2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #include "radv_debug.h"
33 #include "radv_private.h"
34 #include "radv_shader.h"
36 #include "util/disk_cache.h"
37 #include "util/strtod.h"
41 #include <amdgpu_drm.h>
42 #include "winsys/amdgpu/radv_amdgpu_winsys_public.h"
43 #include "ac_llvm_util.h"
44 #include "vk_format.h"
48 #include "addrlib/gfx9/chip/gfx9_enum.h"
49 #include "util/build_id.h"
50 #include "util/debug.h"
51 #include "util/mesa-sha1.h"
54 radv_device_get_cache_uuid(enum radeon_family family
, void *uuid
)
57 unsigned char sha1
[20];
58 unsigned ptr_size
= sizeof(void*);
60 memset(uuid
, 0, VK_UUID_SIZE
);
61 _mesa_sha1_init(&ctx
);
63 if (!disk_cache_get_function_identifier(radv_device_get_cache_uuid
, &ctx
) ||
64 !disk_cache_get_function_identifier(LLVMInitializeAMDGPUTargetInfo
, &ctx
))
67 _mesa_sha1_update(&ctx
, &family
, sizeof(family
));
68 _mesa_sha1_update(&ctx
, &ptr_size
, sizeof(ptr_size
));
69 _mesa_sha1_final(&ctx
, sha1
);
71 memcpy(uuid
, sha1
, VK_UUID_SIZE
);
76 radv_get_driver_uuid(void *uuid
)
78 ac_compute_driver_uuid(uuid
, VK_UUID_SIZE
);
82 radv_get_device_uuid(struct radeon_info
*info
, void *uuid
)
84 ac_compute_device_uuid(info
, uuid
, VK_UUID_SIZE
);
88 radv_get_device_name(enum radeon_family family
, char *name
, size_t name_len
)
90 const char *chip_string
;
91 char llvm_string
[32] = {};
94 case CHIP_TAHITI
: chip_string
= "AMD RADV TAHITI"; break;
95 case CHIP_PITCAIRN
: chip_string
= "AMD RADV PITCAIRN"; break;
96 case CHIP_VERDE
: chip_string
= "AMD RADV CAPE VERDE"; break;
97 case CHIP_OLAND
: chip_string
= "AMD RADV OLAND"; break;
98 case CHIP_HAINAN
: chip_string
= "AMD RADV HAINAN"; break;
99 case CHIP_BONAIRE
: chip_string
= "AMD RADV BONAIRE"; break;
100 case CHIP_KAVERI
: chip_string
= "AMD RADV KAVERI"; break;
101 case CHIP_KABINI
: chip_string
= "AMD RADV KABINI"; break;
102 case CHIP_HAWAII
: chip_string
= "AMD RADV HAWAII"; break;
103 case CHIP_MULLINS
: chip_string
= "AMD RADV MULLINS"; break;
104 case CHIP_TONGA
: chip_string
= "AMD RADV TONGA"; break;
105 case CHIP_ICELAND
: chip_string
= "AMD RADV ICELAND"; break;
106 case CHIP_CARRIZO
: chip_string
= "AMD RADV CARRIZO"; break;
107 case CHIP_FIJI
: chip_string
= "AMD RADV FIJI"; break;
108 case CHIP_POLARIS10
: chip_string
= "AMD RADV POLARIS10"; break;
109 case CHIP_POLARIS11
: chip_string
= "AMD RADV POLARIS11"; break;
110 case CHIP_POLARIS12
: chip_string
= "AMD RADV POLARIS12"; break;
111 case CHIP_STONEY
: chip_string
= "AMD RADV STONEY"; break;
112 case CHIP_VEGAM
: chip_string
= "AMD RADV VEGA M"; break;
113 case CHIP_VEGA10
: chip_string
= "AMD RADV VEGA10"; break;
114 case CHIP_VEGA12
: chip_string
= "AMD RADV VEGA12"; break;
115 case CHIP_RAVEN
: chip_string
= "AMD RADV RAVEN"; break;
116 default: chip_string
= "AMD RADV unknown"; break;
119 snprintf(llvm_string
, sizeof(llvm_string
),
120 " (LLVM %i.%i.%i)", (HAVE_LLVM
>> 8) & 0xff,
121 HAVE_LLVM
& 0xff, MESA_LLVM_VERSION_PATCH
);
122 snprintf(name
, name_len
, "%s%s", chip_string
, llvm_string
);
126 radv_physical_device_init_mem_types(struct radv_physical_device
*device
)
128 STATIC_ASSERT(RADV_MEM_HEAP_COUNT
<= VK_MAX_MEMORY_HEAPS
);
129 uint64_t visible_vram_size
= MIN2(device
->rad_info
.vram_size
,
130 device
->rad_info
.vram_vis_size
);
132 int vram_index
= -1, visible_vram_index
= -1, gart_index
= -1;
133 device
->memory_properties
.memoryHeapCount
= 0;
134 if (device
->rad_info
.vram_size
- visible_vram_size
> 0) {
135 vram_index
= device
->memory_properties
.memoryHeapCount
++;
136 device
->memory_properties
.memoryHeaps
[vram_index
] = (VkMemoryHeap
) {
137 .size
= device
->rad_info
.vram_size
- visible_vram_size
,
138 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
141 if (visible_vram_size
) {
142 visible_vram_index
= device
->memory_properties
.memoryHeapCount
++;
143 device
->memory_properties
.memoryHeaps
[visible_vram_index
] = (VkMemoryHeap
) {
144 .size
= visible_vram_size
,
145 .flags
= VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
148 if (device
->rad_info
.gart_size
> 0) {
149 gart_index
= device
->memory_properties
.memoryHeapCount
++;
150 device
->memory_properties
.memoryHeaps
[gart_index
] = (VkMemoryHeap
) {
151 .size
= device
->rad_info
.gart_size
,
152 .flags
= device
->rad_info
.has_dedicated_vram
? 0 : VK_MEMORY_HEAP_DEVICE_LOCAL_BIT
,
156 STATIC_ASSERT(RADV_MEM_TYPE_COUNT
<= VK_MAX_MEMORY_TYPES
);
157 unsigned type_count
= 0;
158 if (vram_index
>= 0) {
159 device
->mem_type_indices
[type_count
] = RADV_MEM_TYPE_VRAM
;
160 device
->memory_properties
.memoryTypes
[type_count
++] = (VkMemoryType
) {
161 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
,
162 .heapIndex
= vram_index
,
165 if (gart_index
>= 0) {
166 device
->mem_type_indices
[type_count
] = RADV_MEM_TYPE_GTT_WRITE_COMBINE
;
167 device
->memory_properties
.memoryTypes
[type_count
++] = (VkMemoryType
) {
168 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
169 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
170 (device
->rad_info
.has_dedicated_vram
? 0 : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
),
171 .heapIndex
= gart_index
,
174 if (visible_vram_index
>= 0) {
175 device
->mem_type_indices
[type_count
] = RADV_MEM_TYPE_VRAM_CPU_ACCESS
;
176 device
->memory_properties
.memoryTypes
[type_count
++] = (VkMemoryType
) {
177 .propertyFlags
= VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
|
178 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
179 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
,
180 .heapIndex
= visible_vram_index
,
183 if (gart_index
>= 0) {
184 device
->mem_type_indices
[type_count
] = RADV_MEM_TYPE_GTT_CACHED
;
185 device
->memory_properties
.memoryTypes
[type_count
++] = (VkMemoryType
) {
186 .propertyFlags
= VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT
|
187 VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
|
188 VK_MEMORY_PROPERTY_HOST_CACHED_BIT
|
189 (device
->rad_info
.has_dedicated_vram
? 0 : VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT
),
190 .heapIndex
= gart_index
,
193 device
->memory_properties
.memoryTypeCount
= type_count
;
197 radv_handle_env_var_force_family(struct radv_physical_device
*device
)
199 const char *family
= getenv("RADV_FORCE_FAMILY");
205 for (i
= CHIP_TAHITI
; i
< CHIP_LAST
; i
++) {
206 if (!strcmp(family
, ac_get_llvm_processor_name(i
))) {
207 /* Override family and chip_class. */
208 device
->rad_info
.family
= i
;
210 if (i
>= CHIP_VEGA10
)
211 device
->rad_info
.chip_class
= GFX9
;
212 else if (i
>= CHIP_TONGA
)
213 device
->rad_info
.chip_class
= VI
;
214 else if (i
>= CHIP_BONAIRE
)
215 device
->rad_info
.chip_class
= CIK
;
217 device
->rad_info
.chip_class
= SI
;
223 fprintf(stderr
, "radv: Unknown family: %s\n", family
);
228 radv_physical_device_init(struct radv_physical_device
*device
,
229 struct radv_instance
*instance
,
230 drmDevicePtr drm_device
)
232 const char *path
= drm_device
->nodes
[DRM_NODE_RENDER
];
234 drmVersionPtr version
;
238 fd
= open(path
, O_RDWR
| O_CLOEXEC
);
240 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
241 radv_logi("Could not open device '%s'", path
);
243 return vk_error(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
);
246 version
= drmGetVersion(fd
);
250 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
251 radv_logi("Could not get the kernel driver version for device '%s'", path
);
253 return vk_errorf(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
,
254 "failed to get version %s: %m", path
);
257 if (strcmp(version
->name
, "amdgpu")) {
258 drmFreeVersion(version
);
261 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
262 radv_logi("Device '%s' is not using the amdgpu kernel driver.", path
);
264 return VK_ERROR_INCOMPATIBLE_DRIVER
;
266 drmFreeVersion(version
);
268 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
269 radv_logi("Found compatible device '%s'.", path
);
271 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
272 device
->instance
= instance
;
273 assert(strlen(path
) < ARRAY_SIZE(device
->path
));
274 strncpy(device
->path
, path
, ARRAY_SIZE(device
->path
));
276 device
->ws
= radv_amdgpu_winsys_create(fd
, instance
->debug_flags
,
277 instance
->perftest_flags
);
279 result
= vk_error(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
);
283 if (instance
->enabled_extensions
.KHR_display
) {
284 master_fd
= open(drm_device
->nodes
[DRM_NODE_PRIMARY
], O_RDWR
| O_CLOEXEC
);
285 if (master_fd
>= 0) {
286 uint32_t accel_working
= 0;
287 struct drm_amdgpu_info request
= {
288 .return_pointer
= (uintptr_t)&accel_working
,
289 .return_size
= sizeof(accel_working
),
290 .query
= AMDGPU_INFO_ACCEL_WORKING
293 if (drmCommandWrite(master_fd
, DRM_AMDGPU_INFO
, &request
, sizeof (struct drm_amdgpu_info
)) < 0 || !accel_working
) {
300 device
->master_fd
= master_fd
;
301 device
->local_fd
= fd
;
302 device
->ws
->query_info(device
->ws
, &device
->rad_info
);
304 radv_handle_env_var_force_family(device
);
306 radv_get_device_name(device
->rad_info
.family
, device
->name
, sizeof(device
->name
));
308 if (radv_device_get_cache_uuid(device
->rad_info
.family
, device
->cache_uuid
)) {
309 device
->ws
->destroy(device
->ws
);
310 result
= vk_errorf(instance
, VK_ERROR_INITIALIZATION_FAILED
,
311 "cannot generate UUID");
315 /* These flags affect shader compilation. */
316 uint64_t shader_env_flags
=
317 (device
->instance
->perftest_flags
& RADV_PERFTEST_SISCHED
? 0x1 : 0) |
318 (device
->instance
->debug_flags
& RADV_DEBUG_UNSAFE_MATH
? 0x2 : 0);
320 /* The gpu id is already embedded in the uuid so we just pass "radv"
321 * when creating the cache.
323 char buf
[VK_UUID_SIZE
* 2 + 1];
324 disk_cache_format_hex_id(buf
, device
->cache_uuid
, VK_UUID_SIZE
* 2);
325 device
->disk_cache
= disk_cache_create(device
->name
, buf
, shader_env_flags
);
327 if (device
->rad_info
.chip_class
< VI
||
328 device
->rad_info
.chip_class
> GFX9
)
329 fprintf(stderr
, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
331 radv_get_driver_uuid(&device
->device_uuid
);
332 radv_get_device_uuid(&device
->rad_info
, &device
->device_uuid
);
334 if (device
->rad_info
.family
== CHIP_STONEY
||
335 device
->rad_info
.chip_class
>= GFX9
) {
336 device
->has_rbplus
= true;
337 device
->rbplus_allowed
= device
->rad_info
.family
== CHIP_STONEY
||
338 device
->rad_info
.family
== CHIP_VEGA12
||
339 device
->rad_info
.family
== CHIP_RAVEN
;
342 /* The mere presence of CLEAR_STATE in the IB causes random GPU hangs
345 device
->has_clear_state
= device
->rad_info
.chip_class
>= CIK
;
347 device
->cpdma_prefetch_writes_memory
= device
->rad_info
.chip_class
<= VI
;
349 /* Vega10/Raven need a special workaround for a hardware bug. */
350 device
->has_scissor_bug
= device
->rad_info
.family
== CHIP_VEGA10
||
351 device
->rad_info
.family
== CHIP_RAVEN
;
353 /* Out-of-order primitive rasterization. */
354 device
->has_out_of_order_rast
= device
->rad_info
.chip_class
>= VI
&&
355 device
->rad_info
.max_se
>= 2;
356 device
->out_of_order_rast_allowed
= device
->has_out_of_order_rast
&&
357 !(device
->instance
->debug_flags
& RADV_DEBUG_NO_OUT_OF_ORDER
);
359 device
->dcc_msaa_allowed
=
360 (device
->instance
->perftest_flags
& RADV_PERFTEST_DCC_MSAA
);
362 radv_physical_device_init_mem_types(device
);
363 radv_fill_device_extension_table(device
, &device
->supported_extensions
);
365 device
->bus_info
= *drm_device
->businfo
.pci
;
367 if ((device
->instance
->debug_flags
& RADV_DEBUG_INFO
))
368 ac_print_gpu_info(&device
->rad_info
);
370 /* The WSI is structured as a layer on top of the driver, so this has
371 * to be the last part of initialization (at least until we get other
374 result
= radv_init_wsi(device
);
375 if (result
!= VK_SUCCESS
) {
376 device
->ws
->destroy(device
->ws
);
377 vk_error(instance
, result
);
391 radv_physical_device_finish(struct radv_physical_device
*device
)
393 radv_finish_wsi(device
);
394 device
->ws
->destroy(device
->ws
);
395 disk_cache_destroy(device
->disk_cache
);
396 close(device
->local_fd
);
397 if (device
->master_fd
!= -1)
398 close(device
->master_fd
);
402 default_alloc_func(void *pUserData
, size_t size
, size_t align
,
403 VkSystemAllocationScope allocationScope
)
409 default_realloc_func(void *pUserData
, void *pOriginal
, size_t size
,
410 size_t align
, VkSystemAllocationScope allocationScope
)
412 return realloc(pOriginal
, size
);
416 default_free_func(void *pUserData
, void *pMemory
)
421 static const VkAllocationCallbacks default_alloc
= {
423 .pfnAllocation
= default_alloc_func
,
424 .pfnReallocation
= default_realloc_func
,
425 .pfnFree
= default_free_func
,
428 static const struct debug_control radv_debug_options
[] = {
429 {"nofastclears", RADV_DEBUG_NO_FAST_CLEARS
},
430 {"nodcc", RADV_DEBUG_NO_DCC
},
431 {"shaders", RADV_DEBUG_DUMP_SHADERS
},
432 {"nocache", RADV_DEBUG_NO_CACHE
},
433 {"shaderstats", RADV_DEBUG_DUMP_SHADER_STATS
},
434 {"nohiz", RADV_DEBUG_NO_HIZ
},
435 {"nocompute", RADV_DEBUG_NO_COMPUTE_QUEUE
},
436 {"unsafemath", RADV_DEBUG_UNSAFE_MATH
},
437 {"allbos", RADV_DEBUG_ALL_BOS
},
438 {"noibs", RADV_DEBUG_NO_IBS
},
439 {"spirv", RADV_DEBUG_DUMP_SPIRV
},
440 {"vmfaults", RADV_DEBUG_VM_FAULTS
},
441 {"zerovram", RADV_DEBUG_ZERO_VRAM
},
442 {"syncshaders", RADV_DEBUG_SYNC_SHADERS
},
443 {"nosisched", RADV_DEBUG_NO_SISCHED
},
444 {"preoptir", RADV_DEBUG_PREOPTIR
},
445 {"nodynamicbounds", RADV_DEBUG_NO_DYNAMIC_BOUNDS
},
446 {"nooutoforder", RADV_DEBUG_NO_OUT_OF_ORDER
},
447 {"info", RADV_DEBUG_INFO
},
448 {"errors", RADV_DEBUG_ERRORS
},
449 {"startup", RADV_DEBUG_STARTUP
},
450 {"checkir", RADV_DEBUG_CHECKIR
},
451 {"nothreadllvm", RADV_DEBUG_NOTHREADLLVM
},
456 radv_get_debug_option_name(int id
)
458 assert(id
< ARRAY_SIZE(radv_debug_options
) - 1);
459 return radv_debug_options
[id
].string
;
462 static const struct debug_control radv_perftest_options
[] = {
463 {"nobatchchain", RADV_PERFTEST_NO_BATCHCHAIN
},
464 {"sisched", RADV_PERFTEST_SISCHED
},
465 {"localbos", RADV_PERFTEST_LOCAL_BOS
},
466 {"binning", RADV_PERFTEST_BINNING
},
467 {"dccmsaa", RADV_PERFTEST_DCC_MSAA
},
472 radv_get_perftest_option_name(int id
)
474 assert(id
< ARRAY_SIZE(radv_perftest_options
) - 1);
475 return radv_perftest_options
[id
].string
;
479 radv_handle_per_app_options(struct radv_instance
*instance
,
480 const VkApplicationInfo
*info
)
482 const char *name
= info
? info
->pApplicationName
: NULL
;
487 if (!strcmp(name
, "Talos - Linux - 32bit") ||
488 !strcmp(name
, "Talos - Linux - 64bit")) {
489 if (!(instance
->debug_flags
& RADV_DEBUG_NO_SISCHED
)) {
490 /* Force enable LLVM sisched for Talos because it looks
491 * safe and it gives few more FPS.
