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 #include "pipebuffer/pb_buffer.h"
31 #include "amd/common/amd_family.h"
33 #define RADEON_FLUSH_ASYNC (1 << 0)
34 #define RADEON_FLUSH_END_OF_FRAME (1 << 1)
37 enum radeon_bo_layout
{
38 RADEON_LAYOUT_LINEAR
= 0,
40 RADEON_LAYOUT_SQUARETILED
,
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
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),
57 enum radeon_bo_usage
{ /* bitfield */
58 RADEON_USAGE_READ
= 2,
59 RADEON_USAGE_WRITE
= 4,
60 RADEON_USAGE_READWRITE
= RADEON_USAGE_READ
| RADEON_USAGE_WRITE
,
62 /* The winsys ensures that the CS submission will be scheduled after
63 * previously flushed CSs referencing this BO in a conflicting way.
65 RADEON_USAGE_SYNCHRONIZED
= 8
77 enum radeon_value_id
{
78 RADEON_REQUESTED_VRAM_MEMORY
,
79 RADEON_REQUESTED_GTT_MEMORY
,
82 RADEON_BUFFER_WAIT_TIME_NS
,
84 RADEON_NUM_CS_FLUSHES
,
85 RADEON_NUM_BYTES_MOVED
,
89 RADEON_GPU_TEMPERATURE
, /* DRM 2.42.0 */
92 RADEON_GPU_RESET_COUNTER
, /* DRM 2.43.0 */
95 /* Each group of four has the same priority. */
96 enum radeon_bo_priority
{
97 RADEON_PRIO_FENCE
= 0,
99 RADEON_PRIO_SO_FILLED_SIZE
,
102 RADEON_PRIO_IB1
= 4, /* main IB submitted to the kernel */
103 RADEON_PRIO_IB2
, /* IB executed with INDIRECT_BUFFER */
104 RADEON_PRIO_DRAW_INDIRECT
,
105 RADEON_PRIO_INDEX_BUFFER
,
109 RADEON_PRIO_SDMA_BUFFER
,
110 RADEON_PRIO_SDMA_TEXTURE
,
112 RADEON_PRIO_CP_DMA
= 12,
114 RADEON_PRIO_CONST_BUFFER
= 16,
115 RADEON_PRIO_DESCRIPTORS
,
116 RADEON_PRIO_BORDER_COLORS
,
118 RADEON_PRIO_SAMPLER_BUFFER
= 20,
119 RADEON_PRIO_VERTEX_BUFFER
,
121 RADEON_PRIO_SHADER_RW_BUFFER
= 24,
122 RADEON_PRIO_COMPUTE_GLOBAL
,
124 RADEON_PRIO_SAMPLER_TEXTURE
= 28,
125 RADEON_PRIO_SHADER_RW_IMAGE
,
127 RADEON_PRIO_SAMPLER_TEXTURE_MSAA
= 32,
129 RADEON_PRIO_COLOR_BUFFER
= 36,
131 RADEON_PRIO_DEPTH_BUFFER
= 40,
133 RADEON_PRIO_COLOR_BUFFER_MSAA
= 44,
135 RADEON_PRIO_DEPTH_BUFFER_MSAA
= 48,
137 RADEON_PRIO_CMASK
= 52,
140 RADEON_PRIO_SHADER_BINARY
, /* the hw can't hide instruction cache misses */
142 RADEON_PRIO_SHADER_RINGS
= 56,
144 RADEON_PRIO_SCRATCH_BUFFER
= 60,
145 /* 63 is the maximum value */
148 struct winsys_handle
;
149 struct radeon_winsys_ctx
;
151 struct radeon_winsys_cs_chunk
{
152 unsigned cdw
; /* Number of used dwords. */
153 unsigned max_dw
; /* Maximum number of dwords. */
154 uint32_t *buf
; /* The base pointer of the chunk. */
157 struct radeon_winsys_cs
{
158 struct radeon_winsys_cs_chunk current
;
159 struct radeon_winsys_cs_chunk
*prev
;
160 unsigned num_prev
; /* Number of previous chunks. */
161 unsigned max_prev
; /* Space in array pointed to by prev. */
162 unsigned prev_dw
; /* Total number of dwords in previous chunks. */
164 /* Memory usage of the buffer list. These are always 0 for CE and preamble
171 /* PCI info: domain:bus:dev:func */
179 enum radeon_family family
;
180 enum chip_class chip_class
;
181 uint32_t gart_page_size
;
184 uint64_t max_alloc_size
;
185 bool has_dedicated_vram
;
