2 * Copyright © 2015 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
39 #define __gen_validate_value(x) VALGRIND_CHECK_MEM_IS_DEFINED(&(x), sizeof(x))
44 #include "common/gen_device_info.h"
45 #include "blorp/blorp.h"
46 #include "brw_compiler.h"
47 #include "util/macros.h"
48 #include "util/list.h"
49 #include "util/u_vector.h"
50 #include "util/vk_alloc.h"
52 /* Pre-declarations needed for WSI entrypoints */
55 typedef struct xcb_connection_t xcb_connection_t
;
56 typedef uint32_t xcb_visualid_t
;
57 typedef uint32_t xcb_window_t
;
60 struct anv_buffer_view
;
61 struct anv_image_view
;
65 #include <vulkan/vulkan.h>
66 #include <vulkan/vulkan_intel.h>
67 #include <vulkan/vk_icd.h>
69 #include "anv_entrypoints.h"
70 #include "brw_context.h"
73 #include "wsi_common.h"
75 /* Allowing different clear colors requires us to perform a depth resolve at
76 * the end of certain render passes. This is because while slow clears store
77 * the clear color in the HiZ buffer, fast clears (without a resolve) don't.
78 * See the PRMs for examples describing when additional resolves would be
79 * necessary. To enable fast clears without requiring extra resolves, we set
80 * the clear value to a globally-defined one. We could allow different values
81 * if the user doesn't expect coherent data during or after a render passes
82 * (VK_ATTACHMENT_STORE_OP_DONT_CARE), but such users (aside from the CTS)
83 * don't seem to exist yet. In almost all Vulkan applications tested thus far,
84 * 1.0f seems to be the only value used. The only application that doesn't set
85 * this value does so through the usage of an seemingly uninitialized clear
88 #define ANV_HZ_FC_VAL 1.0f
93 #define MAX_VIEWPORTS 16
94 #define MAX_SCISSORS 16
95 #define MAX_PUSH_CONSTANTS_SIZE 128
96 #define MAX_DYNAMIC_BUFFERS 16
99 #define ANV_SVGS_VB_INDEX MAX_VBS
100 #define ANV_DRAWID_VB_INDEX (MAX_VBS + 1)
102 #define anv_printflike(a, b) __attribute__((__format__(__printf__, a, b)))
104 static inline uint32_t
105 align_down_npot_u32(uint32_t v
, uint32_t a
)
110 static inline uint32_t
111 align_u32(uint32_t v
, uint32_t a
)
113 assert(a
!= 0 && a
== (a
& -a
));
114 return (v
+ a
- 1) & ~(a
- 1);
117 static inline uint64_t
118 align_u64(uint64_t v
, uint64_t a
)
120 assert(a
!= 0 && a
== (a
& -a
));
121 return (v
+ a
- 1) & ~(a
- 1);
124 static inline int32_t
125 align_i32(int32_t v
, int32_t a
)
127 assert(a
!= 0 && a
== (a
& -a
));
128 return (v
+ a
- 1) & ~(a
- 1);
131 /** Alignment must be a power of 2. */
133 anv_is_aligned(uintmax_t n
, uintmax_t a
)
135 assert(a
== (a
& -a
));
136 return (n
& (a
- 1)) == 0;
139 static inline uint32_t
140 anv_minify(uint32_t n
, uint32_t levels
)
142 if (unlikely(n
== 0))
145 return MAX2(n
>> levels
, 1);
149 anv_clamp_f(float f
, float min
, float max
)
162 anv_clear_mask(uint32_t *inout_mask
, uint32_t clear_mask
)
164 if (*inout_mask
& clear_mask
) {
165 *inout_mask
&= ~clear_mask
;
172 static inline union isl_color_value
173 vk_to_isl_color(VkClearColorValue color
)
175 return (union isl_color_value
) {
185 #define for_each_bit(b, dword) \
186 for (uint32_t __dword = (dword); \
187 (b) = __builtin_ffs(__dword) - 1, __dword; \
188 __dword &= ~(1 << (b)))
190 #define typed_memcpy(dest, src, count) ({ \
191 STATIC_ASSERT(sizeof(*src) == sizeof(*dest)); \
192 memcpy((dest), (src), (count) * sizeof(*(src))); \
195 /* Whenever we generate an error, pass it through this function. Useful for
196 * debugging, where we can break on it. Only call at error site, not when
197 * propagating errors. Might be useful to plug in a stack trace here.
200 VkResult
__vk_errorf(VkResult error
, const char *file
, int line
, const char *format
, ...);
203 #define vk_error(error) __vk_errorf(error, __FILE__, __LINE__, NULL);
204 #define vk_errorf(error, format, ...) __vk_errorf(error, __FILE__, __LINE__, format, ## __VA_ARGS__);
205 #define anv_debug(format, ...) fprintf(stderr, "debug: " format, ##__VA_ARGS__)
207 #define vk_error(error) error
208 #define vk_errorf(error, format, ...) error
209 #define anv_debug(format, ...)
213 * Warn on ignored extension structs.
215 * The Vulkan spec requires us to ignore unsupported or unknown structs in
216 * a pNext chain. In debug mode, emitting warnings for ignored structs may
217 * help us discover structs that we should not have ignored.
220 * From the Vulkan 1.0.38 spec:
222 * Any component of the implementation (the loader, any enabled layers,
223 * and drivers) must skip over, without processing (other than reading the
224 * sType and pNext members) any chained structures with sType values not
225 * defined by extensions supported by that component.
227 #define anv_debug_ignored_stype(sType) \
228 anv_debug("debug: %s: ignored VkStructureType %u\n", __func__, (sType))
230 void __anv_finishme(const char *file
, int line
, const char *format
, ...)
231 anv_printflike(3, 4);
232 void anv_loge(const char *format
, ...) anv_printflike(1, 2);
233 void anv_loge_v(const char *format
, va_list va
);
236 * Print a FINISHME message, including its source location.
238 #define anv_finishme(format, ...) \
240 static bool reported = false; \
242 __anv_finishme(__FILE__, __LINE__, format, ##__VA_ARGS__); \
247 /* A non-fatal assert. Useful for debugging. */
249 #define anv_assert(x) ({ \
250 if (unlikely(!(x))) \
251 fprintf(stderr, "%s:%d ASSERT: %s\n", __FILE__, __LINE__, #x); \
254 #define anv_assert(x)
258 * If a block of code is annotated with anv_validate, then the block runs only
262 #define anv_validate if (1)
264 #define anv_validate if (0)
267 #define stub_return(v) \
269 anv_finishme("stub %s", __func__); \
275 anv_finishme("stub %s", __func__); \
280 * A dynamically growable, circular buffer. Elements are added at head and
281 * removed from tail. head and tail are free-running uint32_t indices and we
282 * only compute the modulo with size when accessing the array. This way,
283 * number of bytes in the queue is always head - tail, even in case of
290 /* Index into the current validation list. This is used by the
291 * validation list building alrogithm to track which buffers are already
292 * in the validation list so that we can ensure uniqueness.
296 /* Last known offset. This value is provided by the kernel when we
297 * execbuf and is used as the presumed offset for the next bunch of
305 /* We need to set the WRITE flag on winsys bos so GEM will know we're
306 * writing to them and synchronize uses on other rings (eg if the display
307 * server uses the blitter ring).
313 anv_bo_init(struct anv_bo
*bo
, uint32_t gem_handle
, uint64_t size
)
315 bo
->gem_handle
= gem_handle
;
320 bo
->is_winsys_bo
= false;
323 /* Represents a lock-free linked list of "free" things. This is used by
324 * both the block pool and the state pools. Unfortunately, in order to
325 * solve the ABA problem, we can't use a single uint32_t head.
327 union anv_free_list
{
331 /* A simple count that is incremented every time the head changes. */
337 #define ANV_FREE_LIST_EMPTY ((union anv_free_list) { { 1, 0 } })
339 struct anv_block_state
{
349 struct anv_block_pool
{
350 struct anv_device
*device
;
354 /* The offset from the start of the bo to the "center" of the block
355 * pool. Pointers to allocated blocks are given by
356 * bo.map + center_bo_offset + offsets.
