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. */
167 enum radeon_value_id
{
168 RADEON_REQUESTED_VRAM_MEMORY
,
169 RADEON_REQUESTED_GTT_MEMORY
,
170 RADEON_BUFFER_WAIT_TIME_NS
,
172 RADEON_NUM_CS_FLUSHES
,
173 RADEON_NUM_BYTES_MOVED
,
176 RADEON_GPU_TEMPERATURE
, /* DRM 2.42.0 */
179 RADEON_GPU_RESET_COUNTER
, /* DRM 2.43.0 */
182 /* Each group of four has the same priority. */
183 enum radeon_bo_priority
{
184 RADEON_PRIO_FENCE
= 0,
186 RADEON_PRIO_SO_FILLED_SIZE
,
189 RADEON_PRIO_IB1
= 4, /* main IB submitted to the kernel */
190 RADEON_PRIO_IB2
, /* IB executed with INDIRECT_BUFFER */
191 RADEON_PRIO_DRAW_INDIRECT
,
192 RADEON_PRIO_INDEX_BUFFER
,
194 RADEON_PRIO_CP_DMA
= 8,
196 RADEON_PRIO_VCE
= 12,
198 RADEON_PRIO_SDMA_BUFFER
,
199 RADEON_PRIO_SDMA_TEXTURE
,
201 RADEON_PRIO_USER_SHADER
= 16,
202 RADEON_PRIO_INTERNAL_SHADER
, /* fetch shader, etc. */
206 RADEON_PRIO_CONST_BUFFER
= 24,
207 RADEON_PRIO_DESCRIPTORS
,
208 RADEON_PRIO_BORDER_COLORS
,
210 RADEON_PRIO_SAMPLER_BUFFER
= 28,
211 RADEON_PRIO_VERTEX_BUFFER
,
213 RADEON_PRIO_SHADER_RW_BUFFER
= 32,
214 RADEON_PRIO_RINGS_STREAMOUT
,
215 RADEON_PRIO_SCRATCH_BUFFER
,
216 RADEON_PRIO_COMPUTE_GLOBAL
,
218 RADEON_PRIO_SAMPLER_TEXTURE
= 36,
219 RADEON_PRIO_SHADER_RW_IMAGE
,
221 RADEON_PRIO_SAMPLER_TEXTURE_MSAA
= 40,
223 RADEON_PRIO_COLOR_BUFFER
= 44,
225 RADEON_PRIO_DEPTH_BUFFER
= 48,
227 RADEON_PRIO_COLOR_BUFFER_MSAA
= 52,
229 RADEON_PRIO_DEPTH_BUFFER_MSAA
= 56,
231 RADEON_PRIO_CMASK
= 60,
234 /* 63 is the maximum value */
237 struct winsys_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
;
250 enum radeon_family family
;
251 enum chip_class chip_class
;
254 boolean has_virtual_memory
;
255 bool gfx_ib_pad_with_type2
;
258 uint32_t vce_fw_version
;
259 uint32_t vce_harvest_config
;
260 uint32_t clock_crystal_freq
;
263 uint32_t drm_major
; /* version */
265 uint32_t drm_patchlevel
;
269 uint32_t r600_max_quad_pipes
; /* wave size / 16 */
270 uint32_t max_shader_clock
;
271 uint32_t num_good_compute_units
;
272 uint32_t max_se
; /* shader engines */
273 uint32_t max_sh_per_se
; /* shader arrays per shader engine */
275 /* Render backends (color + depth blocks). */
276 uint32_t r300_num_gb_pipes
;
277 uint32_t r300_num_z_pipes
;
278 uint32_t r600_gb_backend_map
; /* R600 harvest config */
279 boolean r600_gb_backend_map_valid
;
280 uint32_t r600_num_banks
;
281 uint32_t num_render_backends
;
282 uint32_t num_tile_pipes
; /* pipe count from PIPE_CONFIG */
283 uint32_t pipe_interleave_bytes
;
284 uint32_t enabled_rb_mask
; /* GCN harvest config */
287 boolean si_tile_mode_array_valid
;
288 uint32_t si_tile_mode_array
[32];
289 boolean cik_macrotile_mode_array_valid
;
290 uint32_t cik_macrotile_mode_array
[16];
293 enum radeon_feature_id
{
294 RADEON_FID_R300_HYPERZ_ACCESS
, /* ZMask + HiZ */
295 RADEON_FID_R300_CMASK_ACCESS
,
298 #define RADEON_SURF_MAX_LEVEL 32
300 #define RADEON_SURF_TYPE_MASK 0xFF
301 #define RADEON_SURF_TYPE_SHIFT 0
302 #define RADEON_SURF_TYPE_1D 0
303 #define RADEON_SURF_TYPE_2D 1
304 #define RADEON_SURF_TYPE_3D 2
305 #define RADEON_SURF_TYPE_CUBEMAP 3
306 #define RADEON_SURF_TYPE_1D_ARRAY 4
307 #define RADEON_SURF_TYPE_2D_ARRAY 5
308 #define RADEON_SURF_MODE_MASK 0xFF
309 #define RADEON_SURF_MODE_SHIFT 8
310 #define RADEON_SURF_MODE_LINEAR 0
311 #define RADEON_SURF_MODE_LINEAR_ALIGNED 1
312 #define RADEON_SURF_MODE_1D 2
313 #define RADEON_SURF_MODE_2D 3
314 #define RADEON_SURF_SCANOUT (1 << 16)
315 #define RADEON_SURF_ZBUFFER (1 << 17)
316 #define RADEON_SURF_SBUFFER (1 << 18)
317 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
318 #define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
