_SCRATCH_BUFFER = 60,
146 /* 63 is the maximum value */
147 };
148
149 struct winsys_handle;
150 struct radeon_winsys_ctx;
151
152 struct radeon_winsys_cs_chunk {
153 unsigned cdw; /* Number of used dwords. */
154 unsigned max_dw; /* Maximum number of dwords. */
155 uint32_t *buf; /* The base pointer of the chunk. */
156 };
157
158 struct radeon_winsys_cs {
159 struct radeon_winsys_cs_chunk current;
160 struct radeon_winsys_cs_chunk *prev;
161 unsigned num_prev; /* Number of previous chunks. */
162 unsigned max_prev; /* Space in array pointed to by prev. */
163 unsigned prev_dw; /* Total number of dwords in previous chunks. */
164
165 /* Memory usage of the buffer list. These are always 0 for CE and preamble
166 * IBs. */
167 uint64_t used_vram;
168 uint64_t used_gart;
169 };
170
171 struct radeon_info {
172 /* PCI info: domain:bus:dev:func */
173 uint32_t pci_domain;
174 uint32_t pci_bus;
175 uint32_t pci_dev;
176 uint32_t pci_func;
177
178 /* Device info. */
179 uint32_t pci_id;
180 enum radeon_family family;
181 enum chip_class chip_class;
182 uint32_t gart_page_size;
183 uint64_t gart_size;
184 uint64_t vram_size;
185 uint64_t max_alloc_size;
186 uint32_t min_alloc_size;
187 bool has_dedicated_vram;
188 bool has_virtual_memory;
189 bool gfx_ib_pad_with_type2;
190 bool has_sdma;
191 bool has_uvd;
192 uint32_t uvd_fw_version;
193 uint32_t vce_fw_version;
194 uint32_t me_fw_version;
195 uint32_t pfp_fw_version;
196 uint32_t ce_fw_version;
197 uint32_t vce_harvest_config;
198 uint32_t clock_crystal_freq;
199
200 /* Kernel info. */
201 uint32_t drm_major; /* version */
202 uint32_t drm_minor;
203 uint32_t drm_patchlevel;
204 bool has_userptr;
205
206 /* Shader cores. */
207 uint32_t r600_max_quad_pipes; /* wave size / 16 */
208 uint32_t max_shader_clock;
209 uint32_t num_good_compute_units;
210 uint32_t max_se; /* shader engines */
211 uint32_t max_sh_per_se; /* shader arrays per shader engine */
212
213 /* Render backends (color + depth blocks). */
214 uint32_t r300_num_gb_pipes;
215 uint32_t r300_num_z_pipes;
216 uint32_t r600_gb_backend_map; /* R600 harvest config */
217 bool r600_gb_backend_map_valid;
218 uint32_t r600_num_banks;
219 uint32_t num_render_backends;
220 uint32_t num_tile_pipes; /* pipe count from PIPE_CONFIG */
221 uint32_t pipe_interleave_bytes;
222 uint32_t enabled_rb_mask; /* GCN harvest config */
223
224 /* Tile modes. */
225 uint32_t si_tile_mode_array[32];
226 uint32_t cik_macrotile_mode_array[16];
227 };
228
229 /* Tiling info for display code, DRI sharing, and other data. */
230 struct radeon_bo_metadata {
231 /* Tiling flags describing the texture layout for display code
232 * and DRI sharing.
233 */
234 enum radeon_bo_layout microtile;
235 enum radeon_bo_layout macrotile;
236 unsigned pipe_config;
237 unsigned bankw;
238 unsigned bankh;
239 unsigned tile_split;
240 unsigned mtilea;
241 unsigned num_banks;
242 unsigned stride;
243 bool scanout;
244
245 /* Additional metadata associated with the buffer, in bytes.
246 * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
247 * Supported by amdgpu only.
