2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 Marek Olšák <maraeo@gmail.com>
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:
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
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. */
24 #ifndef RADEON_WINSYS_H
25 #define RADEON_WINSYS_H
27 /* The public winsys interface header for the radeon driver. */
29 /* R300 features in DRM.
33 * - GB_Z_PEQ_CONFIG on rv350->r4xx
34 * - R500 FG_ALPHA_VALUE
37 * - R500 US_FORMAT regs
38 * - R500 ARGB2101010 colorbuffer
43 #include "pipebuffer/pb_buffer.h"
45 #define RADEON_FLUSH_ASYNC (1 << 0)
46 #define RADEON_FLUSH_KEEP_TILING_FLAGS (1 << 1) /* needs DRM 2.12.0 */
47 #define RADEON_FLUSH_END_OF_FRAME (1 << 2)
50 enum radeon_bo_layout
{
51 RADEON_LAYOUT_LINEAR
= 0,
53 RADEON_LAYOUT_SQUARETILED
,
58 enum radeon_bo_domain
{ /* bitfield */
59 RADEON_DOMAIN_GTT
= 2,
60 RADEON_DOMAIN_VRAM
= 4,
61 RADEON_DOMAIN_VRAM_GTT
= RADEON_DOMAIN_VRAM
| RADEON_DOMAIN_GTT
64 enum radeon_bo_flag
{ /* bitfield */
65 RADEON_FLAG_GTT_WC
= (1 << 0),
66 RADEON_FLAG_CPU_ACCESS
= (1 << 1),
67 RADEON_FLAG_NO_CPU_ACCESS
= (1 << 2),
70 enum radeon_bo_usage
{ /* bitfield */
71 RADEON_USAGE_READ
= 2,
72 RADEON_USAGE_WRITE
= 4,
73 RADEON_USAGE_READWRITE
= RADEON_USAGE_READ
| RADEON_USAGE_WRITE
78 CHIP_R300
, /* R3xx-based cores. */
86 CHIP_R420
, /* R4xx-based cores. */
95 CHIP_RV515
, /* R5xx-based cores. */
166 enum radeon_value_id
{
167 RADEON_REQUESTED_VRAM_MEMORY
,
168 RADEON_REQUESTED_GTT_MEMORY
,
169 RADEON_BUFFER_WAIT_TIME_NS
,
171 RADEON_NUM_CS_FLUSHES
,
172 RADEON_NUM_BYTES_MOVED
,
175 RADEON_GPU_TEMPERATURE
, /* DRM 2.42.0 */
178 RADEON_GPU_RESET_COUNTER
, /* DRM 2.43.0 */
181 /* Each group of four has the same priority. */
182 enum radeon_bo_priority
{
183 RADEON_PRIO_FENCE
= 0,
185 RADEON_PRIO_SO_FILLED_SIZE
,
188 RADEON_PRIO_IB1
= 4, /* main IB submitted to the kernel */
189 RADEON_PRIO_IB2
, /* IB executed with INDIRECT_BUFFER */
190 RADEON_PRIO_DRAW_INDIRECT
,
191 RADEON_PRIO_INDEX_BUFFER
,
193 RADEON_PRIO_CP_DMA
= 8,
195 RADEON_PRIO_VCE
= 12,
197 RADEON_PRIO_SDMA_BUFFER
,
198 RADEON_PRIO_SDMA_TEXTURE
,
200 RADEON_PRIO_USER_SHADER
= 16,
201 RADEON_PRIO_INTERNAL_SHADER
, /* fetch shader, etc. */
205 RADEON_PRIO_CONST_BUFFER
= 24,
206 RADEON_PRIO_DESCRIPTORS
,
207 RADEON_PRIO_BORDER_COLORS
,
209 RADEON_PRIO_SAMPLER_BUFFER
= 28,
210 RADEON_PRIO_VERTEX_BUFFER
,
212 RADEON_PRIO_SHADER_RW_BUFFER
= 32,
213 RADEON_PRIO_RINGS_STREAMOUT
,
214 RADEON_PRIO_SCRATCH_BUFFER
,
215 RADEON_PRIO_COMPUTE_GLOBAL
,
217 RADEON_PRIO_SAMPLER_TEXTURE
= 36,
218 RADEON_PRIO_SHADER_RW_IMAGE
,
220 RADEON_PRIO_SAMPLER_TEXTURE_MSAA
= 40,
222 RADEON_PRIO_COLOR_BUFFER
= 44,
224 RADEON_PRIO_DEPTH_BUFFER
= 48,
226 RADEON_PRIO_COLOR_BUFFER_MSAA
= 52,
