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 #define RADEON_FLUSH_ASYNC (1 << 0)
32 #define RADEON_FLUSH_END_OF_FRAME (1 << 1)
35 enum radeon_bo_layout
{
36 RADEON_LAYOUT_LINEAR
= 0,
38 RADEON_LAYOUT_SQUARETILED
,
43 enum radeon_bo_domain
{ /* bitfield */
44 RADEON_DOMAIN_GTT
= 2,
45 RADEON_DOMAIN_VRAM
= 4,
46 RADEON_DOMAIN_VRAM_GTT
= RADEON_DOMAIN_VRAM
| RADEON_DOMAIN_GTT
49 enum radeon_bo_flag
{ /* bitfield */
50 RADEON_FLAG_GTT_WC
= (1 << 0),
51 RADEON_FLAG_CPU_ACCESS
= (1 << 1),
52 RADEON_FLAG_NO_CPU_ACCESS
= (1 << 2),
55 enum radeon_bo_usage
{ /* bitfield */
56 RADEON_USAGE_READ
= 2,
57 RADEON_USAGE_WRITE
= 4,
58 RADEON_USAGE_READWRITE
= RADEON_USAGE_READ
| RADEON_USAGE_WRITE
63 CHIP_R300
, /* R3xx-based cores. */
71 CHIP_R420
, /* R4xx-based cores. */
80 CHIP_RV515
, /* R5xx-based cores. */
154 enum radeon_value_id
{
155 RADEON_REQUESTED_VRAM_MEMORY
,
156 RADEON_REQUESTED_GTT_MEMORY
,
159 RADEON_BUFFER_WAIT_TIME_NS
,
161 RADEON_NUM_CS_FLUSHES
,
162 RADEON_NUM_BYTES_MOVED
,
165 RADEON_GPU_TEMPERATURE
, /* DRM 2.42.0 */
168 RADEON_GPU_RESET_COUNTER
, /* DRM 2.43.0 */
171 /* Each group of four has the same priority. */
172 enum radeon_bo_priority
{
173 RADEON_PRIO_FENCE
= 0,
175 RADEON_PRIO_SO_FILLED_SIZE
,
178 RADEON_PRIO_IB1
= 4, /* main IB submitted to the kernel */
179 RADEON_PRIO_IB2
, /* IB executed with INDIRECT_BUFFER */
180 RADEON_PRIO_DRAW_INDIRECT
,
181 RADEON_PRIO_INDEX_BUFFER
,
185 RADEON_PRIO_SDMA_BUFFER
,
186 RADEON_PRIO_SDMA_TEXTURE
,
188 RADEON_PRIO_CP_DMA
= 12,
189 RADEON_PRIO_USER_SHADER
,
190 RADEON_PRIO_INTERNAL_SHADER
, /* fetch shader, etc. */
192 RADEON_PRIO_CONST_BUFFER
= 16,
193 RADEON_PRIO_DESCRIPTORS
,
194 RADEON_PRIO_BORDER_COLORS
,
196 RADEON_PRIO_SAMPLER_BUFFER
= 20,
197 RADEON_PRIO_VERTEX_BUFFER
,
199 RADEON_PRIO_SHADER_RW_BUFFER
= 24,
200 RADEON_PRIO_SCRATCH_BUFFER
,
201 RADEON_PRIO_COMPUTE_GLOBAL
,
203 RADEON_PRIO_SAMPLER_TEXTURE
= 28,
204 RADEON_PRIO_SHADER_RW_IMAGE
,
206 RADEON_PRIO_SAMPLER_TEXTURE_MSAA
= 32,
208 RADEON_PRIO_COLOR_BUFFER
= 36,
210 RADEON_PRIO_DEPTH_BUFFER
= 40,
212 RADEON_PRIO_COLOR_BUFFER_MSAA
= 44,
214 RADEON_PRIO_DEPTH_BUFFER_MSAA
= 48,
216 RADEON_PRIO_CMASK
= 52,
219 RADEON_PRIO_SHADER_RINGS
,
220 /* 63 is the maximum value */
223 struct winsys_handle
;
224 struct radeon_winsys_ctx
;