248 */
249 uint32_t size_metadata;
250 uint32_t metadata[64];
251 };
252
253 enum radeon_feature_id {
254 RADEON_FID_R300_HYPERZ_ACCESS, /* ZMask + HiZ */
255 RADEON_FID_R300_CMASK_ACCESS,
256 };
257
258 #define RADEON_SURF_MAX_LEVEL 32
259
260 #define RADEON_SURF_TYPE_MASK 0xFF
261 #define RADEON_SURF_TYPE_SHIFT 0
262 #define RADEON_SURF_TYPE_1D 0
263 #define RADEON_SURF_TYPE_2D 1
264 #define RADEON_SURF_TYPE_3D 2
265 #define RADEON_SURF_TYPE_CUBEMAP 3
266 #define RADEON_SURF_TYPE_1D_ARRAY 4
267 #define RADEON_SURF_TYPE_2D_ARRAY 5
268 #define RADEON_SURF_MODE_MASK 0xFF
269 #define RADEON_SURF_MODE_SHIFT 8
270 #define RADEON_SURF_MODE_LINEAR_ALIGNED 1
271 #define RADEON_SURF_MODE_1D 2
272 #define RADEON_SURF_MODE_2D 3
273 #define RADEON_SURF_SCANOUT (1 << 16)
274 #define RADEON_SURF_ZBUFFER (1 << 17)
275 #define RADEON_SURF_SBUFFER (1 << 18)
276 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
277 #define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
278 #define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
279 #define RADEON_SURF_FMASK (1 << 21)
280 #define RADEON_SURF_DISABLE_DCC (1 << 22)
281 #define RADEON_SURF_TC_COMPATIBLE_HTILE (1 << 23)
282
283 #define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
284 #define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
285 #define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
286
287 struct radeon_surf_level {
288 uint64_t offset;
289 uint64_t slice_size;
290 uint32_t npix_x;
291 uint32_t npix_y;
292 uint32_t npix_z;
293 uint32_t nblk_x;
294 uint32_t nblk_y;
295 uint32_t nblk_z;
296 uint32_t pitch_bytes;
297 uint32_t mode;
298 uint64_t dcc_offset;
299 uint64_t dcc_fast_clear_size;
300 bool dcc_enabled;
301 };
302
303 struct radeon_surf {
304 /* These are inputs to the calculator. */
305 uint32_t npix_x;
306 uint32_t npix_y;
307 uint32_t npix_z;
308 uint32_t blk_w;
309 uint32_t blk_h;
310 uint32_t blk_d;
311 uint32_t array_size;
312 uint32_t last_level;
313 uint32_t bpe;
314 uint32_t nsamples;
315 uint32_t flags;
316
317 /* These are return values. Some of them can be set by the caller, but
318 * they will be treated as hints (e.g. bankw, bankh) and might be
319 * changed by the calculator.
320 */
321 uint64_t bo_size;
322 uint64_t bo_alignment;
323 /* This applies to EG and later. */
324 uint32_t bankw;
325 uint32_t bankh;
326 uint32_t mtilea;
327 uint32_t tile_split;
328 uint32_t stencil_tile_split;
329 struct radeon_surf_level level[RADEON_SURF_MAX_LEVEL];
330 struct radeon_surf_level stencil_level[RADEON_SURF_MAX_LEVEL];
331 uint32_t tiling_index[RADEON_SURF_MAX_LEVEL];
332 uint32_t stencil_tiling_index[RADEON_SURF_MAX_LEVEL];
333 uint32_t pipe_config;
334 uint32_t num_banks;
335 uint32_t macro_tile_index;
336 uint32_t micro_tile_mode; /* displayable, thin, depth, rotated */
337
338 /* Whether the depth miptree or stencil miptree as used by the DB are
339 * adjusted from their TC compatible form to ensure depth/stencil
340 * compatibility. If either is true, the corresponding plane cannot be
341 * sampled from.
342 */
343 bool depth_adjusted;
344 bool stencil_adjusted;
345
346 uint64_t dcc_size;
347 uint64_t dcc_alignment;
348 /* TC-compatible HTILE only. */
349 uint64_t htile_size;
350 uint64_t htile_alignment;
351 };
352
353 struct radeon_bo_list_item {
354 uint64_t bo_size;
355 uint64_t vm_address;
356 uint64_t priority_usage; /* mask of (1 << RADEON_PRIO_*) */
357 };
358
359 struct radeon_winsys {
360 /**
361 * The screen object this winsys was created for
362 */
363 struct pipe_screen *screen;
364
365 /**
366 * Decrement the winsys reference count.
367 *
368 * \param ws The winsys this function is called for.
369 * \return True if the winsys and screen should be destroyed.
370 */
371 bool (*unref)(struct radeon_winsys *ws);
372
373 /**
374 * Destroy this winsys.
375 *
376 * \param ws The winsys this function is called from.
377 */
378 void (*destroy)(struct radeon_winsys *ws);
379
380 /**
381 * Query an info structure from winsys.
382 *
383 * \param ws The winsys this function is called from.
384 * \param info Return structure
385 */
386 void (*query_info)(struct radeon_winsys *ws,
387 struct radeon_info *info);
388
389 /**************************************************************************
390 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
391 *
392 * Remember that gallium gets to choose the interface it needs, and the
393 * window systems must then implement that interface (rather than the
394 * other way around...).