358 uint32_t center_bo_offset
;
360 /* Current memory map of the block pool. This pointer may or may not
361 * point to the actual beginning of the block pool memory. If
362 * anv_block_pool_alloc_back has ever been called, then this pointer
363 * will point to the "center" position of the buffer and all offsets
364 * (negative or positive) given out by the block pool alloc functions
365 * will be valid relative to this pointer.
367 * In particular, map == bo.map + center_offset
373 * Array of mmaps and gem handles owned by the block pool, reclaimed when
374 * the block pool is destroyed.
376 struct u_vector mmap_cleanups
;
380 union anv_free_list free_list
;
381 struct anv_block_state state
;
383 union anv_free_list back_free_list
;
384 struct anv_block_state back_state
;
387 /* Block pools are backed by a fixed-size 2GB memfd */
388 #define BLOCK_POOL_MEMFD_SIZE (1ull << 32)
390 /* The center of the block pool is also the middle of the memfd. This may
391 * change in the future if we decide differently for some reason.
393 #define BLOCK_POOL_MEMFD_CENTER (BLOCK_POOL_MEMFD_SIZE / 2)
395 static inline uint32_t
396 anv_block_pool_size(struct anv_block_pool
*pool
)
398 return pool
->state
.end
+ pool
->back_state
.end
;
407 struct anv_fixed_size_state_pool
{
409 union anv_free_list free_list
;
410 struct anv_block_state block
;
413 #define ANV_MIN_STATE_SIZE_LOG2 6
414 #define ANV_MAX_STATE_SIZE_LOG2 20
416 #define ANV_STATE_BUCKETS (ANV_MAX_STATE_SIZE_LOG2 - ANV_MIN_STATE_SIZE_LOG2 + 1)
418 struct anv_state_pool
{
419 struct anv_block_pool
*block_pool
;
420 struct anv_fixed_size_state_pool buckets
[ANV_STATE_BUCKETS
];
423 struct anv_state_stream_block
;
425 struct anv_state_stream
{
426 struct anv_block_pool
*block_pool
;
428 /* The current working block */
429 struct anv_state_stream_block
*block
;
431 /* Offset at which the current block starts */
433 /* Offset at which to allocate the next state */
435 /* Offset at which the current block ends */
439 #define CACHELINE_SIZE 64
440 #define CACHELINE_MASK 63
443 anv_clflush_range(void *start
, size_t size
)
445 void *p
= (void *) (((uintptr_t) start
) & ~CACHELINE_MASK
);
446 void *end
= start
+ size
;
449 __builtin_ia32_clflush(p
);
455 anv_flush_range(void *start
, size_t size
)
457 __builtin_ia32_mfence();
458 anv_clflush_range(start
, size
);
462 anv_invalidate_range(void *start
, size_t size
)
464 anv_clflush_range(start
, size
);
465 __builtin_ia32_mfence();
468 VkResult
anv_block_pool_init(struct anv_block_pool
*pool
,
469 struct anv_device
*device
, uint32_t block_size
);
470 void anv_block_pool_finish(struct anv_block_pool
*pool
);
471 int32_t anv_block_pool_alloc(struct anv_block_pool
*pool
);
472 int32_t anv_block_pool_alloc_back(struct anv_block_pool
*pool
);
473 void anv_block_pool_free(struct anv_block_pool
*pool
, int32_t offset
);
474 void anv_state_pool_init(struct anv_state_pool
*pool
,
475 struct anv_block_pool
*block_pool
);
476 void anv_state_pool_finish(struct anv_state_pool
*pool
);
477 struct anv_state
anv_state_pool_alloc(struct anv_state_pool
*pool
,
478 size_t state_size
, size_t alignment
);
479 void anv_state_pool_free(struct anv_state_pool
*pool
, struct anv_state state
);
480 void anv_state_stream_init(struct anv_state_stream
*stream
,
481 struct anv_block_pool
*block_pool
);
482 void anv_state_stream_finish(struct anv_state_stream
*stream
);
483 struct anv_state
anv_state_stream_alloc(struct anv_state_stream
*stream
,
484 uint32_t size
, uint32_t alignment
);
487 * Implements a pool of re-usable BOs. The interface is identical to that
488 * of block_pool except that each block is its own BO.
491 struct anv_device
*device
;
496 void anv_bo_pool_init(struct anv_bo_pool
*pool
, struct anv_device
*device
);
497 void anv_bo_pool_finish(struct anv_bo_pool
*pool
);
498 VkResult
anv_bo_pool_alloc(struct anv_bo_pool
*pool
, struct anv_bo
*bo
,
500 void anv_bo_pool_free(struct anv_bo_pool
*pool
, const struct anv_bo
*bo
);
502 struct anv_scratch_bo
{
507 struct anv_scratch_pool
{
508 /* Indexed by Per-Thread Scratch Space number (the hardware value) and stage */
509 struct anv_scratch_bo bos
[16][MESA_SHADER_STAGES
];
512 void anv_scratch_pool_init(struct anv_device
*device
,
513 struct anv_scratch_pool
*pool
);
514 void anv_scratch_pool_finish(struct anv_device
*device
,
515 struct anv_scratch_pool
*pool
);
516 struct anv_bo
*anv_scratch_pool_alloc(struct anv_device
*device
,
517 struct anv_scratch_pool
*pool
,
518 gl_shader_stage stage
,
519 unsigned per_thread_scratch
);
521 struct anv_physical_device
{
522 VK_LOADER_DATA _loader_data
;
524 struct anv_instance
* instance
;
528 struct gen_device_info info
;
529 uint64_t aperture_size
;
530 struct brw_compiler
* compiler
;
531 struct isl_device isl_dev
;
532 int cmd_parser_version
;
535 uint32_t subslice_total
;
537 uint8_t uuid
[VK_UUID_SIZE
];
539 struct wsi_device wsi_device
;
543 struct anv_instance
{
544 VK_LOADER_DATA _loader_data
;
546 VkAllocationCallbacks alloc
;
549 int physicalDeviceCount
;
550 struct anv_physical_device physicalDevice
;
553 VkResult
anv_init_wsi(struct anv_physical_device
*physical_device
);
554 void anv_finish_wsi(struct anv_physical_device
*physical_device
);
557 VK_LOADER_DATA _loader_data
;
559 struct anv_device
* device
;
561 struct anv_state_pool
* pool
;
564 struct anv_pipeline_cache
{
565 struct anv_device
* device
;
566 pthread_mutex_t mutex
;
568 struct hash_table
* cache
;
571 struct anv_pipeline_bind_map
;
573 void anv_pipeline_cache_init(struct anv_pipeline_cache
*cache
,
574 struct anv_device
*device
,
576 void anv_pipeline_cache_finish(struct anv_pipeline_cache
*cache
);
578 struct anv_shader_bin
*
579 anv_pipeline_cache_search(struct anv_pipeline_cache
*cache
,
580 const void *key
, uint32_t key_size
);
581 struct anv_shader_bin
*
582 anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache
*cache
,
583 const void *key_data
, uint32_t key_size
,
584 const void *kernel_data
, uint32_t kernel_size
,
585 const struct brw_stage_prog_data
*prog_data
,
586 uint32_t prog_data_size
,
587 const struct anv_pipeline_bind_map
*bind_map
);
590 VK_LOADER_DATA _loader_data
;
592 VkAllocationCallbacks alloc
;
594 struct anv_instance
* instance
;
596 struct gen_device_info info
;
597 struct isl_device isl_dev
;
600 bool can_chain_batches
;
601 bool robust_buffer_access
;
603 struct anv_bo_pool batch_bo_pool
;
605 struct anv_block_pool dynamic_state_block_pool
;
606 struct anv_state_pool dynamic_state_pool
;
608 struct anv_block_pool instruction_block_pool
;
609 struct anv_state_pool instruction_state_pool
;