226 struct radeon_winsys_cs_chunk
{
227 unsigned cdw
; /* Number of used dwords. */
228 unsigned max_dw
; /* Maximum number of dwords. */
229 uint32_t *buf
; /* The base pointer of the chunk. */
232 struct radeon_winsys_cs
{
233 struct radeon_winsys_cs_chunk current
;
234 struct radeon_winsys_cs_chunk
*prev
;
235 unsigned num_prev
; /* Number of previous chunks. */
236 unsigned max_prev
; /* Space in array pointed to by prev. */
237 unsigned prev_dw
; /* Total number of dwords in previous chunks. */
239 /* Memory usage of the buffer list. These are always 0 for CE and preamble
246 /* PCI info: domain:bus:dev:func */
254 enum radeon_family family
;
255 enum chip_class chip_class
;
256 uint32_t gart_page_size
;
259 uint64_t max_alloc_size
;
260 bool has_dedicated_vram
;
261 bool has_virtual_memory
;
262 bool gfx_ib_pad_with_type2
;
265 uint32_t uvd_fw_version
;
266 uint32_t vce_fw_version
;
267 uint32_t me_fw_version
;
268 uint32_t pfp_fw_version
;
269 uint32_t ce_fw_version
;
270 uint32_t vce_harvest_config
;
271 uint32_t clock_crystal_freq
;
274 uint32_t drm_major
; /* version */
276 uint32_t drm_patchlevel
;
280 uint32_t r600_max_quad_pipes
; /* wave size / 16 */
281 uint32_t max_shader_clock
;
282 uint32_t num_good_compute_units
;
283 uint32_t max_se
; /* shader engines */
284 uint32_t max_sh_per_se
; /* shader arrays per shader engine */
286 /* Render backends (color + depth blocks). */
287 uint32_t r300_num_gb_pipes
;
288 uint32_t r300_num_z_pipes
;
289 uint32_t r600_gb_backend_map
; /* R600 harvest config */
290 bool r600_gb_backend_map_valid
;
291 uint32_t r600_num_banks
;
292 uint32_t num_render_backends
;
293 uint32_t num_tile_pipes
; /* pipe count from PIPE_CONFIG */
294 uint32_t pipe_interleave_bytes
;
295 uint32_t enabled_rb_mask
; /* GCN harvest config */
298 uint32_t si_tile_mode_array
[32];
299 uint32_t cik_macrotile_mode_array
[16];
302 /* Tiling info for display code, DRI sharing, and other data. */
303 struct radeon_bo_metadata
{
304 /* Tiling flags describing the texture layout for display code
307 enum radeon_bo_layout microtile
;
308 enum radeon_bo_layout macrotile
;
309 unsigned pipe_config
;
318 /* Additional metadata associated with the buffer, in bytes.
319 * The maximum size is 64 * 4. This is opaque for the winsys & kernel.
320 * Supported by amdgpu only.