186 bool has_virtual_memory
;
187 bool gfx_ib_pad_with_type2
;
190 uint32_t uvd_fw_version
;
191 uint32_t vce_fw_version
;
192 uint32_t me_fw_version
;
193 uint32_t pfp_fw_version
;
194 uint32_t ce_fw_version
;
195 uint32_t vce_harvest_config
;
196 uint32_t clock_crystal_freq
;
199 uint32_t drm_major
; /* version */
201 uint32_t drm_patchlevel
;
205 uint32_t r600_max_quad_pipes
; /* wave size / 16 */
206 uint32_t max_shader_clock
;
207 uint32_t num_good_compute_units
;
208 uint32_t max_se
; /* shader engines */
209 uint32_t max_sh_per_se
; /* shader arrays per shader engine */
211 /* Render backends (color + depth blocks). */
212 uint32_t r300_num_gb_pipes
;
213 uint32_t r300_num_z_pipes
;
214 uint32_t r600_gb_backend_map
; /* R600 harvest config */
215 bool r600_gb_backend_map_valid
;
216 uint32_t r600_num_banks
;
217 uint32_t num_render_backends
;
218 uint32_t num_tile_pipes
; /* pipe count from PIPE_CONFIG */
219 uint32_t pipe_interleave_bytes
;
220 uint32_t enabled_rb_mask
; /* GCN harvest config */
223 uint32_t si_tile_mode_array
[32];
224 uint32_t cik_macrotile_mode_array
[16];
227 /* Tiling info for display code, DRI sharing, and other data. */
228 struct radeon_bo_metadata
{
229 /* Tiling flags describing the texture layout for display code
232 enum radeon_bo_layout microtile
;
233 enum radeon_bo_layout macrotile
;
234 unsigned pipe_config
;
243 /* Additional metadata associated with the buffer, in bytes.
244 * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
245 * Supported by amdgpu only.
247 uint32_t size_metadata
;
248 uint32_t metadata
[64];
251 enum radeon_feature_id
{
252 RADEON_FID_R300_HYPERZ_ACCESS
, /* ZMask + HiZ */
253 RADEON_FID_R300_CMASK_ACCESS
,
256 #define RADEON_SURF_MAX_LEVEL 32
258 #define RADEON_SURF_TYPE_MASK 0xFF
259 #define RADEON_SURF_TYPE_SHIFT 0
260 #define RADEON_SURF_TYPE_1D 0
261 #define RADEON_SURF_TYPE_2D 1
262 #define RADEON_SURF_TYPE_3D 2
263 #define RADEON_SURF_TYPE_CUBEMAP 3
264 #define RADEON_SURF_TYPE_1D_ARRAY 4
265 #define RADEON_SURF_TYPE_2D_ARRAY 5
266 #define RADEON_SURF_MODE_MASK 0xFF
267 #define RADEON_SURF_MODE_SHIFT 8
268 #define RADEON_SURF_MODE_LINEAR_ALIGNED 1
269 #define RADEON_SURF_MODE_1D 2
270 #define RADEON_SURF_MODE_2D 3
271 #define RADEON_SURF_SCANOUT (1 << 16)
272 #define RADEON_SURF_ZBUFFER (1 << 17)
273 #define RADEON_SURF_SBUFFER (1 << 18)
274 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
275 #define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
276 #define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
277 #define RADEON_SURF_FMASK (1 << 21)
278 #define RADEON_SURF_DISABLE_DCC (1 << 22)
280 #define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
281 #define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
282 #define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
284 struct radeon_surf_level
{
293 uint32_t pitch_bytes
;
296 uint64_t dcc_fast_clear_size
;
301 /* These are inputs to the calculator. */
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.