611 struct anv_block_pool surface_state_block_pool
;
612 struct anv_state_pool surface_state_pool
;
614 struct anv_bo workaround_bo
;
616 struct anv_pipeline_cache blorp_shader_cache
;
617 struct blorp_context blorp
;
619 struct anv_state border_colors
;
621 struct anv_queue queue
;
623 struct anv_scratch_pool scratch_pool
;
625 uint32_t default_mocs
;
627 pthread_mutex_t mutex
;
628 pthread_cond_t queue_submit
;
632 anv_state_flush(struct anv_device
*device
, struct anv_state state
)
634 if (device
->info
.has_llc
)
637 anv_flush_range(state
.map
, state
.alloc_size
);
640 void anv_device_init_blorp(struct anv_device
*device
);
641 void anv_device_finish_blorp(struct anv_device
*device
);
643 VkResult
anv_device_execbuf(struct anv_device
*device
,
644 struct drm_i915_gem_execbuffer2
*execbuf
,
645 struct anv_bo
**execbuf_bos
);
647 void* anv_gem_mmap(struct anv_device
*device
,
648 uint32_t gem_handle
, uint64_t offset
, uint64_t size
, uint32_t flags
);
649 void anv_gem_munmap(void *p
, uint64_t size
);
650 uint32_t anv_gem_create(struct anv_device
*device
, size_t size
);
651 void anv_gem_close(struct anv_device
*device
, uint32_t gem_handle
);
652 uint32_t anv_gem_userptr(struct anv_device
*device
, void *mem
, size_t size
);
653 int anv_gem_wait(struct anv_device
*device
, uint32_t gem_handle
, int64_t *timeout_ns
);
654 int anv_gem_execbuffer(struct anv_device
*device
,
655 struct drm_i915_gem_execbuffer2
*execbuf
);
656 int anv_gem_set_tiling(struct anv_device
*device
, uint32_t gem_handle
,
657 uint32_t stride
, uint32_t tiling
);
658 int anv_gem_create_context(struct anv_device
*device
);
659 int anv_gem_destroy_context(struct anv_device
*device
, int context
);
660 int anv_gem_get_param(int fd
, uint32_t param
);
661 bool anv_gem_get_bit6_swizzle(int fd
, uint32_t tiling
);
662 int anv_gem_get_aperture(int fd
, uint64_t *size
);
663 int anv_gem_handle_to_fd(struct anv_device
*device
, uint32_t gem_handle
);
664 uint32_t anv_gem_fd_to_handle(struct anv_device
*device
, int fd
);
665 int anv_gem_set_caching(struct anv_device
*device
, uint32_t gem_handle
, uint32_t caching
);
666 int anv_gem_set_domain(struct anv_device
*device
, uint32_t gem_handle
,
667 uint32_t read_domains
, uint32_t write_domain
);
669 VkResult
anv_bo_init_new(struct anv_bo
*bo
, struct anv_device
*device
, uint64_t size
);
671 struct anv_reloc_list
{
674 struct drm_i915_gem_relocation_entry
* relocs
;
675 struct anv_bo
** reloc_bos
;
678 VkResult
anv_reloc_list_init(struct anv_reloc_list
*list
,
679 const VkAllocationCallbacks
*alloc
);
680 void anv_reloc_list_finish(struct anv_reloc_list
*list
,
681 const VkAllocationCallbacks
*alloc
);
683 uint64_t anv_reloc_list_add(struct anv_reloc_list
*list
,
684 const VkAllocationCallbacks
*alloc
,
685 uint32_t offset
, struct anv_bo
*target_bo
,
688 struct anv_batch_bo
{
689 /* Link in the anv_cmd_buffer.owned_batch_bos list */
690 struct list_head link
;
694 /* Bytes actually consumed in this batch BO */
697 struct anv_reloc_list relocs
;
701 const VkAllocationCallbacks
* alloc
;
707 struct anv_reloc_list
* relocs
;
709 /* This callback is called (with the associated user data) in the event
710 * that the batch runs out of space.
712 VkResult (*extend_cb
)(struct anv_batch
*, void *);
716 void *anv_batch_emit_dwords(struct anv_batch
*batch
, int num_dwords
);
717 void anv_batch_emit_batch(struct anv_batch
*batch
, struct anv_batch
*other
);
718 uint64_t anv_batch_emit_reloc(struct anv_batch
*batch
,
719 void *location
, struct anv_bo
*bo
, uint32_t offset
);
720 VkResult
anv_device_submit_simple_batch(struct anv_device
*device
,
721 struct anv_batch
*batch
);
728 static inline uint64_t
729 _anv_combine_address(struct anv_batch
*batch
, void *location
,
730 const struct anv_address address
, uint32_t delta
)
732 if (address
.bo
== NULL
) {
733 return address
.offset
+ delta
;
735 assert(batch
->start
<= location
&& location
< batch
->end
);
737 return anv_batch_emit_reloc(batch
, location
, address
.bo
, address
.offset
+ delta
);
741 #define __gen_address_type struct anv_address
742 #define __gen_user_data struct anv_batch
743 #define __gen_combine_address _anv_combine_address
745 /* Wrapper macros needed to work around preprocessor argument issues. In
746 * particular, arguments don't get pre-evaluated if they are concatenated.
747 * This means that, if you pass GENX(3DSTATE_PS) into the emit macro, the
748 * GENX macro won't get evaluated if the emit macro contains "cmd ## foo".
749 * We can work around this easily enough with these helpers.
751 #define __anv_cmd_length(cmd) cmd ## _length
752 #define __anv_cmd_length_bias(cmd) cmd ## _length_bias
753 #define __anv_cmd_header(cmd) cmd ## _header
754 #define __anv_cmd_pack(cmd) cmd ## _pack
755 #define __anv_reg_num(reg) reg ## _num
757 #define anv_pack_struct(dst, struc, ...) do { \
758 struct struc __template = { \
761 __anv_cmd_pack(struc)(NULL, dst, &__template); \
762 VG(VALGRIND_CHECK_MEM_IS_DEFINED(dst, __anv_cmd_length(struc) * 4)); \
765 #define anv_batch_emitn(batch, n, cmd, ...) ({ \
766 void *__dst = anv_batch_emit_dwords(batch, n); \
767 struct cmd __template = { \
768 __anv_cmd_header(cmd), \
769 .DWordLength = n - __anv_cmd_length_bias(cmd), \
772 __anv_cmd_pack(cmd)(batch, __dst, &__template); \
776 #define anv_batch_emit_merge(batch, dwords0, dwords1) \
780 STATIC_ASSERT(ARRAY_SIZE(dwords0) == ARRAY_SIZE(dwords1)); \
781 dw = anv_batch_emit_dwords((batch), ARRAY_SIZE(dwords0)); \
782 for (uint32_t i = 0; i < ARRAY_SIZE(dwords0); i++) \
783 dw[i] = (dwords0)[i] | (dwords1)[i]; \
784 VG(VALGRIND_CHECK_MEM_IS_DEFINED(dw, ARRAY_SIZE(dwords0) * 4));\
787 #define anv_batch_emit(batch, cmd, name) \
788 for (struct cmd name = { __anv_cmd_header(cmd) }, \
789 *_dst = anv_batch_emit_dwords(batch, __anv_cmd_length(cmd)); \
790 __builtin_expect(_dst != NULL, 1); \
791 ({ __anv_cmd_pack(cmd)(batch, _dst, &name); \
792 VG(VALGRIND_CHECK_MEM_IS_DEFINED(_dst, __anv_cmd_length(cmd) * 4)); \
796 #define GEN7_MOCS (struct GEN7_MEMORY_OBJECT_CONTROL_STATE) { \
797 .GraphicsDataTypeGFDT = 0, \
798 .LLCCacheabilityControlLLCCC = 0, \
799 .L3CacheabilityControlL3CC = 1, \
802 #define GEN75_MOCS (struct GEN75_MEMORY_OBJECT_CONTROL_STATE) { \
803 .LLCeLLCCacheabilityControlLLCCC = 0, \
804 .L3CacheabilityControlL3CC = 1, \
807 #define GEN8_MOCS (struct GEN8_MEMORY_OBJECT_CONTROL_STATE) { \
808 .MemoryTypeLLCeLLCCacheabilityControl = WB, \
809 .TargetCache = L3DefertoPATforLLCeLLCselection, \
813 /* Skylake: MOCS is now an index into an array of 62 different caching
814 * configurations programmed by the kernel.