322 uint32_t size_metadata
;
323 uint32_t metadata
[64];
326 enum radeon_feature_id
{
327 RADEON_FID_R300_HYPERZ_ACCESS
, /* ZMask + HiZ */
328 RADEON_FID_R300_CMASK_ACCESS
,
331 #define RADEON_SURF_MAX_LEVEL 32
333 #define RADEON_SURF_TYPE_MASK 0xFF
334 #define RADEON_SURF_TYPE_SHIFT 0
335 #define RADEON_SURF_TYPE_1D 0
336 #define RADEON_SURF_TYPE_2D 1
337 #define RADEON_SURF_TYPE_3D 2
338 #define RADEON_SURF_TYPE_CUBEMAP 3
339 #define RADEON_SURF_TYPE_1D_ARRAY 4
340 #define RADEON_SURF_TYPE_2D_ARRAY 5
341 #define RADEON_SURF_MODE_MASK 0xFF
342 #define RADEON_SURF_MODE_SHIFT 8
343 #define RADEON_SURF_MODE_LINEAR_ALIGNED 1
344 #define RADEON_SURF_MODE_1D 2
345 #define RADEON_SURF_MODE_2D 3
346 #define RADEON_SURF_SCANOUT (1 << 16)
347 #define RADEON_SURF_ZBUFFER (1 << 17)
348 #define RADEON_SURF_SBUFFER (1 << 18)
349 #define RADEON_SURF_Z_OR_SBUFFER (RADEON_SURF_ZBUFFER | RADEON_SURF_SBUFFER)
350 #define RADEON_SURF_HAS_SBUFFER_MIPTREE (1 << 19)
351 #define RADEON_SURF_HAS_TILE_MODE_INDEX (1 << 20)
352 #define RADEON_SURF_FMASK (1 << 21)
353 #define RADEON_SURF_DISABLE_DCC (1 << 22)
355 #define RADEON_SURF_GET(v, field) (((v) >> RADEON_SURF_ ## field ## _SHIFT) & RADEON_SURF_ ## field ## _MASK)
356 #define RADEON_SURF_SET(v, field) (((v) & RADEON_SURF_ ## field ## _MASK) << RADEON_SURF_ ## field ## _SHIFT)
357 #define RADEON_SURF_CLR(v, field) ((v) & ~(RADEON_SURF_ ## field ## _MASK << RADEON_SURF_ ## field ## _SHIFT))
359 struct radeon_surf_level
{
368 uint32_t pitch_bytes
;
371 uint64_t dcc_fast_clear_size
;
376 /* These are inputs to the calculator. */
389 /* These are return values. Some of them can be set by the caller, but
390 * they will be treated as hints (e.g. bankw, bankh) and might be
391 * changed by the calculator.
394 uint64_t bo_alignment
;
395 /* This applies to EG and later. */
400 uint32_t stencil_tile_split
;
401 struct radeon_surf_level level
[RADEON_SURF_MAX_LEVEL
];
402 struct radeon_surf_level stencil_level
[RADEON_SURF_MAX_LEVEL
];
403 uint32_t tiling_index
[RADEON_SURF_MAX_LEVEL
];
404 uint32_t stencil_tiling_index
[RADEON_SURF_MAX_LEVEL
];
405 uint32_t pipe_config
;
407 uint32_t macro_tile_index
;
408 uint32_t micro_tile_mode
; /* displayable, thin, depth, rotated */
410 /* Whether the depth miptree or stencil miptree as used by the DB are
411 * adjusted from their TC compatible form to ensure depth/stencil
412 * compatibility. If either is true, the corresponding plane cannot be
416 bool stencil_adjusted
;
419 uint64_t dcc_alignment
;
422 struct radeon_bo_list_item
{
425 uint64_t priority_usage
; /* mask of (1 << RADEON_PRIO_*) */
428 struct radeon_winsys
{
430 * The screen object this winsys was created for
432 struct pipe_screen
*screen
;
435 * Decrement the winsys reference count.
437 * \param ws The winsys this function is called for.
438 * \return True if the winsys and screen should be destroyed.
440 bool (*unref
)(struct radeon_winsys
*ws
);
443 * Destroy this winsys.
445 * \param ws The winsys this function is called from.
447 void (*destroy
)(struct radeon_winsys
*ws
);
450 * Query an info structure from winsys.
452 * \param ws The winsys this function is called from.
453 * \param info Return structure
455 void (*query_info
)(struct radeon_winsys
*ws
,
456 struct radeon_info
*info
);
458 /**************************************************************************
459 * Buffer management. Buffer attributes are mostly fixed over its lifetime.
461 * Remember that gallium gets to choose the interface it needs, and the
462 * window systems must then implement that interface (rather than the
463 * other way around...).
464 *************************************************************************/
467 * Create a buffer object.