395 *************************************************************************/
396
397 /**
398 * Create a buffer object.
399 *
400 * \param ws The winsys this function is called from.
401 * \param size The size to allocate.
402 * \param alignment An alignment of the buffer in memory.
403 * \param use_reusable_pool Whether the cache buffer manager should be used.
404 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
405 * \return The created buffer object.
406 */
407 struct pb_buffer *(*buffer_create)(struct radeon_winsys *ws,
408 uint64_t size,
409 unsigned alignment,
410 enum radeon_bo_domain domain,
411 enum radeon_bo_flag flags);
412
413 /**
414 * Map the entire data store of a buffer object into the client's address
415 * space.
416 *
417 * \param buf A winsys buffer object to map.
418 * \param cs A command stream to flush if the buffer is referenced by it.
419 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
420 * \return The pointer at the beginning of the buffer.
421 */
422 void *(*buffer_map)(struct pb_buffer *buf,
423 struct radeon_winsys_cs *cs,
424 enum pipe_transfer_usage usage);
425
426 /**
427 * Unmap a buffer object from the client's address space.
428 *
429 * \param buf A winsys buffer object to unmap.
430 */
431 void (*buffer_unmap)(struct pb_buffer *buf);
432
433 /**
434 * Wait for the buffer and return true if the buffer is not used
435 * by the device.
436 *
437 * The timeout of 0 will only return the status.
438 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
439 * is idle.
440 */
441 bool (*buffer_wait)(struct pb_buffer *buf, uint64_t timeout,
442 enum radeon_bo_usage usage);
443
444 /**
445 * Return buffer metadata.
446 * (tiling info for display code, DRI sharing, and other data)
447 *
448 * \param buf A winsys buffer object to get the flags from.
449 * \param md Metadata
450 */
451 void (*buffer_get_metadata)(struct pb_buffer *buf,
452 struct radeon_bo_metadata *md);
453
454 /**
455 * Set buffer metadata.
456 * (tiling info for display code, DRI sharing, and other data)
457 *
458 * \param buf A winsys buffer object to set the flags for.
459 * \param md Metadata
460 */
461 void (*buffer_set_metadata)(struct pb_buffer *buf,
462 struct radeon_bo_metadata *md);
463
464 /**
465 * Get a winsys buffer from a winsys handle. The internal structure
466 * of the handle is platform-specific and only a winsys should access it.
467 *
468 * \param ws The winsys this function is called from.
469 * \param whandle A winsys handle pointer as was received from a state
470 * tracker.
471 * \param stride The returned buffer stride in bytes.
472 */
473 struct pb_buffer *(*buffer_from_handle)(struct radeon_winsys *ws,
474 struct winsys_handle *whandle,
475 unsigned *stride, unsigned *offset);
476
477 /**
478 * Get a winsys buffer from a user pointer. The resulting buffer can't
479 * be exported. Both pointer and size must be page aligned.
480 *
481 * \param ws The winsys this function is called from.
482 * \param pointer User pointer to turn into a buffer object.
483 * \param Size Size in bytes for the new buffer.
484 */
485 struct pb_buffer *(*buffer_from_ptr)(struct radeon_winsys *ws,
486 void *pointer, uint64_t size);
487
488 /**
489 * Whether the buffer was created from a user pointer.
490 *
491 * \param buf A winsys buffer object
492 * \return whether \p buf was created via buffer_from_ptr
493 */
494 bool (*buffer_is_user_ptr)(struct pb_buffer *buf);
495
496 /**
497 * Get a winsys handle from a winsys buffer. The internal structure
498 * of the handle is platform-specific and only a winsys should access it.
499 *
500 * \param buf A winsys buffer object to get the handle from.
501 * \param whandle A winsys handle pointer.
502 * \param stride A stride of the buffer in bytes, for texturing.
503 * \return true on success.
504 */
505 bool (*buffer_get_handle)(struct pb_buffer *buf,
506 unsigned stride, unsigned offset,
507 unsigned slice_size,
508 struct winsys_handle *whandle);
509
510 /**
511 * Return the virtual address of a buffer.
512 *
513 * When virtual memory is not in use, this is the offset relative to the
514 * relocation base (non-zero for sub-allocated buffers).
515 *
516 * \param buf A winsys buffer object
517 * \return virtual address
518 */
519 uint64_t (*buffer_get_virtual_address)(struct pb_buffer *buf);
520
521 /**
522 * Return the offset of this buffer relative to the relocation base.
523 * This is only non-zero for sub-allocated buffers.