493 instance
->perftest_flags
|= RADV_PERFTEST_SISCHED
;
495 } else if (!strcmp(name
, "DOOM_VFR")) {
496 /* Work around a Doom VFR game bug */
497 instance
->debug_flags
|= RADV_DEBUG_NO_DYNAMIC_BOUNDS
;
501 static int radv_get_instance_extension_index(const char *name
)
503 for (unsigned i
= 0; i
< RADV_INSTANCE_EXTENSION_COUNT
; ++i
) {
504 if (strcmp(name
, radv_instance_extensions
[i
].extensionName
) == 0)
511 VkResult
radv_CreateInstance(
512 const VkInstanceCreateInfo
* pCreateInfo
,
513 const VkAllocationCallbacks
* pAllocator
,
514 VkInstance
* pInstance
)
516 struct radv_instance
*instance
;
519 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO
);
521 uint32_t client_version
;
522 if (pCreateInfo
->pApplicationInfo
&&
523 pCreateInfo
->pApplicationInfo
->apiVersion
!= 0) {
524 client_version
= pCreateInfo
->pApplicationInfo
->apiVersion
;
526 radv_EnumerateInstanceVersion(&client_version
);
529 instance
= vk_zalloc2(&default_alloc
, pAllocator
, sizeof(*instance
), 8,
530 VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE
);
532 return vk_error(NULL
, VK_ERROR_OUT_OF_HOST_MEMORY
);
534 instance
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
537 instance
->alloc
= *pAllocator
;
539 instance
->alloc
= default_alloc
;
541 instance
->apiVersion
= client_version
;
542 instance
->physicalDeviceCount
= -1;
544 instance
->debug_flags
= parse_debug_string(getenv("RADV_DEBUG"),
547 instance
->perftest_flags
= parse_debug_string(getenv("RADV_PERFTEST"),
548 radv_perftest_options
);
551 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
552 radv_logi("Created an instance");
554 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
555 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
556 int index
= radv_get_instance_extension_index(ext_name
);
558 if (index
< 0 || !radv_supported_instance_extensions
.extensions
[index
]) {
559 vk_free2(&default_alloc
, pAllocator
, instance
);
560 return vk_error(instance
, VK_ERROR_EXTENSION_NOT_PRESENT
);
563 instance
->enabled_extensions
.extensions
[index
] = true;
566 result
= vk_debug_report_instance_init(&instance
->debug_report_callbacks
);
567 if (result
!= VK_SUCCESS
) {
568 vk_free2(&default_alloc
, pAllocator
, instance
);
569 return vk_error(instance
, result
);
574 VG(VALGRIND_CREATE_MEMPOOL(instance
, 0, false));
576 radv_handle_per_app_options(instance
, pCreateInfo
->pApplicationInfo
);
578 *pInstance
= radv_instance_to_handle(instance
);
583 void radv_DestroyInstance(
584 VkInstance _instance
,
585 const VkAllocationCallbacks
* pAllocator
)
587 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
592 for (int i
= 0; i
< instance
->physicalDeviceCount
; ++i
) {
593 radv_physical_device_finish(instance
->physicalDevices
+ i
);
596 VG(VALGRIND_DESTROY_MEMPOOL(instance
));
600 vk_debug_report_instance_destroy(&instance
->debug_report_callbacks
);
602 vk_free(&instance
->alloc
, instance
);
606 radv_enumerate_devices(struct radv_instance
*instance
)
608 /* TODO: Check for more devices ? */
609 drmDevicePtr devices
[8];
610 VkResult result
= VK_ERROR_INCOMPATIBLE_DRIVER
;
613 instance
->physicalDeviceCount
= 0;
615 max_devices
= drmGetDevices2(0, devices
, ARRAY_SIZE(devices
));
617 if (instance
->debug_flags
& RADV_DEBUG_STARTUP
)
618 radv_logi("Found %d drm nodes", max_devices
);
621 return vk_error(instance
, VK_ERROR_INCOMPATIBLE_DRIVER
);
623 for (unsigned i
= 0; i
< (unsigned)max_devices
; i
++) {
624 if (devices
[i
]->available_nodes
& 1 << DRM_NODE_RENDER
&&
625 devices
[i
]->bustype
== DRM_BUS_PCI
&&
626 devices
[i
]->deviceinfo
.pci
->vendor_id
== ATI_VENDOR_ID
) {
628 result
= radv_physical_device_init(instance
->physicalDevices
+
629 instance
->physicalDeviceCount
,
632 if (result
== VK_SUCCESS
)
633 ++instance
->physicalDeviceCount
;
634 else if (result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
638 drmFreeDevices(devices
, max_devices
);
643 VkResult
radv_EnumeratePhysicalDevices(
644 VkInstance _instance
,
645 uint32_t* pPhysicalDeviceCount
,
646 VkPhysicalDevice
* pPhysicalDevices
)
648 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
651 if (instance
->physicalDeviceCount
< 0) {
652 result
= radv_enumerate_devices(instance
);
653 if (result
!= VK_SUCCESS
&&
654 result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
658 if (!pPhysicalDevices
) {
659 *pPhysicalDeviceCount
= instance
->physicalDeviceCount
;
661 *pPhysicalDeviceCount
= MIN2(*pPhysicalDeviceCount
, instance
->physicalDeviceCount
);
662 for (unsigned i
= 0; i
< *pPhysicalDeviceCount
; ++i
)
663 pPhysicalDevices
[i
] = radv_physical_device_to_handle(instance
->physicalDevices
+ i
);
666 return *pPhysicalDeviceCount
< instance
->physicalDeviceCount
? VK_INCOMPLETE
670 VkResult
radv_EnumeratePhysicalDeviceGroups(
671 VkInstance _instance
,
672 uint32_t* pPhysicalDeviceGroupCount
,
673 VkPhysicalDeviceGroupProperties
* pPhysicalDeviceGroupProperties
)
675 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
678 if (instance
->physicalDeviceCount
< 0) {
679 result
= radv_enumerate_devices(instance
);
680 if (result
!= VK_SUCCESS
&&
681 result
!= VK_ERROR_INCOMPATIBLE_DRIVER
)
685 if (!pPhysicalDeviceGroupProperties
) {
686 *pPhysicalDeviceGroupCount
= instance
->physicalDeviceCount
;
688 *pPhysicalDeviceGroupCount
= MIN2(*pPhysicalDeviceGroupCount
, instance
->physicalDeviceCount
);
689 for (unsigned i
= 0; i
< *pPhysicalDeviceGroupCount
; ++i
) {
690 pPhysicalDeviceGroupProperties
[i
].physicalDeviceCount
= 1;
691 pPhysicalDeviceGroupProperties
[i
].physicalDevices
[0] = radv_physical_device_to_handle(instance
->physicalDevices
+ i
);
692 pPhysicalDeviceGroupProperties
[i
].subsetAllocation
= false;
695 return *pPhysicalDeviceGroupCount
< instance
->physicalDeviceCount
? VK_INCOMPLETE
699 void radv_GetPhysicalDeviceFeatures(
700 VkPhysicalDevice physicalDevice
,
701 VkPhysicalDeviceFeatures
* pFeatures
)
703 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
704 memset(pFeatures
, 0, sizeof(*pFeatures
));
706 *pFeatures
= (VkPhysicalDeviceFeatures
) {
707 .robustBufferAccess
= true,
708 .fullDrawIndexUint32
= true,
709 .imageCubeArray
= true,
710 .independentBlend
= true,
711 .geometryShader
= true,
712 .tessellationShader
= true,
713 .sampleRateShading
= true,
714 .dualSrcBlend
= true,
716 .multiDrawIndirect
= true,
717 .drawIndirectFirstInstance
= true,
719 .depthBiasClamp
= true,
720 .fillModeNonSolid
= true,
725 .multiViewport
= true,
726 .samplerAnisotropy
= true,
727 .textureCompressionETC2
= pdevice
->rad_info
.chip_class
>= GFX9
||
728 pdevice
->rad_info
.family
== CHIP_STONEY
,
729 .textureCompressionASTC_LDR
= false,
730 .textureCompressionBC
= true,
731 .occlusionQueryPrecise
= true,
732 .pipelineStatisticsQuery
= true,
733 .vertexPipelineStoresAndAtomics
= true,
734 .fragmentStoresAndAtomics
= true,
735 .shaderTessellationAndGeometryPointSize
= true,
736 .shaderImageGatherExtended
= true,
737 .shaderStorageImageExtendedFormats
= true,
738 .shaderStorageImageMultisample
= false,
739 .shaderUniformBufferArrayDynamicIndexing
= true,
740 .shaderSampledImageArrayDynamicIndexing
= true,
741 .shaderStorageBufferArrayDynamicIndexing
= true,
742 .shaderStorageImageArrayDynamicIndexing
= true,
743 .shaderStorageImageReadWithoutFormat
= true,
744 .shaderStorageImageWriteWithoutFormat
= true,
745 .shaderClipDistance
= true,
746 .shaderCullDistance
= true,
747 .shaderFloat64
= true,
749 .shaderInt16
= pdevice
->rad_info
.chip_class
>= GFX9
&& HAVE_LLVM
>= 0x700,
750 .sparseBinding
= true,
751 .variableMultisampleRate
= true,
752 .inheritedQueries
= true,
756 void radv_GetPhysicalDeviceFeatures2(
757 VkPhysicalDevice physicalDevice
,
758 VkPhysicalDeviceFeatures2KHR
*pFeatures
)
760 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
761 vk_foreach_struct(ext
, pFeatures
->pNext
) {
762 switch (ext
->sType
) {
763 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTER_FEATURES_KHR
: {
764 VkPhysicalDeviceVariablePointerFeaturesKHR
*features
= (void *)ext
;
765 features
->variablePointersStorageBuffer
= true;
766 features
->variablePointers
= false;
769 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES_KHR
: {
770 VkPhysicalDeviceMultiviewFeaturesKHR
*features
= (VkPhysicalDeviceMultiviewFeaturesKHR
*)ext
;
771 features
->multiview
= true;
772 features
->multiviewGeometryShader
= true;
773 features
->multiviewTessellationShader
= true;
776 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETER_FEATURES
: {
777 VkPhysicalDeviceShaderDrawParameterFeatures
*features
=
778 (VkPhysicalDeviceShaderDrawParameterFeatures
*)ext
;
779 features
->shaderDrawParameters
= true;
782 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES
: {
783 VkPhysicalDeviceProtectedMemoryFeatures
*features
=
784 (VkPhysicalDeviceProtectedMemoryFeatures
*)ext
;
785 features
->protectedMemory
= false;
788 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES
: {
789 VkPhysicalDevice16BitStorageFeatures
*features
=
790 (VkPhysicalDevice16BitStorageFeatures
*)ext
;
791 bool enabled
= HAVE_LLVM
>= 0x0700 && pdevice
->rad_info
.chip_class
>= VI
;
792 features
->storageBuffer16BitAccess
= enabled
;
793 features
->uniformAndStorageBuffer16BitAccess
= enabled
;
794 features
->storagePushConstant16
= enabled
;
795 features
->storageInputOutput16
= enabled
;
798 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES
: {
799 VkPhysicalDeviceSamplerYcbcrConversionFeatures
*features
=
800 (VkPhysicalDeviceSamplerYcbcrConversionFeatures
*)ext
;
801 features
->samplerYcbcrConversion
= false;
804 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT
: {
805 VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*features
=
806 (VkPhysicalDeviceDescriptorIndexingFeaturesEXT
*)ext
;
807 features
->shaderInputAttachmentArrayDynamicIndexing
= true;
808 features
->shaderUniformTexelBufferArrayDynamicIndexing
= true;
809 features
->shaderStorageTexelBufferArrayDynamicIndexing
= true;
810 features
->shaderUniformBufferArrayNonUniformIndexing
= false;
811 features
->shaderSampledImageArrayNonUniformIndexing
= false;
812 features
->shaderStorageBufferArrayNonUniformIndexing
= false;
813 features
->shaderStorageImageArrayNonUniformIndexing
= false;
814 features
->shaderInputAttachmentArrayNonUniformIndexing
= false;
815 features
->shaderUniformTexelBufferArrayNonUniformIndexing
= false;
816 features
->shaderStorageTexelBufferArrayNonUniformIndexing
= false;
817 features
->descriptorBindingUniformBufferUpdateAfterBind
= true;
818 features
->descriptorBindingSampledImageUpdateAfterBind
= true;
819 features
->descriptorBindingStorageImageUpdateAfterBind
= true;
820 features
->descriptorBindingStorageBufferUpdateAfterBind
= true;
821 features
->descriptorBindingUniformTexelBufferUpdateAfterBind
= true;
822 features
->descriptorBindingStorageTexelBufferUpdateAfterBind
= true;
823 features
->descriptorBindingUpdateUnusedWhilePending
= true;
824 features
->descriptorBindingPartiallyBound
= true;
825 features
->descriptorBindingVariableDescriptorCount
= true;
826 features
->runtimeDescriptorArray
= true;
829 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT
: {
830 VkPhysicalDeviceConditionalRenderingFeaturesEXT
*features
=
831 (VkPhysicalDeviceConditionalRenderingFeaturesEXT
*)ext
;
832 features
->conditionalRendering
= true;
833 features
->inheritedConditionalRendering
= false;
836 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT
: {
837 VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*features
=
838 (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT
*)ext
;
839 features
->vertexAttributeInstanceRateDivisor
= VK_TRUE
;
840 features
->vertexAttributeInstanceRateZeroDivisor
= VK_TRUE
;
843 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT
: {
844 VkPhysicalDeviceTransformFeedbackFeaturesEXT
*features
=
845 (VkPhysicalDeviceTransformFeedbackFeaturesEXT
*)ext
;
846 features
->transformFeedback
= true;
847 features
->geometryStreams
= true;
854 return radv_GetPhysicalDeviceFeatures(physicalDevice
, &pFeatures
->features
);
857 void radv_GetPhysicalDeviceProperties(
858 VkPhysicalDevice physicalDevice
,
859 VkPhysicalDeviceProperties
* pProperties
)
861 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
862 VkSampleCountFlags sample_counts
= 0xf;
864 /* make sure that the entire descriptor set is addressable with a signed
865 * 32-bit int. So the sum of all limits scaled by descriptor size has to
866 * be at most 2 GiB. the combined image & samples object count as one of
867 * both. This limit is for the pipeline layout, not for the set layout, but
868 * there is no set limit, so we just set a pipeline limit. I don't think
869 * any app is going to hit this soon. */
870 size_t max_descriptor_set_size
= ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS
) /
871 (32 /* uniform buffer, 32 due to potential space wasted on alignment */ +
872 32 /* storage buffer, 32 due to potential space wasted on alignment */ +
873 32 /* sampler, largest when combined with image */ +
874 64 /* sampled image */ +
875 64 /* storage image */);
877 VkPhysicalDeviceLimits limits
= {
878 .maxImageDimension1D
= (1 << 14),
879 .maxImageDimension2D
= (1 << 14),
880 .maxImageDimension3D
= (1 << 11),
881 .maxImageDimensionCube
= (1 << 14),
882 .maxImageArrayLayers
= (1 << 11),
883 .maxTexelBufferElements
= 128 * 1024 * 1024,
884 .maxUniformBufferRange
= UINT32_MAX
,
885 .maxStorageBufferRange
= UINT32_MAX
,
886 .maxPushConstantsSize
= MAX_PUSH_CONSTANTS_SIZE
,
887 .maxMemoryAllocationCount
= UINT32_MAX
,
888 .maxSamplerAllocationCount
= 64 * 1024,
889 .bufferImageGranularity
= 64, /* A cache line */
890 .sparseAddressSpaceSize
= 0xffffffffu
, /* buffer max size */
891 .maxBoundDescriptorSets
= MAX_SETS
,
892 .maxPerStageDescriptorSamplers
= max_descriptor_set_size
,
893 .maxPerStageDescriptorUniformBuffers
= max_descriptor_set_size
,
894 .maxPerStageDescriptorStorageBuffers
= max_descriptor_set_size
,
895 .maxPerStageDescriptorSampledImages
= max_descriptor_set_size
,
896 .maxPerStageDescriptorStorageImages
= max_descriptor_set_size
,
897 .maxPerStageDescriptorInputAttachments
= max_descriptor_set_size
,
898 .maxPerStageResources
= max_descriptor_set_size
,
899 .maxDescriptorSetSamplers
= max_descriptor_set_size
,
900 .maxDescriptorSetUniformBuffers
= max_descriptor_set_size
,
901 .maxDescriptorSetUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
,
902 .maxDescriptorSetStorageBuffers
= max_descriptor_set_size
,
903 .maxDescriptorSetStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
,
904 .maxDescriptorSetSampledImages
= max_descriptor_set_size
,
905 .maxDescriptorSetStorageImages
= max_descriptor_set_size
,
906 .maxDescriptorSetInputAttachments
= max_descriptor_set_size
,
907 .maxVertexInputAttributes
= 32,
908 .maxVertexInputBindings
= 32,
909 .maxVertexInputAttributeOffset
= 2047,
910 .maxVertexInputBindingStride
= 2048,
911 .maxVertexOutputComponents
= 128,
912 .maxTessellationGenerationLevel
= 64,
913 .maxTessellationPatchSize
= 32,
914 .maxTessellationControlPerVertexInputComponents
= 128,
915 .maxTessellationControlPerVertexOutputComponents
= 128,
916 .maxTessellationControlPerPatchOutputComponents
= 120,
917 .maxTessellationControlTotalOutputComponents
= 4096,
918 .maxTessellationEvaluationInputComponents
= 128,
919 .maxTessellationEvaluationOutputComponents
= 128,
920 .maxGeometryShaderInvocations
= 127,
921 .maxGeometryInputComponents
= 64,
922 .maxGeometryOutputComponents
= 128,
923 .maxGeometryOutputVertices
= 256,
924 .maxGeometryTotalOutputComponents
= 1024,
925 .maxFragmentInputComponents
= 128,
926 .maxFragmentOutputAttachments
= 8,
927 .maxFragmentDualSrcAttachments
= 1,
928 .maxFragmentCombinedOutputResources
= 8,
929 .maxComputeSharedMemorySize
= 32768,
930 .maxComputeWorkGroupCount
= { 65535, 65535, 65535 },
931 .maxComputeWorkGroupInvocations
= 2048,
932 .maxComputeWorkGroupSize
= {
937 .subPixelPrecisionBits
= 4 /* FIXME */,
938 .subTexelPrecisionBits
= 4 /* FIXME */,
939 .mipmapPrecisionBits
= 4 /* FIXME */,
940 .maxDrawIndexedIndexValue
= UINT32_MAX
,
941 .maxDrawIndirectCount
= UINT32_MAX
,
942 .maxSamplerLodBias
= 16,
943 .maxSamplerAnisotropy
= 16,
944 .maxViewports
= MAX_VIEWPORTS
,
945 .maxViewportDimensions
= { (1 << 14), (1 << 14) },
946 .viewportBoundsRange
= { INT16_MIN
, INT16_MAX
},
947 .viewportSubPixelBits
= 8,
948 .minMemoryMapAlignment
= 4096, /* A page */
949 .minTexelBufferOffsetAlignment
= 1,
950 .minUniformBufferOffsetAlignment
= 4,
951 .minStorageBufferOffsetAlignment
= 4,
952 .minTexelOffset
= -32,
953 .maxTexelOffset
= 31,
954 .minTexelGatherOffset
= -32,
955 .maxTexelGatherOffset
= 31,
956 .minInterpolationOffset
= -2,
957 .maxInterpolationOffset
= 2,
958 .subPixelInterpolationOffsetBits
= 8,
959 .maxFramebufferWidth
= (1 << 14),
960 .maxFramebufferHeight
= (1 << 14),
961 .maxFramebufferLayers
= (1 << 10),
962 .framebufferColorSampleCounts
= sample_counts
,
963 .framebufferDepthSampleCounts
= sample_counts
,
964 .framebufferStencilSampleCounts
= sample_counts
,
965 .framebufferNoAttachmentsSampleCounts
= sample_counts
,
966 .maxColorAttachments
= MAX_RTS
,
967 .sampledImageColorSampleCounts
= sample_counts
,
968 .sampledImageIntegerSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
969 .sampledImageDepthSampleCounts
= sample_counts
,
970 .sampledImageStencilSampleCounts
= sample_counts
,
971 .storageImageSampleCounts
= VK_SAMPLE_COUNT_1_BIT
,
972 .maxSampleMaskWords
= 1,
973 .timestampComputeAndGraphics
= true,
974 .timestampPeriod
= 1000000.0 / pdevice
->rad_info
.clock_crystal_freq
,
975 .maxClipDistances
= 8,
976 .maxCullDistances
= 8,
977 .maxCombinedClipAndCullDistances
= 8,
978 .discreteQueuePriorities
= 2,
979 .pointSizeRange
= { 0.125, 255.875 },
980 .lineWidthRange
= { 0.0, 7.9921875 },
981 .pointSizeGranularity
= (1.0 / 8.0),
982 .lineWidthGranularity
= (1.0 / 128.0),
983 .strictLines
= false, /* FINISHME */
984 .standardSampleLocations
= true,
985 .optimalBufferCopyOffsetAlignment
= 128,
986 .optimalBufferCopyRowPitchAlignment
= 128,
987 .nonCoherentAtomSize
= 64,
990 *pProperties
= (VkPhysicalDeviceProperties
) {
991 .apiVersion
= radv_physical_device_api_version(pdevice
),
992 .driverVersion
= vk_get_driver_version(),
993 .vendorID
= ATI_VENDOR_ID
,
994 .deviceID
= pdevice
->rad_info
.pci_id
,
995 .deviceType
= pdevice
->rad_info
.has_dedicated_vram
? VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
: VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
,
997 .sparseProperties
= {0},
1000 strcpy(pProperties
->deviceName
, pdevice
->name
);
1001 memcpy(pProperties
->pipelineCacheUUID
, pdevice
->cache_uuid
, VK_UUID_SIZE
);
1004 void radv_GetPhysicalDeviceProperties2(
1005 VkPhysicalDevice physicalDevice
,
1006 VkPhysicalDeviceProperties2KHR
*pProperties
)
1008 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
1009 radv_GetPhysicalDeviceProperties(physicalDevice
, &pProperties
->properties
);
1011 vk_foreach_struct(ext
, pProperties
->pNext
) {
1012 switch (ext
->sType
) {
1013 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR
: {
1014 VkPhysicalDevicePushDescriptorPropertiesKHR
*properties
=
1015 (VkPhysicalDevicePushDescriptorPropertiesKHR
*) ext
;
1016 properties
->maxPushDescriptors
= MAX_PUSH_DESCRIPTORS
;
1019 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES_KHR
: {
1020 VkPhysicalDeviceIDPropertiesKHR
*properties
= (VkPhysicalDeviceIDPropertiesKHR
*)ext
;
1021 memcpy(properties
->driverUUID
, pdevice
->driver_uuid
, VK_UUID_SIZE
);
1022 memcpy(properties
->deviceUUID
, pdevice
->device_uuid
, VK_UUID_SIZE
);
1023 properties
->deviceLUIDValid
= false;
1026 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES_KHR
: {
1027 VkPhysicalDeviceMultiviewPropertiesKHR
*properties
= (VkPhysicalDeviceMultiviewPropertiesKHR
*)ext
;
1028 properties
->maxMultiviewViewCount
= MAX_VIEWS
;
1029 properties
->maxMultiviewInstanceIndex
= INT_MAX
;
1032 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES_KHR
: {
1033 VkPhysicalDevicePointClippingPropertiesKHR
*properties
=
1034 (VkPhysicalDevicePointClippingPropertiesKHR
*)ext
;
1035 properties
->pointClippingBehavior
= VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES_KHR
;
1038 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DISCARD_RECTANGLE_PROPERTIES_EXT
: {
1039 VkPhysicalDeviceDiscardRectanglePropertiesEXT
*properties
=
1040 (VkPhysicalDeviceDiscardRectanglePropertiesEXT
*)ext
;
1041 properties
->maxDiscardRectangles
= MAX_DISCARD_RECTANGLES
;
1044 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT
: {
1045 VkPhysicalDeviceExternalMemoryHostPropertiesEXT
*properties
=
1046 (VkPhysicalDeviceExternalMemoryHostPropertiesEXT
*) ext
;
1047 properties
->minImportedHostPointerAlignment
= 4096;
1050 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES
: {
1051 VkPhysicalDeviceSubgroupProperties
*properties
=
1052 (VkPhysicalDeviceSubgroupProperties
*)ext
;
1053 properties
->subgroupSize
= 64;
1054 properties
->supportedStages
= VK_SHADER_STAGE_ALL
;
1055 /* TODO: Enable VK_SUBGROUP_FEATURE_VOTE_BIT when wwm
1058 properties
->supportedOperations
=
1059 VK_SUBGROUP_FEATURE_ARITHMETIC_BIT
|
1060 VK_SUBGROUP_FEATURE_BASIC_BIT
|
1061 VK_SUBGROUP_FEATURE_BALLOT_BIT
|
1062 VK_SUBGROUP_FEATURE_QUAD_BIT
;
1063 if (pdevice
->rad_info
.chip_class
>= VI
) {
1064 properties
->supportedOperations
|=
1065 VK_SUBGROUP_FEATURE_SHUFFLE_BIT
|
1066 VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT
;
1068 properties
->quadOperationsInAllStages
= true;
1071 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES
: {
1072 VkPhysicalDeviceMaintenance3Properties
*properties
=
1073 (VkPhysicalDeviceMaintenance3Properties
*)ext
;
1074 /* Make sure everything is addressable by a signed 32-bit int, and
1075 * our largest descriptors are 96 bytes. */
1076 properties
->maxPerSetDescriptors
= (1ull << 31) / 96;
1077 /* Our buffer size fields allow only this much */
1078 properties
->maxMemoryAllocationSize
= 0xFFFFFFFFull
;
1081 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT
: {
1082 VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT
*properties
=
1083 (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT
*)ext
;
1084 /* GFX6-8 only support single channel min/max filter. */
1085 properties
->filterMinmaxImageComponentMapping
= pdevice
->rad_info
.chip_class
>= GFX9
;
1086 properties
->filterMinmaxSingleComponentFormats
= true;
1089 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_CORE_PROPERTIES_AMD
: {
1090 VkPhysicalDeviceShaderCorePropertiesAMD
*properties
=
1091 (VkPhysicalDeviceShaderCorePropertiesAMD
*)ext
;
1093 /* Shader engines. */
1094 properties
->shaderEngineCount
=
1095 pdevice
->rad_info
.max_se
;
1096 properties
->shaderArraysPerEngineCount
=
1097 pdevice
->rad_info
.max_sh_per_se
;
1098 properties
->computeUnitsPerShaderArray
=
1099 pdevice
->rad_info
.num_good_compute_units
/
1100 (pdevice
->rad_info
.max_se
*
1101 pdevice
->rad_info
.max_sh_per_se
);
1102 properties
->simdPerComputeUnit
= 4;
1103 properties
->wavefrontsPerSimd
=
1104 pdevice
->rad_info
.family
== CHIP_TONGA
||
1105 pdevice
->rad_info
.family
== CHIP_ICELAND
||
1106 pdevice
->rad_info
.family
== CHIP_POLARIS10
||
1107 pdevice
->rad_info
.family
== CHIP_POLARIS11
||
1108 pdevice
->rad_info
.family
== CHIP_POLARIS12
||
1109 pdevice
->rad_info
.family
== CHIP_VEGAM
? 8 : 10;
1110 properties
->wavefrontSize
= 64;
1113 properties
->sgprsPerSimd
=
1114 radv_get_num_physical_sgprs(pdevice
);
1115 properties
->minSgprAllocation
=
1116 pdevice
->rad_info
.chip_class
>= VI
? 16 : 8;
1117 properties
->maxSgprAllocation
=
1118 pdevice
->rad_info
.family
== CHIP_TONGA
||
1119 pdevice
->rad_info
.family
== CHIP_ICELAND
? 96 : 104;
1120 properties
->sgprAllocationGranularity
=
1121 pdevice
->rad_info
.chip_class
>= VI
? 16 : 8;
1124 properties
->vgprsPerSimd
= RADV_NUM_PHYSICAL_VGPRS
;
1125 properties
->minVgprAllocation
= 4;
1126 properties
->maxVgprAllocation
= 256;
1127 properties
->vgprAllocationGranularity
= 4;
1130 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT
: {
1131 VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*properties
=
1132 (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT
*)ext
;
1133 properties
->maxVertexAttribDivisor
= UINT32_MAX
;
1136 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT
: {
