Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Portable RT: Fixes for shader-binding-table assembly (sbt data) #1993

Merged
merged 9 commits into from
Jan 30, 2025
Merged

Portable RT: Fixes for shader-binding-table assembly (sbt data) #1993

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
2a55bdd
Account for all bind-points when tracking bound pipelines
fabian-lunarg Jan 27, 2025
0950541
Adjust internal compute-shader: pass-through data that could not be m…
fabian-lunarg Jan 27, 2025
db9cc6a
Keep track of opaque/assigned acceleration-structure addresses
fabian-lunarg Jan 27, 2025
277d50b
Rebind previously bound compute-pipelines, if any
fabian-lunarg Jan 27, 2025
aae9d06
Adjust internal compute-shader: pass-through data that could not be m…
fabian-lunarg Jan 27, 2025
e136f23
rework I/O address calculation, account for data-payloads and pass-th…
fabian-lunarg Jan 27, 2025
f12a476
Drop GL_GOOGLE_include_directive from internal compute-shader, not used
fabian-lunarg Jan 28, 2025
a848a37
Use accelerationStructureCaptureReplay only when allocator supports o…
fabian-lunarg Jan 28, 2025
a68aeea
work on review-comments
fabian-lunarg Jan 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
74 changes: 57 additions & 17 deletions framework/decode/vulkan_address_replacer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,7 +147,7 @@ VulkanAddressReplacer::VulkanAddressReplacer(const VulkanDeviceInfo*
const encode::VulkanDeviceTable* device_table,
const encode::VulkanInstanceTable* instance_table,
const decode::CommonObjectInfoTable& object_table) :
device_table_(device_table)
device_table_(device_table), object_table_(&object_table)
{
GFXRECON_ASSERT(device_info != nullptr && device_table != nullptr && instance_table != nullptr);
physical_device_info_ = object_table.GetVkPhysicalDeviceInfo(device_info->parent_id);
Expand Down Expand Up @@ -354,8 +354,8 @@ void VulkanAddressReplacer::ProcessCmdTraceRays(
}
auto input_addresses = reinterpret_cast<VkDeviceAddress*>(pipeline_context_sbt.input_handle_buffer.mapped_data);

std::unordered_map<const VkStridedDeviceAddressRegionKHR*, uint32_t> num_handles_map;
uint32_t num_addresses = 0;
std::unordered_map<const VkStridedDeviceAddressRegionKHR*, uint32_t> num_addresses_map;
uint32_t num_addresses = 0;

{
const auto handle_size_capture = static_cast<uint32_t>(util::aligned_value(
Expand All @@ -365,10 +365,18 @@ void VulkanAddressReplacer::ProcessCmdTraceRays(
{
if (region != nullptr && region->size != 0 && region->stride != 0)
{
num_handles_map[region] = region->size / region->stride;
num_addresses_map[region] = region->size / capture_ray_properties_->shaderGroupHandleSize;

for (uint32_t offset = 0; offset < region->size; offset += region->stride)
uint32_t capture_handle_size = capture_ray_properties_->shaderGroupHandleSize;
if (region->stride > capture_handle_size)
{
uint32_t payload_size = region->stride - capture_handle_size;
GFXRECON_LOG_DEBUG("Extra data in sbt: %d", payload_size);
}

for (uint32_t offset = 0; offset < region->size; offset += capture_handle_size)
{
// input-address are handles and extra data
input_addresses[num_addresses++] = region->deviceAddress + offset;
}
}
Expand Down Expand Up @@ -430,15 +438,17 @@ void VulkanAddressReplacer::ProcessCmdTraceRays(
uint32_t num_handles_limit = region->size / region->stride;
auto group_size = static_cast<uint32_t>(util::aligned_value(
num_handles_limit * handle_size_aligned, replay_ray_properties_->shaderGroupBaseAlignment));
sbt_offset += group_size;

// adjust group-size
region->size = group_size;
region->stride = handle_size_aligned;
// increase group-size/stride, if required
region->size = std::max<VkDeviceSize>(group_size, region->size);
region->stride = std::max<VkDeviceSize>(handle_size_aligned, region->stride);

sbt_offset += region->size;
}
}
// raygen: stride == size
raygen_sbt->size = raygen_sbt->stride = replay_ray_properties_->shaderGroupBaseAlignment;
raygen_sbt->size = raygen_sbt->stride =
util::aligned_value(raygen_sbt->stride, replay_ray_properties_->shaderGroupBaseAlignment);

if (!create_buffer(shadow_buf_context,
sbt_offset,
Expand All @@ -457,7 +467,7 @@ void VulkanAddressReplacer::ProcessCmdTraceRays(
{
if (region != nullptr && region->size != 0 && region->stride != 0)
{
uint32_t num_handles = num_handles_map[region];
uint32_t num_handles = num_addresses_map[region];