319 uint64_t bo_alignment
;
320 /* This applies to EG and later. */
325 uint32_t stencil_tile_split
;
326 struct radeon_surf_level level
[RADEON_SURF_MAX_LEVEL
];
327 struct radeon_surf_level stencil_level
[RADEON_SURF_MAX_LEVEL
];
328 uint32_t tiling_index
[RADEON_SURF_MAX_LEVEL
];
329 uint32_t stencil_tiling_index
[RADEON_SURF_MAX_LEVEL
];
330 uint32_t pipe_config
;
332 uint32_t macro_tile_index
;
333 uint32_t micro_tile_mode
; /* displayable, thin, depth, rotated */
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
341 bool stencil_adjusted
;
344 uint64_t dcc_alignment
;
347 struct radeon_bo_list_item
{
350 uint64_t priority_usage
; /* mask of (1 << RADEON_PRIO_*) */
353 struct radeon_winsys
{
355 * The screen object this winsys was created for
357 struct pipe_screen
*screen
;
360 * Decrement the winsys reference count.
362 * \param ws The winsys this function is called for.
363 * \return True if the winsys and screen should be destroyed.
365 bool (*unref
)(struct radeon_winsys
*ws
);
368 * Destroy this winsys.
370 * \param ws The winsys this function is called from.
372 void (*destroy
)(struct radeon_winsys
*ws
);
375 * Query an info structure from winsys.
377 * \param ws The winsys this function is called from.
378 * \param info Return structure
380 void (*query_info
)(struct radeon_winsys
*ws
,
381 struct radeon_info
*info
);
383 /**************************************************************************
384 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
386 * Remember that gallium gets to choose the interface it needs, and the
387 * window systems must then implement that interface (rather than the
388 * other way around...).
389 *************************************************************************/
392 * Create a buffer object.
394 * \param ws The winsys this function is called from.
395 * \param size The size to allocate.
396 * \param alignment An alignment of the buffer in memory.
397 * \param use_reusable_pool Whether the cache buffer manager should be used.
398 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
399 * \return The created buffer object.
401 struct pb_buffer
*(*buffer_create
)(struct radeon_winsys
*ws
,
404 enum radeon_bo_domain domain
,
405 enum radeon_bo_flag flags
);
408 * Map the entire data store of a buffer object into the client's address
411 * \param buf A winsys buffer object to map.
412 * \param cs A command stream to flush if the buffer is referenced by it.
413 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
414 * \return The pointer at the beginning of the buffer.
416 void *(*buffer_map
)(struct pb_buffer
*buf
,
417 struct radeon_winsys_cs
*cs
,
418 enum pipe_transfer_usage usage
);
421 * Unmap a buffer object from the client's address space.
423 * \param buf A winsys buffer object to unmap.
425 void (*buffer_unmap
)(struct pb_buffer
*buf
);
428 * Wait for the buffer and return true if the buffer is not used
431 * The timeout of 0 will only return the status.
432 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
435 bool (*buffer_wait
)(struct pb_buffer
*buf
, uint64_t timeout
,
436 enum radeon_bo_usage usage
);
439 * Return buffer metadata.
440 * (tiling info for display code, DRI sharing, and other data)
442 * \param buf A winsys buffer object to get the flags from.
445 void (*buffer_get_metadata
)(struct pb_buffer
*buf
,
446 struct radeon_bo_metadata
*md
);
449 * Set buffer metadata.
450 * (tiling info for display code, DRI sharing, and other data)
452 * \param buf A winsys buffer object to set the flags for.
455 void (*buffer_set_metadata
)(struct pb_buffer
*buf
,
456 struct radeon_bo_metadata
*md
);
459 * Get a winsys buffer from a winsys handle. The internal structure
460 * of the handle is platform-specific and only a winsys should access it.
462 * \param ws The winsys this function is called from.
463 * \param whandle A winsys handle pointer as was received from a state
465 * \param stride The returned buffer stride in bytes.
467 struct pb_buffer
*(*buffer_from_handle
)(struct radeon_winsys
*ws
,
468 struct winsys_handle
*whandle
,
469 unsigned *stride
, unsigned *offset
);
472 * Get a winsys buffer from a user pointer. The resulting buffer can't
473 * be exported. Both pointer and size must be page aligned.
475 * \param ws The winsys this function is called from.
476 * \param pointer User pointer to turn into a buffer object.
477 * \param Size Size in bytes for the new buffer.
479 struct pb_buffer
*(*buffer_from_ptr
)(struct radeon_winsys
*ws
,
480 void *pointer
, uint64_t size
);
483 * Whether the buffer was created from a user pointer.