228 RADEON_PRIO_DEPTH_BUFFER_MSAA
= 56,
230 RADEON_PRIO_CMASK
= 60,
233 /* 63 is the maximum value */
236 struct winsys_handle
;
237 struct radeon_winsys_cs_handle
;
238 struct radeon_winsys_ctx
;
240 struct radeon_winsys_cs
{
241 unsigned cdw
; /* Number of used dwords. */
242 unsigned max_dw
; /* Maximum number of dwords. */
243 uint32_t *buf
; /* The command buffer. */
244 enum ring_type ring_type
;
249 enum radeon_family family
;
250 enum chip_class chip_class
;
254 uint32_t max_compute_units
;
256 uint32_t max_sh_per_se
;
258 uint32_t drm_major
; /* version */
260 uint32_t drm_patchlevel
;
263 uint32_t vce_fw_version
;
265 bool gfx_ib_pad_with_type2
;
267 uint32_t r300_num_gb_pipes
;
268 uint32_t r300_num_z_pipes
;
270 uint32_t r600_num_backends
;
271 uint32_t r600_clock_crystal_freq
;
272 uint32_t r600_tiling_config
;
273 uint32_t r600_num_tile_pipes
;
274 uint32_t r600_max_pipes
;
275 boolean r600_virtual_address
;
276 boolean r600_has_dma
;
278 uint32_t r600_backend_map
;
279 boolean r600_backend_map_valid
;
281 boolean si_tile_mode_array_valid
;
282 uint32_t si_tile_mode_array
[32];
283 uint32_t si_backend_enabled_mask
;
285 boolean cik_macrotile_mode_array_valid
;
286 uint32_t cik_macrotile_mode_array
[16];
287 uint32_t vce_harvest_config
;
290 enum radeon_feature_id
{
291 RADEON_FID_R300_HYPERZ_ACCESS
, /* ZMask + HiZ */
292 RADEON_FID_R300_CMASK_ACCESS
,
295 #define RADEON_SURF_MAX_LEVEL 32
297 #define RADEON_SURF_TYPE_MASK 0xFF
298 #define RADEON_SURF_TYPE_SHIFT 0
299 #define RADEON_SURF_TYPE_1D 0
300 #define RADEON_SURF_TYPE_2D 1
301 #define RADEON_SURF_TYPE_3D 2
302 #define RADEON_SURF_TYPE_CUBEMAP 3
303 #define RADEON_SURF_TYPE_1D_ARRAY 4
304 #define RADEON_SURF_TYPE_2D_ARRAY 5
305 #define RADEON_SURF_MODE_MASK 0xFF
306 #define RADEON_SURF_MODE_SHIFT 8
307 #define RADEON_SURF_MODE_LINEAR 0
308 #define RADEON_SURF_MODE_LINEAR_ALIGNED 1
309 #define RADEON_SURF_MODE_1D 2
310 #define RADEON_SURF_MODE_2D 3
311 #define RADEON_SURF_SCANOUT (1 << 16)
312 #define RADEON_SURF_ZBUFFER (1 << 17)
313 #define RADEON_SURF_SBUFFER (1 << 18)
314 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
315 #define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
316 #define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
317 #define RADEON_SURF_FMASK (1 << 21)
319 #define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
320 #define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
321 #define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
323 struct radeon_surf_level
{
332 uint32_t pitch_bytes
;
337 /* These are inputs to the calculator. */
350 /* These are return values. Some of them can be set by the caller, but
351 * they will be treated as hints (e.g. bankw, bankh) and might be
352 * changed by the calculator.