// assign shadow-sbt-address
region->deviceAddress = shadow_buf_context.device_address + sbt_offset;
Expand Down Expand Up @@ -497,7 +507,20 @@ void VulkanAddressReplacer::ProcessCmdTraceRays(
VK_ACCESS_SHADER_READ_BIT);
}

// set previous push-constant data
// set previous compute-pipeline, if any
if (command_buffer_info->bound_pipelines.count(VK_PIPELINE_BIND_POINT_COMPUTE))
{
auto* previous_pipeline = object_table_->GetVkPipelineInfo(
command_buffer_info->bound_pipelines.at(VK_PIPELINE_BIND_POINT_COMPUTE));
GFXRECON_ASSERT(previous_pipeline);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

previous_pipeline MUST be valid?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yes. at this point we already checked the bound_pipelines map and know a compute-pipeline was previously bound (rare'ish case anyway, since the command-buffer is used to record a raytracing pipeline). so in fact, we only assert that it can also be mapped.

Copy link
Contributor Author

@fabian-lunarg fabian-lunarg Jan 27, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

double-checked the assumption, I think it's correct. the only caveat I could think of if a compute-pipeline would have been bound and then unbound again. but that's not allowed ("pipeline must be a valid VkPipeline handle")

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@antonio-lunarg is your concern resolved?

if (previous_pipeline != nullptr)
{
device_table_->CmdBindPipeline(
command_buffer_info->handle, VK_PIPELINE_BIND_POINT_COMPUTE, previous_pipeline->handle);
}
}

// set previous push-constant data, if any
if (!command_buffer_info->push_constant_data.empty())
{
device_table_->CmdPushConstants(command_buffer_info->handle,
Expand Down Expand Up @@ -578,13 +601,17 @@ void VulkanAddressReplacer::ProcessCmdBuildAccelerationStructuresKHR(
&build_size_info);
}

bool as_buffer_usable = false;

// retrieve VkAccelerationStructureKHR -> VkBuffer -> check/correct size
auto* acceleration_structure_info =
address_tracker.GetAccelerationStructureByHandle(build_geometry_info.dstAccelerationStructure);
GFXRECON_ASSERT(acceleration_structure_info != nullptr);
auto* buffer_info = address_tracker.GetBufferByHandle(acceleration_structure_info->buffer);
bool as_buffer_usable =
buffer_info != nullptr && buffer_info->size >= build_size_info.accelerationStructureSize;
if (acceleration_structure_info != nullptr)
{
auto* buffer_info = address_tracker.GetBufferByHandle(acceleration_structure_info->buffer);
as_buffer_usable =
buffer_info != nullptr && buffer_info->size >= build_size_info.accelerationStructureSize;
}

// determine required size of scratch-buffer
uint32_t scratch_size = build_geometry_info.mode == VK_BUILD_ACCELERATION_STRUCTURE_MODE_BUILD_KHR
Expand Down Expand Up @@ -811,7 +838,20 @@ void VulkanAddressReplacer::ProcessCmdBuildAccelerationStructuresKHR(
VK_ACCESS_SHADER_READ_BIT);
}

// set previous push-constant data
// set previous compute-pipeline, if any
if (command_buffer_info->bound_pipelines.count(VK_PIPELINE_BIND_POINT_COMPUTE))
{
auto* previous_pipeline = object_table_->GetVkPipelineInfo(
command_buffer_info->bound_pipelines.at(VK_PIPELINE_BIND_POINT_COMPUTE));
GFXRECON_ASSERT(previous_pipeline);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@antonio-lunarg is your concern resolved?

if (previous_pipeline != nullptr)
{
device_table_->CmdBindPipeline(
command_buffer_info->handle, VK_PIPELINE_BIND_POINT_COMPUTE, previous_pipeline->handle);
}
}

// set previous push-constant data, if any
if (!command_buffer_info->push_constant_data.empty())
{
device_table_->CmdPushConstants(command_buffer_info->handle,
Expand Down
1 change: 1 addition & 0 deletions framework/decode/vulkan_address_replacer.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -298,6 +298,7 @@ class VulkanAddressReplacer
bool swap_acceleration_structure_handle(VkAccelerationStructureKHR& handle);

const encode::VulkanDeviceTable* device_table_ = nullptr;
const decode::CommonObjectInfoTable* object_table_ = nullptr;
VkPhysicalDeviceMemoryProperties memory_properties_ = {};
std::optional<VkPhysicalDeviceRayTracingPipelinePropertiesKHR> capture_ray_properties_{}, replay_ray_properties_{};
std::optional<VkPhysicalDeviceAccelerationStructurePropertiesKHR> replay_acceleration_structure_properties_{};
Expand Down
Loading
Loading