469 * \param ws The winsys this function is called from.
470 * \param size The size to allocate.
471 * \param alignment An alignment of the buffer in memory.
472 * \param use_reusable_pool Whether the cache buffer manager should be used.
473 * \param domain A bitmask of the RADEON_DOMAIN_* flags.
474 * \return The created buffer object.
476 struct pb_buffer
*(*buffer_create
)(struct radeon_winsys
*ws
,
479 enum radeon_bo_domain domain
,
480 enum radeon_bo_flag flags
);
483 * Map the entire data store of a buffer object into the client's address
486 * \param buf A winsys buffer object to map.
487 * \param cs A command stream to flush if the buffer is referenced by it.
488 * \param usage A bitmask of the PIPE_TRANSFER_* flags.
489 * \return The pointer at the beginning of the buffer.
491 void *(*buffer_map
)(struct pb_buffer
*buf
,
492 struct radeon_winsys_cs
*cs
,
493 enum pipe_transfer_usage usage
);
496 * Unmap a buffer object from the client's address space.
498 * \param buf A winsys buffer object to unmap.
500 void (*buffer_unmap
)(struct pb_buffer
*buf
);
503 * Wait for the buffer and return true if the buffer is not used
506 * The timeout of 0 will only return the status.
507 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the buffer
510 bool (*buffer_wait
)(struct pb_buffer
*buf
, uint64_t timeout
,
511 enum radeon_bo_usage usage
);
514 * Return buffer metadata.
515 * (tiling info for display code, DRI sharing, and other data)
517 * \param buf A winsys buffer object to get the flags from.
520 void (*buffer_get_metadata
)(struct pb_buffer
*buf
,
521 struct radeon_bo_metadata
*md
);
524 * Set buffer metadata.
525 * (tiling info for display code, DRI sharing, and other data)
527 * \param buf A winsys buffer object to set the flags for.
530 void (*buffer_set_metadata
)(struct pb_buffer
*buf
,
531 struct radeon_bo_metadata
*md
);
534 * Get a winsys buffer from a winsys handle. The internal structure
535 * of the handle is platform-specific and only a winsys should access it.
537 * \param ws The winsys this function is called from.
538 * \param whandle A winsys handle pointer as was received from a state
540 * \param stride The returned buffer stride in bytes.
542 struct pb_buffer
*(*buffer_from_handle
)(struct radeon_winsys
*ws
,
543 struct winsys_handle
*whandle
,
544 unsigned *stride
, unsigned *offset
);
547 * Get a winsys buffer from a user pointer. The resulting buffer can't
548 * be exported. Both pointer and size must be page aligned.
550 * \param ws The winsys this function is called from.
551 * \param pointer User pointer to turn into a buffer object.
552 * \param Size Size in bytes for the new buffer.
554 struct pb_buffer
*(*buffer_from_ptr
)(struct radeon_winsys
*ws
,
555 void *pointer
, uint64_t size
);
558 * Whether the buffer was created from a user pointer.
560 * \param buf A winsys buffer object
561 * \return whether \p buf was created via buffer_from_ptr
563 bool (*buffer_is_user_ptr
)(struct pb_buffer
*buf
);
566 * Get a winsys handle from a winsys buffer. The internal structure
567 * of the handle is platform-specific and only a winsys should access it.
569 * \param buf A winsys buffer object to get the handle from.
570 * \param whandle A winsys handle pointer.
571 * \param stride A stride of the buffer in bytes, for texturing.
572 * \return true on success.
574 bool (*buffer_get_handle
)(struct pb_buffer
*buf
,
575 unsigned stride
, unsigned offset
,
577 struct winsys_handle
*whandle
);
580 * Return the virtual address of a buffer.
582 * \param buf A winsys buffer object
583 * \return virtual address
585 uint64_t (*buffer_get_virtual_address
)(struct pb_buffer
*buf
);
588 * Query the initial placement of the buffer from the kernel driver.