355 uint64_t bo_alignment
;
356 /* This applies to EG and later. */
361 uint32_t stencil_tile_split
;
362 uint64_t stencil_offset
;
363 struct radeon_surf_level level
[RADEON_SURF_MAX_LEVEL
];
364 struct radeon_surf_level stencil_level
[RADEON_SURF_MAX_LEVEL
];
365 uint32_t tiling_index
[RADEON_SURF_MAX_LEVEL
];
366 uint32_t stencil_tiling_index
[RADEON_SURF_MAX_LEVEL
];
367 uint32_t pipe_config
;
371 struct radeon_winsys
{
373 * The screen object this winsys was created for
375 struct pipe_screen
*screen
;
378 * Decrement the winsys reference count.
380 * \param ws The winsys this function is called for.
381 * \return True if the winsys and screen should be destroyed.
383 bool (*unref
)(struct radeon_winsys
*ws
);
386 * Destroy this winsys.
388 * \param ws The winsys this function is called from.
390 void (*destroy
)(struct radeon_winsys
*ws
);
393 * Query an info structure from winsys.
395 * \param ws The winsys this function is called from.
396 * \param info Return structure
398 void (*query_info
)(struct radeon_winsys
*ws
,
399 struct radeon_info
*info
);
401 /**************************************************************************
402 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
404 * Remember that gallium gets to choose the interface it needs, and the
405 * window systems must then implement that interface (rather than the
406 * other way around...).
407 *************************************************************************/
410 * Create a buffer object.
412 * \param ws The winsys this function is called from.
413 * \param size The size to allocate.
414 * \param alignment An alignment of the buffer in memory.
415 * \param use_reusable_pool Whether the cache buffer manager should be used.
416 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
417 * \return The created buffer object.
419 struct pb_buffer
*(*buffer_create
)(struct radeon_winsys
*ws
,
422 boolean use_reusable_pool
,
423 enum radeon_bo_domain domain
,
424 enum radeon_bo_flag flags
);
426 struct radeon_winsys_cs_handle
*(*buffer_get_cs_handle
)(
427 struct pb_buffer
*buf
);
430 * Map the entire data store of a buffer object into the client's address
433 * \param buf A winsys buffer object to map.
434 * \param cs A command stream to flush if the buffer is referenced by it.
435 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
436 * \return The pointer at the beginning of the buffer.
438 void *(*buffer_map
)(struct radeon_winsys_cs_handle
*buf
,
439 struct radeon_winsys_cs
*cs
,
440 enum pipe_transfer_usage usage
);
443 * Unmap a buffer object from the client's address space.
445 * \param buf A winsys buffer object to unmap.
447 void (*buffer_unmap
)(struct radeon_winsys_cs_handle
*buf
);
450 * Wait for the buffer and return true if the buffer is not used
453 * The timeout of 0 will only return the status.
454 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
457 bool (*buffer_wait
)(struct pb_buffer
*buf
, uint64_t timeout
,
458 enum radeon_bo_usage usage
);
461 * Return tiling flags describing a memory layout of a buffer object.
463 * \param buf A winsys buffer object to get the flags from.
464 * \param macrotile A pointer to the return value of the microtile flag.
465 * \param microtile A pointer to the return value of the macrotile flag.
467 * \note microtile and macrotile are not bitmasks!
469 void (*buffer_get_tiling
)(struct pb_buffer
*buf
,
470 enum radeon_bo_layout
*microtile
,
471 enum radeon_bo_layout
*macrotile
,
472 unsigned *bankw
, unsigned *bankh
,
473 unsigned *tile_split
,
474 unsigned *stencil_tile_split
,
479 * Set tiling flags describing a memory layout of a buffer object.
481 * \param buf A winsys buffer object to set the flags for.
482 * \param cs A command stream to flush if the buffer is referenced by it.
483 * \param macrotile A macrotile flag.
484 * \param microtile A microtile flag.
