Mesa (main): radv: Support address capture and replay.
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Mon Jul 5 17:52:16 UTC 2021
Module: Mesa
Branch: main
Commit: 406c87b779d7abf4f23b99beb045f95a31a1e8e0
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=406c87b779d7abf4f23b99beb045f95a31a1e8e0
Author: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
Date: Fri Apr 30 14:22:15 2021 +0200
radv: Support address capture and replay.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10570>
---
src/amd/vulkan/radv_device.c | 42 ++++++++++++++++++++-------
src/amd/vulkan/radv_radeon_winsys.h | 1 +
src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c | 10 +++++--
3 files changed, 40 insertions(+), 13 deletions(-)
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index a796ca1fc44..6a74cc01f20 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1202,7 +1202,7 @@ radv_get_physical_device_features_1_2(struct radv_physical_device *pdevice,
f->separateDepthStencilLayouts = true;
f->hostQueryReset = true;
f->timelineSemaphore = true, f->bufferDeviceAddress = true;
- f->bufferDeviceAddressCaptureReplay = false;
+ f->bufferDeviceAddressCaptureReplay = true;
f->bufferDeviceAddressMultiDevice = false;
f->vulkanMemoryModel = true;
f->vulkanMemoryModelDeviceScope = true;
@@ -1335,9 +1335,9 @@ radv_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT: {
VkPhysicalDeviceBufferDeviceAddressFeaturesEXT *features =
(VkPhysicalDeviceBufferDeviceAddressFeaturesEXT *)ext;
- features->bufferDeviceAddress = true;
- features->bufferDeviceAddressCaptureReplay = false;
- features->bufferDeviceAddressMultiDevice = false;
+ CORE_FEATURE(1, 2, bufferDeviceAddress);
+ CORE_FEATURE(1, 2, bufferDeviceAddressCaptureReplay);
+ CORE_FEATURE(1, 2, bufferDeviceAddressMultiDevice);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES: {
@@ -5296,6 +5296,12 @@ radv_alloc_memory(struct radv_device *device, const VkMemoryAllocateInfo *pAlloc
if (priority_ext)
priority_float = priority_ext->priority;
+ uint64_t replay_address = 0;
+ const VkMemoryOpaqueCaptureAddressAllocateInfo *replay_info =
+ vk_find_struct_const(pAllocateInfo->pNext, MEMORY_OPAQUE_CAPTURE_ADDRESS_ALLOCATE_INFO);
+ if (replay_info && replay_info->opaqueCaptureAddress)
+ replay_address = replay_info->opaqueCaptureAddress;
+
unsigned priority = MIN2(RADV_BO_PRIORITY_APPLICATION_MAX - 1,
(int)(priority_float * RADV_BO_PRIORITY_APPLICATION_MAX));
@@ -5369,6 +5375,10 @@ radv_alloc_memory(struct radv_device *device, const VkMemoryAllocateInfo *pAlloc
}
}
+ const VkMemoryAllocateFlagsInfo *flags_info = vk_find_struct_const(pAllocateInfo->pNext, MEMORY_ALLOCATE_FLAGS_INFO);
+ if (flags_info && flags_info->flags & VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_CAPTURE_REPLAY_BIT)
+ flags |= RADEON_FLAG_REPLAYABLE;
+
if (device->overallocation_disallowed) {
uint64_t total_size =
device->physical_device->memory_properties.memoryHeaps[heap_index].size;
@@ -5385,7 +5395,7 @@ radv_alloc_memory(struct radv_device *device, const VkMemoryAllocateInfo *pAlloc
result = device->ws->buffer_create(device->ws, alloc_size,
device->physical_device->rad_info.max_alignment, domain,
- flags, priority, 0, &mem->bo);
+ flags, priority, replay_address, &mem->bo);
if (result != VK_SUCCESS) {
if (device->overallocation_disallowed) {