1137 VkPhysicalDeviceDescriptorIndexingPropertiesEXT
*properties
=
1138 (VkPhysicalDeviceDescriptorIndexingPropertiesEXT
*)ext
;
1139 properties
->maxUpdateAfterBindDescriptorsInAllPools
= UINT32_MAX
/ 64;
1140 properties
->shaderUniformBufferArrayNonUniformIndexingNative
= false;
1141 properties
->shaderSampledImageArrayNonUniformIndexingNative
= false;
1142 properties
->shaderStorageBufferArrayNonUniformIndexingNative
= false;
1143 properties
->shaderStorageImageArrayNonUniformIndexingNative
= false;
1144 properties
->shaderInputAttachmentArrayNonUniformIndexingNative
= false;
1145 properties
->robustBufferAccessUpdateAfterBind
= false;
1146 properties
->quadDivergentImplicitLod
= false;
1148 size_t max_descriptor_set_size
= ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS
) /
1149 (32 /* uniform buffer, 32 due to potential space wasted on alignment */ +
1150 32 /* storage buffer, 32 due to potential space wasted on alignment */ +
1151 32 /* sampler, largest when combined with image */ +
1152 64 /* sampled image */ +
1153 64 /* storage image */);
1154 properties
->maxPerStageDescriptorUpdateAfterBindSamplers
= max_descriptor_set_size
;
1155 properties
->maxPerStageDescriptorUpdateAfterBindUniformBuffers
= max_descriptor_set_size
;
1156 properties
->maxPerStageDescriptorUpdateAfterBindStorageBuffers
= max_descriptor_set_size
;
1157 properties
->maxPerStageDescriptorUpdateAfterBindSampledImages
= max_descriptor_set_size
;
1158 properties
->maxPerStageDescriptorUpdateAfterBindStorageImages
= max_descriptor_set_size
;
1159 properties
->maxPerStageDescriptorUpdateAfterBindInputAttachments
= max_descriptor_set_size
;
1160 properties
->maxPerStageUpdateAfterBindResources
= max_descriptor_set_size
;
1161 properties
->maxDescriptorSetUpdateAfterBindSamplers
= max_descriptor_set_size
;
1162 properties
->maxDescriptorSetUpdateAfterBindUniformBuffers
= max_descriptor_set_size
;
1163 properties
->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic
= MAX_DYNAMIC_UNIFORM_BUFFERS
;
1164 properties
->maxDescriptorSetUpdateAfterBindStorageBuffers
= max_descriptor_set_size
;
1165 properties
->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic
= MAX_DYNAMIC_STORAGE_BUFFERS
;
1166 properties
->maxDescriptorSetUpdateAfterBindSampledImages
= max_descriptor_set_size
;
1167 properties
->maxDescriptorSetUpdateAfterBindStorageImages
= max_descriptor_set_size
;
1168 properties
->maxDescriptorSetUpdateAfterBindInputAttachments
= max_descriptor_set_size
;
1171 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES
: {
1172 VkPhysicalDeviceProtectedMemoryProperties
*properties
=
1173 (VkPhysicalDeviceProtectedMemoryProperties
*)ext
;
1174 properties
->protectedNoFault
= false;
1177 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONSERVATIVE_RASTERIZATION_PROPERTIES_EXT
: {
1178 VkPhysicalDeviceConservativeRasterizationPropertiesEXT
*properties
=
1179 (VkPhysicalDeviceConservativeRasterizationPropertiesEXT
*)ext
;
1180 properties
->primitiveOverestimationSize
= 0;
1181 properties
->maxExtraPrimitiveOverestimationSize
= 0;
1182 properties
->extraPrimitiveOverestimationSizeGranularity
= 0;
1183 properties
->primitiveUnderestimation
= VK_FALSE
;
1184 properties
->conservativePointAndLineRasterization
= VK_FALSE
;
1185 properties
->degenerateTrianglesRasterized
= VK_FALSE
;
1186 properties
->degenerateLinesRasterized
= VK_FALSE
;
1187 properties
->fullyCoveredFragmentShaderInputVariable
= VK_FALSE
;
1188 properties
->conservativeRasterizationPostDepthCoverage
= VK_FALSE
;
1191 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PCI_BUS_INFO_PROPERTIES_EXT
: {
1192 VkPhysicalDevicePCIBusInfoPropertiesEXT
*properties
=
1193 (VkPhysicalDevicePCIBusInfoPropertiesEXT
*)ext
;
1194 properties
->pciDomain
= pdevice
->bus_info
.domain
;
1195 properties
->pciBus
= pdevice
->bus_info
.bus
;
1196 properties
->pciDevice
= pdevice
->bus_info
.dev
;
1197 properties
->pciFunction
= pdevice
->bus_info
.func
;
1200 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR
: {
1201 VkPhysicalDeviceDriverPropertiesKHR
*driver_props
=
1202 (VkPhysicalDeviceDriverPropertiesKHR
*) ext
;
1204 driver_props
->driverID
= VK_DRIVER_ID_MESA_RADV_KHR
;
1205 memset(driver_props
->driverName
, 0, VK_MAX_DRIVER_NAME_SIZE_KHR
);
1206 strcpy(driver_props
->driverName
, "radv");
1208 memset(driver_props
->driverInfo
, 0, VK_MAX_DRIVER_INFO_SIZE_KHR
);
1209 snprintf(driver_props
->driverInfo
, VK_MAX_DRIVER_INFO_SIZE_KHR
,
1210 "Mesa " PACKAGE_VERSION MESA_GIT_SHA1
1212 (HAVE_LLVM
>> 8) & 0xff, HAVE_LLVM
& 0xff,
1213 MESA_LLVM_VERSION_PATCH
);
1215 driver_props
->conformanceVersion
= (VkConformanceVersionKHR
) {
1223 case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT
: {
1224 VkPhysicalDeviceTransformFeedbackPropertiesEXT
*properties
=
1225 (VkPhysicalDeviceTransformFeedbackPropertiesEXT
*)ext
;
1226 properties
->maxTransformFeedbackStreams
= MAX_SO_STREAMS
;
1227 properties
->maxTransformFeedbackBuffers
= MAX_SO_BUFFERS
;
1228 properties
->maxTransformFeedbackBufferSize
= UINT32_MAX
;
1229 properties
->maxTransformFeedbackStreamDataSize
= 512;
1230 properties
->maxTransformFeedbackBufferDataSize
= UINT32_MAX
;
1231 properties
->maxTransformFeedbackBufferDataStride
= 512;
1232 properties
->transformFeedbackQueries
= true;
1233 properties
->transformFeedbackStreamsLinesTriangles
= false;
1234 properties
->transformFeedbackRasterizationStreamSelect
= false;
1235 properties
->transformFeedbackDraw
= true;
1244 static void radv_get_physical_device_queue_family_properties(
1245 struct radv_physical_device
* pdevice
,
1247 VkQueueFamilyProperties
** pQueueFamilyProperties
)
1249 int num_queue_families
= 1;
1251 if (pdevice
->rad_info
.num_compute_rings
> 0 &&
1252 !(pdevice
->instance
->debug_flags
& RADV_DEBUG_NO_COMPUTE_QUEUE
))
1253 num_queue_families
++;
1255 if (pQueueFamilyProperties
== NULL
) {
1256 *pCount
= num_queue_families
;
1265 *pQueueFamilyProperties
[idx
] = (VkQueueFamilyProperties
) {
1266 .queueFlags
= VK_QUEUE_GRAPHICS_BIT
|
1267 VK_QUEUE_COMPUTE_BIT
|
1268 VK_QUEUE_TRANSFER_BIT
|
1269 VK_QUEUE_SPARSE_BINDING_BIT
,
1271 .timestampValidBits
= 64,
1272 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
1277 if (pdevice
->rad_info
.num_compute_rings
> 0 &&
1278 !(pdevice
->instance
->debug_flags
& RADV_DEBUG_NO_COMPUTE_QUEUE
)) {
1279 if (*pCount
> idx
) {
1280 *pQueueFamilyProperties
[idx
] = (VkQueueFamilyProperties
) {
1281 .queueFlags
= VK_QUEUE_COMPUTE_BIT
|
1282 VK_QUEUE_TRANSFER_BIT
|
1283 VK_QUEUE_SPARSE_BINDING_BIT
,
1284 .queueCount
= pdevice
->rad_info
.num_compute_rings
,
1285 .timestampValidBits
= 64,
1286 .minImageTransferGranularity
= (VkExtent3D
) { 1, 1, 1 },
1294 void radv_GetPhysicalDeviceQueueFamilyProperties(
1295 VkPhysicalDevice physicalDevice
,
1297 VkQueueFamilyProperties
* pQueueFamilyProperties
)
1299 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
1300 if (!pQueueFamilyProperties
) {
1301 return radv_get_physical_device_queue_family_properties(pdevice
, pCount
, NULL
);
1304 VkQueueFamilyProperties
*properties
[] = {
1305 pQueueFamilyProperties
+ 0,
1306 pQueueFamilyProperties
+ 1,
1307 pQueueFamilyProperties
+ 2,
1309 radv_get_physical_device_queue_family_properties(pdevice
, pCount
, properties
);
1310 assert(*pCount
<= 3);
1313 void radv_GetPhysicalDeviceQueueFamilyProperties2(
1314 VkPhysicalDevice physicalDevice
,
1316 VkQueueFamilyProperties2KHR
*pQueueFamilyProperties
)
1318 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
1319 if (!pQueueFamilyProperties
) {
1320 return radv_get_physical_device_queue_family_properties(pdevice
, pCount
, NULL
);
1323 VkQueueFamilyProperties
*properties
[] = {
1324 &pQueueFamilyProperties
[0].queueFamilyProperties
,
1325 &pQueueFamilyProperties
[1].queueFamilyProperties
,
1326 &pQueueFamilyProperties
[2].queueFamilyProperties
,
1328 radv_get_physical_device_queue_family_properties(pdevice
, pCount
, properties
);
1329 assert(*pCount
<= 3);
1332 void radv_GetPhysicalDeviceMemoryProperties(
1333 VkPhysicalDevice physicalDevice
,
1334 VkPhysicalDeviceMemoryProperties
*pMemoryProperties
)
1336 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
1338 *pMemoryProperties
= physical_device
->memory_properties
;
1341 void radv_GetPhysicalDeviceMemoryProperties2(
1342 VkPhysicalDevice physicalDevice
,
1343 VkPhysicalDeviceMemoryProperties2KHR
*pMemoryProperties
)
1345 return radv_GetPhysicalDeviceMemoryProperties(physicalDevice
,
1346 &pMemoryProperties
->memoryProperties
);
1349 VkResult
radv_GetMemoryHostPointerPropertiesEXT(
1351 VkExternalMemoryHandleTypeFlagBitsKHR handleType
,
1352 const void *pHostPointer
,
1353 VkMemoryHostPointerPropertiesEXT
*pMemoryHostPointerProperties
)
1355 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1359 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT
: {
1360 const struct radv_physical_device
*physical_device
= device
->physical_device
;
1361 uint32_t memoryTypeBits
= 0;
1362 for (int i
= 0; i
< physical_device
->memory_properties
.memoryTypeCount
; i
++) {
1363 if (physical_device
->mem_type_indices
[i
] == RADV_MEM_TYPE_GTT_CACHED
) {
1364 memoryTypeBits
= (1 << i
);
1368 pMemoryHostPointerProperties
->memoryTypeBits
= memoryTypeBits
;
1372 return VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
;
1376 static enum radeon_ctx_priority
1377 radv_get_queue_global_priority(const VkDeviceQueueGlobalPriorityCreateInfoEXT
*pObj
)
1379 /* Default to MEDIUM when a specific global priority isn't requested */
1381 return RADEON_CTX_PRIORITY_MEDIUM
;
1383 switch(pObj
->globalPriority
) {
1384 case VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT
:
1385 return RADEON_CTX_PRIORITY_REALTIME
;
1386 case VK_QUEUE_GLOBAL_PRIORITY_HIGH_EXT
:
1387 return RADEON_CTX_PRIORITY_HIGH
;
1388 case VK_QUEUE_GLOBAL_PRIORITY_MEDIUM_EXT
:
1389 return RADEON_CTX_PRIORITY_MEDIUM
;
1390 case VK_QUEUE_GLOBAL_PRIORITY_LOW_EXT
:
1391 return RADEON_CTX_PRIORITY_LOW
;
1393 unreachable("Illegal global priority value");
1394 return RADEON_CTX_PRIORITY_INVALID
;
1399 radv_queue_init(struct radv_device
*device
, struct radv_queue
*queue
,
1400 uint32_t queue_family_index
, int idx
,
1401 VkDeviceQueueCreateFlags flags
,
1402 const VkDeviceQueueGlobalPriorityCreateInfoEXT
*global_priority
)
1404 queue
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1405 queue
->device
= device
;
1406 queue
->queue_family_index
= queue_family_index
;
1407 queue
->queue_idx
= idx
;
1408 queue
->priority
= radv_get_queue_global_priority(global_priority
);
1409 queue
->flags
= flags
;
1411 queue
->hw_ctx
= device
->ws
->ctx_create(device
->ws
, queue
->priority
);
1413 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1419 radv_queue_finish(struct radv_queue
*queue
)
1422 queue
->device
->ws
->ctx_destroy(queue
->hw_ctx
);
1424 if (queue
->initial_full_flush_preamble_cs
)
1425 queue
->device
->ws
->cs_destroy(queue
->initial_full_flush_preamble_cs
);
1426 if (queue
->initial_preamble_cs
)
1427 queue
->device
->ws
->cs_destroy(queue
->initial_preamble_cs
);
1428 if (queue
->continue_preamble_cs
)
1429 queue
->device
->ws
->cs_destroy(queue
->continue_preamble_cs
);
1430 if (queue
->descriptor_bo
)
1431 queue
->device
->ws
->buffer_destroy(queue
->descriptor_bo
);
1432 if (queue
->scratch_bo
)
1433 queue
->device
->ws
->buffer_destroy(queue
->scratch_bo
);
1434 if (queue
->esgs_ring_bo
)
1435 queue
->device
->ws
->buffer_destroy(queue
->esgs_ring_bo
);
1436 if (queue
->gsvs_ring_bo
)
1437 queue
->device
->ws
->buffer_destroy(queue
->gsvs_ring_bo
);
1438 if (queue
->tess_rings_bo
)
1439 queue
->device
->ws
->buffer_destroy(queue
->tess_rings_bo
);
1440 if (queue
->compute_scratch_bo
)
1441 queue
->device
->ws
->buffer_destroy(queue
->compute_scratch_bo
);
1445 radv_bo_list_init(struct radv_bo_list
*bo_list
)
1447 pthread_mutex_init(&bo_list
->mutex
, NULL
);
1448 bo_list
->list
.count
= bo_list
->capacity
= 0;
1449 bo_list
->list
.bos
= NULL
;
1453 radv_bo_list_finish(struct radv_bo_list
*bo_list
)
1455 free(bo_list
->list
.bos
);
1456 pthread_mutex_destroy(&bo_list
->mutex
);
1459 static VkResult
radv_bo_list_add(struct radv_device
*device
,
1460 struct radeon_winsys_bo
*bo
)
1462 struct radv_bo_list
*bo_list
= &device
->bo_list
;
1464 if (unlikely(!device
->use_global_bo_list
))
1467 pthread_mutex_lock(&bo_list
->mutex
);
1468 if (bo_list
->list
.count
== bo_list
->capacity
) {
1469 unsigned capacity
= MAX2(4, bo_list
->capacity
* 2);
1470 void *data
= realloc(bo_list
->list
.bos
, capacity
* sizeof(struct radeon_winsys_bo
*));
1473 pthread_mutex_unlock(&bo_list
->mutex
);
1474 return VK_ERROR_OUT_OF_HOST_MEMORY
;
1477 bo_list
->list
.bos
= (struct radeon_winsys_bo
**)data
;
1478 bo_list
->capacity
= capacity
;
1481 bo_list
->list
.bos
[bo_list
->list
.count
++] = bo
;
1482 pthread_mutex_unlock(&bo_list
->mutex
);
1486 static void radv_bo_list_remove(struct radv_device
*device
,
1487 struct radeon_winsys_bo
*bo
)
1489 struct radv_bo_list
*bo_list
= &device
->bo_list
;
1491 if (unlikely(!device
->use_global_bo_list
))
1494 pthread_mutex_lock(&bo_list
->mutex
);
1495 for(unsigned i
= 0; i
< bo_list
->list
.count
; ++i
) {
1496 if (bo_list
->list
.bos
[i
] == bo
) {
1497 bo_list
->list
.bos
[i
] = bo_list
->list
.bos
[bo_list
->list
.count
- 1];
1498 --bo_list
->list
.count
;
1502 pthread_mutex_unlock(&bo_list
->mutex
);
1506 radv_device_init_gs_info(struct radv_device
*device
)
1508 device
->gs_table_depth
= ac_get_gs_table_depth(device
->physical_device
->rad_info
.chip_class
,
1509 device
->physical_device
->rad_info
.family
);
1512 static int radv_get_device_extension_index(const char *name
)
1514 for (unsigned i
= 0; i
< RADV_DEVICE_EXTENSION_COUNT
; ++i
) {
1515 if (strcmp(name
, radv_device_extensions
[i
].extensionName
) == 0)
1522 radv_get_int_debug_option(const char *name
, int default_value
)
1529 result
= default_value
;
1533 result
= strtol(str
, &endptr
, 0);
1534 if (str
== endptr
) {
1535 /* No digits founs. */
1536 result
= default_value
;
1543 VkResult
radv_CreateDevice(
1544 VkPhysicalDevice physicalDevice
,
1545 const VkDeviceCreateInfo
* pCreateInfo
,
1546 const VkAllocationCallbacks
* pAllocator
,
1549 RADV_FROM_HANDLE(radv_physical_device
, physical_device
, physicalDevice
);
1551 struct radv_device
*device
;
1553 bool keep_shader_info
= false;
1555 /* Check enabled features */
1556 if (pCreateInfo
->pEnabledFeatures
) {
1557 VkPhysicalDeviceFeatures supported_features
;
1558 radv_GetPhysicalDeviceFeatures(physicalDevice
, &supported_features
);
1559 VkBool32
*supported_feature
= (VkBool32
*)&supported_features
;
1560 VkBool32
*enabled_feature
= (VkBool32
*)pCreateInfo
->pEnabledFeatures
;
1561 unsigned num_features
= sizeof(VkPhysicalDeviceFeatures
) / sizeof(VkBool32
);
1562 for (uint32_t i
= 0; i
< num_features
; i
++) {
1563 if (enabled_feature
[i
] && !supported_feature
[i
])
1564 return vk_error(physical_device
->instance
, VK_ERROR_FEATURE_NOT_PRESENT
);
1568 device
= vk_zalloc2(&physical_device
->instance
->alloc
, pAllocator
,
1570 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1572 return vk_error(physical_device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
1574 device
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
1575 device
->instance
= physical_device
->instance
;
1576 device
->physical_device
= physical_device
;
1578 device
->ws
= physical_device
->ws
;
1580 device
->alloc
= *pAllocator
;
1582 device
->alloc
= physical_device
->instance
->alloc
;
1584 for (uint32_t i
= 0; i
< pCreateInfo
->enabledExtensionCount
; i
++) {
1585 const char *ext_name
= pCreateInfo
->ppEnabledExtensionNames
[i
];
1586 int index
= radv_get_device_extension_index(ext_name
);
1587 if (index
< 0 || !physical_device
->supported_extensions
.extensions
[index
]) {
1588 vk_free(&device
->alloc
, device
);
1589 return vk_error(physical_device
->instance
, VK_ERROR_EXTENSION_NOT_PRESENT
);
1592 device
->enabled_extensions
.extensions
[index
] = true;
1595 keep_shader_info
= device
->enabled_extensions
.AMD_shader_info
;
1597 /* With update after bind we can't attach bo's to the command buffer
1598 * from the descriptor set anymore, so we have to use a global BO list.
1600 device
->use_global_bo_list
=
1601 device
->enabled_extensions
.EXT_descriptor_indexing
;
1603 mtx_init(&device
->shader_slab_mutex
, mtx_plain
);
1604 list_inithead(&device
->shader_slabs
);
1606 radv_bo_list_init(&device
->bo_list
);
1608 for (unsigned i
= 0; i
< pCreateInfo
->queueCreateInfoCount
; i
++) {
1609 const VkDeviceQueueCreateInfo
*queue_create
= &pCreateInfo
->pQueueCreateInfos
[i
];
1610 uint32_t qfi
= queue_create
->queueFamilyIndex
;
1611 const VkDeviceQueueGlobalPriorityCreateInfoEXT
*global_priority
=
1612 vk_find_struct_const(queue_create
->pNext
, DEVICE_QUEUE_GLOBAL_PRIORITY_CREATE_INFO_EXT
);
1614 assert(!global_priority
|| device
->physical_device
->rad_info
.has_ctx_priority
);
1616 device
->queues
[qfi
] = vk_alloc(&device
->alloc
,
1617 queue_create
->queueCount
* sizeof(struct radv_queue
), 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
1618 if (!device
->queues
[qfi
]) {
1619 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
1623 memset(device
->queues
[qfi
], 0, queue_create
->queueCount
* sizeof(struct radv_queue
));
1625 device
->queue_count
[qfi
] = queue_create
->queueCount
;
1627 for (unsigned q
= 0; q
< queue_create
->queueCount
; q
++) {
1628 result
= radv_queue_init(device
, &device
->queues
[qfi
][q
],
1629 qfi
, q
, queue_create
->flags
,
1631 if (result
!= VK_SUCCESS
)
1636 device
->pbb_allowed
= device
->physical_device
->rad_info
.chip_class
>= GFX9
&&
1637 ((device
->instance
->perftest_flags
& RADV_PERFTEST_BINNING
) ||
1638 device
->physical_device
->rad_info
.family
== CHIP_RAVEN
);
1640 /* Disabled and not implemented for now. */
1641 device
->dfsm_allowed
= device
->pbb_allowed
&&
1642 device
->physical_device
->rad_info
.family
== CHIP_RAVEN
;
1645 device
->always_use_syncobj
= device
->physical_device
->rad_info
.has_syncobj_wait_for_submit
;
1648 /* The maximum number of scratch waves. Scratch space isn't divided
1649 * evenly between CUs. The number is only a function of the number of CUs.
1650 * We can decrease the constant to decrease the scratch buffer size.
1652 * sctx->scratch_waves must be >= the maximum possible size of
1653 * 1 threadgroup, so that the hw doesn't hang from being unable
1656 * The recommended value is 4 per CU at most. Higher numbers don't
1657 * bring much benefit, but they still occupy chip resources (think
1658 * async compute). I've seen ~2% performance difference between 4 and 32.
1660 uint32_t max_threads_per_block
= 2048;
1661 device
->scratch_waves
= MAX2(32 * physical_device
->rad_info
.num_good_compute_units
,
1662 max_threads_per_block
/ 64);
1664 device
->dispatch_initiator
= S_00B800_COMPUTE_SHADER_EN(1);
1666 if (device
->physical_device
->rad_info
.chip_class
>= CIK
) {
1667 /* If the KMD allows it (there is a KMD hw register for it),
1668 * allow launching waves out-of-order.
1670 device
->dispatch_initiator
|= S_00B800_ORDER_MODE(1);
1673 radv_device_init_gs_info(device
);
1675 device
->tess_offchip_block_dw_size
=
1676 device
->physical_device
->rad_info
.family
== CHIP_HAWAII
? 4096 : 8192;
1677 device
->has_distributed_tess
=
1678 device
->physical_device
->rad_info
.chip_class
>= VI
&&
1679 device
->physical_device
->rad_info
.max_se
>= 2;
1681 if (getenv("RADV_TRACE_FILE")) {
1682 const char *filename
= getenv("RADV_TRACE_FILE");
1684 keep_shader_info
= true;
1686 if (!radv_init_trace(device
))
1689 fprintf(stderr
, "*****************************************************************************\n");
1690 fprintf(stderr
, "* WARNING: RADV_TRACE_FILE is costly and should only be used for debugging! *\n");
1691 fprintf(stderr
, "*****************************************************************************\n");
1693 fprintf(stderr
, "Trace file will be dumped to %s\n", filename
);
1694 radv_dump_enabled_options(device
, stderr
);
1697 device
->keep_shader_info
= keep_shader_info
;
1699 result
= radv_device_init_meta(device
);
1700 if (result
!= VK_SUCCESS
)
1703 radv_device_init_msaa(device
);
1705 for (int family
= 0; family
< RADV_MAX_QUEUE_FAMILIES
; ++family
) {
1706 device
->empty_cs
[family
] = device
->ws
->cs_create(device
->ws
, family
);
1708 case RADV_QUEUE_GENERAL
:
1709 radeon_emit(device
->empty_cs
[family
], PKT3(PKT3_CONTEXT_CONTROL
, 1, 0));
1710 radeon_emit(device
->empty_cs
[family
], CONTEXT_CONTROL_LOAD_ENABLE(1));
1711 radeon_emit(device
->empty_cs
[family
], CONTEXT_CONTROL_SHADOW_ENABLE(1));
1713 case RADV_QUEUE_COMPUTE
:
1714 radeon_emit(device
->empty_cs
[family
], PKT3(PKT3_NOP
, 0, 0));
1715 radeon_emit(device
->empty_cs
[family
], 0);
1718 device
->ws
->cs_finalize(device
->empty_cs
[family
]);
1721 if (device
->physical_device
->rad_info
.chip_class
>= CIK
)
1722 cik_create_gfx_config(device
);
1724 VkPipelineCacheCreateInfo ci
;
1725 ci
.sType
= VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO
;
1728 ci
.pInitialData
= NULL
;
1729 ci
.initialDataSize
= 0;
1731 result
= radv_CreatePipelineCache(radv_device_to_handle(device
),
1733 if (result
!= VK_SUCCESS
)
1736 device
->mem_cache
= radv_pipeline_cache_from_handle(pc
);
1738 device
->force_aniso
=
1739 MIN2(16, radv_get_int_debug_option("RADV_TEX_ANISO", -1));
1740 if (device
->force_aniso
>= 0) {
1741 fprintf(stderr
, "radv: Forcing anisotropy filter to %ix\n",
1742 1 << util_logbase2(device
->force_aniso
));
1745 *pDevice
= radv_device_to_handle(device
);
1749 radv_device_finish_meta(device
);
1751 radv_bo_list_finish(&device
->bo_list
);
1753 if (device
->trace_bo
)
1754 device
->ws
->buffer_destroy(device
->trace_bo
);
1756 if (device
->gfx_init
)
1757 device
->ws
->buffer_destroy(device
->gfx_init
);
1759 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
1760 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1761 radv_queue_finish(&device
->queues
[i
][q
]);
1762 if (device
->queue_count
[i
])
1763 vk_free(&device
->alloc
, device
->queues
[i
]);
1766 vk_free(&device
->alloc
, device
);
1770 void radv_DestroyDevice(
1772 const VkAllocationCallbacks
* pAllocator
)
1774 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1779 if (device
->trace_bo
)
1780 device
->ws
->buffer_destroy(device
->trace_bo
);
1782 if (device
->gfx_init
)
1783 device
->ws
->buffer_destroy(device
->gfx_init
);
1785 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
1786 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++)
1787 radv_queue_finish(&device
->queues
[i
][q
]);
1788 if (device
->queue_count
[i
])
1789 vk_free(&device
->alloc
, device
->queues
[i
]);
1790 if (device
->empty_cs
[i
])
1791 device
->ws
->cs_destroy(device
->empty_cs
[i
]);
1793 radv_device_finish_meta(device
);
1795 VkPipelineCache pc
= radv_pipeline_cache_to_handle(device
->mem_cache
);
1796 radv_DestroyPipelineCache(radv_device_to_handle(device
), pc
, NULL
);
1798 radv_destroy_shader_slabs(device
);
1800 radv_bo_list_finish(&device
->bo_list
);
1801 vk_free(&device
->alloc
, device
);
1804 VkResult
radv_EnumerateInstanceLayerProperties(
1805 uint32_t* pPropertyCount
,
1806 VkLayerProperties
* pProperties
)
1808 if (pProperties
== NULL
) {
1809 *pPropertyCount
= 0;
1813 /* None supported at this time */
1814 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1817 VkResult
radv_EnumerateDeviceLayerProperties(
1818 VkPhysicalDevice physicalDevice
,
1819 uint32_t* pPropertyCount
,
1820 VkLayerProperties
* pProperties
)
1822 if (pProperties
== NULL
) {
1823 *pPropertyCount
= 0;
1827 /* None supported at this time */
1828 return vk_error(NULL
, VK_ERROR_LAYER_NOT_PRESENT
);
1831 void radv_GetDeviceQueue2(
1833 const VkDeviceQueueInfo2
* pQueueInfo
,
1836 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1837 struct radv_queue
*queue
;
1839 queue
= &device
->queues
[pQueueInfo
->queueFamilyIndex
][pQueueInfo
->queueIndex
];
1840 if (pQueueInfo
->flags
!= queue
->flags
) {
1841 /* From the Vulkan 1.1.70 spec:
1843 * "The queue returned by vkGetDeviceQueue2 must have the same
1844 * flags value from this structure as that used at device
1845 * creation time in a VkDeviceQueueCreateInfo instance. If no
1846 * matching flags were specified at device creation time then
1847 * pQueue will return VK_NULL_HANDLE."