817 #define GEN9_MOCS (struct GEN9_MEMORY_OBJECT_CONTROL_STATE) { \
818 /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */ \
819 .IndextoMOCSTables = 2 \
822 #define GEN9_MOCS_PTE { \
823 /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */ \
824 .IndextoMOCSTables = 1 \
827 struct anv_device_memory
{
830 VkDeviceSize map_size
;
835 * Header for Vertex URB Entry (VUE)
837 struct anv_vue_header
{
839 uint32_t RTAIndex
; /* RenderTargetArrayIndex */
840 uint32_t ViewportIndex
;
844 struct anv_descriptor_set_binding_layout
{
846 /* The type of the descriptors in this binding */
847 VkDescriptorType type
;
850 /* Number of array elements in this binding */
853 /* Index into the flattend descriptor set */
854 uint16_t descriptor_index
;
856 /* Index into the dynamic state array for a dynamic buffer */
857 int16_t dynamic_offset_index
;
859 /* Index into the descriptor set buffer views */
860 int16_t buffer_index
;
863 /* Index into the binding table for the associated surface */
864 int16_t surface_index
;
866 /* Index into the sampler table for the associated sampler */
867 int16_t sampler_index
;
869 /* Index into the image table for the associated image */
871 } stage
[MESA_SHADER_STAGES
];
873 /* Immutable samplers (or NULL if no immutable samplers) */
874 struct anv_sampler
**immutable_samplers
;
877 struct anv_descriptor_set_layout
{
878 /* Number of bindings in this descriptor set */
879 uint16_t binding_count
;
881 /* Total size of the descriptor set with room for all array entries */
884 /* Shader stages affected by this descriptor set */
885 uint16_t shader_stages
;
887 /* Number of buffers in this descriptor set */
888 uint16_t buffer_count
;
890 /* Number of dynamic offsets used by this descriptor set */
891 uint16_t dynamic_offset_count
;
893 /* Bindings in this descriptor set */
894 struct anv_descriptor_set_binding_layout binding
[0];
897 struct anv_descriptor
{
898 VkDescriptorType type
;
902 struct anv_image_view
*image_view
;
903 struct anv_sampler
*sampler
;
906 struct anv_buffer_view
*buffer_view
;
910 struct anv_descriptor_set
{
911 const struct anv_descriptor_set_layout
*layout
;
913 uint32_t buffer_count
;
914 struct anv_buffer_view
*buffer_views
;
915 struct anv_descriptor descriptors
[0];
918 struct anv_buffer_view
{
919 enum isl_format format
; /**< VkBufferViewCreateInfo::format */
921 uint32_t offset
; /**< Offset into bo. */
922 uint64_t range
; /**< VkBufferViewCreateInfo::range */
924 struct anv_state surface_state
;
925 struct anv_state storage_surface_state
;
926 struct anv_state writeonly_storage_surface_state
;
928 struct brw_image_param storage_image_param
;
931 struct anv_descriptor_pool
{
936 struct anv_state_stream surface_state_stream
;
937 void *surface_state_free_list
;
943 anv_descriptor_set_layout_size(const struct anv_descriptor_set_layout
*layout
);
946 anv_descriptor_set_write_image_view(struct anv_descriptor_set
*set
,
947 VkDescriptorType type
,
948 VkImageView _image_view
,
954 anv_descriptor_set_write_buffer_view(struct anv_descriptor_set
*set
,
955 VkDescriptorType type
,
956 struct anv_buffer_view
*buffer_view
,
961 anv_descriptor_set_write_buffer(struct anv_descriptor_set
*set
,
962 struct anv_device
*device
,
963 struct anv_state_stream
*alloc_stream
,
964 VkDescriptorType type
,
965 struct anv_buffer
*buffer
,
972 anv_descriptor_set_create(struct anv_device
*device
,
973 struct anv_descriptor_pool
*pool
,
974 const struct anv_descriptor_set_layout
*layout
,
975 struct anv_descriptor_set
**out_set
);
978 anv_descriptor_set_destroy(struct anv_device
*device
,
979 struct anv_descriptor_pool
*pool
,
980 struct anv_descriptor_set
*set
);
982 #define ANV_DESCRIPTOR_SET_COLOR_ATTACHMENTS UINT8_MAX
984 struct anv_pipeline_binding
{
985 /* The descriptor set this surface corresponds to. The special value of
986 * ANV_DESCRIPTOR_SET_COLOR_ATTACHMENTS indicates that the offset refers
987 * to a color attachment and not a regular descriptor.
991 /* Binding in the descriptor set */
994 /* Index in the binding */
997 /* Input attachment index (relative to the subpass) */
998 uint8_t input_attachment_index
;
1000 /* For a storage image, whether it is write-only */
1004 struct anv_pipeline_layout
{
1006 struct anv_descriptor_set_layout
*layout
;
1007 uint32_t dynamic_offset_start
;
1013 bool has_dynamic_offsets
;
1014 } stage
[MESA_SHADER_STAGES
];
1016 unsigned char sha1
[20];
1020 struct anv_device
* device
;
1023 VkBufferUsageFlags usage
;
1025 /* Set when bound */
1027 VkDeviceSize offset
;
1030 enum anv_cmd_dirty_bits
{
1031 ANV_CMD_DIRTY_DYNAMIC_VIEWPORT
= 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
1032 ANV_CMD_DIRTY_DYNAMIC_SCISSOR
= 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
1033 ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
= 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
1034 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
= 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
1035 ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
= 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
1036 ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
= 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
1037 ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
= 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
1038 ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
= 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
1039 ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
= 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
1040 ANV_CMD_DIRTY_DYNAMIC_ALL
= (1 << 9) - 1,
1041 ANV_CMD_DIRTY_PIPELINE
= 1 << 9,
1042 ANV_CMD_DIRTY_INDEX_BUFFER
= 1 << 10,
1043 ANV_CMD_DIRTY_RENDER_TARGETS
= 1 << 11,
1045 typedef uint32_t anv_cmd_dirty_mask_t
;
1047 enum anv_pipe_bits
{
1048 ANV_PIPE_DEPTH_CACHE_FLUSH_BIT
= (1 << 0),
1049 ANV_PIPE_STALL_AT_SCOREBOARD_BIT
= (1 << 1),
1050 ANV_PIPE_STATE_CACHE_INVALIDATE_BIT
= (1 << 2),
1051 ANV_PIPE_CONSTANT_CACHE_INVALIDATE_BIT
= (1 << 3),
1052 ANV_PIPE_VF_CACHE_INVALIDATE_BIT
= (1 << 4),
1053 ANV_PIPE_DATA_CACHE_FLUSH_BIT
= (1 << 5),
1054 ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT
= (1 << 10),
1055 ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT
= (1 << 11),
1056 ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT
= (1 << 12),
1057 ANV_PIPE_DEPTH_STALL_BIT
= (1 << 13),
1058 ANV_PIPE_CS_STALL_BIT
= (1 << 20),
1060 /* This bit does not exist directly in PIPE_CONTROL. Instead it means that
1061 * a flush has happened but not a CS stall. The next time we do any sort
1062 * of invalidation we need to insert a CS stall at that time. Otherwise,
1063 * we would have to CS stall on every flush which could be bad.