485 * \param stride A stride of the buffer in bytes, for texturing.
487 * \note microtile and macrotile are not bitmasks!
489 void (*buffer_set_tiling
)(struct pb_buffer
*buf
,
490 struct radeon_winsys_cs
*rcs
,
491 enum radeon_bo_layout microtile
,
492 enum radeon_bo_layout macrotile
,
493 unsigned pipe_config
,
494 unsigned bankw
, unsigned bankh
,
496 unsigned stencil_tile_split
,
497 unsigned mtilea
, unsigned num_banks
,
502 * Get a winsys buffer from a winsys handle. The internal structure
503 * of the handle is platform-specific and only a winsys should access it.
505 * \param ws The winsys this function is called from.
506 * \param whandle A winsys handle pointer as was received from a state
508 * \param stride The returned buffer stride in bytes.
510 struct pb_buffer
*(*buffer_from_handle
)(struct radeon_winsys
*ws
,
511 struct winsys_handle
*whandle
,
515 * Get a winsys buffer from a user pointer. The resulting buffer can't
516 * be exported. Both pointer and size must be page aligned.
518 * \param ws The winsys this function is called from.
519 * \param pointer User pointer to turn into a buffer object.
520 * \param Size Size in bytes for the new buffer.
522 struct pb_buffer
*(*buffer_from_ptr
)(struct radeon_winsys
*ws
,
523 void *pointer
, unsigned size
);
526 * Get a winsys handle from a winsys buffer. The internal structure
527 * of the handle is platform-specific and only a winsys should access it.
529 * \param buf A winsys buffer object to get the handle from.
530 * \param whandle A winsys handle pointer.
531 * \param stride A stride of the buffer in bytes, for texturing.
532 * \return TRUE on success.
534 boolean (*buffer_get_handle
)(struct pb_buffer
*buf
,
536 struct winsys_handle
*whandle
);
539 * Return the virtual address of a buffer.
541 * \param buf A winsys buffer object
542 * \return virtual address
544 uint64_t (*buffer_get_virtual_address
)(struct radeon_winsys_cs_handle
*buf
);
547 * Query the initial placement of the buffer from the kernel driver.
549 enum radeon_bo_domain (*buffer_get_initial_domain
)(struct radeon_winsys_cs_handle
*buf
);
551 /**************************************************************************
552 * Command submission.
554 * Each pipe context should create its own command stream and submit
555 * commands independently of other contexts.
556 *************************************************************************/
559 * Create a command submission context.
560 * Various command streams can be submitted to the same context.
562 struct radeon_winsys_ctx
*(*ctx_create
)(struct radeon_winsys
*ws
);
567 void (*ctx_destroy
)(struct radeon_winsys_ctx
*ctx
);
570 * Query a GPU reset status.
572 enum pipe_reset_status (*ctx_query_reset_status
)(struct radeon_winsys_ctx
*ctx
);
575 * Create a command stream.
577 * \param ctx The submission context
578 * \param ring_type The ring type (GFX, DMA, UVD)
579 * \param flush Flush callback function associated with the command stream.
580 * \param user User pointer that will be passed to the flush callback.
581 * \param trace_buf Trace buffer when tracing is enabled
583 struct radeon_winsys_cs
*(*cs_create
)(struct radeon_winsys_ctx
*ctx
,
584 enum ring_type ring_type
,
585 void (*flush
)(void *ctx
, unsigned flags
,
586 struct pipe_fence_handle
**fence
),
588 struct radeon_winsys_cs_handle
*trace_buf
);
591 * Destroy a command stream.
593 * \param cs A command stream to destroy.
595 void (*cs_destroy
)(struct radeon_winsys_cs
*cs
);
598 * Add a buffer. Each buffer used by a CS must be added using this function.
600 * \param cs Command stream
602 * \param usage Whether the buffer is used for read and/or write.
603 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
604 * \param priority A higher number means a greater chance of being
605 * placed in the requested domain. 15 is the maximum.
606 * \return Buffer index.