1849 *pQueue
= VK_NULL_HANDLE
;
1853 *pQueue
= radv_queue_to_handle(queue
);
1856 void radv_GetDeviceQueue(
1858 uint32_t queueFamilyIndex
,
1859 uint32_t queueIndex
,
1862 const VkDeviceQueueInfo2 info
= (VkDeviceQueueInfo2
) {
1863 .sType
= VK_STRUCTURE_TYPE_DEVICE_QUEUE_INFO_2
,
1864 .queueFamilyIndex
= queueFamilyIndex
,
1865 .queueIndex
= queueIndex
1868 radv_GetDeviceQueue2(_device
, &info
, pQueue
);
1872 fill_geom_tess_rings(struct radv_queue
*queue
,
1874 bool add_sample_positions
,
1875 uint32_t esgs_ring_size
,
1876 struct radeon_winsys_bo
*esgs_ring_bo
,
1877 uint32_t gsvs_ring_size
,
1878 struct radeon_winsys_bo
*gsvs_ring_bo
,
1879 uint32_t tess_factor_ring_size
,
1880 uint32_t tess_offchip_ring_offset
,
1881 uint32_t tess_offchip_ring_size
,
1882 struct radeon_winsys_bo
*tess_rings_bo
)
1884 uint64_t esgs_va
= 0, gsvs_va
= 0;
1885 uint64_t tess_va
= 0, tess_offchip_va
= 0;
1886 uint32_t *desc
= &map
[4];
1889 esgs_va
= radv_buffer_get_va(esgs_ring_bo
);
1891 gsvs_va
= radv_buffer_get_va(gsvs_ring_bo
);
1892 if (tess_rings_bo
) {
1893 tess_va
= radv_buffer_get_va(tess_rings_bo
);
1894 tess_offchip_va
= tess_va
+ tess_offchip_ring_offset
;
1897 /* stride 0, num records - size, add tid, swizzle, elsize4,
1900 desc
[1] = S_008F04_BASE_ADDRESS_HI(esgs_va
>> 32) |
1901 S_008F04_STRIDE(0) |
1902 S_008F04_SWIZZLE_ENABLE(true);
1903 desc
[2] = esgs_ring_size
;
1904 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1905 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1906 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1907 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1908 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1909 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1910 S_008F0C_ELEMENT_SIZE(1) |
1911 S_008F0C_INDEX_STRIDE(3) |
1912 S_008F0C_ADD_TID_ENABLE(true);
1915 /* GS entry for ES->GS ring */
1916 /* stride 0, num records - size, elsize0,
1919 desc
[1] = S_008F04_BASE_ADDRESS_HI(esgs_va
>> 32)|
1920 S_008F04_STRIDE(0) |
1921 S_008F04_SWIZZLE_ENABLE(false);
1922 desc
[2] = esgs_ring_size
;
1923 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1924 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1925 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1926 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1927 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1928 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1929 S_008F0C_ELEMENT_SIZE(0) |
1930 S_008F0C_INDEX_STRIDE(0) |
1931 S_008F0C_ADD_TID_ENABLE(false);
1934 /* VS entry for GS->VS ring */
1935 /* stride 0, num records - size, elsize0,
1938 desc
[1] = S_008F04_BASE_ADDRESS_HI(gsvs_va
>> 32)|
1939 S_008F04_STRIDE(0) |
1940 S_008F04_SWIZZLE_ENABLE(false);
1941 desc
[2] = gsvs_ring_size
;
1942 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1943 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1944 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1945 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1946 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1947 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1948 S_008F0C_ELEMENT_SIZE(0) |
1949 S_008F0C_INDEX_STRIDE(0) |
1950 S_008F0C_ADD_TID_ENABLE(false);
1953 /* stride gsvs_itemsize, num records 64
1954 elsize 4, index stride 16 */
1955 /* shader will patch stride and desc[2] */
1957 desc
[1] = S_008F04_BASE_ADDRESS_HI(gsvs_va
>> 32)|
1958 S_008F04_STRIDE(0) |
1959 S_008F04_SWIZZLE_ENABLE(true);
1961 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1962 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1963 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1964 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1965 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1966 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1967 S_008F0C_ELEMENT_SIZE(1) |
1968 S_008F0C_INDEX_STRIDE(1) |
1969 S_008F0C_ADD_TID_ENABLE(true);
1973 desc
[1] = S_008F04_BASE_ADDRESS_HI(tess_va
>> 32) |
1974 S_008F04_STRIDE(0) |
1975 S_008F04_SWIZZLE_ENABLE(false);
1976 desc
[2] = tess_factor_ring_size
;
1977 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1978 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1979 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1980 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1981 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1982 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1983 S_008F0C_ELEMENT_SIZE(0) |
1984 S_008F0C_INDEX_STRIDE(0) |
1985 S_008F0C_ADD_TID_ENABLE(false);
1988 desc
[0] = tess_offchip_va
;
1989 desc
[1] = S_008F04_BASE_ADDRESS_HI(tess_offchip_va
>> 32) |
1990 S_008F04_STRIDE(0) |
1991 S_008F04_SWIZZLE_ENABLE(false);
1992 desc
[2] = tess_offchip_ring_size
;
1993 desc
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1994 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1995 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1996 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1997 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1998 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
) |
1999 S_008F0C_ELEMENT_SIZE(0) |
2000 S_008F0C_INDEX_STRIDE(0) |
2001 S_008F0C_ADD_TID_ENABLE(false);
2004 /* add sample positions after all rings */
2005 memcpy(desc
, queue
->device
->sample_locations_1x
, 8);
2007 memcpy(desc
, queue
->device
->sample_locations_2x
, 16);
2009 memcpy(desc
, queue
->device
->sample_locations_4x
, 32);
2011 memcpy(desc
, queue
->device
->sample_locations_8x
, 64);
2013 memcpy(desc
, queue
->device
->sample_locations_16x
, 128);
2017 radv_get_hs_offchip_param(struct radv_device
*device
, uint32_t *max_offchip_buffers_p
)
2019 bool double_offchip_buffers
= device
->physical_device
->rad_info
.chip_class
>= CIK
&&
2020 device
->physical_device
->rad_info
.family
!= CHIP_CARRIZO
&&
2021 device
->physical_device
->rad_info
.family
!= CHIP_STONEY
;
2022 unsigned max_offchip_buffers_per_se
= double_offchip_buffers
? 128 : 64;
2023 unsigned max_offchip_buffers
;
2024 unsigned offchip_granularity
;
2025 unsigned hs_offchip_param
;
2029 * This must be one less than the maximum number due to a hw limitation.
2030 * Various hardware bugs in SI, CIK, and GFX9 need this.
2033 * Vega10 should limit max_offchip_buffers to 508 (4 * 127).
2034 * Gfx7 should limit max_offchip_buffers to 508
2035 * Gfx6 should limit max_offchip_buffers to 126 (2 * 63)
2037 * Follow AMDVLK here.
2039 if (device
->physical_device
->rad_info
.family
== CHIP_VEGA10
||
2040 device
->physical_device
->rad_info
.chip_class
== CIK
||
2041 device
->physical_device
->rad_info
.chip_class
== SI
)
2042 --max_offchip_buffers_per_se
;
2044 max_offchip_buffers
= max_offchip_buffers_per_se
*
2045 device
->physical_device
->rad_info
.max_se
;
2047 switch (device
->tess_offchip_block_dw_size
) {
2052 offchip_granularity
= V_03093C_X_8K_DWORDS
;
2055 offchip_granularity
= V_03093C_X_4K_DWORDS
;
2059 switch (device
->physical_device
->rad_info
.chip_class
) {
2061 max_offchip_buffers
= MIN2(max_offchip_buffers
, 126);
2067 max_offchip_buffers
= MIN2(max_offchip_buffers
, 508);
2071 *max_offchip_buffers_p
= max_offchip_buffers
;
2072 if (device
->physical_device
->rad_info
.chip_class
>= CIK
) {
2073 if (device
->physical_device
->rad_info
.chip_class
>= VI
)
2074 --max_offchip_buffers
;
2076 S_03093C_OFFCHIP_BUFFERING(max_offchip_buffers
) |
2077 S_03093C_OFFCHIP_GRANULARITY(offchip_granularity
);
2080 S_0089B0_OFFCHIP_BUFFERING(max_offchip_buffers
);
2082 return hs_offchip_param
;
2086 radv_emit_gs_ring_sizes(struct radv_queue
*queue
, struct radeon_cmdbuf
*cs
,
2087 struct radeon_winsys_bo
*esgs_ring_bo
,
2088 uint32_t esgs_ring_size
,
2089 struct radeon_winsys_bo
*gsvs_ring_bo
,
2090 uint32_t gsvs_ring_size
)
2092 if (!esgs_ring_bo
&& !gsvs_ring_bo
)
2096 radv_cs_add_buffer(queue
->device
->ws
, cs
, esgs_ring_bo
);
2099 radv_cs_add_buffer(queue
->device
->ws
, cs
, gsvs_ring_bo
);
2101 if (queue
->device
->physical_device
->rad_info
.chip_class
>= CIK
) {
2102 radeon_set_uconfig_reg_seq(cs
, R_030900_VGT_ESGS_RING_SIZE
, 2);
2103 radeon_emit(cs
, esgs_ring_size
>> 8);
2104 radeon_emit(cs
, gsvs_ring_size
>> 8);
2106 radeon_set_config_reg_seq(cs
, R_0088C8_VGT_ESGS_RING_SIZE
, 2);
2107 radeon_emit(cs
, esgs_ring_size
>> 8);
2108 radeon_emit(cs
, gsvs_ring_size
>> 8);
2113 radv_emit_tess_factor_ring(struct radv_queue
*queue
, struct radeon_cmdbuf
*cs
,
2114 unsigned hs_offchip_param
, unsigned tf_ring_size
,
2115 struct radeon_winsys_bo
*tess_rings_bo
)
2122 tf_va
= radv_buffer_get_va(tess_rings_bo
);
2124 radv_cs_add_buffer(queue
->device
->ws
, cs
, tess_rings_bo
);
2126 if (queue
->device
->physical_device
->rad_info
.chip_class
>= CIK
) {
2127 radeon_set_uconfig_reg(cs
, R_030938_VGT_TF_RING_SIZE
,
2128 S_030938_SIZE(tf_ring_size
/ 4));
2129 radeon_set_uconfig_reg(cs
, R_030940_VGT_TF_MEMORY_BASE
,
2131 if (queue
->device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
2132 radeon_set_uconfig_reg(cs
, R_030944_VGT_TF_MEMORY_BASE_HI
,
2133 S_030944_BASE_HI(tf_va
>> 40));
2135 radeon_set_uconfig_reg(cs
, R_03093C_VGT_HS_OFFCHIP_PARAM
,
2138 radeon_set_config_reg(cs
, R_008988_VGT_TF_RING_SIZE
,
2139 S_008988_SIZE(tf_ring_size
/ 4));
2140 radeon_set_config_reg(cs
, R_0089B8_VGT_TF_MEMORY_BASE
,
2142 radeon_set_config_reg(cs
, R_0089B0_VGT_HS_OFFCHIP_PARAM
,
2148 radv_emit_compute_scratch(struct radv_queue
*queue
, struct radeon_cmdbuf
*cs
,
2149 struct radeon_winsys_bo
*compute_scratch_bo
)
2151 uint64_t scratch_va
;
2153 if (!compute_scratch_bo
)
2156 scratch_va
= radv_buffer_get_va(compute_scratch_bo
);
2158 radv_cs_add_buffer(queue
->device
->ws
, cs
, compute_scratch_bo
);
2160 radeon_set_sh_reg_seq(cs
, R_00B900_COMPUTE_USER_DATA_0
, 2);
2161 radeon_emit(cs
, scratch_va
);
2162 radeon_emit(cs
, S_008F04_BASE_ADDRESS_HI(scratch_va
>> 32) |
2163 S_008F04_SWIZZLE_ENABLE(1));
2167 radv_emit_global_shader_pointers(struct radv_queue
*queue
,
2168 struct radeon_cmdbuf
*cs
,
2169 struct radeon_winsys_bo
*descriptor_bo
)
2176 va
= radv_buffer_get_va(descriptor_bo
);
2178 radv_cs_add_buffer(queue
->device
->ws
, cs
, descriptor_bo
);
2180 if (queue
->device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
2181 uint32_t regs
[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0
,
2182 R_00B130_SPI_SHADER_USER_DATA_VS_0
,
2183 R_00B208_SPI_SHADER_USER_DATA_ADDR_LO_GS
,
2184 R_00B408_SPI_SHADER_USER_DATA_ADDR_LO_HS
};
2186 for (int i
= 0; i
< ARRAY_SIZE(regs
); ++i
) {
2187 radv_emit_shader_pointer(queue
->device
, cs
, regs
[i
],
2191 uint32_t regs
[] = {R_00B030_SPI_SHADER_USER_DATA_PS_0
,
2192 R_00B130_SPI_SHADER_USER_DATA_VS_0
,
2193 R_00B230_SPI_SHADER_USER_DATA_GS_0
,
2194 R_00B330_SPI_SHADER_USER_DATA_ES_0
,
2195 R_00B430_SPI_SHADER_USER_DATA_HS_0
,
2196 R_00B530_SPI_SHADER_USER_DATA_LS_0
};
2198 for (int i
= 0; i
< ARRAY_SIZE(regs
); ++i
) {
2199 radv_emit_shader_pointer(queue
->device
, cs
, regs
[i
],
2206 radv_init_graphics_state(struct radeon_cmdbuf
*cs
, struct radv_queue
*queue
)
2208 struct radv_device
*device
= queue
->device
;
2210 if (device
->gfx_init
) {
2211 uint64_t va
= radv_buffer_get_va(device
->gfx_init
);
2213 radeon_emit(cs
, PKT3(PKT3_INDIRECT_BUFFER_CIK
, 2, 0));
2214 radeon_emit(cs
, va
);
2215 radeon_emit(cs
, va
>> 32);
2216 radeon_emit(cs
, device
->gfx_init_size_dw
& 0xffff);
2218 radv_cs_add_buffer(device
->ws
, cs
, device
->gfx_init
);
2220 struct radv_physical_device
*physical_device
= device
->physical_device
;
2221 si_emit_graphics(physical_device
, cs
);
2226 radv_init_compute_state(struct radeon_cmdbuf
*cs
, struct radv_queue
*queue
)
2228 struct radv_physical_device
*physical_device
= queue
->device
->physical_device
;
2229 si_emit_compute(physical_device
, cs
);
2233 radv_get_preamble_cs(struct radv_queue
*queue
,
2234 uint32_t scratch_size
,
2235 uint32_t compute_scratch_size
,
2236 uint32_t esgs_ring_size
,
2237 uint32_t gsvs_ring_size
,
2238 bool needs_tess_rings
,
2239 bool needs_sample_positions
,
2240 struct radeon_cmdbuf
**initial_full_flush_preamble_cs
,
2241 struct radeon_cmdbuf
**initial_preamble_cs
,
2242 struct radeon_cmdbuf
**continue_preamble_cs
)
2244 struct radeon_winsys_bo
*scratch_bo
= NULL
;
2245 struct radeon_winsys_bo
*descriptor_bo
= NULL
;
2246 struct radeon_winsys_bo
*compute_scratch_bo
= NULL
;
2247 struct radeon_winsys_bo
*esgs_ring_bo
= NULL
;
2248 struct radeon_winsys_bo
*gsvs_ring_bo
= NULL
;
2249 struct radeon_winsys_bo
*tess_rings_bo
= NULL
;
2250 struct radeon_cmdbuf
*dest_cs
[3] = {0};
2251 bool add_tess_rings
= false, add_sample_positions
= false;
2252 unsigned tess_factor_ring_size
= 0, tess_offchip_ring_size
= 0;
2253 unsigned max_offchip_buffers
;
2254 unsigned hs_offchip_param
= 0;
2255 unsigned tess_offchip_ring_offset
;
2256 uint32_t ring_bo_flags
= RADEON_FLAG_NO_CPU_ACCESS
| RADEON_FLAG_NO_INTERPROCESS_SHARING
;
2257 if (!queue
->has_tess_rings
) {
2258 if (needs_tess_rings
)
2259 add_tess_rings
= true;
2261 if (!queue
->has_sample_positions
) {
2262 if (needs_sample_positions
)
2263 add_sample_positions
= true;
2265 tess_factor_ring_size
= 32768 * queue
->device
->physical_device
->rad_info
.max_se
;
2266 hs_offchip_param
= radv_get_hs_offchip_param(queue
->device
,
2267 &max_offchip_buffers
);
2268 tess_offchip_ring_offset
= align(tess_factor_ring_size
, 64 * 1024);
2269 tess_offchip_ring_size
= max_offchip_buffers
*
2270 queue
->device
->tess_offchip_block_dw_size
* 4;
2272 if (scratch_size
<= queue
->scratch_size
&&
2273 compute_scratch_size
<= queue
->compute_scratch_size
&&
2274 esgs_ring_size
<= queue
->esgs_ring_size
&&
2275 gsvs_ring_size
<= queue
->gsvs_ring_size
&&
2276 !add_tess_rings
&& !add_sample_positions
&&
2277 queue
->initial_preamble_cs
) {
2278 *initial_full_flush_preamble_cs
= queue
->initial_full_flush_preamble_cs
;
2279 *initial_preamble_cs
= queue
->initial_preamble_cs
;
2280 *continue_preamble_cs
= queue
->continue_preamble_cs
;
2281 if (!scratch_size
&& !compute_scratch_size
&& !esgs_ring_size
&& !gsvs_ring_size
)
2282 *continue_preamble_cs
= NULL
;
2286 if (scratch_size
> queue
->scratch_size
) {
2287 scratch_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2295 scratch_bo
= queue
->scratch_bo
;
2297 if (compute_scratch_size
> queue
->compute_scratch_size
) {
2298 compute_scratch_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2299 compute_scratch_size
,
2303 if (!compute_scratch_bo
)
2307 compute_scratch_bo
= queue
->compute_scratch_bo
;
2309 if (esgs_ring_size
> queue
->esgs_ring_size
) {
2310 esgs_ring_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2318 esgs_ring_bo
= queue
->esgs_ring_bo
;
2319 esgs_ring_size
= queue
->esgs_ring_size
;
2322 if (gsvs_ring_size
> queue
->gsvs_ring_size
) {
2323 gsvs_ring_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2331 gsvs_ring_bo
= queue
->gsvs_ring_bo
;
2332 gsvs_ring_size
= queue
->gsvs_ring_size
;
2335 if (add_tess_rings
) {
2336 tess_rings_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2337 tess_offchip_ring_offset
+ tess_offchip_ring_size
,
2344 tess_rings_bo
= queue
->tess_rings_bo
;
2347 if (scratch_bo
!= queue
->scratch_bo
||
2348 esgs_ring_bo
!= queue
->esgs_ring_bo
||
2349 gsvs_ring_bo
!= queue
->gsvs_ring_bo
||
2350 tess_rings_bo
!= queue
->tess_rings_bo
||
2351 add_sample_positions
) {
2353 if (gsvs_ring_bo
|| esgs_ring_bo
||
2354 tess_rings_bo
|| add_sample_positions
) {
2355 size
= 112; /* 2 dword + 2 padding + 4 dword * 6 */
2356 if (add_sample_positions
)
2357 size
+= 256; /* 32+16+8+4+2+1 samples * 4 * 2 = 248 bytes. */
2359 else if (scratch_bo
)
2360 size
= 8; /* 2 dword */
2362 descriptor_bo
= queue
->device
->ws
->buffer_create(queue
->device
->ws
,
2366 RADEON_FLAG_CPU_ACCESS
|
2367 RADEON_FLAG_NO_INTERPROCESS_SHARING
|
2368 RADEON_FLAG_READ_ONLY
);
2372 descriptor_bo
= queue
->descriptor_bo
;
2374 for(int i
= 0; i
< 3; ++i
) {
2375 struct radeon_cmdbuf
*cs
= NULL
;
2376 cs
= queue
->device
->ws
->cs_create(queue
->device
->ws
,
2377 queue
->queue_family_index
? RING_COMPUTE
: RING_GFX
);
2384 radv_cs_add_buffer(queue
->device
->ws
, cs
, scratch_bo
);
2386 /* Emit initial configuration. */
2387 switch (queue
->queue_family_index