590 enum radeon_bo_domain (*buffer_get_initial_domain
)(struct pb_buffer
*buf
);
592 /**************************************************************************
593 * Command submission.
595 * Each pipe context should create its own command stream and submit
596 * commands independently of other contexts.
597 *************************************************************************/
600 * Create a command submission context.
601 * Various command streams can be submitted to the same context.
603 struct radeon_winsys_ctx
*(*ctx_create
)(struct radeon_winsys
*ws
);
608 void (*ctx_destroy
)(struct radeon_winsys_ctx
*ctx
);
611 * Query a GPU reset status.
613 enum pipe_reset_status (*ctx_query_reset_status
)(struct radeon_winsys_ctx
*ctx
);
616 * Create a command stream.
618 * \param ctx The submission context
619 * \param ring_type The ring type (GFX, DMA, UVD)
620 * \param flush Flush callback function associated with the command stream.
621 * \param user User pointer that will be passed to the flush callback.
623 struct radeon_winsys_cs
*(*cs_create
)(struct radeon_winsys_ctx
*ctx
,
624 enum ring_type ring_type
,
625 void (*flush
)(void *ctx
, unsigned flags
,
626 struct pipe_fence_handle
**fence
),
630 * Add a constant engine IB to a graphics CS. This makes the graphics CS
631 * from "cs_create" a group of two IBs that share a buffer list and are
634 * The returned constant CS is only a stream for writing packets to the new
635 * IB. Calling other winsys functions with it is not allowed, not even
638 * In order to add buffers and check memory usage, use the graphics CS.
639 * In order to flush it, use the graphics CS, which will flush both IBs.
640 * Destroying the graphics CS will destroy both of them.
642 * \param cs The graphics CS from "cs_create" that will hold the buffer
643 * list and will be used for flushing.
645 struct radeon_winsys_cs
*(*cs_add_const_ib
)(struct radeon_winsys_cs
*cs
);
648 * Add a constant engine preamble IB to a graphics CS. This add an extra IB
649 * in similar manner to cs_add_const_ib. This should always be called after
652 * The returned IB is a constant engine IB that only gets flushed if the
655 * \param cs The graphics CS from "cs_create" that will hold the buffer
656 * list and will be used for flushing.
658 struct radeon_winsys_cs
*(*cs_add_const_preamble_ib
)(struct radeon_winsys_cs
*cs
);
660 * Destroy a command stream.
662 * \param cs A command stream to destroy.
664 void (*cs_destroy
)(struct radeon_winsys_cs
*cs
);
667 * Add a buffer. Each buffer used by a CS must be added using this function.
669 * \param cs Command stream
671 * \param usage Whether the buffer is used for read and/or write.
672 * \param domain Bitmask of the RADEON_DOMAIN_* flags.
673 * \param priority A higher number means a greater chance of being
674 * placed in the requested domain. 15 is the maximum.
675 * \return Buffer index.
677 unsigned (*cs_add_buffer
)(struct radeon_winsys_cs
*cs
,
678 struct pb_buffer
*buf
,
679 enum radeon_bo_usage usage
,
680 enum radeon_bo_domain domain
,
681 enum radeon_bo_priority priority
);
684 * Return the index of an already-added buffer.
686 * \param cs Command stream
688 * \return The buffer index, or -1 if the buffer has not been added.
690 int (*cs_lookup_buffer
)(struct radeon_winsys_cs
*cs
,
691 struct pb_buffer
*buf
);
694 * Return true if there is enough memory in VRAM and GTT for the buffers
695 * added so far. If the validation fails, all buffers which have
696 * been added since the last call of cs_validate will be removed and
697 * the CS will be flushed (provided there are still any buffers).
699 * \param cs A command stream to validate.
701 bool (*cs_validate
)(struct radeon_winsys_cs
*cs
);
704 * Check whether the given number of dwords is available in the IB.
705 * Optionally chain a new chunk of the IB if necessary and supported.