1065 ANV_PIPE_NEEDS_CS_STALL_BIT
= (1 << 21),
1068 #define ANV_PIPE_FLUSH_BITS ( \
1069 ANV_PIPE_DEPTH_CACHE_FLUSH_BIT | \
1070 ANV_PIPE_DATA_CACHE_FLUSH_BIT | \
1071 ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT)
1073 #define ANV_PIPE_STALL_BITS ( \
1074 ANV_PIPE_STALL_AT_SCOREBOARD_BIT | \
1075 ANV_PIPE_DEPTH_STALL_BIT | \
1076 ANV_PIPE_CS_STALL_BIT)
1078 #define ANV_PIPE_INVALIDATE_BITS ( \
1079 ANV_PIPE_STATE_CACHE_INVALIDATE_BIT | \
1080 ANV_PIPE_CONSTANT_CACHE_INVALIDATE_BIT | \
1081 ANV_PIPE_VF_CACHE_INVALIDATE_BIT | \
1082 ANV_PIPE_DATA_CACHE_FLUSH_BIT | \
1083 ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT | \
1084 ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT)
1086 struct anv_vertex_binding
{
1087 struct anv_buffer
* buffer
;
1088 VkDeviceSize offset
;
1091 struct anv_push_constants
{
1092 /* Current allocated size of this push constants data structure.
1093 * Because a decent chunk of it may not be used (images on SKL, for
1094 * instance), we won't actually allocate the entire structure up-front.
1098 /* Push constant data provided by the client through vkPushConstants */
1099 uint8_t client_data
[MAX_PUSH_CONSTANTS_SIZE
];
1101 /* Our hardware only provides zero-based vertex and instance id so, in
1102 * order to satisfy the vulkan requirements, we may have to push one or
1103 * both of these into the shader.
1105 uint32_t base_vertex
;
1106 uint32_t base_instance
;
1108 /* Offsets and ranges for dynamically bound buffers */
1112 } dynamic
[MAX_DYNAMIC_BUFFERS
];
1114 /* Image data for image_load_store on pre-SKL */
1115 struct brw_image_param images
[MAX_IMAGES
];
1118 struct anv_dynamic_state
{
1121 VkViewport viewports
[MAX_VIEWPORTS
];
1126 VkRect2D scissors
[MAX_SCISSORS
];
1137 float blend_constants
[4];
1147 } stencil_compare_mask
;
1152 } stencil_write_mask
;
1157 } stencil_reference
;
1160 extern const struct anv_dynamic_state default_dynamic_state
;
1162 void anv_dynamic_state_copy(struct anv_dynamic_state
*dest
,
1163 const struct anv_dynamic_state
*src
,
1164 uint32_t copy_mask
);
1167 * Attachment state when recording a renderpass instance.
1169 * The clear value is valid only if there exists a pending clear.
1171 struct anv_attachment_state
{
1172 enum isl_aux_usage aux_usage
;
1173 enum isl_aux_usage input_aux_usage
;
1174 struct anv_state color_rt_state
;
1175 struct anv_state input_att_state
;
1177 VkImageLayout current_layout
;
1178 VkImageAspectFlags pending_clear_aspects
;
1180 VkClearValue clear_value
;
1181 bool clear_color_is_zero_one
;
1184 /** State required while building cmd buffer */
1185 struct anv_cmd_state
{
1186 /* PIPELINE_SELECT.PipelineSelection */
1187 uint32_t current_pipeline
;
1188 const struct gen_l3_config
* current_l3_config
;
1190 anv_cmd_dirty_mask_t dirty
;
1191 anv_cmd_dirty_mask_t compute_dirty
;
1192 enum anv_pipe_bits pending_pipe_bits
;
1193 uint32_t num_workgroups_offset
;
1194 struct anv_bo
*num_workgroups_bo
;
1195 VkShaderStageFlags descriptors_dirty
;
1196 VkShaderStageFlags push_constants_dirty
;
1197 uint32_t scratch_size
;
1198 struct anv_pipeline
* pipeline
;
1199 struct anv_pipeline
* compute_pipeline
;
1200 struct anv_framebuffer
* framebuffer
;
1201 struct anv_render_pass
* pass
;
1202 struct anv_subpass
* subpass
;
1203 VkRect2D render_area
;
1204 uint32_t restart_index
;
1205 struct anv_vertex_binding vertex_bindings
[MAX_VBS
];
1206 struct anv_descriptor_set
* descriptors
[MAX_SETS
];
1207 VkShaderStageFlags push_constant_stages
;
1208 struct anv_push_constants
* push_constants
[MESA_SHADER_STAGES
];
1209 struct anv_state binding_tables
[MESA_SHADER_STAGES
];
1210 struct anv_state samplers
[MESA_SHADER_STAGES
];
1211 struct anv_dynamic_state dynamic
;
1215 * Whether or not the gen8 PMA fix is enabled. We ensure that, at the top
1216 * of any command buffer it is disabled by disabling it in EndCommandBuffer
1217 * and before invoking the secondary in ExecuteCommands.
1219 bool pma_fix_enabled
;
1222 * Whether or not we know for certain that HiZ is enabled for the current
1223 * subpass. If, for whatever reason, we are unsure as to whether HiZ is
1224 * enabled or not, this will be false.
1229 * Array length is anv_cmd_state::pass::attachment_count. Array content is
1230 * valid only when recording a render pass instance.
1232 struct anv_attachment_state
* attachments
;
1235 * Surface states for color render targets. These are stored in a single
1236 * flat array. For depth-stencil attachments, the surface state is simply
1239 struct anv_state render_pass_states
;
1242 * A null surface state of the right size to match the framebuffer. This
1243 * is one of the states in render_pass_states.
1245 struct anv_state null_surface_state
;
1248 struct anv_buffer
* index_buffer
;
1249 uint32_t index_type
; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */
1250 uint32_t index_offset
;
1254 struct anv_cmd_pool
{
1255 VkAllocationCallbacks alloc
;
1256 struct list_head cmd_buffers
;
1259 #define ANV_CMD_BUFFER_BATCH_SIZE 8192
1261 enum anv_cmd_buffer_exec_mode
{
1262 ANV_CMD_BUFFER_EXEC_MODE_PRIMARY
,
1263 ANV_CMD_BUFFER_EXEC_MODE_EMIT
,
1264 ANV_CMD_BUFFER_EXEC_MODE_GROW_AND_EMIT
,
1265 ANV_CMD_BUFFER_EXEC_MODE_CHAIN
,
1266 ANV_CMD_BUFFER_EXEC_MODE_COPY_AND_CHAIN
,
1269 struct anv_cmd_buffer
{
1270 VK_LOADER_DATA _loader_data
;
1272 struct anv_device
* device
;
1274 struct anv_cmd_pool
* pool
;
1275 struct list_head pool_link
;
1277 struct anv_batch batch
;
1279 /* Fields required for the actual chain of anv_batch_bo's.
1281 * These fields are initialized by anv_cmd_buffer_init_batch_bo_chain().
1283 struct list_head batch_bos
;
1284 enum anv_cmd_buffer_exec_mode exec_mode
;
1286 /* A vector of anv_batch_bo pointers for every batch or surface buffer
1287 * referenced by this command buffer
1289 * initialized by anv_cmd_buffer_init_batch_bo_chain()
1291 struct u_vector seen_bbos
;
1293 /* A vector of int32_t's for every block of binding tables.