) {
2388 case RADV_QUEUE_GENERAL
:
2389 radv_init_graphics_state(cs
, queue
);
2391 case RADV_QUEUE_COMPUTE
:
2392 radv_init_compute_state(cs
, queue
);
2394 case RADV_QUEUE_TRANSFER
:
2398 if (descriptor_bo
!= queue
->descriptor_bo
) {
2399 uint32_t *map
= (uint32_t*)queue
->device
->ws
->buffer_map(descriptor_bo
);
2402 uint64_t scratch_va
= radv_buffer_get_va(scratch_bo
);
2403 uint32_t rsrc1
= S_008F04_BASE_ADDRESS_HI(scratch_va
>> 32) |
2404 S_008F04_SWIZZLE_ENABLE(1);
2405 map
[0] = scratch_va
;
2409 if (esgs_ring_bo
|| gsvs_ring_bo
|| tess_rings_bo
||
2410 add_sample_positions
)
2411 fill_geom_tess_rings(queue
, map
, add_sample_positions
,
2412 esgs_ring_size
, esgs_ring_bo
,
2413 gsvs_ring_size
, gsvs_ring_bo
,
2414 tess_factor_ring_size
,
2415 tess_offchip_ring_offset
,
2416 tess_offchip_ring_size
,
2419 queue
->device
->ws
->buffer_unmap(descriptor_bo
);
2422 if (esgs_ring_bo
|| gsvs_ring_bo
|| tess_rings_bo
) {
2423 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
2424 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
2425 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
2426 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_FLUSH
) | EVENT_INDEX(0));
2429 radv_emit_gs_ring_sizes(queue
, cs
, esgs_ring_bo
, esgs_ring_size
,
2430 gsvs_ring_bo
, gsvs_ring_size
);
2431 radv_emit_tess_factor_ring(queue
, cs
, hs_offchip_param
,
2432 tess_factor_ring_size
, tess_rings_bo
);
2433 radv_emit_global_shader_pointers(queue
, cs
, descriptor_bo
);
2434 radv_emit_compute_scratch(queue
, cs
, compute_scratch_bo
);
2437 si_cs_emit_cache_flush(cs
,
2438 queue
->device
->physical_device
->rad_info
.chip_class
,
2440 queue
->queue_family_index
== RING_COMPUTE
&&
2441 queue
->device
->physical_device
->rad_info
.chip_class
>= CIK
,
2442 (queue
->queue_family_index
== RADV_QUEUE_COMPUTE
? RADV_CMD_FLAG_CS_PARTIAL_FLUSH
: (RADV_CMD_FLAG_CS_PARTIAL_FLUSH
| RADV_CMD_FLAG_PS_PARTIAL_FLUSH
)) |
2443 RADV_CMD_FLAG_INV_ICACHE
|
2444 RADV_CMD_FLAG_INV_SMEM_L1
|
2445 RADV_CMD_FLAG_INV_VMEM_L1
|
2446 RADV_CMD_FLAG_INV_GLOBAL_L2
|
2447 RADV_CMD_FLAG_START_PIPELINE_STATS
, 0);
2448 } else if (i
== 1) {
2449 si_cs_emit_cache_flush(cs
,
2450 queue
->device
->physical_device
->rad_info
.chip_class
,
2452 queue
->queue_family_index
== RING_COMPUTE
&&
2453 queue
->device
->physical_device
->rad_info
.chip_class
>= CIK
,
2454 RADV_CMD_FLAG_INV_ICACHE
|
2455 RADV_CMD_FLAG_INV_SMEM_L1
|
2456 RADV_CMD_FLAG_INV_VMEM_L1
|
2457 RADV_CMD_FLAG_INV_GLOBAL_L2
|
2458 RADV_CMD_FLAG_START_PIPELINE_STATS
, 0);
2461 if (!queue
->device
->ws
->cs_finalize(cs
))
2465 if (queue
->initial_full_flush_preamble_cs
)
2466 queue
->device
->ws
->cs_destroy(queue
->initial_full_flush_preamble_cs
);
2468 if (queue
->initial_preamble_cs
)
2469 queue
->device
->ws
->cs_destroy(queue
->initial_preamble_cs
);
2471 if (queue
->continue_preamble_cs
)
2472 queue
->device
->ws
->cs_destroy(queue
->continue_preamble_cs
);
2474 queue
->initial_full_flush_preamble_cs
= dest_cs
[0];
2475 queue
->initial_preamble_cs
= dest_cs
[1];
2476 queue
->continue_preamble_cs
= dest_cs
[2];
2478 if (scratch_bo
!= queue
->scratch_bo
) {
2479 if (queue
->scratch_bo
)
2480 queue
->device
->ws
->buffer_destroy(queue
->scratch_bo
);
2481 queue
->scratch_bo
= scratch_bo
;
2482 queue
->scratch_size
= scratch_size
;
2485 if (compute_scratch_bo
!= queue
->compute_scratch_bo
) {
2486 if (queue
->compute_scratch_bo
)
2487 queue
->device
->ws
->buffer_destroy(queue
->compute_scratch_bo
);
2488 queue
->compute_scratch_bo
= compute_scratch_bo
;
2489 queue
->compute_scratch_size
= compute_scratch_size
;
2492 if (esgs_ring_bo
!= queue
->esgs_ring_bo
) {
2493 if (queue
->esgs_ring_bo
)
2494 queue
->device
->ws
->buffer_destroy(queue
->esgs_ring_bo
);
2495 queue
->esgs_ring_bo
= esgs_ring_bo
;
2496 queue
->esgs_ring_size
= esgs_ring_size
;
2499 if (gsvs_ring_bo
!= queue
->gsvs_ring_bo
) {
2500 if (queue
->gsvs_ring_bo
)
2501 queue
->device
->ws
->buffer_destroy(queue
->gsvs_ring_bo
);
2502 queue
->gsvs_ring_bo
= gsvs_ring_bo
;
2503 queue
->gsvs_ring_size
= gsvs_ring_size
;
2506 if (tess_rings_bo
!= queue
->tess_rings_bo
) {
2507 queue
->tess_rings_bo
= tess_rings_bo
;
2508 queue
->has_tess_rings
= true;
2511 if (descriptor_bo
!= queue
->descriptor_bo
) {
2512 if (queue
->descriptor_bo
)
2513 queue
->device
->ws
->buffer_destroy(queue
->descriptor_bo
);
2515 queue
->descriptor_bo
= descriptor_bo
;
2518 if (add_sample_positions
)
2519 queue
->has_sample_positions
= true;
2521 *initial_full_flush_preamble_cs
= queue
->initial_full_flush_preamble_cs
;
2522 *initial_preamble_cs
= queue
->initial_preamble_cs
;
2523 *continue_preamble_cs
= queue
->continue_preamble_cs
;
2524 if (!scratch_size
&& !compute_scratch_size
&& !esgs_ring_size
&& !gsvs_ring_size
)
2525 *continue_preamble_cs
= NULL
;
2528 for (int i
= 0; i
< ARRAY_SIZE(dest_cs
); ++i
)
2530 queue
->device
->ws
->cs_destroy(dest_cs
[i
]);
2531 if (descriptor_bo
&& descriptor_bo
!= queue
->descriptor_bo
)
2532 queue
->device
->ws
->buffer_destroy(descriptor_bo
);
2533 if (scratch_bo
&& scratch_bo
!= queue
->scratch_bo
)
2534 queue
->device
->ws
->buffer_destroy(scratch_bo
);
2535 if (compute_scratch_bo
&& compute_scratch_bo
!= queue
->compute_scratch_bo
)
2536 queue
->device
->ws
->buffer_destroy(compute_scratch_bo
);
2537 if (esgs_ring_bo
&& esgs_ring_bo
!= queue
->esgs_ring_bo
)
2538 queue
->device
->ws
->buffer_destroy(esgs_ring_bo
);
2539 if (gsvs_ring_bo
&& gsvs_ring_bo
!= queue
->gsvs_ring_bo
)
2540 queue
->device
->ws
->buffer_destroy(gsvs_ring_bo
);
2541 if (tess_rings_bo
&& tess_rings_bo
!= queue
->tess_rings_bo
)
2542 queue
->device
->ws
->buffer_destroy(tess_rings_bo
);
2543 return vk_error(queue
->device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
2546 static VkResult
radv_alloc_sem_counts(struct radv_instance
*instance
,
2547 struct radv_winsys_sem_counts
*counts
,
2549 const VkSemaphore
*sems
,
2553 int syncobj_idx
= 0, sem_idx
= 0;
2555 if (num_sems
== 0 && _fence
== VK_NULL_HANDLE
)
2558 for (uint32_t i
= 0; i
< num_sems
; i
++) {
2559 RADV_FROM_HANDLE(radv_semaphore
, sem
, sems
[i
]);
2561 if (sem
->temp_syncobj
|| sem
->syncobj
)
2562 counts
->syncobj_count
++;
2564 counts
->sem_count
++;
2567 if (_fence
!= VK_NULL_HANDLE
) {
2568 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
2569 if (fence
->temp_syncobj
|| fence
->syncobj
)
2570 counts
->syncobj_count
++;
2573 if (counts
->syncobj_count
) {
2574 counts
->syncobj
= (uint32_t *)malloc(sizeof(uint32_t) * counts
->syncobj_count
);
2575 if (!counts
->syncobj
)
2576 return vk_error(instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2579 if (counts
->sem_count
) {
2580 counts
->sem
= (struct radeon_winsys_sem
**)malloc(sizeof(struct radeon_winsys_sem
*) * counts
->sem_count
);
2582 free(counts
->syncobj
);
2583 return vk_error(instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2587 for (uint32_t i
= 0; i
< num_sems
; i
++) {
2588 RADV_FROM_HANDLE(radv_semaphore
, sem
, sems
[i
]);
2590 if (sem
->temp_syncobj
) {
2591 counts
->syncobj
[syncobj_idx
++] = sem
->temp_syncobj
;
2593 else if (sem
->syncobj
)
2594 counts
->syncobj
[syncobj_idx
++] = sem
->syncobj
;
2597 counts
->sem
[sem_idx
++] = sem
->sem
;
2601 if (_fence
!= VK_NULL_HANDLE
) {
2602 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
2603 if (fence
->temp_syncobj
)
2604 counts
->syncobj
[syncobj_idx
++] = fence
->temp_syncobj
;
2605 else if (fence
->syncobj
)
2606 counts
->syncobj
[syncobj_idx
++] = fence
->syncobj
;
2613 radv_free_sem_info(struct radv_winsys_sem_info
*sem_info
)
2615 free(sem_info
->wait
.syncobj
);
2616 free(sem_info
->wait
.sem
);
2617 free(sem_info
->signal
.syncobj
);
2618 free(sem_info
->signal
.sem
);
2622 static void radv_free_temp_syncobjs(struct radv_device
*device
,
2624 const VkSemaphore
*sems
)
2626 for (uint32_t i
= 0; i
< num_sems
; i
++) {
2627 RADV_FROM_HANDLE(radv_semaphore
, sem
, sems
[i
]);
2629 if (sem
->temp_syncobj
) {
2630 device
->ws
->destroy_syncobj(device
->ws
, sem
->temp_syncobj
);
2631 sem
->temp_syncobj
= 0;
2637 radv_alloc_sem_info(struct radv_instance
*instance
,
2638 struct radv_winsys_sem_info
*sem_info
,
2640 const VkSemaphore
*wait_sems
,
2641 int num_signal_sems
,
2642 const VkSemaphore
*signal_sems
,
2646 memset(sem_info
, 0, sizeof(*sem_info
));
2648 ret
= radv_alloc_sem_counts(instance
, &sem_info
->wait
, num_wait_sems
, wait_sems
, VK_NULL_HANDLE
, true);
2651 ret
= radv_alloc_sem_counts(instance
, &sem_info
->signal
, num_signal_sems
, signal_sems
, fence
, false);
2653 radv_free_sem_info(sem_info
);
2655 /* caller can override these */
2656 sem_info
->cs_emit_wait
= true;
2657 sem_info
->cs_emit_signal
= true;
2661 /* Signals fence as soon as all the work currently put on queue is done. */
2662 static VkResult
radv_signal_fence(struct radv_queue
*queue
,
2663 struct radv_fence
*fence
)
2667 struct radv_winsys_sem_info sem_info
;
2669 result
= radv_alloc_sem_info(queue
->device
->instance
, &sem_info
, 0, NULL
, 0, NULL
,
2670 radv_fence_to_handle(fence
));
2671 if (result
!= VK_SUCCESS
)
2674 ret
= queue
->device
->ws
->cs_submit(queue
->hw_ctx
, queue
->queue_idx
,
2675 &queue
->device
->empty_cs
[queue
->queue_family_index
],
2676 1, NULL
, NULL
, &sem_info
, NULL
,
2677 false, fence
->fence
);
2678 radv_free_sem_info(&sem_info
);
2681 return vk_error(queue
->device
->instance
, VK_ERROR_DEVICE_LOST
);
2686 VkResult
radv_QueueSubmit(
2688 uint32_t submitCount
,
2689 const VkSubmitInfo
* pSubmits
,
2692 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
2693 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
2694 struct radeon_winsys_fence
*base_fence
= fence
? fence
->fence
: NULL
;
2695 struct radeon_winsys_ctx
*ctx
= queue
->hw_ctx
;
2697 uint32_t max_cs_submission
= queue
->device
->trace_bo
? 1 : UINT32_MAX
;
2698 uint32_t scratch_size
= 0;
2699 uint32_t compute_scratch_size
= 0;
2700 uint32_t esgs_ring_size
= 0, gsvs_ring_size
= 0;
2701 struct radeon_cmdbuf
*initial_preamble_cs
= NULL
, *initial_flush_preamble_cs
= NULL
, *continue_preamble_cs
= NULL
;
2703 bool fence_emitted
= false;
2704 bool tess_rings_needed
= false;
2705 bool sample_positions_needed
= false;
2707 /* Do this first so failing to allocate scratch buffers can't result in
2708 * partially executed submissions. */
2709 for (uint32_t i
= 0; i
< submitCount
; i
++) {
2710 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
2711 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
,
2712 pSubmits
[i
].pCommandBuffers
[j
]);
2714 scratch_size
= MAX2(scratch_size
, cmd_buffer
->scratch_size_needed
);
2715 compute_scratch_size
= MAX2(compute_scratch_size
,
2716 cmd_buffer
->compute_scratch_size_needed
);
2717 esgs_ring_size
= MAX2(esgs_ring_size
, cmd_buffer
->esgs_ring_size_needed
);
2718 gsvs_ring_size
= MAX2(gsvs_ring_size
, cmd_buffer
->gsvs_ring_size_needed
);
2719 tess_rings_needed
|= cmd_buffer
->tess_rings_needed
;
2720 sample_positions_needed
|= cmd_buffer
->sample_positions_needed
;
2724 result
= radv_get_preamble_cs(queue
, scratch_size
, compute_scratch_size
,
2725 esgs_ring_size
, gsvs_ring_size
, tess_rings_needed
,
2726 sample_positions_needed
, &initial_flush_preamble_cs
,
2727 &initial_preamble_cs
, &continue_preamble_cs
);
2728 if (result
!= VK_SUCCESS
)
2731 for (uint32_t i
= 0; i
< submitCount
; i
++) {
2732 struct radeon_cmdbuf
**cs_array
;
2733 bool do_flush
= !i
|| pSubmits
[i
].pWaitDstStageMask
;
2734 bool can_patch
= true;
2736 struct radv_winsys_sem_info sem_info
;
2738 result
= radv_alloc_sem_info(queue
->device
->instance
,
2740 pSubmits
[i
].waitSemaphoreCount
,
2741 pSubmits
[i
].pWaitSemaphores
,
2742 pSubmits
[i
].signalSemaphoreCount
,
2743 pSubmits
[i
].pSignalSemaphores
,
2745 if (result
!= VK_SUCCESS
)
2748 if (!pSubmits
[i
].commandBufferCount
) {
2749 if (pSubmits
[i
].waitSemaphoreCount
|| pSubmits
[i
].signalSemaphoreCount
) {
2750 ret
= queue
->device
->ws
->cs_submit(ctx
, queue
->queue_idx
,
2751 &queue
->device
->empty_cs
[queue
->queue_family_index
],
2756 radv_loge("failed to submit CS %d\n", i
);
2759 fence_emitted
= true;
2761 radv_free_sem_info(&sem_info
);
2765 cs_array
= malloc(sizeof(struct radeon_cmdbuf
*) *
2766 (pSubmits
[i
].commandBufferCount
));
2768 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
++) {
2769 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
,
2770 pSubmits
[i
].pCommandBuffers
[j
]);
2771 assert(cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
);
2773 cs_array
[j
] = cmd_buffer
->cs
;
2774 if ((cmd_buffer
->usage_flags
& VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT
))
2777 cmd_buffer
->status
= RADV_CMD_BUFFER_STATUS_PENDING
;
2780 for (uint32_t j
= 0; j
< pSubmits
[i
].commandBufferCount
; j
+= advance
) {
2781 struct radeon_cmdbuf
*initial_preamble
= (do_flush
&& !j
) ? initial_flush_preamble_cs
: initial_preamble_cs
;
2782 const struct radv_winsys_bo_list
*bo_list
= NULL
;
2784 advance
= MIN2(max_cs_submission
,
2785 pSubmits
[i
].commandBufferCount
- j
);
2787 if (queue
->device
->trace_bo
)
2788 *queue
->device
->trace_id_ptr
= 0;
2790 sem_info
.cs_emit_wait
= j
== 0;
2791 sem_info
.cs_emit_signal
= j
+ advance
== pSubmits
[i
].commandBufferCount
;
2793 if (unlikely(queue
->device
->use_global_bo_list
)) {
2794 pthread_mutex_lock(&queue
->device
->bo_list
.mutex
);
2795 bo_list
= &queue
->device
->bo_list
.list
;
2798 ret
= queue
->device
->ws
->cs_submit(ctx
, queue
->queue_idx
, cs_array
+ j
,
2799 advance
, initial_preamble
, continue_preamble_cs
,
2801 can_patch
, base_fence
);
2803 if (unlikely(queue
->device
->use_global_bo_list
))
2804 pthread_mutex_unlock(&queue
->device
->bo_list
.mutex
);
2807 radv_loge("failed to submit CS %d\n", i
);
2810 fence_emitted
= true;
2811 if (queue
->device
->trace_bo
) {
2812 radv_check_gpu_hangs(queue
, cs_array
[j
]);
2816 radv_free_temp_syncobjs(queue
->device
,
2817 pSubmits
[i
].waitSemaphoreCount
,
2818 pSubmits
[i
].pWaitSemaphores
);
2819 radv_free_sem_info(&sem_info
);
2824 if (!fence_emitted
) {
2825 result
= radv_signal_fence(queue
, fence
);
2826 if (result
!= VK_SUCCESS
)
2829 fence
->submitted
= true;
2835 VkResult
radv_QueueWaitIdle(
2838 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
2840 queue
->device
->ws
->ctx_wait_idle(queue
->hw_ctx
,
2841 radv_queue_family_to_ring(queue
->queue_family_index
),
2846 VkResult
radv_DeviceWaitIdle(
2849 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2851 for (unsigned i
= 0; i
< RADV_MAX_QUEUE_FAMILIES
; i
++) {
2852 for (unsigned q
= 0; q
< device
->queue_count
[i
]; q
++) {
2853 radv_QueueWaitIdle(radv_queue_to_handle(&device
->queues
[i
][q
]));
2859 VkResult
radv_EnumerateInstanceExtensionProperties(
2860 const char* pLayerName
,
2861 uint32_t* pPropertyCount
,
2862 VkExtensionProperties
* pProperties
)
2864 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
2866 for (int i
= 0; i
< RADV_INSTANCE_EXTENSION_COUNT
; i
++) {
2867 if (radv_supported_instance_extensions
.extensions
[i
]) {
2868 vk_outarray_append(&out
, prop
) {
2869 *prop
= radv_instance_extensions
[i
];
2874 return vk_outarray_status(&out
);
2877 VkResult
radv_EnumerateDeviceExtensionProperties(
2878 VkPhysicalDevice physicalDevice
,
2879 const char* pLayerName
,
2880 uint32_t* pPropertyCount
,
2881 VkExtensionProperties
* pProperties
)
2883 RADV_FROM_HANDLE(radv_physical_device
, device
, physicalDevice
);
2884 VK_OUTARRAY_MAKE(out
, pProperties
, pPropertyCount
);
2886 for (int i
= 0; i
< RADV_DEVICE_EXTENSION_COUNT
; i
++) {
2887 if (device
->supported_extensions
.extensions
[i
]) {
2888 vk_outarray_append(&out
, prop
) {
2889 *prop
= radv_device_extensions
[i
];
2894 return vk_outarray_status(&out
);
2897 PFN_vkVoidFunction
radv_GetInstanceProcAddr(
2898 VkInstance _instance
,
2901 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
2903 return radv_lookup_entrypoint_checked(pName
,
2904 instance
? instance
->apiVersion
: 0,
2905 instance
? &instance
->enabled_extensions
: NULL
,
2909 /* The loader wants us to expose a second GetInstanceProcAddr function
2910 * to work around certain LD_PRELOAD issues seen in apps.