524 *
525 * This is only supported in the radeon winsys, since amdgpu uses virtual
526 * addresses in submissions even for the video engines.
527 *
528 * \param buf A winsys buffer object
529 * \return the offset for relocations
530 */
531 unsigned (*buffer_get_reloc_offset)(struct pb_buffer *buf);
532
533 /**
534 * Query the initial placement of the buffer from the kernel driver.
535 */
536 enum radeon_bo_domain (*buffer_get_initial_domain)(struct pb_buffer *buf);
537
538 /**************************************************************************
539 * Command submission.
540 *
541 * Each pipe context should create its own command stream and submit
542 * commands independently of other contexts.
543 *************************************************************************/
544
545 /**
546 * Create a command submission context.
547 * Various command streams can be submitted to the same context.
548 */
549 struct radeon_winsys_ctx *(*ctx_create)(struct radeon_winsys *ws);
550
551 /**
552 * Destroy a context.
553 */
554 void (*ctx_destroy)(struct radeon_winsys_ctx *ctx);
555
556 /**
557 * Query a GPU reset status.
558 */
559 enum pipe_reset_status (*ctx_query_reset_status)(struct radeon_winsys_ctx *ctx);
560
561 /**
562 * Create a command stream.
563 *
564 * \param ctx The submission context
565 * \param ring_type The ring type (GFX, DMA, UVD)
566 * \param flush Flush callback function associated with the command stream.
567 * \param user User pointer that will be passed to the flush callback.
568 */
569 struct radeon_winsys_cs *(*cs_create)(struct radeon_winsys_ctx *ctx,
570 enum ring_type ring_type,
571 void (*flush)(void *ctx, unsigned flags,
572 struct pipe_fence_handle **fence),
573 void *flush_ctx);
574
575 /**
576 * Add a constant engine IB to a graphics CS. This makes the graphics CS
577 * from "cs_create" a group of two IBs that share a buffer list and are
578 * flushed together.
579 *
580 * The returned constant CS is only a stream for writing packets to the new
581 * IB. Calling other winsys functions with it is not allowed, not even
582 * "cs_destroy".
583 *
584 * In order to add buffers and check memory usage, use the graphics CS.
585 * In order to flush it, use the graphics CS, which will flush both IBs.
586 * Destroying the graphics CS will destroy both of them.
587 *
588 * \param cs The graphics CS from "cs_create" that will hold the buffer
589 * list and will be used for flushing.
590 */
591 struct radeon_winsys_cs *(*cs_add_const_ib)(struct radeon_winsys_cs *cs);
592
593 /**
594 * Add a constant engine preamble IB to a graphics CS. This add an extra IB
595 * in similar manner to cs_add_const_ib. This should always be called after
596 * cs_add_const_ib.
597 *
598 * The returned IB is a constant engine IB that only gets flushed if the
599 * context changed.
600 *
601 * \param cs The graphics CS from "cs_create" that will hold the buffer
602 * list and will be used for flushing.
603 */
604 struct radeon_winsys_cs *(*cs_add_const_preamble_ib)(struct radeon_winsys_cs *cs);
605 /**
606 * Destroy a command stream.
607 *
608 * \param cs A command stream to destroy.
609 */
610 void (*cs_destroy)(struct radeon_winsys_cs *cs);
611
612 /**
613 * Add a buffer. Each buffer used by a CS must be added using this function.
614 *
615 * \param cs Command stream
616 * \param buf Buffer
617 * \param usage Whether the buffer is used for read and/or write.
618 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
619 * \param priority A higher number means a greater chance of being
620 * placed in the requested domain. 15 is the maximum.
621 * \return Buffer index.
622 */
623 unsigned (*cs_add_buffer)(struct radeon_winsys_cs *cs,
624 struct pb_buffer *buf,
625 enum radeon_bo_usage usage,
626 enum radeon_bo_domain domain,
627 enum radeon_bo_priority priority);
628
629 /**
630 * Return the index of an already-added buffer.
631 *
632 * Not supported on amdgpu. Drivers with GPUVM should not care about
633 * buffer indices.
634 *
635 * \param cs Command stream
636 * \param buf Buffer
637 * \return The buffer index, or -1 if the buffer has not been added.
638 */
639 int (*cs_lookup_buffer)(struct radeon_winsys_cs *cs,
640 struct pb_buffer *buf);
641
642 /**
643 * Return true if there is enough memory in VRAM and GTT for the buffers
644 * added so far. If the validation fails, all buffers which have
645 * been added since the last call of cs_validate will be removed and
646 * the CS will be flushed (provided there are still any buffers).