485 * \param buf A winsys buffer object
486 * \return whether \p buf was created via buffer_from_ptr
488 bool (*buffer_is_user_ptr
)(struct pb_buffer
*buf
);
491 * Get a winsys handle from a winsys buffer. The internal structure
492 * of the handle is platform-specific and only a winsys should access it.
494 * \param buf A winsys buffer object to get the handle from.
495 * \param whandle A winsys handle pointer.
496 * \param stride A stride of the buffer in bytes, for texturing.
497 * \return true on success.
499 bool (*buffer_get_handle
)(struct pb_buffer
*buf
,
500 unsigned stride
, unsigned offset
,
502 struct winsys_handle
*whandle
);
505 * Return the virtual address of a buffer.
507 * \param buf A winsys buffer object
508 * \return virtual address
510 uint64_t (*buffer_get_virtual_address
)(struct pb_buffer
*buf
);
513 * Query the initial placement of the buffer from the kernel driver.
515 enum radeon_bo_domain (*buffer_get_initial_domain
)(struct pb_buffer
*buf
);
517 /**************************************************************************
518 * Command submission.
520 * Each pipe context should create its own command stream and submit
521 * commands independently of other contexts.
522 *************************************************************************/
525 * Create a command submission context.
526 * Various command streams can be submitted to the same context.
528 struct radeon_winsys_ctx
*(*ctx_create
)(struct radeon_winsys
*ws
);
533 void (*ctx_destroy
)(struct radeon_winsys_ctx
*ctx
);
536 * Query a GPU reset status.
538 enum pipe_reset_status (*ctx_query_reset_status
)(struct radeon_winsys_ctx
*ctx
);
541 * Create a command stream.
543 * \param ctx The submission context
544 * \param ring_type The ring type (GFX, DMA, UVD)
545 * \param flush Flush callback function associated with the command stream.
546 * \param user User pointer that will be passed to the flush callback.
548 struct radeon_winsys_cs
*(*cs_create
)(struct radeon_winsys_ctx
*ctx
,
549 enum ring_type ring_type
,
550 void (*flush
)(void *ctx
, unsigned flags
,
551 struct pipe_fence_handle
**fence
),
555 * Add a constant engine IB to a graphics CS. This makes the graphics CS
556 * from "cs_create" a group of two IBs that share a buffer list and are
559 * The returned constant CS is only a stream for writing packets to the new
560 * IB. Calling other winsys functions with it is not allowed, not even
563 * In order to add buffers and check memory usage, use the graphics CS.
564 * In order to flush it, use the graphics CS, which will flush both IBs.
565 * Destroying the graphics CS will destroy both of them.
567 * \param cs The graphics CS from "cs_create" that will hold the buffer
568 * list and will be used for flushing.
570 struct radeon_winsys_cs
*(*cs_add_const_ib
)(struct radeon_winsys_cs
*cs
);
573 * Add a constant engine preamble IB to a graphics CS. This add an extra IB
574 * in similar manner to cs_add_const_ib. This should always be called after
577 * The returned IB is a constant engine IB that only gets flushed if the
580 * \param cs The graphics CS from "cs_create" that will hold the buffer
581 * list and will be used for flushing.
583 struct radeon_winsys_cs
*(*cs_add_const_preamble_ib
)(struct radeon_winsys_cs
*cs
);
585 * Destroy a command stream.
587 * \param cs A command stream to destroy.
589 void (*cs_destroy
)(struct radeon_winsys_cs
*cs
);
592 * Add a buffer. Each buffer used by a CS must be added using this function.
594 * \param cs Command stream
596 * \param usage Whether the buffer is used for read and/or write.
597 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
598 * \param priority A higher number means a greater chance of being
599 * placed in the requested domain. 15 is the maximum.
600 * \return Buffer index.
602 unsigned (*cs_add_buffer
)(struct radeon_winsys_cs
*cs
,
603 struct pb_buffer
*buf
,
604 enum radeon_bo_usage usage
,
605 enum radeon_bo_domain domain
,
606 enum radeon_bo_priority priority
);
609 * Return the index of an already-added buffer.
611 * Not supported on amdgpu. Drivers with GPUVM should not care about
614 * \param cs Command stream
616 * \return The buffer index, or -1 if the buffer has not been added.