2913 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
2914 VkInstance instance
,
2918 VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vk_icdGetInstanceProcAddr(
2919 VkInstance instance
,
2922 return radv_GetInstanceProcAddr(instance
, pName
);
2925 PFN_vkVoidFunction
radv_GetDeviceProcAddr(
2929 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2931 return radv_lookup_entrypoint_checked(pName
,
2932 device
->instance
->apiVersion
,
2933 &device
->instance
->enabled_extensions
,
2934 &device
->enabled_extensions
);
2937 bool radv_get_memory_fd(struct radv_device
*device
,
2938 struct radv_device_memory
*memory
,
2941 struct radeon_bo_metadata metadata
;
2943 if (memory
->image
) {
2944 radv_init_metadata(device
, memory
->image
, &metadata
);
2945 device
->ws
->buffer_set_metadata(memory
->bo
, &metadata
);
2948 return device
->ws
->buffer_get_fd(device
->ws
, memory
->bo
,
2952 static VkResult
radv_alloc_memory(struct radv_device
*device
,
2953 const VkMemoryAllocateInfo
* pAllocateInfo
,
2954 const VkAllocationCallbacks
* pAllocator
,
2955 VkDeviceMemory
* pMem
)
2957 struct radv_device_memory
*mem
;
2959 enum radeon_bo_domain domain
;
2961 enum radv_mem_type mem_type_index
= device
->physical_device
->mem_type_indices
[pAllocateInfo
->memoryTypeIndex
];
2963 assert(pAllocateInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO
);
2965 if (pAllocateInfo
->allocationSize
== 0) {
2966 /* Apparently, this is allowed */
2967 *pMem
= VK_NULL_HANDLE
;
2971 const VkImportMemoryFdInfoKHR
*import_info
=
2972 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_FD_INFO_KHR
);
2973 const VkMemoryDedicatedAllocateInfoKHR
*dedicate_info
=
2974 vk_find_struct_const(pAllocateInfo
->pNext
, MEMORY_DEDICATED_ALLOCATE_INFO_KHR
);
2975 const VkExportMemoryAllocateInfoKHR
*export_info
=
2976 vk_find_struct_const(pAllocateInfo
->pNext
, EXPORT_MEMORY_ALLOCATE_INFO_KHR
);
2977 const VkImportMemoryHostPointerInfoEXT
*host_ptr_info
=
2978 vk_find_struct_const(pAllocateInfo
->pNext
, IMPORT_MEMORY_HOST_POINTER_INFO_EXT
);
2980 const struct wsi_memory_allocate_info
*wsi_info
=
2981 vk_find_struct_const(pAllocateInfo
->pNext
, WSI_MEMORY_ALLOCATE_INFO_MESA
);
2983 mem
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*mem
), 8,
2984 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2986 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
2988 if (wsi_info
&& wsi_info
->implicit_sync
)
2989 flags
|= RADEON_FLAG_IMPLICIT_SYNC
;
2991 if (dedicate_info
) {
2992 mem
->image
= radv_image_from_handle(dedicate_info
->image
);
2993 mem
->buffer
= radv_buffer_from_handle(dedicate_info
->buffer
);
2999 mem
->user_ptr
= NULL
;
3002 assert(import_info
->handleType
==
3003 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
||
3004 import_info
->handleType
==
3005 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
3006 mem
->bo
= device
->ws
->buffer_from_fd(device
->ws
, import_info
->fd
,
3009 result
= VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
;
3012 close(import_info
->fd
);
3014 } else if (host_ptr_info
) {
3015 assert(host_ptr_info
->handleType
== VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT
);
3016 assert(mem_type_index
== RADV_MEM_TYPE_GTT_CACHED
);
3017 mem
->bo
= device
->ws
->buffer_from_ptr(device
->ws
, host_ptr_info
->pHostPointer
,
3018 pAllocateInfo
->allocationSize
);
3020 result
= VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
;
3023 mem
->user_ptr
= host_ptr_info
->pHostPointer
;
3026 uint64_t alloc_size
= align_u64(pAllocateInfo
->allocationSize
, 4096);
3027 if (mem_type_index
== RADV_MEM_TYPE_GTT_WRITE_COMBINE
||
3028 mem_type_index
== RADV_MEM_TYPE_GTT_CACHED
)
3029 domain
= RADEON_DOMAIN_GTT
;
3031 domain
= RADEON_DOMAIN_VRAM
;
3033 if (mem_type_index
== RADV_MEM_TYPE_VRAM
)
3034 flags
|= RADEON_FLAG_NO_CPU_ACCESS
;
3036 flags
|= RADEON_FLAG_CPU_ACCESS
;
3038 if (mem_type_index
== RADV_MEM_TYPE_GTT_WRITE_COMBINE
)
3039 flags
|= RADEON_FLAG_GTT_WC
;
3041 if (!dedicate_info
&& !import_info
&& (!export_info
|| !export_info
->handleTypes
))
3042 flags
|= RADEON_FLAG_NO_INTERPROCESS_SHARING
;
3044 mem
->bo
= device
->ws
->buffer_create(device
->ws
, alloc_size
, device
->physical_device
->rad_info
.max_alignment
,
3048 result
= VK_ERROR_OUT_OF_DEVICE_MEMORY
;
3051 mem
->type_index
= mem_type_index
;
3054 result
= radv_bo_list_add(device
, mem
->bo
);
3055 if (result
!= VK_SUCCESS
)
3058 *pMem
= radv_device_memory_to_handle(mem
);
3063 device
->ws
->buffer_destroy(mem
->bo
);
3065 vk_free2(&device
->alloc
, pAllocator
, mem
);
3070 VkResult
radv_AllocateMemory(
3072 const VkMemoryAllocateInfo
* pAllocateInfo
,
3073 const VkAllocationCallbacks
* pAllocator
,
3074 VkDeviceMemory
* pMem
)
3076 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3077 return radv_alloc_memory(device
, pAllocateInfo
, pAllocator
, pMem
);
3080 void radv_FreeMemory(
3082 VkDeviceMemory _mem
,
3083 const VkAllocationCallbacks
* pAllocator
)
3085 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3086 RADV_FROM_HANDLE(radv_device_memory
, mem
, _mem
);
3091 radv_bo_list_remove(device
, mem
->bo
);
3092 device
->ws
->buffer_destroy(mem
->bo
);
3095 vk_free2(&device
->alloc
, pAllocator
, mem
);
3098 VkResult
radv_MapMemory(
3100 VkDeviceMemory _memory
,
3101 VkDeviceSize offset
,
3103 VkMemoryMapFlags flags
,
3106 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3107 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
3115 *ppData
= mem
->user_ptr
;
3117 *ppData
= device
->ws
->buffer_map(mem
->bo
);
3124 return vk_error(device
->instance
, VK_ERROR_MEMORY_MAP_FAILED
);
3127 void radv_UnmapMemory(
3129 VkDeviceMemory _memory
)
3131 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3132 RADV_FROM_HANDLE(radv_device_memory
, mem
, _memory
);
3137 if (mem
->user_ptr
== NULL
)
3138 device
->ws
->buffer_unmap(mem
->bo
);
3141 VkResult
radv_FlushMappedMemoryRanges(
3143 uint32_t memoryRangeCount
,
3144 const VkMappedMemoryRange
* pMemoryRanges
)
3149 VkResult
radv_InvalidateMappedMemoryRanges(
3151 uint32_t memoryRangeCount
,
3152 const VkMappedMemoryRange
* pMemoryRanges
)
3157 void radv_GetBufferMemoryRequirements(
3160 VkMemoryRequirements
* pMemoryRequirements
)
3162 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3163 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
3165 pMemoryRequirements
->memoryTypeBits
= (1u << device
->physical_device
->memory_properties
.memoryTypeCount
) - 1;
3167 if (buffer
->flags
& VK_BUFFER_CREATE_SPARSE_BINDING_BIT
)
3168 pMemoryRequirements
->alignment
= 4096;
3170 pMemoryRequirements
->alignment
= 16;
3172 pMemoryRequirements
->size
= align64(buffer
->size
, pMemoryRequirements
->alignment
);
3175 void radv_GetBufferMemoryRequirements2(
3177 const VkBufferMemoryRequirementsInfo2KHR
* pInfo
,
3178 VkMemoryRequirements2KHR
* pMemoryRequirements
)
3180 radv_GetBufferMemoryRequirements(device
, pInfo
->buffer
,
3181 &pMemoryRequirements
->memoryRequirements
);
3182 RADV_FROM_HANDLE(radv_buffer
, buffer
, pInfo
->buffer
);
3183 vk_foreach_struct(ext
, pMemoryRequirements
->pNext
) {
3184 switch (ext
->sType
) {
3185 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR
: {
3186 VkMemoryDedicatedRequirementsKHR
*req
=
3187 (VkMemoryDedicatedRequirementsKHR
*) ext
;
3188 req
->requiresDedicatedAllocation
= buffer
->shareable
;
3189 req
->prefersDedicatedAllocation
= req
->requiresDedicatedAllocation
;
3198 void radv_GetImageMemoryRequirements(
3201 VkMemoryRequirements
* pMemoryRequirements
)
3203 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3204 RADV_FROM_HANDLE(radv_image
, image
, _image
);
3206 pMemoryRequirements
->memoryTypeBits
= (1u << device
->physical_device
->memory_properties
.memoryTypeCount
) - 1;
3208 pMemoryRequirements
->size
= image
->size
;
3209 pMemoryRequirements
->alignment
= image
->alignment
;
3212 void radv_GetImageMemoryRequirements2(
3214 const VkImageMemoryRequirementsInfo2KHR
* pInfo
,
3215 VkMemoryRequirements2KHR
* pMemoryRequirements
)
3217 radv_GetImageMemoryRequirements(device
, pInfo
->image
,
3218 &pMemoryRequirements
->memoryRequirements
);
3220 RADV_FROM_HANDLE(radv_image
, image
, pInfo
->image
);
3222 vk_foreach_struct(ext
, pMemoryRequirements
->pNext
) {
3223 switch (ext
->sType
) {
3224 case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS_KHR
: {
3225 VkMemoryDedicatedRequirementsKHR
*req
=
3226 (VkMemoryDedicatedRequirementsKHR
*) ext
;
3227 req
->requiresDedicatedAllocation
= image
->shareable
;
3228 req
->prefersDedicatedAllocation
= req
->requiresDedicatedAllocation
;
3237 void radv_GetImageSparseMemoryRequirements(
3240 uint32_t* pSparseMemoryRequirementCount
,
3241 VkSparseImageMemoryRequirements
* pSparseMemoryRequirements
)
3246 void radv_GetImageSparseMemoryRequirements2(
3248 const VkImageSparseMemoryRequirementsInfo2KHR
* pInfo
,
3249 uint32_t* pSparseMemoryRequirementCount
,
3250 VkSparseImageMemoryRequirements2KHR
* pSparseMemoryRequirements
)
3255 void radv_GetDeviceMemoryCommitment(
3257 VkDeviceMemory memory
,
3258 VkDeviceSize
* pCommittedMemoryInBytes
)
3260 *pCommittedMemoryInBytes
= 0;
3263 VkResult
radv_BindBufferMemory2(VkDevice device
,
3264 uint32_t bindInfoCount
,
3265 const VkBindBufferMemoryInfoKHR
*pBindInfos
)
3267 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
3268 RADV_FROM_HANDLE(radv_device_memory
, mem
, pBindInfos
[i
].memory
);
3269 RADV_FROM_HANDLE(radv_buffer
, buffer
, pBindInfos
[i
].buffer
);
3272 buffer
->bo
= mem
->bo
;
3273 buffer
->offset
= pBindInfos
[i
].memoryOffset
;
3281 VkResult
radv_BindBufferMemory(
3284 VkDeviceMemory memory
,
3285 VkDeviceSize memoryOffset
)
3287 const VkBindBufferMemoryInfoKHR info
= {
3288 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO_KHR
,
3291 .memoryOffset
= memoryOffset
3294 return radv_BindBufferMemory2(device
, 1, &info
);
3297 VkResult
radv_BindImageMemory2(VkDevice device
,
3298 uint32_t bindInfoCount
,
3299 const VkBindImageMemoryInfoKHR
*pBindInfos
)
3301 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
3302 RADV_FROM_HANDLE(radv_device_memory
, mem
, pBindInfos
[i
].memory
);
3303 RADV_FROM_HANDLE(radv_image
, image
, pBindInfos
[i
].image
);
3306 image
->bo
= mem
->bo
;
3307 image
->offset
= pBindInfos
[i
].memoryOffset
;
3317 VkResult
radv_BindImageMemory(
3320 VkDeviceMemory memory
,
3321 VkDeviceSize memoryOffset
)
3323 const VkBindImageMemoryInfoKHR info
= {
3324 .sType
= VK_STRUCTURE_TYPE_BIND_BUFFER_MEMORY_INFO_KHR
,
3327 .memoryOffset
= memoryOffset
3330 return radv_BindImageMemory2(device
, 1, &info
);
3335 radv_sparse_buffer_bind_memory(struct radv_device
*device
,
3336 const VkSparseBufferMemoryBindInfo
*bind
)
3338 RADV_FROM_HANDLE(radv_buffer
, buffer
, bind
->buffer
);
3340 for (uint32_t i
= 0; i
< bind
->bindCount
; ++i
) {
3341 struct radv_device_memory
*mem
= NULL
;
3343 if (bind
->pBinds
[i
].memory
!= VK_NULL_HANDLE
)
3344 mem
= radv_device_memory_from_handle(bind
->pBinds
[i
].memory
);
3346 device
->ws
->buffer_virtual_bind(buffer
->bo
,
3347 bind
->pBinds
[i
].resourceOffset
,
3348 bind
->pBinds
[i
].size
,
3349 mem
? mem
->bo
: NULL
,
3350 bind
->pBinds
[i
].memoryOffset
);
3355 radv_sparse_image_opaque_bind_memory(struct radv_device
*device
,
3356 const VkSparseImageOpaqueMemoryBindInfo
*bind
)
3358 RADV_FROM_HANDLE(radv_image
, image
, bind
->image
);
3360 for (uint32_t i
= 0; i
< bind
->bindCount
; ++i
) {
3361 struct radv_device_memory
*mem
= NULL
;
3363 if (bind
->pBinds
[i
].memory
!= VK_NULL_HANDLE
)
3364 mem
= radv_device_memory_from_handle(bind
->pBinds
[i
].memory
);
3366 device
->ws
->buffer_virtual_bind(image
->bo
,
3367 bind
->pBinds
[i
].resourceOffset
,
3368 bind
->pBinds
[i
].size
,
3369 mem
? mem
->bo
: NULL
,
3370 bind
->pBinds
[i
].memoryOffset
);
3374 VkResult
radv_QueueBindSparse(
3376 uint32_t bindInfoCount
,
3377 const VkBindSparseInfo
* pBindInfo
,
3380 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
3381 RADV_FROM_HANDLE(radv_queue
, queue
, _queue
);
3382 struct radeon_winsys_fence
*base_fence
= fence
? fence
->fence
: NULL
;
3383 bool fence_emitted
= false;
3387 for (uint32_t i
= 0; i
< bindInfoCount
; ++i
) {
3388 struct radv_winsys_sem_info sem_info
;
3389 for (uint32_t j
= 0; j
< pBindInfo
[i
].bufferBindCount
; ++j
) {
3390 radv_sparse_buffer_bind_memory(queue
->device
,
3391 pBindInfo
[i
].pBufferBinds
+ j
);
3394 for (uint32_t j
= 0; j
< pBindInfo
[i
].imageOpaqueBindCount
; ++j
) {
3395 radv_sparse_image_opaque_bind_memory(queue
->device
,
3396 pBindInfo
[i
].pImageOpaqueBinds
+ j
);
3400 result
= radv_alloc_sem_info(queue
->device
->instance
,
3402 pBindInfo
[i
].waitSemaphoreCount
,
3403 pBindInfo
[i
].pWaitSemaphores
,
3404 pBindInfo
[i
].signalSemaphoreCount
,
3405 pBindInfo
[i
].pSignalSemaphores
,
3407 if (result
!= VK_SUCCESS
)
3410 if (pBindInfo
[i
].waitSemaphoreCount
|| pBindInfo
[i
].signalSemaphoreCount
) {
3411 ret
= queue
->device
->ws
->cs_submit(queue
->hw_ctx
, queue
->queue_idx
,
3412 &queue
->device
->empty_cs
[queue
->queue_family_index
],
3417 radv_loge("failed to submit CS %d\n", i
);
3421 fence_emitted
= true;
3423 fence
->submitted
= true;
3426 radv_free_sem_info(&sem_info
);
3431 if (!fence_emitted
) {
3432 result
= radv_signal_fence(queue
, fence
);
3433 if (result
!= VK_SUCCESS
)
3436 fence
->submitted
= true;
3442 VkResult
radv_CreateFence(
3444 const VkFenceCreateInfo
* pCreateInfo
,
3445 const VkAllocationCallbacks
* pAllocator
,
3448 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3449 const VkExportFenceCreateInfoKHR
*export
=
3450 vk_find_struct_const(pCreateInfo
->pNext
, EXPORT_FENCE_CREATE_INFO_KHR
);
3451 VkExternalFenceHandleTypeFlagsKHR handleTypes
=
3452 export
? export
->handleTypes
: 0;
3454 struct radv_fence
*fence
= vk_alloc2(&device
->alloc
, pAllocator
,
3456 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3459 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3461 fence
->fence_wsi
= NULL
;
3462 fence
->submitted
= false;
3463 fence
->signalled
= !!(pCreateInfo
->flags
& VK_FENCE_CREATE_SIGNALED_BIT
);
3464 fence
->temp_syncobj
= 0;
3465 if (device
->always_use_syncobj
|| handleTypes
) {
3466 int ret
= device
->ws
->create_syncobj(device
->ws
, &fence
->syncobj
);
3468 vk_free2(&device
->alloc
, pAllocator
, fence
);
3469 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3471 if (pCreateInfo
->flags
& VK_FENCE_CREATE_SIGNALED_BIT
) {
3472 device
->ws
->signal_syncobj(device
->ws
, fence
->syncobj
);
3474 fence
->fence
= NULL
;
3476 fence
->fence
= device
->ws
->create_fence();
3477 if (!fence
->fence
) {
3478 vk_free2(&device
->alloc
, pAllocator
, fence
);
3479 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3484 *pFence
= radv_fence_to_handle(fence
);
3489 void radv_DestroyFence(
3492 const VkAllocationCallbacks
* pAllocator
)
3494 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3495 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
3500 if (fence
->temp_syncobj
)
3501 device
->ws
->destroy_syncobj(device
->ws
, fence
->temp_syncobj
);
3503 device
->ws
->destroy_syncobj(device
->ws
, fence
->syncobj
);
3505 device
->ws
->destroy_fence(fence
->fence
);
3506 if (fence
->fence_wsi
)
3507 fence
->fence_wsi
->destroy(fence
->fence_wsi
);
3508 vk_free2(&device
->alloc
, pAllocator
, fence
);
3512 static uint64_t radv_get_current_time()
3515 clock_gettime(CLOCK_MONOTONIC
, &tv
);
3516 return tv
.tv_nsec
+ tv
.tv_sec
*1000000000ull;
3519 static uint64_t radv_get_absolute_timeout(uint64_t timeout
)
3521 uint64_t current_time
= radv_get_current_time();
3523 timeout
= MIN2(UINT64_MAX
- current_time
, timeout
);
3525 return current_time
+ timeout
;
3529 static bool radv_all_fences_plain_and_submitted(uint32_t fenceCount
, const VkFence
*pFences
)
3531 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3532 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3533 if (fence
->fence
== NULL
|| fence
->syncobj
||
3534 fence
->temp_syncobj
||
3535 (!fence
->signalled
&& !fence
->submitted
))
3541 static bool radv_all_fences_syncobj(uint32_t fenceCount
, const VkFence
*pFences
)
3543 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3544 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3545 if (fence
->syncobj
== 0 && fence
->temp_syncobj
== 0)
3551 VkResult
radv_WaitForFences(
3553 uint32_t fenceCount
,
3554 const VkFence
* pFences
,
3558 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3559 timeout
= radv_get_absolute_timeout(timeout
);
3561 if (device
->always_use_syncobj
&&
3562 radv_all_fences_syncobj(fenceCount
, pFences
))
3564 uint32_t *handles
= malloc(sizeof(uint32_t) * fenceCount
);
3566 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3568 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3569 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3570 handles
[i
] = fence
->temp_syncobj
? fence
->temp_syncobj
: fence
->syncobj
;
3573 bool success
= device
->ws
->wait_syncobj(device
->ws
, handles
, fenceCount
, waitAll
, timeout
);
3576 return success
? VK_SUCCESS
: VK_TIMEOUT
;
3579 if (!waitAll
&& fenceCount
> 1) {
3580 /* Not doing this by default for waitAll, due to needing to allocate twice. */
3581 if (device
->physical_device
->rad_info
.drm_minor
>= 10 && radv_all_fences_plain_and_submitted(fenceCount
, pFences
)) {
3582 uint32_t wait_count
= 0;
3583 struct radeon_winsys_fence
**fences
= malloc(sizeof(struct radeon_winsys_fence
*) * fenceCount
);
3585 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3587 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3588 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3590 if (fence
->signalled
) {
3595 fences
[wait_count
++] = fence
->fence
;
3598 bool success
= device
->ws
->fences_wait(device
->ws
, fences
, wait_count
,
3599 waitAll
, timeout
- radv_get_current_time());
3602 return success
? VK_SUCCESS
: VK_TIMEOUT
;
3605 while(radv_get_current_time() <= timeout
) {
3606 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3607 if (radv_GetFenceStatus(_device
, pFences
[i
]) == VK_SUCCESS
)
3614 for (uint32_t i
= 0; i
< fenceCount
; ++i
) {
3615 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3616 bool expired
= false;
3618 if (fence
->temp_syncobj
) {
3619 if (!device
->ws
->wait_syncobj(device
->ws
, &fence
->temp_syncobj
, 1, true, timeout
))
3624 if (fence
->syncobj
) {
3625 if (!device
->ws
->wait_syncobj(device
->ws
, &fence
->syncobj
, 1, true, timeout
))
3630 if (fence
->signalled
)
3634 if (!fence
->submitted
) {
3635 while(radv_get_current_time() <= timeout
&&
3639 if (!fence
->submitted
)
3642 /* Recheck as it may have been set by
3643 * submitting operations. */
3645 if (fence
->signalled
)
3649 expired
= device
->ws
->fence_wait(device
->ws
,
3656 if (fence
->fence_wsi
) {
3657 VkResult result
= fence
->fence_wsi
->wait(fence
->fence_wsi
, timeout
);
3658 if (result
!= VK_SUCCESS
)
3662 fence
->signalled
= true;
3668 VkResult
radv_ResetFences(VkDevice _device
,
3669 uint32_t fenceCount
,
3670 const VkFence
*pFences
)
3672 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3674 for (unsigned i
= 0; i
< fenceCount
; ++i
) {
3675 RADV_FROM_HANDLE(radv_fence
, fence
, pFences
[i
]);
3676 fence
->submitted
= fence
->signalled
= false;
3678 /* Per spec, we first restore the permanent payload, and then reset, so
3679 * having a temp syncobj should not skip resetting the permanent syncobj. */
3680 if (fence
->temp_syncobj
) {
3681 device
->ws
->destroy_syncobj(device
->ws
, fence
->temp_syncobj
);
3682 fence
->temp_syncobj
= 0;
3685 if (fence
->syncobj
) {
3686 device
->ws
->reset_syncobj(device
->ws
, fence
->syncobj
);
3693 VkResult
radv_GetFenceStatus(VkDevice _device
, VkFence _fence
)
3695 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3696 RADV_FROM_HANDLE(radv_fence
, fence
, _fence
);
3698 if (fence
->temp_syncobj
) {
3699 bool success
= device
->ws
->wait_syncobj(device
->ws
, &fence
->temp_syncobj
, 1, true, 0);
3700 return success
? VK_SUCCESS
: VK_NOT_READY
;
3703 if (fence
->syncobj
) {
3704 bool success
= device
->ws
->wait_syncobj(device
->ws
, &fence
->syncobj
, 1, true, 0);
3705 return success
? VK_SUCCESS
: VK_NOT_READY
;
3708 if (fence
->signalled
)
3710 if (!fence
->submitted
)
3711 return VK_NOT_READY
;
3713 if (!device
->ws
->fence_wait(device
->ws
, fence
->fence
, false, 0))
3714 return VK_NOT_READY
;
3716 if (fence
->fence_wsi
) {
3717 VkResult result
= fence
->fence_wsi
->wait(fence
->fence_wsi
, 0);
3719 if (result
!= VK_SUCCESS
) {
3720 if (result
== VK_TIMEOUT
)
3721 return VK_NOT_READY
;
3729 // Queue semaphore functions
3731 VkResult
radv_CreateSemaphore(
3733 const VkSemaphoreCreateInfo
* pCreateInfo
,
3734 const VkAllocationCallbacks
* pAllocator
,
3735 VkSemaphore
* pSemaphore
)
3737 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3738 const VkExportSemaphoreCreateInfoKHR
*export
=
3739 vk_find_struct_const(pCreateInfo
->pNext
, EXPORT_SEMAPHORE_CREATE_INFO_KHR
);
3740 VkExternalSemaphoreHandleTypeFlagsKHR handleTypes
=
3741 export
? export
->handleTypes
: 0;
3743 struct radv_semaphore
*sem
= vk_alloc2(&device
->alloc
, pAllocator
,
3745 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3747 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3749 sem
->temp_syncobj
= 0;
3750 /* create a syncobject if we are going to export this semaphore */
3751 if (device
->always_use_syncobj
|| handleTypes
) {
3752 assert (device
->physical_device
->rad_info
.has_syncobj
);
3753 int ret
= device
->ws
->create_syncobj(device
->ws
, &sem
->syncobj
);
3755 vk_free2(&device
->alloc
, pAllocator
, sem
);
3756 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3760 sem
->sem
= device
->ws
->create_sem(device
->ws
);
3762 vk_free2(&device
->alloc
, pAllocator
, sem
);
3763 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3768 *pSemaphore
= radv_semaphore_to_handle(sem
);
3772 void radv_DestroySemaphore(
3774 VkSemaphore _semaphore
,
3775 const VkAllocationCallbacks
* pAllocator
)
3777 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3778 RADV_FROM_HANDLE(radv_semaphore
, sem
, _semaphore
);
3783 device
->ws
->destroy_syncobj(device
->ws
, sem
->syncobj
);
3785 device
->ws
->destroy_sem(sem
->sem
);
3786 vk_free2(&device
->alloc
, pAllocator
, sem
);
3789 VkResult
radv_CreateEvent(
3791 const VkEventCreateInfo
* pCreateInfo
,
3792 const VkAllocationCallbacks
* pAllocator
,
3795 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3796 struct radv_event
*event
= vk_alloc2(&device
->alloc
, pAllocator
,
3798 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3801 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3803 event
->bo
= device
->ws
->buffer_create(device
->ws
, 8, 8,
3805 RADEON_FLAG_VA_UNCACHED
| RADEON_FLAG_CPU_ACCESS
| RADEON_FLAG_NO_INTERPROCESS_SHARING
);
3807 vk_free2(&device
->alloc
, pAllocator
, event
);
3808 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
3811 event
->map
= (uint64_t*)device
->ws
->buffer_map(event
->bo
);
3813 *pEvent
= radv_event_to_handle(event
);
3818 void radv_DestroyEvent(
3821 const VkAllocationCallbacks
* pAllocator
)
3823 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3824 RADV_FROM_HANDLE(radv_event
, event
, _event
);
3828 device
->ws
->buffer_destroy(event
->bo
);
3829 vk_free2(&device
->alloc
, pAllocator
, event
);
3832 VkResult
radv_GetEventStatus(
3836 RADV_FROM_HANDLE(radv_event
, event
, _event
);
3838 if (*event
->map
== 1)
3839 return VK_EVENT_SET
;
3840 return VK_EVENT_RESET
;
3843 VkResult
radv_SetEvent(
3847 RADV_FROM_HANDLE(radv_event
, event
, _event
);
3853 VkResult
radv_ResetEvent(
3857 RADV_FROM_HANDLE(radv_event
, event
, _event
);
3863 VkResult
radv_CreateBuffer(
3865 const VkBufferCreateInfo
* pCreateInfo
,
3866 const VkAllocationCallbacks
* pAllocator
,
3869 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3870 struct radv_buffer
*buffer
;
3872 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO
);
3874 buffer
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*buffer
), 8,
3875 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
3877 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
3879 buffer
->size
= pCreateInfo
->size
;
3880 buffer
->usage
= pCreateInfo
->usage
;
3883 buffer
->flags
= pCreateInfo
->flags
;
3885 buffer
->shareable
= vk_find_struct_const(pCreateInfo
->pNext
,
3886 EXTERNAL_MEMORY_BUFFER_CREATE_INFO_KHR
) != NULL
;
3888 if (pCreateInfo
->flags
& VK_BUFFER_CREATE_SPARSE_BINDING_BIT
) {
3889 buffer
->bo
= device
->ws
->buffer_create(device
->ws
,
3890 align64(buffer
->size
, 4096),
3891 4096, 0, RADEON_FLAG_VIRTUAL
);
3893 vk_free2(&device
->alloc
, pAllocator
, buffer
);
3894 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
3898 *pBuffer
= radv_buffer_to_handle(buffer
);
3903 void radv_DestroyBuffer(
3906 const VkAllocationCallbacks
* pAllocator
)
3908 RADV_FROM_HANDLE(radv_device
, device
, _device
);
3909 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
3914 if (buffer
->flags
& VK_BUFFER_CREATE_SPARSE_BINDING_BIT
)
3915 device
->ws
->buffer_destroy(buffer
->bo
);
3917 vk_free2(&device
->alloc
, pAllocator
, buffer
);
3920 static inline unsigned
3921 si_tile_mode_index(const struct radv_image
*image
, unsigned level
, bool stencil
)
3924 return image
->surface
.u
.legacy
.stencil_tiling_index
[level
];
3926 return image
->surface
.u
.legacy
.tiling_index
[level
];
3929 static uint32_t radv_surface_max_layer_count(struct radv_image_view
*iview
)
3931 return iview
->type
== VK_IMAGE_VIEW_TYPE_3D
? iview
->extent
.depth
: (iview
->base_layer
+ iview
->layer_count
);
3935 radv_init_dcc_control_reg(struct radv_device
*device
,
3936 struct radv_image_view
*iview
)
3938 unsigned max_uncompressed_block_size
= V_028C78_MAX_BLOCK_SIZE_256B
;
3939 unsigned min_compressed_block_size
= V_028C78_MIN_BLOCK_SIZE_32B
;
3940 unsigned max_compressed_block_size
;
3941 unsigned independent_64b_blocks
;
3943 if (!radv_image_has_dcc(iview
->image
))
3946 if (iview
->image
->info
.samples
> 1) {
3947 if (iview
->image
->surface
.bpe
== 1)
3948 max_uncompressed_block_size
= V_028C78_MAX_BLOCK_SIZE_64B
;
3949 else if (iview
->image
->surface
.bpe
== 2)
3950 max_uncompressed_block_size
= V_028C78_MAX_BLOCK_SIZE_128B
;
3953 if (!device
->physical_device
->rad_info
.has_dedicated_vram
) {
3954 /* amdvlk: [min-compressed-block-size] should be set to 32 for
3955 * dGPU and 64 for APU because all of our APUs to date use
3956 * DIMMs which have a request granularity size of 64B while all
3957 * other chips have a 32B request size.