1295 * initialized by anv_cmd_buffer_init_batch_bo_chain()
1297 struct u_vector bt_blocks
;
1300 struct anv_reloc_list surface_relocs
;
1301 /** Last seen surface state block pool center bo offset */
1302 uint32_t last_ss_pool_center
;
1304 /* Serial for tracking buffer completion */
1307 /* Stream objects for storing temporary data */
1308 struct anv_state_stream surface_state_stream
;
1309 struct anv_state_stream dynamic_state_stream
;
1311 VkCommandBufferUsageFlags usage_flags
;
1312 VkCommandBufferLevel level
;
1314 struct anv_cmd_state state
;
1317 VkResult
anv_cmd_buffer_init_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1318 void anv_cmd_buffer_fini_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1319 void anv_cmd_buffer_reset_batch_bo_chain(struct anv_cmd_buffer
*cmd_buffer
);
1320 void anv_cmd_buffer_end_batch_buffer(struct anv_cmd_buffer
*cmd_buffer
);
1321 void anv_cmd_buffer_add_secondary(struct anv_cmd_buffer
*primary
,
1322 struct anv_cmd_buffer
*secondary
);
1323 void anv_cmd_buffer_prepare_execbuf(struct anv_cmd_buffer
*cmd_buffer
);
1324 VkResult
anv_cmd_buffer_execbuf(struct anv_device
*device
,
1325 struct anv_cmd_buffer
*cmd_buffer
);
1327 VkResult
anv_cmd_buffer_reset(struct anv_cmd_buffer
*cmd_buffer
);
1330 anv_cmd_buffer_ensure_push_constants_size(struct anv_cmd_buffer
*cmd_buffer
,
1331 gl_shader_stage stage
, uint32_t size
);
1332 #define anv_cmd_buffer_ensure_push_constant_field(cmd_buffer, stage, field) \
1333 anv_cmd_buffer_ensure_push_constants_size(cmd_buffer, stage, \
1334 (offsetof(struct anv_push_constants, field) + \
1335 sizeof(cmd_buffer->state.push_constants[0]->field)))
1337 struct anv_state
anv_cmd_buffer_emit_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
1338 const void *data
, uint32_t size
, uint32_t alignment
);
1339 struct anv_state
anv_cmd_buffer_merge_dynamic(struct anv_cmd_buffer
*cmd_buffer
,
1340 uint32_t *a
, uint32_t *b
,
1341 uint32_t dwords
, uint32_t alignment
);
1344 anv_cmd_buffer_surface_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1346 anv_cmd_buffer_alloc_binding_table(struct anv_cmd_buffer
*cmd_buffer
,
1347 uint32_t entries
, uint32_t *state_offset
);
1349 anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer
*cmd_buffer
);
1351 anv_cmd_buffer_alloc_dynamic_state(struct anv_cmd_buffer
*cmd_buffer
,
1352 uint32_t size
, uint32_t alignment
);
1355 anv_cmd_buffer_new_binding_table_block(struct anv_cmd_buffer
*cmd_buffer
);
1357 void gen8_cmd_buffer_emit_viewport(struct anv_cmd_buffer
*cmd_buffer
);
1358 void gen8_cmd_buffer_emit_depth_viewport(struct anv_cmd_buffer
*cmd_buffer
,
1359 bool depth_clamp_enable
);
1360 void gen7_cmd_buffer_emit_scissor(struct anv_cmd_buffer
*cmd_buffer
);
1362 void anv_cmd_buffer_setup_attachments(struct anv_cmd_buffer
*cmd_buffer
,
1363 struct anv_render_pass
*pass
,
1364 struct anv_framebuffer
*framebuffer
,
1365 const VkClearValue
*clear_values
);
1367 void anv_cmd_buffer_emit_state_base_address(struct anv_cmd_buffer
*cmd_buffer
);
1370 anv_cmd_buffer_push_constants(struct anv_cmd_buffer
*cmd_buffer
,
1371 gl_shader_stage stage
);
1373 anv_cmd_buffer_cs_push_constants(struct anv_cmd_buffer
*cmd_buffer
);
1375 void anv_cmd_buffer_clear_subpass(struct anv_cmd_buffer
*cmd_buffer
);
1376 void anv_cmd_buffer_resolve_subpass(struct anv_cmd_buffer
*cmd_buffer
);
1378 const struct anv_image_view
*
1379 anv_cmd_buffer_get_depth_stencil_view(const struct anv_cmd_buffer
*cmd_buffer
);
1382 anv_cmd_buffer_alloc_blorp_binding_table(struct anv_cmd_buffer
*cmd_buffer
,
1383 uint32_t num_entries
,
1384 uint32_t *state_offset
);
1386 void anv_cmd_buffer_dump(struct anv_cmd_buffer
*cmd_buffer
);
1388 enum anv_fence_state
{
1389 /** Indicates that this is a new (or newly reset fence) */
1390 ANV_FENCE_STATE_RESET
,
1392 /** Indicates that this fence has been submitted to the GPU but is still
1393 * (as far as we know) in use by the GPU.
1395 ANV_FENCE_STATE_SUBMITTED
,
1397 ANV_FENCE_STATE_SIGNALED
,
1402 struct drm_i915_gem_execbuffer2 execbuf
;
1403 struct drm_i915_gem_exec_object2 exec2_objects
[1];
1404 enum anv_fence_state state
;
1409 struct anv_state state
;
1412 struct anv_shader_module
{
1413 unsigned char sha1
[20];
1418 void anv_hash_shader(unsigned char *hash
, const void *key
, size_t key_size
,
1419 struct anv_shader_module
*module
,
1420 const char *entrypoint
,
1421 const struct anv_pipeline_layout
*pipeline_layout
,
1422 const VkSpecializationInfo
*spec_info
);
1424 static inline gl_shader_stage
1425 vk_to_mesa_shader_stage(VkShaderStageFlagBits vk_stage
)
1427 assert(__builtin_popcount(vk_stage
) == 1);
1428 return ffs(vk_stage
) - 1;
1431 static inline VkShaderStageFlagBits
1432 mesa_to_vk_shader_stage(gl_shader_stage mesa_stage
)
1434 return (1 << mesa_stage
);
1437 #define ANV_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
1439 #define anv_foreach_stage(stage, stage_bits) \
1440 for (gl_shader_stage stage, \
1441 __tmp = (gl_shader_stage)((stage_bits) & ANV_STAGE_MASK); \
1442 stage = __builtin_ffs(__tmp) - 1, __tmp; \
1443 __tmp &= ~(1 << (stage)))
1445 struct anv_pipeline_bind_map
{
1446 uint32_t surface_count
;
1447 uint32_t sampler_count
;
1448 uint32_t image_count
;
1450 struct anv_pipeline_binding
* surface_to_descriptor
;
1451 struct anv_pipeline_binding
* sampler_to_descriptor
;
1454 struct anv_shader_bin_key
{
1459 struct anv_shader_bin
{
1462 const struct anv_shader_bin_key
*key
;
1464 struct anv_state kernel
;
1465 uint32_t kernel_size
;
1467 const struct brw_stage_prog_data
*prog_data
;
1468 uint32_t prog_data_size
;
1470 struct anv_pipeline_bind_map bind_map
;
1472 /* Prog data follows, then params, then the key, all aligned to 8-bytes */
1475 struct anv_shader_bin
*
1476 anv_shader_bin_create(struct anv_device
*device
,
1477 const void *key
, uint32_t key_size
,
1478 const void *kernel
, uint32_t kernel_size
,
1479 const struct brw_stage_prog_data
*prog_data
,
1480 uint32_t prog_data_size
, const void *prog_data_param
,
1481 const struct anv_pipeline_bind_map
*bind_map
);
1484 anv_shader_bin_destroy(struct anv_device
*device
, struct anv_shader_bin
*shader
);
1487 anv_shader_bin_ref(struct anv_shader_bin
*shader
)
1489 assert(shader
->ref_cnt
>= 1);
1490 __sync_fetch_and_add(&shader
->ref_cnt
, 1);
1494 anv_shader_bin_unref(struct anv_device
*device
, struct anv_shader_bin
*shader
)
1496 assert(shader
->ref_cnt
>= 1);
1497 if (__sync_fetch_and_add(&shader
->ref_cnt
, -1) == 1)
1498 anv_shader_bin_destroy(device
, shader
);
1501 struct anv_pipeline