319 #define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
320 #define RADEON_SURF_FMASK (1 << 21)
322 #define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
323 #define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
324 #define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
326 struct radeon_surf_level
{
335 uint32_t pitch_bytes
;
341 /* These are inputs to the calculator. */
354 /* These are return values. Some of them can be set by the caller, but
355 * they will be treated as hints (e.g. bankw, bankh) and might be
356 * changed by the calculator.
359 uint64_t bo_alignment
;
360 /* This applies to EG and later. */
365 uint32_t stencil_tile_split
;
366 uint64_t stencil_offset
;
367 struct radeon_surf_level level
[RADEON_SURF_MAX_LEVEL
];
368 struct radeon_surf_level stencil_level
[RADEON_SURF_MAX_LEVEL
];
369 uint32_t tiling_index
[RADEON_SURF_MAX_LEVEL
];
370 uint32_t stencil_tiling_index
[RADEON_SURF_MAX_LEVEL
];
371 uint32_t pipe_config
;
375 uint64_t dcc_alignment
;
378 struct radeon_bo_list_item
{
379 struct pb_buffer
*buf
;
381 uint64_t priority_usage
; /* mask of (1 << RADEON_PRIO_*) */
384 struct radeon_winsys
{
386 * The screen object this winsys was created for
388 struct pipe_screen
*screen
;
391 * Decrement the winsys reference count.
393 * \param ws The winsys this function is called for.
394 * \return True if the winsys and screen should be destroyed.
396 bool (*unref
)(struct radeon_winsys
*ws
);
399 * Destroy this winsys.
401 * \param ws The winsys this function is called from.
403 void (*destroy
)(struct radeon_winsys
*ws
);
406 * Query an info structure from winsys.
408 * \param ws The winsys this function is called from.
409 * \param info Return structure
411 void (*query_info
)(struct radeon_winsys
*ws
,
412 struct radeon_info
*info
);
414 /**************************************************************************
415 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
417 * Remember that gallium gets to choose the interface it needs, and the
418 * window systems must then implement that interface (rather than the
419 * other way around...).
420 *************************************************************************/
423 * Create a buffer object.
425 * \param ws The winsys this function is called from.
426 * \param size The size to allocate.
427 * \param alignment An alignment of the buffer in memory.
428 * \param use_reusable_pool Whether the cache buffer manager should be used.
429 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
430 * \return The created buffer object.
432 struct pb_buffer
*(*buffer_create
)(struct radeon_winsys
*ws
,
435 boolean use_reusable_pool
,
436 enum radeon_bo_domain domain
,
437 enum radeon_bo_flag flags
);
440 * Map the entire data store of a buffer object into the client's address
443 * \param buf A winsys buffer object to map.
444 * \param cs A command stream to flush if the buffer is referenced by it.
445 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
446 * \return The pointer at the beginning of the buffer.
448 void *(*buffer_map
)(struct pb_buffer
*buf
,
449 struct radeon_winsys_cs
*cs
,
450 enum pipe_transfer_usage usage
);
453 * Unmap a buffer object from the client's address space.
455 * \param buf A winsys buffer object to unmap.
457 void (*buffer_unmap
)(struct pb_buffer
*buf
);
460 * Wait for the buffer and return true if the buffer is not used
463 * The timeout of 0 will only return the status.
464 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
467 bool (*buffer_wait
)(struct pb_buffer
*buf
, uint64_t timeout
,
468 enum radeon_bo_usage usage
);
471 * Return tiling flags describing a memory layout of a buffer object.