3959 min_compressed_block_size
= V_028C78_MIN_BLOCK_SIZE_64B
;
3962 if (iview
->image
->usage
& (VK_IMAGE_USAGE_SAMPLED_BIT
|
3963 VK_IMAGE_USAGE_TRANSFER_SRC_BIT
|
3964 VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT
)) {
3965 /* If this DCC image is potentially going to be used in texture
3966 * fetches, we need some special settings.
3968 independent_64b_blocks
= 1;
3969 max_compressed_block_size
= V_028C78_MAX_BLOCK_SIZE_64B
;
3971 /* MAX_UNCOMPRESSED_BLOCK_SIZE must be >=
3972 * MAX_COMPRESSED_BLOCK_SIZE. Set MAX_COMPRESSED_BLOCK_SIZE as
3973 * big as possible for better compression state.
3975 independent_64b_blocks
= 0;
3976 max_compressed_block_size
= max_uncompressed_block_size
;
3979 return S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size
) |
3980 S_028C78_MAX_COMPRESSED_BLOCK_SIZE(max_compressed_block_size
) |
3981 S_028C78_MIN_COMPRESSED_BLOCK_SIZE(min_compressed_block_size
) |
3982 S_028C78_INDEPENDENT_64B_BLOCKS(independent_64b_blocks
);
3986 radv_initialise_color_surface(struct radv_device
*device
,
3987 struct radv_color_buffer_info
*cb
,
3988 struct radv_image_view
*iview
)
3990 const struct vk_format_description
*desc
;
3991 unsigned ntype
, format
, swap
, endian
;
3992 unsigned blend_clamp
= 0, blend_bypass
= 0;
3994 const struct radeon_surf
*surf
= &iview
->image
->surface
;
3996 desc
= vk_format_description(iview
->vk_format
);
3998 memset(cb
, 0, sizeof(*cb
));
4000 /* Intensity is implemented as Red, so treat it that way. */
4001 cb
->cb_color_attrib
= S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == VK_SWIZZLE_1
);
4003 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
;
4005 cb
->cb_color_base
= va
>> 8;
4007 if (device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
4008 struct gfx9_surf_meta_flags meta
;
4009 if (iview
->image
->dcc_offset
)
4010 meta
= iview
->image
->surface
.u
.gfx9
.dcc
;
4012 meta
= iview
->image
->surface
.u
.gfx9
.cmask
;
4014 cb
->cb_color_attrib
|= S_028C74_COLOR_SW_MODE(iview
->image
->surface
.u
.gfx9
.surf
.swizzle_mode
) |
4015 S_028C74_FMASK_SW_MODE(iview
->image
->surface
.u
.gfx9
.fmask
.swizzle_mode
) |
4016 S_028C74_RB_ALIGNED(meta
.rb_aligned
) |
4017 S_028C74_PIPE_ALIGNED(meta
.pipe_aligned
);
4019 cb
->cb_color_base
+= iview
->image
->surface
.u
.gfx9
.surf_offset
>> 8;
4020 cb
->cb_color_base
|= iview
->image
->surface
.tile_swizzle
;
4022 const struct legacy_surf_level
*level_info
= &surf
->u
.legacy
.level
[iview
->base_mip
];
4023 unsigned pitch_tile_max
, slice_tile_max
, tile_mode_index
;
4025 cb
->cb_color_base
+= level_info
->offset
>> 8;
4026 if (level_info
->mode
== RADEON_SURF_MODE_2D
)
4027 cb
->cb_color_base
|= iview
->image
->surface
.tile_swizzle
;
4029 pitch_tile_max
= level_info
->nblk_x
/ 8 - 1;
4030 slice_tile_max
= (level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1;
4031 tile_mode_index
= si_tile_mode_index(iview
->image
, iview
->base_mip
, false);
4033 cb
->cb_color_pitch
= S_028C64_TILE_MAX(pitch_tile_max
);
4034 cb
->cb_color_slice
= S_028C68_TILE_MAX(slice_tile_max
);
4035 cb
->cb_color_cmask_slice
= iview
->image
->cmask
.slice_tile_max
;
4037 cb
->cb_color_attrib
|= S_028C74_TILE_MODE_INDEX(tile_mode_index
);
4039 if (radv_image_has_fmask(iview
->image
)) {
4040 if (device
->physical_device
->rad_info
.chip_class
>= CIK
)
4041 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(iview
->image
->fmask
.pitch_in_pixels
/ 8 - 1);
4042 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(iview
->image
->fmask
.tile_mode_index
);
4043 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(iview
->image
->fmask
.slice_tile_max
);
4045 /* This must be set for fast clear to work without FMASK. */
4046 if (device
->physical_device
->rad_info
.chip_class
>= CIK
)
4047 cb
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(pitch_tile_max
);
4048 cb
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index
);
4049 cb
->cb_color_fmask_slice
= S_028C88_TILE_MAX(slice_tile_max
);
4053 /* CMASK variables */
4054 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
;
4055 va
+= iview
->image
->cmask
.offset
;
4056 cb
->cb_color_cmask
= va
>> 8;
4058 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
;
4059 va
+= iview
->image
->dcc_offset
;
4060 cb
->cb_dcc_base
= va
>> 8;
4061 cb
->cb_dcc_base
|= iview
->image
->surface
.tile_swizzle
;
4063 uint32_t max_slice
= radv_surface_max_layer_count(iview
) - 1;
4064 cb
->cb_color_view
= S_028C6C_SLICE_START(iview
->base_layer
) |
4065 S_028C6C_SLICE_MAX(max_slice
);
4067 if (iview
->image
->info
.samples
> 1) {
4068 unsigned log_samples
= util_logbase2(iview
->image
->info
.samples
);
4070 cb
->cb_color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
4071 S_028C74_NUM_FRAGMENTS(log_samples
);
4074 if (radv_image_has_fmask(iview
->image
)) {
4075 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
+ iview
->image
->fmask
.offset
;
4076 cb
->cb_color_fmask
= va
>> 8;
4077 cb
->cb_color_fmask
|= iview
->image
->fmask
.tile_swizzle
;
4079 cb
->cb_color_fmask
= cb
->cb_color_base
;
4082 ntype
= radv_translate_color_numformat(iview
->vk_format
,
4084 vk_format_get_first_non_void_channel(iview
->vk_format
));
4085 format
= radv_translate_colorformat(iview
->vk_format
);
4086 if (format
== V_028C70_COLOR_INVALID
|| ntype
== ~0u)
4087 radv_finishme("Illegal color\n");
4088 swap
= radv_translate_colorswap(iview
->vk_format
, FALSE
);
4089 endian
= radv_colorformat_endian_swap(format
);
4091 /* blend clamp should be set for all NORM/SRGB types */
4092 if (ntype
== V_028C70_NUMBER_UNORM
||
4093 ntype
== V_028C70_NUMBER_SNORM
||
4094 ntype
== V_028C70_NUMBER_SRGB
)
4097 /* set blend bypass according to docs if SINT/UINT or
4098 8/24 COLOR variants */
4099 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
4100 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
4101 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
4106 if ((ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
) &&
4107 (format
== V_028C70_COLOR_8
||
4108 format
== V_028C70_COLOR_8_8
||
4109 format
== V_028C70_COLOR_8_8_8_8
))
4110 ->color_is_int8
= true;
4112 cb
->cb_color_info
= S_028C70_FORMAT(format
) |
4113 S_028C70_COMP_SWAP(swap
) |
4114 S_028C70_BLEND_CLAMP(blend_clamp
) |
4115 S_028C70_BLEND_BYPASS(blend_bypass
) |
4116 S_028C70_SIMPLE_FLOAT(1) |
4117 S_028C70_ROUND_MODE(ntype
!= V_028C70_NUMBER_UNORM
&&
4118 ntype
!= V_028C70_NUMBER_SNORM
&&
4119 ntype
!= V_028C70_NUMBER_SRGB
&&
4120 format
!= V_028C70_COLOR_8_24
&&
4121 format
!= V_028C70_COLOR_24_8
) |
4122 S_028C70_NUMBER_TYPE(ntype
) |
4123 S_028C70_ENDIAN(endian
);
4124 if (radv_image_has_fmask(iview
->image
)) {
4125 cb
->cb_color_info
|= S_028C70_COMPRESSION(1);
4126 if (device
->physical_device
->rad_info
.chip_class
== SI
) {
4127 unsigned fmask_bankh
= util_logbase2(iview
->image
->fmask
.bank_height
);
4128 cb
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh
);
4132 if (radv_image_has_cmask(iview
->image
) &&
4133 !(device
->instance
->debug_flags
& RADV_DEBUG_NO_FAST_CLEARS
))
4134 cb
->cb_color_info
|= S_028C70_FAST_CLEAR(1);
4136 if (radv_dcc_enabled(iview
->image
, iview
->base_mip
))
4137 cb
->cb_color_info
|= S_028C70_DCC_ENABLE(1);
4139 cb
->cb_dcc_control
= radv_init_dcc_control_reg(device
, iview
);
4141 /* This must be set for fast clear to work without FMASK. */
4142 if (!radv_image_has_fmask(iview
->image
) &&
4143 device
->physical_device
->rad_info
.chip_class
== SI
) {
4144 unsigned bankh
= util_logbase2(iview
->image
->surface
.u
.legacy
.bankh
);
4145 cb
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(bankh
);
4148 if (device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
4149 unsigned mip0_depth
= iview
->image
->type
== VK_IMAGE_TYPE_3D
?
4150 (iview
->extent
.depth
- 1) : (iview
->image
->info
.array_size
- 1);
4152 cb
->cb_color_view
|= S_028C6C_MIP_LEVEL(iview
->base_mip
);
4153 cb
->cb_color_attrib
|= S_028C74_MIP0_DEPTH(mip0_depth
) |
4154 S_028C74_RESOURCE_TYPE(iview
->image
->surface
.u
.gfx9
.resource_type
);
4155 cb
->cb_color_attrib2
= S_028C68_MIP0_WIDTH(iview
->extent
.width
- 1) |
4156 S_028C68_MIP0_HEIGHT(iview
->extent
.height
- 1) |
4157 S_028C68_MAX_MIP(iview
->image
->info
.levels
- 1);
4162 radv_calc_decompress_on_z_planes(struct radv_device
*device
,
4163 struct radv_image_view
*iview
)
4165 unsigned max_zplanes
= 0;
4167 assert(radv_image_is_tc_compat_htile(iview
->image
));
4169 if (device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
4170 /* Default value for 32-bit depth surfaces. */
4173 if (iview
->vk_format
== VK_FORMAT_D16_UNORM
&&
4174 iview
->image
->info
.samples
> 1)
4177 max_zplanes
= max_zplanes
+ 1;
4179 if (iview
->vk_format
== VK_FORMAT_D16_UNORM
) {
4180 /* Do not enable Z plane compression for 16-bit depth
4181 * surfaces because isn't supported on GFX8. Only
4182 * 32-bit depth surfaces are supported by the hardware.
4183 * This allows to maintain shader compatibility and to
4184 * reduce the number of depth decompressions.
4188 if (iview
->image
->info
.samples
<= 1)
4190 else if (iview
->image
->info
.samples
<= 4)
4201 radv_initialise_ds_surface(struct radv_device
*device
,
4202 struct radv_ds_buffer_info
*ds
,
4203 struct radv_image_view
*iview
)
4205 unsigned level
= iview
->base_mip
;
4206 unsigned format
, stencil_format
;
4207 uint64_t va
, s_offs
, z_offs
;
4208 bool stencil_only
= false;
4209 memset(ds
, 0, sizeof(*ds
));
4210 switch (iview
->image
->vk_format
) {
4211 case VK_FORMAT_D24_UNORM_S8_UINT
:
4212 case VK_FORMAT_X8_D24_UNORM_PACK32
:
4213 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
4214 ds
->offset_scale
= 2.0f
;
4216 case VK_FORMAT_D16_UNORM
:
4217 case VK_FORMAT_D16_UNORM_S8_UINT
:
4218 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
4219 ds
->offset_scale
= 4.0f
;
4221 case VK_FORMAT_D32_SFLOAT
:
4222 case VK_FORMAT_D32_SFLOAT_S8_UINT
:
4223 ds
->pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
4224 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
4225 ds
->offset_scale
= 1.0f
;
4227 case VK_FORMAT_S8_UINT
:
4228 stencil_only
= true;
4234 format
= radv_translate_dbformat(iview
->image
->vk_format
);
4235 stencil_format
= iview
->image
->surface
.has_stencil
?
4236 V_028044_STENCIL_8
: V_028044_STENCIL_INVALID
;
4238 uint32_t max_slice
= radv_surface_max_layer_count(iview
) - 1;
4239 ds
->db_depth_view
= S_028008_SLICE_START(iview
->base_layer
) |
4240 S_028008_SLICE_MAX(max_slice
);
4242 ds
->db_htile_data_base
= 0;
4243 ds
->db_htile_surface
= 0;
4245 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
;
4246 s_offs
= z_offs
= va
;
4248 if (device
->physical_device
->rad_info
.chip_class
>= GFX9
) {
4249 assert(iview
->image
->surface
.u
.gfx9
.surf_offset
== 0);
4250 s_offs
+= iview
->image
->surface
.u
.gfx9
.stencil_offset
;
4252 ds
->db_z_info
= S_028038_FORMAT(format
) |
4253 S_028038_NUM_SAMPLES(util_logbase2(iview
->image
->info
.samples
)) |
4254 S_028038_SW_MODE(iview
->image
->surface
.u
.gfx9
.surf
.swizzle_mode
) |
4255 S_028038_MAXMIP(iview
->image
->info
.levels
- 1) |
4256 S_028038_ZRANGE_PRECISION(1);
4257 ds
->db_stencil_info
= S_02803C_FORMAT(stencil_format
) |
4258 S_02803C_SW_MODE(iview
->image
->surface
.u
.gfx9
.stencil
.swizzle_mode
);
4260 ds
->db_z_info2
= S_028068_EPITCH(iview
->image
->surface
.u
.gfx9
.surf
.epitch
);
4261 ds
->db_stencil_info2
= S_02806C_EPITCH(iview
->image
->surface
.u
.gfx9
.stencil
.epitch
);
4262 ds
->db_depth_view
|= S_028008_MIPID(level
);
4264 ds
->db_depth_size
= S_02801C_X_MAX(iview
->image
->info
.width
- 1) |
4265 S_02801C_Y_MAX(iview
->image
->info
.height
- 1);
4267 if (radv_htile_enabled(iview
->image
, level
)) {
4268 ds
->db_z_info
|= S_028038_TILE_SURFACE_ENABLE(1);
4270 if (radv_image_is_tc_compat_htile(iview
->image
)) {
4271 unsigned max_zplanes
=
4272 radv_calc_decompress_on_z_planes(device
, iview
);
4274 ds
->db_z_info
|= S_028038_DECOMPRESS_ON_N_ZPLANES(max_zplanes
) |
4275 S_028038_ITERATE_FLUSH(1);
4276 ds
->db_stencil_info
|= S_02803C_ITERATE_FLUSH(1);
4279 if (!iview
->image
->surface
.has_stencil
)
4280 /* Use all of the htile_buffer for depth if there's no stencil. */
4281 ds
->db_stencil_info
|= S_02803C_TILE_STENCIL_DISABLE(1);
4282 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
+
4283 iview
->image
->htile_offset
;
4284 ds
->db_htile_data_base
= va
>> 8;
4285 ds
->db_htile_surface
= S_028ABC_FULL_CACHE(1) |
4286 S_028ABC_PIPE_ALIGNED(iview
->image
->surface
.u
.gfx9
.htile
.pipe_aligned
) |
4287 S_028ABC_RB_ALIGNED(iview
->image
->surface
.u
.gfx9
.htile
.rb_aligned
);
4290 const struct legacy_surf_level
*level_info
= &iview
->image
->surface
.u
.legacy
.level
[level
];
4293 level_info
= &iview
->image
->surface
.u
.legacy
.stencil_level
[level
];
4295 z_offs
+= iview
->image
->surface
.u
.legacy
.level
[level
].offset
;
4296 s_offs
+= iview
->image
->surface
.u
.legacy
.stencil_level
[level
].offset
;
4298 ds
->db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(!radv_image_is_tc_compat_htile(iview
->image
));
4299 ds
->db_z_info
= S_028040_FORMAT(format
) | S_028040_ZRANGE_PRECISION(1);
4300 ds
->db_stencil_info
= S_028044_FORMAT(stencil_format
);
4302 if (iview
->image
->info
.samples
> 1)
4303 ds
->db_z_info
|= S_028040_NUM_SAMPLES(util_logbase2(iview
->image
->info
.samples
));
4305 if (device
->physical_device
->rad_info
.chip_class
>= CIK
) {
4306 struct radeon_info
*info
= &device
->physical_device
->rad_info
;
4307 unsigned tiling_index
= iview
->image
->surface
.u
.legacy
.tiling_index
[level
];
4308 unsigned stencil_index
= iview
->image
->surface
.u
.legacy
.stencil_tiling_index
[level
];
4309 unsigned macro_index
= iview
->image
->surface
.u
.legacy
.macro_tile_index
;
4310 unsigned tile_mode
= info
->si_tile_mode_array
[tiling_index
];
4311 unsigned stencil_tile_mode
= info
->si_tile_mode_array
[stencil_index
];
4312 unsigned macro_mode
= info
->cik_macrotile_mode_array
[macro_index
];
4315 tile_mode
= stencil_tile_mode
;
4317 ds
->db_depth_info
|=
4318 S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode
)) |
4319 S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode
)) |
4320 S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode
)) |
4321 S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode
)) |
4322 S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode
)) |
4323 S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode
));
4324 ds
->db_z_info
|= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode
));
4325 ds
->db_stencil_info
|= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode
));
4327 unsigned tile_mode_index
= si_tile_mode_index(iview
->image
, level
, false);
4328 ds
->db_z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
4329 tile_mode_index
= si_tile_mode_index(iview
->image
, level
, true);
4330 ds
->db_stencil_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
4332 ds
->db_z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
4335 ds
->db_depth_size
= S_028058_PITCH_TILE_MAX((level_info
->nblk_x
/ 8) - 1) |
4336 S_028058_HEIGHT_TILE_MAX((level_info
->nblk_y
/ 8) - 1);
4337 ds
->db_depth_slice
= S_02805C_SLICE_TILE_MAX((level_info
->nblk_x
* level_info
->nblk_y
) / 64 - 1);
4339 if (radv_htile_enabled(iview
->image
, level
)) {
4340 ds
->db_z_info
|= S_028040_TILE_SURFACE_ENABLE(1);
4342 if (!iview
->image
->surface
.has_stencil
&&
4343 !radv_image_is_tc_compat_htile(iview
->image
))
4344 /* Use all of the htile_buffer for depth if there's no stencil. */
4345 ds
->db_stencil_info
|= S_028044_TILE_STENCIL_DISABLE(1);
4347 va
= radv_buffer_get_va(iview
->bo
) + iview
->image
->offset
+
4348 iview
->image
->htile_offset
;
4349 ds
->db_htile_data_base
= va
>> 8;
4350 ds
->db_htile_surface
= S_028ABC_FULL_CACHE(1);
4352 if (radv_image_is_tc_compat_htile(iview
->image
)) {
4353 unsigned max_zplanes
=
4354 radv_calc_decompress_on_z_planes(device
, iview
);
4356 ds
->db_htile_surface
|= S_028ABC_TC_COMPATIBLE(1);
4357 ds
->db_z_info
|= S_028040_DECOMPRESS_ON_N_ZPLANES(max_zplanes
);
4362 ds
->db_z_read_base
= ds
->db_z_write_base
= z_offs
>> 8;
4363 ds
->db_stencil_read_base
= ds
->db_stencil_write_base
= s_offs
>> 8;
4366 VkResult
radv_CreateFramebuffer(
4368 const VkFramebufferCreateInfo
* pCreateInfo
,
4369 const VkAllocationCallbacks
* pAllocator
,
4370 VkFramebuffer
* pFramebuffer
)
4372 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4373 struct radv_framebuffer
*framebuffer
;
4375 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO
);
4377 size_t size
= sizeof(*framebuffer
) +
4378 sizeof(struct radv_attachment_info
) * pCreateInfo
->attachmentCount
;
4379 framebuffer
= vk_alloc2(&device
->alloc
, pAllocator
, size
, 8,
4380 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
4381 if (framebuffer
== NULL
)
4382 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
4384 framebuffer
->attachment_count
= pCreateInfo
->attachmentCount
;
4385 framebuffer
->width
= pCreateInfo
->width
;
4386 framebuffer
->height
= pCreateInfo
->height
;
4387 framebuffer
->layers
= pCreateInfo
->layers
;
4388 for (uint32_t i
= 0; i
< pCreateInfo
->attachmentCount
; i
++) {
4389 VkImageView _iview
= pCreateInfo
->pAttachments
[i
];
4390 struct radv_image_view
*iview
= radv_image_view_from_handle(_iview
);
4391 framebuffer
->attachments
[i
].attachment
= iview
;
4392 if (iview
->aspect_mask
& VK_IMAGE_ASPECT_COLOR_BIT
) {
4393 radv_initialise_color_surface(device
, &framebuffer
->attachments
[i
].cb
, iview
);
4394 } else if (iview
->aspect_mask
& (VK_IMAGE_ASPECT_DEPTH_BIT
| VK_IMAGE_ASPECT_STENCIL_BIT
)) {
4395 radv_initialise_ds_surface(device
, &framebuffer
->attachments
[i
].ds
, iview
);
4397 framebuffer
->width
= MIN2(framebuffer
->width
, iview
->extent
.width
);
4398 framebuffer
->height
= MIN2(framebuffer
->height
, iview
->extent
.height
);
4399 framebuffer
->layers
= MIN2(framebuffer
->layers
, radv_surface_max_layer_count(iview
));
4402 *pFramebuffer
= radv_framebuffer_to_handle(framebuffer
);
4406 void radv_DestroyFramebuffer(
4409 const VkAllocationCallbacks
* pAllocator
)
4411 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4412 RADV_FROM_HANDLE(radv_framebuffer
, fb
, _fb
);
4416 vk_free2(&device
->alloc
, pAllocator
, fb
);
4419 static unsigned radv_tex_wrap(VkSamplerAddressMode address_mode
)
4421 switch (address_mode
) {
4422 case VK_SAMPLER_ADDRESS_MODE_REPEAT
:
4423 return V_008F30_SQ_TEX_WRAP
;
4424 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT
:
4425 return V_008F30_SQ_TEX_MIRROR
;
4426 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE
:
4427 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
4428 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER
:
4429 return V_008F30_SQ_TEX_CLAMP_BORDER
;
4430 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE
:
4431 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
4433 unreachable("illegal tex wrap mode");
4439 radv_tex_compare(VkCompareOp op
)
4442 case VK_COMPARE_OP_NEVER
:
4443 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
4444 case VK_COMPARE_OP_LESS
:
4445 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
4446 case VK_COMPARE_OP_EQUAL
:
4447 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
4448 case VK_COMPARE_OP_LESS_OR_EQUAL
:
4449 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
4450 case VK_COMPARE_OP_GREATER
:
4451 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
4452 case VK_COMPARE_OP_NOT_EQUAL
:
4453 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
4454 case VK_COMPARE_OP_GREATER_OR_EQUAL
:
4455 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
4456 case VK_COMPARE_OP_ALWAYS
:
4457 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
4459 unreachable("illegal compare mode");
4465 radv_tex_filter(VkFilter filter
, unsigned max_ansio
)
4468 case VK_FILTER_NEAREST
:
4469 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_POINT
:
4470 V_008F38_SQ_TEX_XY_FILTER_POINT
);
4471 case VK_FILTER_LINEAR
:
4472 return (max_ansio
> 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_BILINEAR
:
4473 V_008F38_SQ_TEX_XY_FILTER_BILINEAR
);
4474 case VK_FILTER_CUBIC_IMG
:
4476 fprintf(stderr
, "illegal texture filter");
4482 radv_tex_mipfilter(VkSamplerMipmapMode mode
)
4485 case VK_SAMPLER_MIPMAP_MODE_NEAREST
:
4486 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
4487 case VK_SAMPLER_MIPMAP_MODE_LINEAR
:
4488 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