{
1502 struct anv_device
* device
;
1503 struct anv_batch batch
;
1504 uint32_t batch_data
[512];
1505 struct anv_reloc_list batch_relocs
;
1506 uint32_t dynamic_state_mask
;
1507 struct anv_dynamic_state dynamic_state
;
1509 struct anv_pipeline_layout
* layout
;
1511 bool needs_data_cache
;
1513 struct anv_shader_bin
* shaders
[MESA_SHADER_STAGES
];
1516 const struct gen_l3_config
* l3_config
;
1517 uint32_t total_size
;
1520 VkShaderStageFlags active_stages
;
1521 struct anv_state blend_state
;
1524 uint32_t binding_stride
[MAX_VBS
];
1525 bool instancing_enable
[MAX_VBS
];
1526 bool primitive_restart
;
1529 uint32_t cs_right_mask
;
1532 bool depth_test_enable
;
1533 bool writes_stencil
;
1534 bool stencil_test_enable
;
1535 bool depth_clamp_enable
;
1540 uint32_t depth_stencil_state
[3];
1546 uint32_t wm_depth_stencil
[3];
1550 uint32_t wm_depth_stencil
[4];
1553 uint32_t interface_descriptor_data
[8];
1557 anv_pipeline_has_stage(const struct anv_pipeline
*pipeline
,
1558 gl_shader_stage stage
)
1560 return (pipeline
->active_stages
& mesa_to_vk_shader_stage(stage
)) != 0;
1563 #define ANV_DECL_GET_PROG_DATA_FUNC(prefix, stage) \
1564 static inline const struct brw_##prefix##_prog_data * \
1565 get_##prefix##_prog_data(const struct anv_pipeline *pipeline) \
1567 if (anv_pipeline_has_stage(pipeline, stage)) { \
1568 return (const struct brw_##prefix##_prog_data *) \
1569 pipeline->shaders[stage]->prog_data; \
1575 ANV_DECL_GET_PROG_DATA_FUNC(vs
, MESA_SHADER_VERTEX
)
1576 ANV_DECL_GET_PROG_DATA_FUNC(tcs
, MESA_SHADER_TESS_CTRL
)
1577 ANV_DECL_GET_PROG_DATA_FUNC(tes
, MESA_SHADER_TESS_EVAL
)
1578 ANV_DECL_GET_PROG_DATA_FUNC(gs
, MESA_SHADER_GEOMETRY
)
1579 ANV_DECL_GET_PROG_DATA_FUNC(wm
, MESA_SHADER_FRAGMENT
)
1580 ANV_DECL_GET_PROG_DATA_FUNC(cs
, MESA_SHADER_COMPUTE
)
1582 static inline const struct brw_vue_prog_data
*
1583 anv_pipeline_get_last_vue_prog_data(const struct anv_pipeline
*pipeline
)
1585 if (anv_pipeline_has_stage(pipeline
, MESA_SHADER_GEOMETRY
))
1586 return &get_gs_prog_data(pipeline
)->base
;
1587 else if (anv_pipeline_has_stage(pipeline
, MESA_SHADER_TESS_EVAL
))
1588 return &get_tes_prog_data(pipeline
)->base
;
1590 return &get_vs_prog_data(pipeline
)->base
;
1594 anv_pipeline_init(struct anv_pipeline
*pipeline
, struct anv_device
*device
,
1595 struct anv_pipeline_cache
*cache
,
1596 const VkGraphicsPipelineCreateInfo
*pCreateInfo
,
1597 const VkAllocationCallbacks
*alloc
);
1600 anv_pipeline_compile_cs(struct anv_pipeline
*pipeline
,
1601 struct anv_pipeline_cache
*cache
,
1602 const VkComputePipelineCreateInfo
*info
,
1603 struct anv_shader_module
*module
,
1604 const char *entrypoint
,
1605 const VkSpecializationInfo
*spec_info
);
1608 enum isl_format isl_format
:16;
1609 struct isl_swizzle swizzle
;
1613 anv_get_format(const struct gen_device_info
*devinfo
, VkFormat format
,
1614 VkImageAspectFlags aspect
, VkImageTiling tiling
);
1616 static inline enum isl_format
1617 anv_get_isl_format(const struct gen_device_info
*devinfo
, VkFormat vk_format
,
1618 VkImageAspectFlags aspect
, VkImageTiling tiling
)
1620 return anv_get_format(devinfo
, vk_format
, aspect
, tiling
).isl_format
;
1623 static inline struct isl_swizzle
1624 anv_swizzle_for_render(struct isl_swizzle swizzle
)
1626 /* Sometimes the swizzle will have alpha map to one. We do this to fake
1627 * RGB as RGBA for texturing
1629 assert(swizzle
.a
== ISL_CHANNEL_SELECT_ONE
||
1630 swizzle
.a
== ISL_CHANNEL_SELECT_ALPHA
);
1632 /* But it doesn't matter what we render to that channel */
1633 swizzle
.a
= ISL_CHANNEL_SELECT_ALPHA
;
1639 anv_pipeline_setup_l3_config(struct anv_pipeline
*pipeline
, bool needs_slm
);
1642 * Subsurface of an anv_image.
1644 struct anv_surface
{
1645 /** Valid only if isl_surf::size > 0. */
1646 struct isl_surf isl
;
1649 * Offset from VkImage's base address, as bound by vkBindImageMemory().
1656 /* The original VkFormat provided by the client. This may not match any
1657 * of the actual surface formats.
1660 VkImageAspectFlags aspects
;
1663 uint32_t array_size
;
1664 uint32_t samples
; /**< VkImageCreateInfo::samples */
1665 VkImageUsageFlags usage
; /**< Superset of VkImageCreateInfo::usage. */
1666 VkImageTiling tiling
; /** VkImageCreateInfo::tiling */
1671 /* Set when bound */
1673 VkDeviceSize offset
;
1678 * For each foo, anv_image::foo_surface is valid if and only if
1679 * anv_image::aspects has a foo aspect.
1681 * The hardware requires that the depth buffer and stencil buffer be
1682 * separate surfaces. From Vulkan's perspective, though, depth and stencil
1683 * reside in the same VkImage. To satisfy both the hardware and Vulkan, we
1684 * allocate the depth and stencil buffers as separate surfaces in the same
1688 struct anv_surface color_surface
;
1691 struct anv_surface depth_surface
;
1692 struct anv_surface stencil_surface
;
1697 * For color images, this is the aux usage for this image when not used as a
1700 * For depth/stencil images, this is set to ISL_AUX_USAGE_HIZ if the image
1703 enum isl_aux_usage aux_usage
;
1705 struct anv_surface aux_surface
;
1708 /* Returns true if a HiZ-enabled depth buffer can be sampled from. */
1710 anv_can_sample_with_hiz(uint8_t gen
, uint32_t samples
)
1712 return gen
>= 8 && samples
== 1;
1716 anv_gen8_hiz_op_resolve(struct anv_cmd_buffer
*cmd_buffer
,
1717 const struct anv_image
*image
,
1718 enum blorp_hiz_op op
);
1720 static inline uint32_t
1721 anv_get_layerCount(const struct anv_image
*image
,
1722 const VkImageSubresourceRange
*range
)
1724 return range
->layerCount
== VK_REMAINING_ARRAY_LAYERS
?
1725 image
->array_size
- range
->baseArrayLayer
: range
->layerCount
;
1728 static inline uint32_t
1729 anv_get_levelCount(const struct anv_image
*image
,
1730 const VkImageSubresourceRange
*range
)
1732 return range
->levelCount
== VK_REMAINING_MIP_LEVELS
?
1733 image
->levels
- range
->baseMipLevel
: range
->levelCount
;
1737 struct anv_image_view
{
1738 const struct anv_image
*image
; /**< VkImageViewCreateInfo::image */
1740 uint32_t offset
; /**< Offset into bo. */
1742 struct isl_view isl
;
1744 VkImageAspectFlags aspect_mask
;
1746 VkExtent3D extent
; /**< Extent of VkImageViewCreateInfo::baseMipLevel. */
1748 /** RENDER_SURFACE_STATE when using image as a sampler surface. */
1749 struct anv_state sampler_surface_state
;
1752 * RENDER_SURFACE_STATE when using image as a storage image. Separate states
1753 * for write-only and readable, using the real format for write-only and the
1754 * lowered format for readable.