608 unsigned (*cs_add_buffer
)(struct radeon_winsys_cs
*cs
,
609 struct radeon_winsys_cs_handle
*buf
,
610 enum radeon_bo_usage usage
,
611 enum radeon_bo_domain domain
,
612 enum radeon_bo_priority priority
);
615 * Return the index of an already-added buffer.
617 * \param cs Command stream
619 * \return The buffer index, or -1 if the buffer has not been added.
621 int (*cs_lookup_buffer
)(struct radeon_winsys_cs
*cs
,
622 struct radeon_winsys_cs_handle
*buf
);
625 * Return TRUE if there is enough memory in VRAM and GTT for the buffers
626 * added so far. If the validation fails, all buffers which have
627 * been added since the last call of cs_validate will be removed and
628 * the CS will be flushed (provided there are still any buffers).
630 * \param cs A command stream to validate.
632 boolean (*cs_validate
)(struct radeon_winsys_cs
*cs
);
635 * Return TRUE if there is enough memory in VRAM and GTT for the buffers
638 * \param cs A command stream to validate.
639 * \param vram VRAM memory size pending to be use
640 * \param gtt GTT memory size pending to be use
642 boolean (*cs_memory_below_limit
)(struct radeon_winsys_cs
*cs
, uint64_t vram
, uint64_t gtt
);
645 * Flush a command stream.
647 * \param cs A command stream to flush.
648 * \param flags, RADEON_FLUSH_ASYNC or 0.
649 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
650 * after the CS and is returned through this parameter.
651 * \param cs_trace_id A unique identifier of the cs, used for tracing.
653 void (*cs_flush
)(struct radeon_winsys_cs
*cs
,
655 struct pipe_fence_handle
**fence
,
656 uint32_t cs_trace_id
);
659 * Return TRUE if a buffer is referenced by a command stream.
661 * \param cs A command stream.
662 * \param buf A winsys buffer.
664 boolean (*cs_is_buffer_referenced
)(struct radeon_winsys_cs
*cs
,
665 struct radeon_winsys_cs_handle
*buf
,
666 enum radeon_bo_usage usage
);
669 * Request access to a feature for a command stream.
671 * \param cs A command stream.
672 * \param fid Feature ID, one of RADEON_FID_*
673 * \param enable Whether to enable or disable the feature.
675 boolean (*cs_request_feature
)(struct radeon_winsys_cs
*cs
,
676 enum radeon_feature_id fid
,
679 * Make sure all asynchronous flush of the cs have completed
681 * \param cs A command stream.
683 void (*cs_sync_flush
)(struct radeon_winsys_cs
*cs
);
686 * Wait for the fence and return true if the fence has been signalled.
687 * The timeout of 0 will only return the status.
688 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
691 bool (*fence_wait
)(struct radeon_winsys
*ws
,
692 struct pipe_fence_handle
*fence
,
696 * Reference counting for fences.
698 void (*fence_reference
)(struct pipe_fence_handle
**dst
,
699 struct pipe_fence_handle
*src
);
704 * \param ws The winsys this function is called from.
705 * \param surf Surface structure ptr
707 int (*surface_init
)(struct radeon_winsys
*ws
,
708 struct radeon_surf
*surf
);
711 * Find best values for a surface
713 * \param ws The winsys this function is called from.
714 * \param surf Surface structure ptr
716 int (*surface_best
)(struct radeon_winsys
*ws
,
717 struct radeon_surf
*surf
);
719 uint64_t (*query_value
)(struct radeon_winsys
*ws
,
720 enum radeon_value_id value
);
722 bool (*read_registers
)(struct radeon_winsys
*ws
, unsigned reg_offset
,
723 unsigned num_registers
, uint32_t *out
);
727 static inline void radeon_emit(struct radeon_winsys_cs
*cs
, uint32_t value
)
729 cs
->buf
[cs
->cdw
++] = value
;
732 static inline void radeon_emit_array(struct radeon_winsys_cs
*cs
,
733 const uint32_t *values
, unsigned count
)
735 memcpy(cs
->buf
+cs
->cdw
, values
, count
* 4);