4490 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
4495 radv_tex_bordercolor(VkBorderColor bcolor
)
4498 case VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK
:
4499 case VK_BORDER_COLOR_INT_TRANSPARENT_BLACK
:
4500 return V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
4501 case VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK
:
4502 case VK_BORDER_COLOR_INT_OPAQUE_BLACK
:
4503 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK
;
4504 case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE
:
4505 case VK_BORDER_COLOR_INT_OPAQUE_WHITE
:
4506 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE
;
4514 radv_tex_aniso_filter(unsigned filter
)
4528 radv_tex_filter_mode(VkSamplerReductionModeEXT mode
)
4531 case VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT
:
4532 return SQ_IMG_FILTER_MODE_BLEND
;
4533 case VK_SAMPLER_REDUCTION_MODE_MIN_EXT
:
4534 return SQ_IMG_FILTER_MODE_MIN
;
4535 case VK_SAMPLER_REDUCTION_MODE_MAX_EXT
:
4536 return SQ_IMG_FILTER_MODE_MAX
;
4544 radv_get_max_anisotropy(struct radv_device
*device
,
4545 const VkSamplerCreateInfo
*pCreateInfo
)
4547 if (device
->force_aniso
>= 0)
4548 return device
->force_aniso
;
4550 if (pCreateInfo
->anisotropyEnable
&&
4551 pCreateInfo
->maxAnisotropy
> 1.0f
)
4552 return (uint32_t)pCreateInfo
->maxAnisotropy
;
4558 radv_init_sampler(struct radv_device
*device
,
4559 struct radv_sampler
*sampler
,
4560 const VkSamplerCreateInfo
*pCreateInfo
)
4562 uint32_t max_aniso
= radv_get_max_anisotropy(device
, pCreateInfo
);
4563 uint32_t max_aniso_ratio
= radv_tex_aniso_filter(max_aniso
);
4564 bool is_vi
= (device
->physical_device
->rad_info
.chip_class
>= VI
);
4565 unsigned filter_mode
= SQ_IMG_FILTER_MODE_BLEND
;
4567 const struct VkSamplerReductionModeCreateInfoEXT
*sampler_reduction
=
4568 vk_find_struct_const(pCreateInfo
->pNext
,
4569 SAMPLER_REDUCTION_MODE_CREATE_INFO_EXT
);
4570 if (sampler_reduction
)
4571 filter_mode
= radv_tex_filter_mode(sampler_reduction
->reductionMode
);
4573 sampler
->state
[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo
->addressModeU
)) |
4574 S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo
->addressModeV
)) |
4575 S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo
->addressModeW
)) |
4576 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio
) |
4577 S_008F30_DEPTH_COMPARE_FUNC(radv_tex_compare(pCreateInfo
->compareOp
)) |
4578 S_008F30_FORCE_UNNORMALIZED(pCreateInfo
->unnormalizedCoordinates
? 1 : 0) |
4579 S_008F30_ANISO_THRESHOLD(max_aniso_ratio
>> 1) |
4580 S_008F30_ANISO_BIAS(max_aniso_ratio
) |
4581 S_008F30_DISABLE_CUBE_WRAP(0) |
4582 S_008F30_COMPAT_MODE(is_vi
) |
4583 S_008F30_FILTER_MODE(filter_mode
));
4584 sampler
->state
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(pCreateInfo
->minLod
, 0, 15), 8)) |
4585 S_008F34_MAX_LOD(S_FIXED(CLAMP(pCreateInfo
->maxLod
, 0, 15), 8)) |
4586 S_008F34_PERF_MIP(max_aniso_ratio
? max_aniso_ratio
+ 6 : 0));
4587 sampler
->state
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(pCreateInfo
->mipLodBias
, -16, 16), 8)) |
4588 S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo
->magFilter
, max_aniso
)) |
4589 S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo
->minFilter
, max_aniso
)) |
4590 S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo
->mipmapMode
)) |
4591 S_008F38_MIP_POINT_PRECLAMP(0) |
4592 S_008F38_DISABLE_LSB_CEIL(device
->physical_device
->rad_info
.chip_class
<= VI
) |
4593 S_008F38_FILTER_PREC_FIX(1) |
4594 S_008F38_ANISO_OVERRIDE(is_vi
));
4595 sampler
->state
[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
4596 S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(pCreateInfo
->borderColor
)));
4599 VkResult
radv_CreateSampler(
4601 const VkSamplerCreateInfo
* pCreateInfo
,
4602 const VkAllocationCallbacks
* pAllocator
,
4603 VkSampler
* pSampler
)
4605 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4606 struct radv_sampler
*sampler
;
4608 assert(pCreateInfo
->sType
== VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO
);
4610 sampler
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*sampler
), 8,
4611 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
4613 return vk_error(device
->instance
, VK_ERROR_OUT_OF_HOST_MEMORY
);
4615 radv_init_sampler(device
, sampler
, pCreateInfo
);
4616 *pSampler
= radv_sampler_to_handle(sampler
);
4621 void radv_DestroySampler(
4624 const VkAllocationCallbacks
* pAllocator
)
4626 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4627 RADV_FROM_HANDLE(radv_sampler
, sampler
, _sampler
);
4631 vk_free2(&device
->alloc
, pAllocator
, sampler
);
4634 /* vk_icd.h does not declare this function, so we declare it here to
4635 * suppress Wmissing-prototypes.
4637 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
4638 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
);
4640 PUBLIC VKAPI_ATTR VkResult VKAPI_CALL
4641 vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t *pSupportedVersion
)
4643 /* For the full details on loader interface versioning, see
4644 * <https://github.com/KhronosGroup/Vulkan-LoaderAndValidationLayers/blob/master/loader/LoaderAndLayerInterface.md>.
4645 * What follows is a condensed summary, to help you navigate the large and
4646 * confusing official doc.
4648 * - Loader interface v0 is incompatible with later versions. We don't
4651 * - In loader interface v1:
4652 * - The first ICD entrypoint called by the loader is
4653 * vk_icdGetInstanceProcAddr(). The ICD must statically expose this
4655 * - The ICD must statically expose no other Vulkan symbol unless it is
4656 * linked with -Bsymbolic.
4657 * - Each dispatchable Vulkan handle created by the ICD must be
4658 * a pointer to a struct whose first member is VK_LOADER_DATA. The
4659 * ICD must initialize VK_LOADER_DATA.loadMagic to ICD_LOADER_MAGIC.
4660 * - The loader implements vkCreate{PLATFORM}SurfaceKHR() and
4661 * vkDestroySurfaceKHR(). The ICD must be capable of working with
4662 * such loader-managed surfaces.
4664 * - Loader interface v2 differs from v1 in:
4665 * - The first ICD entrypoint called by the loader is
4666 * vk_icdNegotiateLoaderICDInterfaceVersion(). The ICD must
4667 * statically expose this entrypoint.
4669 * - Loader interface v3 differs from v2 in:
4670 * - The ICD must implement vkCreate{PLATFORM}SurfaceKHR(),
4671 * vkDestroySurfaceKHR(), and other API which uses VKSurfaceKHR,
4672 * because the loader no longer does so.
4674 *pSupportedVersion
= MIN2(*pSupportedVersion
, 3u);
4678 VkResult
radv_GetMemoryFdKHR(VkDevice _device
,
4679 const VkMemoryGetFdInfoKHR
*pGetFdInfo
,
4682 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4683 RADV_FROM_HANDLE(radv_device_memory
, memory
, pGetFdInfo
->memory
);
4685 assert(pGetFdInfo
->sType
== VK_STRUCTURE_TYPE_MEMORY_GET_FD_INFO_KHR
);
4687 /* At the moment, we support only the below handle types. */
4688 assert(pGetFdInfo
->handleType
==
4689 VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
||
4690 pGetFdInfo
->handleType
==
4691 VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
);
4693 bool ret
= radv_get_memory_fd(device
, memory
, pFD
);
4695 return vk_error(device
->instance
, VK_ERROR_OUT_OF_DEVICE_MEMORY
);
4699 VkResult
radv_GetMemoryFdPropertiesKHR(VkDevice _device
,
4700 VkExternalMemoryHandleTypeFlagBitsKHR handleType
,
4702 VkMemoryFdPropertiesKHR
*pMemoryFdProperties
)
4704 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4706 switch (handleType
) {
4707 case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT
:
4708 pMemoryFdProperties
->memoryTypeBits
= (1 << RADV_MEM_TYPE_COUNT
) - 1;
4712 /* The valid usage section for this function says:
4714 * "handleType must not be one of the handle types defined as
4717 * So opaque handle types fall into the default "unsupported" case.
4719 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4723 static VkResult
radv_import_opaque_fd(struct radv_device
*device
,
4727 uint32_t syncobj_handle
= 0;
4728 int ret
= device
->ws
->import_syncobj(device
->ws
, fd
, &syncobj_handle
);
4730 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4733 device
->ws
->destroy_syncobj(device
->ws
, *syncobj
);
4735 *syncobj
= syncobj_handle
;
4741 static VkResult
radv_import_sync_fd(struct radv_device
*device
,
4745 /* If we create a syncobj we do it locally so that if we have an error, we don't
4746 * leave a syncobj in an undetermined state in the fence. */
4747 uint32_t syncobj_handle
= *syncobj
;
4748 if (!syncobj_handle
) {
4749 int ret
= device
->ws
->create_syncobj(device
->ws
, &syncobj_handle
);
4751 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4756 device
->ws
->signal_syncobj(device
->ws
, syncobj_handle
);
4758 int ret
= device
->ws
->import_syncobj_from_sync_file(device
->ws
, syncobj_handle
, fd
);
4760 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4763 *syncobj
= syncobj_handle
;
4770 VkResult
radv_ImportSemaphoreFdKHR(VkDevice _device
,
4771 const VkImportSemaphoreFdInfoKHR
*pImportSemaphoreFdInfo
)
4773 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4774 RADV_FROM_HANDLE(radv_semaphore
, sem
, pImportSemaphoreFdInfo
->semaphore
);
4775 uint32_t *syncobj_dst
= NULL
;
4777 if (pImportSemaphoreFdInfo
->flags
& VK_SEMAPHORE_IMPORT_TEMPORARY_BIT_KHR
) {
4778 syncobj_dst
= &sem
->temp_syncobj
;
4780 syncobj_dst
= &sem
->syncobj
;
4783 switch(pImportSemaphoreFdInfo
->handleType
) {
4784 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
:
4785 return radv_import_opaque_fd(device
, pImportSemaphoreFdInfo
->fd
, syncobj_dst
);
4786 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT_KHR
:
4787 return radv_import_sync_fd(device
, pImportSemaphoreFdInfo
->fd
, syncobj_dst
);
4789 unreachable("Unhandled semaphore handle type");
4793 VkResult
radv_GetSemaphoreFdKHR(VkDevice _device
,
4794 const VkSemaphoreGetFdInfoKHR
*pGetFdInfo
,
4797 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4798 RADV_FROM_HANDLE(radv_semaphore
, sem
, pGetFdInfo
->semaphore
);
4800 uint32_t syncobj_handle
;
4802 if (sem
->temp_syncobj
)
4803 syncobj_handle
= sem
->temp_syncobj
;
4805 syncobj_handle
= sem
->syncobj
;
4807 switch(pGetFdInfo
->handleType
) {
4808 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
:
4809 ret
= device
->ws
->export_syncobj(device
->ws
, syncobj_handle
, pFd
);
4811 case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT_KHR
:
4812 ret
= device
->ws
->export_syncobj_to_sync_file(device
->ws
, syncobj_handle
, pFd
);
4814 if (sem
->temp_syncobj
) {
4815 close (sem
->temp_syncobj
);
4816 sem
->temp_syncobj
= 0;
4818 device
->ws
->reset_syncobj(device
->ws
, syncobj_handle
);
4823 unreachable("Unhandled semaphore handle type");
4827 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4831 void radv_GetPhysicalDeviceExternalSemaphoreProperties(
4832 VkPhysicalDevice physicalDevice
,
4833 const VkPhysicalDeviceExternalSemaphoreInfoKHR
* pExternalSemaphoreInfo
,
4834 VkExternalSemaphorePropertiesKHR
* pExternalSemaphoreProperties
)
4836 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
4838 /* Require has_syncobj_wait_for_submit for the syncobj signal ioctl introduced at virtually the same time */
4839 if (pdevice
->rad_info
.has_syncobj_wait_for_submit
&&
4840 (pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
||
4841 pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT_KHR
)) {
4842 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT_KHR
;
4843 pExternalSemaphoreProperties
->compatibleHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
| VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT_KHR
;
4844 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT_KHR
|
4845 VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT_KHR
;
4846 } else if (pExternalSemaphoreInfo
->handleType
== VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
) {
4847 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
;
4848 pExternalSemaphoreProperties
->compatibleHandleTypes
= VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
;
4849 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT_KHR
|
4850 VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT_KHR
;
4852 pExternalSemaphoreProperties
->exportFromImportedHandleTypes
= 0;
4853 pExternalSemaphoreProperties
->compatibleHandleTypes
= 0;
4854 pExternalSemaphoreProperties
->externalSemaphoreFeatures
= 0;
4858 VkResult
radv_ImportFenceFdKHR(VkDevice _device
,
4859 const VkImportFenceFdInfoKHR
*pImportFenceFdInfo
)
4861 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4862 RADV_FROM_HANDLE(radv_fence
, fence
, pImportFenceFdInfo
->fence
);
4863 uint32_t *syncobj_dst
= NULL
;
4866 if (pImportFenceFdInfo
->flags
& VK_FENCE_IMPORT_TEMPORARY_BIT_KHR
) {
4867 syncobj_dst
= &fence
->temp_syncobj
;
4869 syncobj_dst
= &fence
->syncobj
;
4872 switch(pImportFenceFdInfo
->handleType
) {
4873 case VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
:
4874 return radv_import_opaque_fd(device
, pImportFenceFdInfo
->fd
, syncobj_dst
);
4875 case VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT_KHR
:
4876 return radv_import_sync_fd(device
, pImportFenceFdInfo
->fd
, syncobj_dst
);
4878 unreachable("Unhandled fence handle type");
4882 VkResult
radv_GetFenceFdKHR(VkDevice _device
,
4883 const VkFenceGetFdInfoKHR
*pGetFdInfo
,
4886 RADV_FROM_HANDLE(radv_device
, device
, _device
);
4887 RADV_FROM_HANDLE(radv_fence
, fence
, pGetFdInfo
->fence
);
4889 uint32_t syncobj_handle
;
4891 if (fence
->temp_syncobj
)
4892 syncobj_handle
= fence
->temp_syncobj
;
4894 syncobj_handle
= fence
->syncobj
;
4896 switch(pGetFdInfo
->handleType
) {
4897 case VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
:
4898 ret
= device
->ws
->export_syncobj(device
->ws
, syncobj_handle
, pFd
);
4900 case VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT_KHR
:
4901 ret
= device
->ws
->export_syncobj_to_sync_file(device
->ws
, syncobj_handle
, pFd
);
4903 if (fence
->temp_syncobj
) {
4904 close (fence
->temp_syncobj
);
4905 fence
->temp_syncobj
= 0;
4907 device
->ws
->reset_syncobj(device
->ws
, syncobj_handle
);
4912 unreachable("Unhandled fence handle type");
4916 return vk_error(device
->instance
, VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR
);
4920 void radv_GetPhysicalDeviceExternalFenceProperties(
4921 VkPhysicalDevice physicalDevice
,
4922 const VkPhysicalDeviceExternalFenceInfoKHR
* pExternalFenceInfo
,
4923 VkExternalFencePropertiesKHR
* pExternalFenceProperties
)
4925 RADV_FROM_HANDLE(radv_physical_device
, pdevice
, physicalDevice
);
4927 if (pdevice
->rad_info
.has_syncobj_wait_for_submit
&&
4928 (pExternalFenceInfo
->handleType
== VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
||
4929 pExternalFenceInfo
->handleType
== VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT_KHR
)) {
4930 pExternalFenceProperties
->exportFromImportedHandleTypes
= VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
| VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT_KHR
;
4931 pExternalFenceProperties
->compatibleHandleTypes
= VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT_KHR
| VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT_KHR
;
4932 pExternalFenceProperties
->externalFenceFeatures
= VK_EXTERNAL_FENCE_FEATURE_EXPORTABLE_BIT_KHR
|
4933 VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT_KHR
;
4935 pExternalFenceProperties
->exportFromImportedHandleTypes
= 0;
4936 pExternalFenceProperties
->compatibleHandleTypes
= 0;
4937 pExternalFenceProperties
->externalFenceFeatures
= 0;
4942 radv_CreateDebugReportCallbackEXT(VkInstance _instance
,
4943 const VkDebugReportCallbackCreateInfoEXT
* pCreateInfo
,
4944 const VkAllocationCallbacks
* pAllocator
,
4945 VkDebugReportCallbackEXT
* pCallback
)
4947 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
4948 return vk_create_debug_report_callback(&instance
->debug_report_callbacks
,
4949 pCreateInfo
, pAllocator
, &instance
->alloc
,
4954 radv_DestroyDebugReportCallbackEXT(VkInstance _instance
,
4955 VkDebugReportCallbackEXT _callback
,
4956 const VkAllocationCallbacks
* pAllocator
)
4958 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
4959 vk_destroy_debug_report_callback(&instance
->debug_report_callbacks
,
4960 _callback
, pAllocator
, &instance
->alloc
);
4964 radv_DebugReportMessageEXT(VkInstance _instance
,
4965 VkDebugReportFlagsEXT flags
,
4966 VkDebugReportObjectTypeEXT objectType
,
4969 int32_t messageCode
,
4970 const char* pLayerPrefix
,
4971 const char* pMessage
)
4973 RADV_FROM_HANDLE(radv_instance
, instance
, _instance
);
4974 vk_debug_report(&instance
->debug_report_callbacks
, flags
, objectType
,
4975 object
, location
, messageCode
, pLayerPrefix
, pMessage
);
4979 radv_GetDeviceGroupPeerMemoryFeatures(
4982 uint32_t localDeviceIndex
,
4983 uint32_t remoteDeviceIndex
,
4984 VkPeerMemoryFeatureFlags
* pPeerMemoryFeatures
)
4986 assert(localDeviceIndex
== remoteDeviceIndex
);
4988 *pPeerMemoryFeatures
= VK_PEER_MEMORY_FEATURE_COPY_SRC_BIT
|
4989 VK_PEER_MEMORY_FEATURE_COPY_DST_BIT
|
4990 VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT
|
4991 VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT
;
4994 static const VkTimeDomainEXT radv_time_domains
[] = {
4995 VK_TIME_DOMAIN_DEVICE_EXT
,
4996 VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT
,
4997 VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT
,
5000 VkResult
radv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
5001 VkPhysicalDevice physicalDevice
,
5002 uint32_t *pTimeDomainCount
,
5003 VkTimeDomainEXT
*pTimeDomains
)
5006 VK_OUTARRAY_MAKE(out
, pTimeDomains
, pTimeDomainCount
);
5008 for (d
= 0; d
< ARRAY_SIZE(radv_time_domains
); d
++) {
5009 vk_outarray_append(&out
, i
) {
5010 *i
= radv_time_domains
[d
];
5014 return vk_outarray_status(&out
);
5018 radv_clock_gettime(clockid_t clock_id
)
5020 struct timespec current
;
5023 ret
= clock_gettime(clock_id
, ¤t
);
5024 if (ret
< 0 && clock_id
== CLOCK_MONOTONIC_RAW
)
5025 ret
= clock_gettime(CLOCK_MONOTONIC
, ¤t
);
5029 return (uint64_t) current
.tv_sec
* 1000000000ULL + current
.tv_nsec
;
5032 VkResult
radv_GetCalibratedTimestampsEXT(
5034 uint32_t timestampCount
,
5035 const VkCalibratedTimestampInfoEXT
*pTimestampInfos
,
5036 uint64_t *pTimestamps
,
5037 uint64_t *pMaxDeviation
)
5039 RADV_FROM_HANDLE(radv_device
, device
, _device
);
5040 uint32_t clock_crystal_freq
= device
->physical_device
->rad_info
.clock_crystal_freq
;
5042 uint64_t begin
, end
;
5043 uint64_t max_clock_period
= 0;
5045 begin
= radv_clock_gettime(CLOCK_MONOTONIC_RAW
);
5047 for (d
= 0; d
< timestampCount
; d
++) {
5048 switch (pTimestampInfos
[d
].timeDomain
) {
5049 case VK_TIME_DOMAIN_DEVICE_EXT
:
5050 pTimestamps
[d
] = device
->ws
->query_value(device
->ws
,
5052 uint64_t device_period
= DIV_ROUND_UP(1000000, clock_crystal_freq
);
5053 max_clock_period
= MAX2(max_clock_period
, device_period
);
5055 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT
:
5056 pTimestamps
[d
] = radv_clock_gettime(CLOCK_MONOTONIC
);
5057 max_clock_period
= MAX2(max_clock_period
, 1);
5060 case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT
:
5061 pTimestamps
[d
] = begin
;
5069 end
= radv_clock_gettime(CLOCK_MONOTONIC_RAW
);
5072 * The maximum deviation is the sum of the interval over which we
5073 * perform the sampling and the maximum period of any sampled
5074 * clock. That's because the maximum skew between any two sampled
5075 * clock edges is when the sampled clock with the largest period is
5076 * sampled at the end of that period but right at the beginning of the
5077 * sampling interval and some other clock is sampled right at the
5078 * begining of its sampling period and right at the end of the
5079 * sampling interval. Let's assume the GPU has the longest clock
5080 * period and that the application is sampling GPU and monotonic:
5083 * w x y z 0 1 2 3 4 5 6 7 8 9 a b c d e f
5084 * Raw -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
5088 * GPU -----_____-----_____-----_____-----_____
5091 * x y z 0 1 2 3 4 5 6 7 8 9 a b c
5092 * Monotonic -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
5094 * Interval <----------------->
5095 * Deviation <-------------------------->
5099 * m = read(monotonic) 2
5102 * We round the sample interval up by one tick to cover sampling error
5103 * in the interval clock
5106 uint64_t sample_interval
= end
- begin
+ 1;
5108 *pMaxDeviation
= sample_interval
+ max_clock_period
;