1756 struct anv_state storage_surface_state
;
1757 struct anv_state writeonly_storage_surface_state
;
1759 struct brw_image_param storage_image_param
;
1762 struct anv_image_create_info
{
1763 const VkImageCreateInfo
*vk_info
;
1765 /** An opt-in bitmask which filters an ISL-mapping of the Vulkan tiling. */
1766 isl_tiling_flags_t isl_tiling_flags
;
1771 VkResult
anv_image_create(VkDevice _device
,
1772 const struct anv_image_create_info
*info
,
1773 const VkAllocationCallbacks
* alloc
,
1776 const struct anv_surface
*
1777 anv_image_get_surface_for_aspect_mask(const struct anv_image
*image
,
1778 VkImageAspectFlags aspect_mask
);
1781 anv_isl_format_for_descriptor_type(VkDescriptorType type
);
1783 static inline struct VkExtent3D
1784 anv_sanitize_image_extent(const VkImageType imageType
,
1785 const struct VkExtent3D imageExtent
)
1787 switch (imageType
) {
1788 case VK_IMAGE_TYPE_1D
:
1789 return (VkExtent3D
) { imageExtent
.width
, 1, 1 };
1790 case VK_IMAGE_TYPE_2D
:
1791 return (VkExtent3D
) { imageExtent
.width
, imageExtent
.height
, 1 };
1792 case VK_IMAGE_TYPE_3D
:
1795 unreachable("invalid image type");
1799 static inline struct VkOffset3D
1800 anv_sanitize_image_offset(const VkImageType imageType
,
1801 const struct VkOffset3D imageOffset
)
1803 switch (imageType
) {
1804 case VK_IMAGE_TYPE_1D
:
1805 return (VkOffset3D
) { imageOffset
.x
, 0, 0 };
1806 case VK_IMAGE_TYPE_2D
:
1807 return (VkOffset3D
) { imageOffset
.x
, imageOffset
.y
, 0 };
1808 case VK_IMAGE_TYPE_3D
:
1811 unreachable("invalid image type");
1816 void anv_fill_buffer_surface_state(struct anv_device
*device
,
1817 struct anv_state state
,
1818 enum isl_format format
,
1819 uint32_t offset
, uint32_t range
,
1822 void anv_image_view_fill_image_param(struct anv_device
*device
,
1823 struct anv_image_view
*view
,
1824 struct brw_image_param
*param
);
1825 void anv_buffer_view_fill_image_param(struct anv_device
*device
,
1826 struct anv_buffer_view
*view
,
1827 struct brw_image_param
*param
);
1829 struct anv_sampler
{
1833 struct anv_framebuffer
{
1838 uint32_t attachment_count
;
1839 struct anv_image_view
* attachments
[0];
1842 struct anv_subpass
{
1843 uint32_t input_count
;
1844 uint32_t * input_attachments
;
1845 uint32_t color_count
;
1846 uint32_t * color_attachments
;
1847 uint32_t * resolve_attachments
;
1849 /* TODO: Consider storing the depth/stencil VkAttachmentReference
1850 * instead of its two structure members (below) individually.
1852 uint32_t depth_stencil_attachment
;
1853 VkImageLayout depth_stencil_layout
;
1855 /** Subpass has a depth/stencil self-dependency */
1856 bool has_ds_self_dep
;
1858 /** Subpass has at least one resolve attachment */
1862 enum anv_subpass_usage
{
1863 ANV_SUBPASS_USAGE_DRAW
= (1 << 0),
1864 ANV_SUBPASS_USAGE_INPUT
= (1 << 1),
1865 ANV_SUBPASS_USAGE_RESOLVE_SRC
= (1 << 2),
1866 ANV_SUBPASS_USAGE_RESOLVE_DST
= (1 << 3),
1869 struct anv_render_pass_attachment
{
1870 /* TODO: Consider using VkAttachmentDescription instead of storing each of
1871 * its members individually.
1875 VkImageUsageFlags usage
;
1876 VkAttachmentLoadOp load_op
;
1877 VkAttachmentStoreOp store_op
;
1878 VkAttachmentLoadOp stencil_load_op
;
1879 VkImageLayout initial_layout
;
1880 VkImageLayout final_layout
;
1882 /* An array, indexed by subpass id, of how the attachment will be used. */
1883 enum anv_subpass_usage
* subpass_usage
;
1885 /* The subpass id in which the attachment will be used last. */
1886 uint32_t last_subpass_idx
;
1889 struct anv_render_pass
{
1890 uint32_t attachment_count
;
1891 uint32_t subpass_count
;
1892 uint32_t * subpass_attachments
;
1893 enum anv_subpass_usage
* subpass_usages
;
1894 struct anv_render_pass_attachment
* attachments
;
1895 struct anv_subpass subpasses
[0];
1898 struct anv_query_pool_slot
{
1904 struct anv_query_pool
{
1910 void *anv_lookup_entrypoint(const struct gen_device_info
*devinfo
,
1913 void anv_dump_image_to_ppm(struct anv_device
*device
,
1914 struct anv_image
*image
, unsigned miplevel
,
1915 unsigned array_layer
, VkImageAspectFlagBits aspect
,
1916 const char *filename
);
1918 enum anv_dump_action
{
1919 ANV_DUMP_FRAMEBUFFERS_BIT
= 0x1,
1922 void anv_dump_start(struct anv_device
*device
, enum anv_dump_action actions
);
1923 void anv_dump_finish(void);
1925 void anv_dump_add_framebuffer(struct anv_cmd_buffer
*cmd_buffer
,
1926 struct anv_framebuffer
*fb
);
1928 #define ANV_DEFINE_HANDLE_CASTS(__anv_type, __VkType) \
1930 static inline struct __anv_type * \
1931 __anv_type ## _from_handle(__VkType _handle) \
1933 return (struct __anv_type *) _handle; \
1936 static inline __VkType \
1937 __anv_type ## _to_handle(struct __anv_type *_obj) \
1939 return (__VkType) _obj; \
1942 #define ANV_DEFINE_NONDISP_HANDLE_CASTS(__anv_type, __VkType) \
1944 static inline struct __anv_type * \
1945 __anv_type ## _from_handle(__VkType _handle) \
1947 return (struct __anv_type *)(uintptr_t) _handle; \
1950 static inline __VkType \
1951 __anv_type ## _to_handle(struct __anv_type *_obj) \
1953 return (__VkType)(uintptr_t) _obj; \
1956 #define ANV_FROM_HANDLE(__anv_type, __name, __handle) \
1957 struct __anv_type *__name = __anv_type ## _from_handle(__handle)
1959 ANV_DEFINE_HANDLE_CASTS(anv_cmd_buffer
, VkCommandBuffer
)
1960 ANV_DEFINE_HANDLE_CASTS(anv_device
, VkDevice
)
1961 ANV_DEFINE_HANDLE_CASTS(anv_instance
, VkInstance
)
1962 ANV_DEFINE_HANDLE_CASTS(anv_physical_device
, VkPhysicalDevice
)
1963 ANV_DEFINE_HANDLE_CASTS(anv_queue
, VkQueue
)
1965 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_cmd_pool
, VkCommandPool
)
1966 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer
, VkBuffer
)
1967 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer_view
, VkBufferView
)
1968 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_pool
, VkDescriptorPool
)
1969 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set
, VkDescriptorSet
)
1970 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set_layout
, VkDescriptorSetLayout
)
1971 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_device_memory
, VkDeviceMemory
)
1972 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_fence
, VkFence
)
1973 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_event
, VkEvent
)
1974 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_framebuffer
, VkFramebuffer
)
1975 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image
, VkImage
)
1976 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image_view
, VkImageView
);
1977 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_cache
, VkPipelineCache
)
1978 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline
, VkPipeline
)
1979 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_layout
, VkPipelineLayout
)
1980 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_query_pool
, VkQueryPool
)
1981 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_render_pass
, VkRenderPass
)
1982 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_sampler
, VkSampler
)
1983 ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_shader_module
, VkShaderModule
)
1985 /* Gen-specific function declarations */
1987 # include "anv_genX.h"
1989 # define genX(x) gen7_##x
1990 # include "anv_genX.h"
1992 # define genX(x) gen75_##x
1993 # include "anv_genX.h"
1995 # define genX(x) gen8_##x
1996 # include "anv_genX.h"
1998 # define genX(x) gen9_##x
1999 # include "anv_genX.h"
2003 #endif /* ANV_PRIVATE_H */