647 *
648 * \param cs A command stream to validate.
649 */
650 bool (*cs_validate)(struct radeon_winsys_cs *cs);
651
652 /**
653 * Check whether the given number of dwords is available in the IB.
654 * Optionally chain a new chunk of the IB if necessary and supported.
655 *
656 * \param cs A command stream.
657 * \param dw Number of CS dwords requested by the caller.
658 */
659 bool (*cs_check_space)(struct radeon_winsys_cs *cs, unsigned dw);
660
661 /**
662 * Return the buffer list.
663 *
664 * This is the buffer list as passed to the kernel, i.e. it only contains
665 * the parent buffers of sub-allocated buffers.
666 *
667 * \param cs Command stream
668 * \param list Returned buffer list. Set to NULL to query the count only.
669 * \return The buffer count.
670 */
671 unsigned (*cs_get_buffer_list)(struct radeon_winsys_cs *cs,
672 struct radeon_bo_list_item *list);
673
674 /**
675 * Flush a command stream.
676 *
677 * \param cs A command stream to flush.
678 * \param flags, RADEON_FLUSH_ASYNC or 0.
679 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
680 * after the CS and is returned through this parameter.
681 * \return Negative POSIX error code or 0 for success.
682 * Asynchronous submissions never return an error.
683 */
684 int (*cs_flush)(struct radeon_winsys_cs *cs,
685 unsigned flags,
686 struct pipe_fence_handle **fence);
687
688 /**
689 * Create a fence before the CS is flushed.
690 * The user must flush manually to complete the initializaton of the fence.
691 * The fence must not be used before the flush.
692 */
693 struct pipe_fence_handle *(*cs_get_next_fence)(struct radeon_winsys_cs *cs);
694
695 /**
696 * Return true if a buffer is referenced by a command stream.
697 *
698 * \param cs A command stream.
699 * \param buf A winsys buffer.
700 */
701 bool (*cs_is_buffer_referenced)(struct radeon_winsys_cs *cs,
702 struct pb_buffer *buf,
703 enum radeon_bo_usage usage);
704
705 /**
706 * Request access to a feature for a command stream.
707 *
708 * \param cs A command stream.
709 * \param fid Feature ID, one of RADEON_FID_*
710 * \param enable Whether to enable or disable the feature.
711 */
712 bool (*cs_request_feature)(struct radeon_winsys_cs *cs,
713 enum radeon_feature_id fid,
714 bool enable);
715 /**
716 * Make sure all asynchronous flush of the cs have completed
717 *
718 * \param cs A command stream.
719 */
720 void (*cs_sync_flush)(struct radeon_winsys_cs *cs);
721
722 /**
723 * Wait for the fence and return true if the fence has been signalled.
724 * The timeout of 0 will only return the status.
725 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
726 * is signalled.
727 */
728 bool (*fence_wait)(struct radeon_winsys *ws,
729 struct pipe_fence_handle *fence,
730 uint64_t timeout);
731
732 /**
733 * Reference counting for fences.
734 */
735 void (*fence_reference)(struct pipe_fence_handle **dst,
736 struct pipe_fence_handle *src);
737
738 /**
739 * Initialize surface
740 *
741 * \param ws The winsys this function is called from.
742 * \param surf Surface structure ptr
743 */
744 int (*surface_init)(struct radeon_winsys *ws,
745 struct radeon_surf *surf);
746
747 /**
748 * Find best values for a surface
749 *
750 * \param ws The winsys this function is called from.
751 * \param surf Surface structure ptr
752 */
753 int (*surface_best)(struct radeon_winsys *ws,
754 struct radeon_surf *surf);
755
756 uint64_t (*query_value)(struct radeon_winsys *ws,
757 enum radeon_value_id value);
758
759 bool (*read_registers)(struct radeon_winsys *ws, unsigned reg_offset,
760 unsigned num_registers, uint32_t *out);
761 };
762
763 static inline bool radeon_emitted(struct radeon_winsys_cs *cs, unsigned num_dw)
764 {
765 return cs && (cs->prev_dw + cs->current.cdw > num_dw);
766 }
767
768 static inline void radeon_emit(struct radeon_winsys_cs *cs, uint32_t value)
769 {
770 cs->current.buf[cs->current.cdw++] = value;
771 }
772
773 static inline void radeon_emit_array(struct radeon_winsys_cs *cs,
774 const uint32_t *values, unsigned count)
775 {
776 memcpy(cs->current.buf + cs->current.cdw, values, count * 4);
777 cs->current.cdw += count;
778 }
779
780 #endif