707 * \param cs A command stream.
708 * \param dw Number of CS dwords requested by the caller.
710 bool (*cs_check_space
)(struct radeon_winsys_cs
*cs
, unsigned dw
);
713 * Return the buffer list.
715 * \param cs Command stream
716 * \param list Returned buffer list. Set to NULL to query the count only.
717 * \return The buffer count.
719 unsigned (*cs_get_buffer_list
)(struct radeon_winsys_cs
*cs
,
720 struct radeon_bo_list_item
*list
);
723 * Flush a command stream.
725 * \param cs A command stream to flush.
726 * \param flags, RADEON_FLUSH_ASYNC or 0.
727 * \param fence Pointer to a fence. If non-NULL, a fence is inserted
728 * after the CS and is returned through this parameter.
729 * \return Negative POSIX error code or 0 for success.
730 * Asynchronous submissions never return an error.
732 int (*cs_flush
)(struct radeon_winsys_cs
*cs
,
734 struct pipe_fence_handle
**fence
);
737 * Create a fence before the CS is flushed.
738 * The user must flush manually to complete the initializaton of the fence.
739 * The fence must not be used before the flush.
741 struct pipe_fence_handle
*(*cs_get_next_fence
)(struct radeon_winsys_cs
*cs
);
744 * Return true if a buffer is referenced by a command stream.
746 * \param cs A command stream.
747 * \param buf A winsys buffer.
749 bool (*cs_is_buffer_referenced
)(struct radeon_winsys_cs
*cs
,
750 struct pb_buffer
*buf
,
751 enum radeon_bo_usage usage
);
754 * Request access to a feature for a command stream.
756 * \param cs A command stream.
757 * \param fid Feature ID, one of RADEON_FID_*
758 * \param enable Whether to enable or disable the feature.
760 bool (*cs_request_feature
)(struct radeon_winsys_cs
*cs
,
761 enum radeon_feature_id fid
,
764 * Make sure all asynchronous flush of the cs have completed
766 * \param cs A command stream.
768 void (*cs_sync_flush
)(struct radeon_winsys_cs
*cs
);
771 * Wait for the fence and return true if the fence has been signalled.
772 * The timeout of 0 will only return the status.
773 * The timeout of PIPE_TIMEOUT_INFINITE will always wait until the fence
776 bool (*fence_wait
)(struct radeon_winsys
*ws
,
777 struct pipe_fence_handle
*fence
,
781 * Reference counting for fences.
783 void (*fence_reference
)(struct pipe_fence_handle
**dst
,
784 struct pipe_fence_handle
*src
);
789 * \param ws The winsys this function is called from.
790 * \param surf Surface structure ptr
792 int (*surface_init
)(struct radeon_winsys
*ws
,
793 struct radeon_surf
*surf
);
796 * Find best values for a surface
798 * \param ws The winsys this function is called from.
799 * \param surf Surface structure ptr
801 int (*surface_best
)(struct radeon_winsys
*ws
,
802 struct radeon_surf
*surf
);
804 uint64_t (*query_value
)(struct radeon_winsys
*ws
,
805 enum radeon_value_id value
);
807 bool (*read_registers
)(struct radeon_winsys
*ws
, unsigned reg_offset
,
808 unsigned num_registers
, uint32_t *out
);
811 static inline bool radeon_emitted(struct radeon_winsys_cs
*cs
, unsigned num_dw
)
813 return cs
&& (cs
->prev_dw
+ cs
->current
.cdw
> num_dw
);
816 static inline void radeon_emit(struct radeon_winsys_cs
*cs
, uint32_t value
)
818 cs
->current
.buf
[cs
->current
.cdw
++] = value
;
821 static inline void radeon_emit_array(struct radeon_winsys_cs
*cs
,
822 const uint32_t *values
, unsigned count
)
824 memcpy(cs
->current
.buf
+ cs
->current
.cdw
, values
, count
* 4);
825 cs
->current
.cdw
+= count
;