@@ -6385,9 +6395,19 @@ radv_CreateBuffer(VkDevice _device, const VkBufferCreateInfo *pCreateInfo,
vk_find_struct_const(pCreateInfo->pNext, EXTERNAL_MEMORY_BUFFER_CREATE_INFO) != NULL;
if (pCreateInfo->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT) {
- VkResult result =
- device->ws->buffer_create(device->ws, align64(buffer->size, 4096), 4096, 0,
- RADEON_FLAG_VIRTUAL, RADV_BO_PRIORITY_VIRTUAL, 0, &buffer->bo);
+ enum radeon_bo_flag flags = RADEON_FLAG_VIRTUAL;
+ if (pCreateInfo->flags & VK_BUFFER_CREATE_DEVICE_ADDRESS_CAPTURE_REPLAY_BIT)
+ flags |= RADEON_FLAG_REPLAYABLE;
+
+ uint64_t replay_address = 0;
+ const VkBufferOpaqueCaptureAddressCreateInfo *replay_info =
+ vk_find_struct_const(pCreateInfo->pNext, BUFFER_OPAQUE_CAPTURE_ADDRESS_CREATE_INFO);
+ if (replay_info && replay_info->opaqueCaptureAddress)
+ replay_address = replay_info->opaqueCaptureAddress;
+
+ VkResult result = device->ws->buffer_create(device->ws, align64(buffer->size, 4096), 4096, 0,
+ flags, RADV_BO_PRIORITY_VIRTUAL,
+ replay_address, &buffer->bo);
if (result != VK_SUCCESS) {
radv_destroy_buffer(device, pAllocator, buffer);
return vk_error(device->instance, result);
@@ -6421,14 +6441,16 @@ radv_GetBufferDeviceAddress(VkDevice device, const VkBufferDeviceAddressInfo *pI
uint64_t
radv_GetBufferOpaqueCaptureAddress(VkDevice device, const VkBufferDeviceAddressInfo *pInfo)
{
- return 0;
+ RADV_FROM_HANDLE(radv_buffer, buffer, pInfo->buffer);
+ return buffer->bo ? radv_buffer_get_va(buffer->bo) + buffer->offset : 0;
}
uint64_t
radv_GetDeviceMemoryOpaqueCaptureAddress(VkDevice device,
const VkDeviceMemoryOpaqueCaptureAddressInfo *pInfo)
{
- return 0;
+ RADV_FROM_HANDLE(radv_device_memory, mem, pInfo->memory);
+ return radv_buffer_get_va(mem->bo);
}
static inline unsigned
diff --git a/src/amd/vulkan/radv_radeon_winsys.h b/src/amd/vulkan/radv_radeon_winsys.h
index 2a50556cab6..9f12b6040c3 100644
--- a/src/amd/vulkan/radv_radeon_winsys.h
+++ b/src/amd/vulkan/radv_radeon_winsys.h
@@ -63,6 +63,7 @@ enum radeon_bo_flag { /* bitfield */
RADEON_FLAG_32BIT = (1 << 8),
RADEON_FLAG_PREFER_LOCAL_BO = (1 << 9),
RADEON_FLAG_ZERO_VRAM = (1 << 10),
+ RADEON_FLAG_REPLAYABLE = (1 << 11),
};
enum radeon_ctx_priority {
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
index baec9066a14..3bea2f38028 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
@@ -420,9 +420,13 @@ radv_amdgpu_winsys_bo_create(struct radeon_winsys *_ws, uint64_t size, unsigned
if (size >= ws->info.pte_fragment_size)
virt_alignment = MAX2(virt_alignment, ws->info.pte_fragment_size);
- r = amdgpu_va_range_alloc(
- ws->dev, amdgpu_gpu_va_range_general, size, virt_alignment, replay_address, &va, &va_handle,
- (flags & RADEON_FLAG_32BIT ? AMDGPU_VA_RANGE_32_BIT : 0) | AMDGPU_VA_RANGE_HIGH);
+ assert(!replay_address || (flags & RADEON_FLAG_REPLAYABLE));
+
+ const uint64_t va_flags = AMDGPU_VA_RANGE_HIGH |
+ (flags & RADEON_FLAG_32BIT ? AMDGPU_VA_RANGE_32_BIT : 0) |
+ (flags & RADEON_FLAG_REPLAYABLE ? AMDGPU_VA_RANGE_REPLAYABLE : 0);
+ r = amdgpu_va_range_alloc(ws->dev, amdgpu_gpu_va_range_general, size, virt_alignment, replay_address,
+ &va, &va_handle, va_flags);
if (r) {
result =
replay_address ? VK_ERROR_INVALID_OPAQUE_CAPTURE_ADDRESS : VK_ERROR_OUT_OF_DEVICE_MEMORY;
More information about the mesa-commit
mailing list