618 int (*cs_lookup_buffer
)(struct radeon_winsys_cs
*cs
,
619 struct pb_buffer
*buf
);
622 * Return true if there is enough memory in VRAM and GTT for the buffers
623 * added so far. If the validation fails, all buffers which have
624 * been added since the last call of cs_validate will be removed and
625 * the CS will be flushed (provided there are still any buffers).
627 * \param cs A command stream to validate.
629 bool (*cs_validate
)(struct radeon_winsys_cs
*cs
);
632 * Check whether the given number of dwords is available in the IB.
633 * Optionally chain a new chunk of the IB if necessary and supported.
635 * \param cs A command stream.
636 * \param dw Number of CS dwords requested by the caller.
638 bool (*cs_check_space
)(struct radeon_winsys_cs
*cs
, unsigned dw
);
641 * Return the buffer list.
643 * \param cs Command stream
644 * \param list Returned buffer list. Set to NULL to query the count only.
645 * \return The buffer count.
647 unsigned (*cs_get_buffer_list
)(struct radeon_winsys_cs
*cs
,
648 struct radeon_bo_list_item
*list
);
651 * Flush a command stream.
653 * \param cs A command stream to flush.
654 * \param flags, RADEON_FLUSH_ASYNC or 0.
655 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
656 * after the CS and is returned through this parameter.
657 * \return Negative POSIX error code or 0 for success.
658 * Asynchronous submissions never return an error.
660 int (*cs_flush
)(struct radeon_winsys_cs
*cs
,
662 struct pipe_fence_handle
**fence
);
665 * Create a fence before the CS is flushed.
666 * The user must flush manually to complete the initializaton of the fence.
667 * The fence must not be used before the flush.
669 struct pipe_fence_handle
*(*cs_get_next_fence
)(struct radeon_winsys_cs
*cs
);
672 * Return true if a buffer is referenced by a command stream.
674 * \param cs A command stream.
675 * \param buf A winsys buffer.
677 bool (*cs_is_buffer_referenced
)(struct radeon_winsys_cs
*cs
,
678 struct pb_buffer
*buf
,
679 enum radeon_bo_usage usage
);
682 * Request access to a feature for a command stream.
684 * \param cs A command stream.
685 * \param fid Feature ID, one of RADEON_FID_*
686 * \param enable Whether to enable or disable the feature.
688 bool (*cs_request_feature
)(struct radeon_winsys_cs
*cs
,
689 enum radeon_feature_id fid
,
692 * Make sure all asynchronous flush of the cs have completed
694 * \param cs A command stream.
696 void (*cs_sync_flush
)(struct radeon_winsys_cs
*cs
);
699 * Wait for the fence and return true if the fence has been signalled.
700 * The timeout of 0 will only return the status.
701 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
704 bool (*fence_wait
)(struct radeon_winsys
*ws
,
705 struct pipe_fence_handle
*fence
,
709 * Reference counting for fences.
711 void (*fence_reference
)(struct pipe_fence_handle
**dst
,
712 struct pipe_fence_handle
*src
);
717 * \param ws The winsys this function is called from.
718 * \param surf Surface structure ptr
720 int (*surface_init
)(struct radeon_winsys
*ws
,
721 struct radeon_surf
*surf
);
724 * Find best values for a surface
726 * \param ws The winsys this function is called from.
727 * \param surf Surface structure ptr
729 int (*surface_best
)(struct radeon_winsys
*ws
,
730 struct radeon_surf
*surf
);
732 uint64_t (*query_value
)(struct radeon_winsys
*ws
,
733 enum radeon_value_id value
);
735 bool (*read_registers
)(struct radeon_winsys
*ws
, unsigned reg_offset
,
736 unsigned num_registers
, uint32_t *out
);
739 static inline bool radeon_emitted(struct radeon_winsys_cs
*cs
, unsigned num_dw
)
741 return cs
&& (cs
->prev_dw
+ cs
->current
.cdw
> num_dw
);
744 static inline void radeon_emit(struct radeon_winsys_cs
*cs
, uint32_t value
)
746 cs
->current
.buf
[cs
->current
.cdw
++] = value
;
749 static inline void radeon_emit_array(struct radeon_winsys_cs
*cs
,
750 const uint32_t *values
, unsigned count
)
752 memcpy(cs
->current
.buf
+ cs
->current
.cdw
, values
, count
* 4);
753 cs
->current
.cdw
+= count
;