473 * \param buf A winsys buffer object to get the flags from.
474 * \param macrotile A pointer to the return value of the microtile flag.
475 * \param microtile A pointer to the return value of the macrotile flag.
477 * \note microtile and macrotile are not bitmasks!
479 void (*buffer_get_tiling
)(struct pb_buffer
*buf
,
480 enum radeon_bo_layout
*microtile
,
481 enum radeon_bo_layout
*macrotile
,
482 unsigned *bankw
, unsigned *bankh
,
483 unsigned *tile_split
,
484 unsigned *stencil_tile_split
,
489 * Set tiling flags describing a memory layout of a buffer object.
491 * \param buf A winsys buffer object to set the flags for.
492 * \param cs A command stream to flush if the buffer is referenced by it.
493 * \param macrotile A macrotile flag.
494 * \param microtile A microtile flag.
495 * \param stride A stride of the buffer in bytes, for texturing.
497 * \note microtile and macrotile are not bitmasks!
499 void (*buffer_set_tiling
)(struct pb_buffer
*buf
,
500 struct radeon_winsys_cs
*rcs
,
501 enum radeon_bo_layout microtile
,
502 enum radeon_bo_layout macrotile
,
503 unsigned pipe_config
,
504 unsigned bankw
, unsigned bankh
,
506 unsigned stencil_tile_split
,
507 unsigned mtilea
, unsigned num_banks
,
512 * Get a winsys buffer from a winsys handle. The internal structure
513 * of the handle is platform-specific and only a winsys should access it.
515 * \param ws The winsys this function is called from.
516 * \param whandle A winsys handle pointer as was received from a state
518 * \param stride The returned buffer stride in bytes.
520 struct pb_buffer
*(*buffer_from_handle
)(struct radeon_winsys
*ws
,
521 struct winsys_handle
*whandle
,
525 * Get a winsys buffer from a user pointer. The resulting buffer can't
526 * be exported. Both pointer and size must be page aligned.
528 * \param ws The winsys this function is called from.
529 * \param pointer User pointer to turn into a buffer object.
530 * \param Size Size in bytes for the new buffer.
532 struct pb_buffer
*(*buffer_from_ptr
)(struct radeon_winsys
*ws
,
533 void *pointer
, unsigned size
);
536 * Whether the buffer was created from a user pointer.
538 * \param buf A winsys buffer object
539 * \return whether \p buf was created via buffer_from_ptr
541 bool (*buffer_is_user_ptr
)(struct pb_buffer
*buf
);
544 * Get a winsys handle from a winsys buffer. The internal structure
545 * of the handle is platform-specific and only a winsys should access it.
547 * \param buf A winsys buffer object to get the handle from.
548 * \param whandle A winsys handle pointer.
549 * \param stride A stride of the buffer in bytes, for texturing.
550 * \return TRUE on success.
552 boolean (*buffer_get_handle
)(struct pb_buffer
*buf
,
554 struct winsys_handle
*whandle
);
557 * Return the virtual address of a buffer.
559 * \param buf A winsys buffer object
560 * \return virtual address
562 uint64_t (*buffer_get_virtual_address
)(struct pb_buffer
*buf
);
565 * Query the initial placement of the buffer from the kernel driver.
567 enum radeon_bo_domain (*buffer_get_initial_domain
)(struct pb_buffer
*buf
);
569 /**************************************************************************
570 * Command submission.
572 * Each pipe context should create its own command stream and submit
573 * commands independently of other contexts.
574 *************************************************************************/
577 * Create a command submission context.
578 * Various command streams can be submitted to the same context.
580 struct radeon_winsys_ctx
*(*ctx_create
)(struct radeon_winsys
*ws
);
585 void (*ctx_destroy
)(struct radeon_winsys_ctx
*ctx
);
588 * Query a GPU reset status.
590 enum pipe_reset_status (*ctx_query_reset_status
)(struct radeon_winsys_ctx
*ctx
);
593 * Create a command stream.
595 * \param ctx The submission context
596 * \param ring_type The ring type (GFX, DMA, UVD)
597 * \param flush Flush callback function associated with the command stream.
598 * \param user User pointer that will be passed to the flush callback.
599 * \param trace_buf Trace buffer when tracing is enabled
601 struct radeon_winsys_cs
*(*cs_create
)(struct radeon_winsys_ctx
*ctx
,
602 enum ring_type ring_type
,
603 void (*flush
)(void *ctx
, unsigned flags
,
604 struct pipe_fence_handle
**fence
),
606 struct pb_buffer
*trace_buf
);
609 * Destroy a command stream.
611 * \param cs A command stream to destroy.
613 void (*cs_destroy
)(struct radeon_winsys_cs
*cs
);
616 * Add a buffer. Each buffer used by a CS must be added using this function.
618 * \param cs Command stream
620 * \param usage Whether the buffer is used for read and/or write.
621 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
622 * \param priority A higher number means a greater chance of being
623 * placed in the requested domain. 15 is the maximum.
624 * \return Buffer index.
626 unsigned (*cs_add_buffer
)(struct radeon_winsys_cs
*cs
,
627 struct pb_buffer
*buf
,
628 enum radeon_bo_usage usage
,
629 enum radeon_bo_domain domain
,
630 enum radeon_bo_priority priority
);
633 * Return the index of an already-added buffer.
635 * \param cs Command stream
637 * \return The buffer index, or -1 if the buffer has not been added.
639 int (*cs_lookup_buffer
)(struct radeon_winsys_cs
*cs
,
640 struct pb_buffer
*buf
);
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).
648 * \param cs A command stream to validate.
650 boolean (*cs_validate
)(struct radeon_winsys_cs
*cs
);
653 * Return TRUE if there is enough memory in VRAM and GTT for the buffers
656 * \param cs A command stream to validate.
657 * \param vram VRAM memory size pending to be use
658 * \param gtt GTT memory size pending to be use
660 boolean (*cs_memory_below_limit
)(struct radeon_winsys_cs
*cs
, uint64_t vram
, uint64_t gtt
);
663 * Return the buffer list.
665 * \param cs Command stream
666 * \param list Returned buffer list. Set to NULL to query the count only.
667 * \return The buffer count.
669 unsigned (*cs_get_buffer_list
)(struct radeon_winsys_cs
*cs
,
670 struct radeon_bo_list_item
*list
);
673 * Flush a command stream.
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 * \param cs_trace_id A unique identifier of the cs, used for tracing.
681 void (*cs_flush
)(struct radeon_winsys_cs
*cs
,
683 struct pipe_fence_handle
**fence
,
684 uint32_t cs_trace_id
);
687 * Return TRUE if a buffer is referenced by a command stream.
689 * \param cs A command stream.
690 * \param buf A winsys buffer.
692 boolean (*cs_is_buffer_referenced
)(struct radeon_winsys_cs
*cs
,
693 struct pb_buffer
*buf
,
694 enum radeon_bo_usage usage
);
697 * Request access to a feature for a command stream.
699 * \param cs A command stream.
700 * \param fid Feature ID, one of RADEON_FID_*
701 * \param enable Whether to enable or disable the feature.
703 boolean (*cs_request_feature
)(struct radeon_winsys_cs
*cs
,
704 enum radeon_feature_id fid
,
707 * Make sure all asynchronous flush of the cs have completed
709 * \param cs A command stream.
711 void (*cs_sync_flush
)(struct radeon_winsys_cs
*cs
);
714 * Wait for the fence and return true if the fence has been signalled.
715 * The timeout of 0 will only return the status.
716 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
719 bool (*fence_wait
)(struct radeon_winsys
*ws
,
720 struct pipe_fence_handle
*fence
,
724 * Reference counting for fences.
726 void (*fence_reference
)(struct pipe_fence_handle
**dst
,
727 struct pipe_fence_handle
*src
);
732 * \param ws The winsys this function is called from.
733 * \param surf Surface structure ptr
735 int (*surface_init
)(struct radeon_winsys
*ws
,
736 struct radeon_surf
*surf
);
739 * Find best values for a surface
741 * \param ws The winsys this function is called from.
742 * \param surf Surface structure ptr
744 int (*surface_best
)(struct radeon_winsys
*ws
,
745 struct radeon_surf
*surf
);
747 uint64_t (*query_value
)(struct radeon_winsys
*ws
,
748 enum radeon_value_id value
);
750 bool (*read_registers
)(struct radeon_winsys
*ws
, unsigned reg_offset
,
751 unsigned num_registers
, uint32_t *out
);
755 static inline void radeon_emit(struct radeon_winsys_cs
*cs
, uint32_t value
)
757 cs
->buf
[cs
->cdw
++] = value
;
760 static inline void radeon_emit_array(struct radeon_winsys_cs
*cs
,
761 const uint32_t *values
, unsigned count
)
763 memcpy(cs
->buf
+cs
->cdw
, values
, count
* 4);