mirrors
/
vkd3d-proton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vkd3d: Attempt to reuse application indirect command buffer. 

Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
This commit is contained in:
Hans-Kristian Arntzen 2022-06-10 14:35:21 +02:00
parent 9e45c72256
commit e17a7cb40c
1 changed files with 129 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

206
libs/vkd3d/command.c
View File

@ -9647,6 +9647,7 @@ static void d3d12_command_list_execute_indirect_state_template(
struct d3d12_resource *count_buffer, UINT64 count_buffer_offset)
{
const struct vkd3d_vk_device_procs *vk_procs = &list->device->vk_procs;
const VkPhysicalDeviceDeviceGeneratedCommandsPropertiesNV *props;
struct vkd3d_scratch_allocation preprocess_allocation;
struct vkd3d_scratch_allocation stream_allocation;
struct vkd3d_scratch_allocation count_allocation;
@ -9657,6 +9658,8 @@ static void d3d12_command_list_execute_indirect_state_template(
VkDeviceSize preprocess_size;
VkPipeline current_pipeline;
VkMemoryBarrier barrier;
bool require_ibo_update;
bool require_patch;
unsigned int i;
HRESULT hr;
@ -9672,6 +9675,8 @@ static void d3d12_command_list_execute_indirect_state_template(
current_pipeline = list->current_pipeline;
memset(&patch_args, 0, sizeof(patch_args));
patch_args.debug_tag = 0; /* Modify to non-zero value as desired when debugging. */
if (FAILED(hr = d3d12_command_signature_allocate_preprocess_memory_for_list(
list, signature, current_pipeline,
max_command_count, &preprocess_allocation, &preprocess_size)))
@ -9680,87 +9685,120 @@ static void d3d12_command_list_execute_indirect_state_template(
return;
}
if (FAILED(hr = d3d12_command_signature_allocate_stream_memory_for_list(
list, signature, max_command_count, &stream_allocation)))
/* If everything regarding alignment works out, we can just reuse the app indirect buffer instead. */
require_ibo_update = false;
require_patch = false;
/* Bind IBO. If we always update the IBO indirectly, do not validate the index buffer here.
* We can render fine even with a NULL IBO bound. */
for (i = 0; i < signature->desc.NumArgumentDescs; i++)
{
WARN("Failed to allocate stream memory.\n");
return;
if (signature->desc.pArgumentDescs[i].Type == D3D12_INDIRECT_ARGUMENT_TYPE_INDEX_BUFFER_VIEW)
{
require_ibo_update = true;
break;
}
}
if (count_buffer)
/* - Stride can mismatch, i.e. we need internal alignment of arguments.
* - Min required alignment on the indirect buffer itself might be too strict.
* - Min required alignment on count buffer might be too strict.
* - We require debugging.
* - Temporary: IBO type rewrite is required. TODO: Use index type LUT feature. */
props = &list->device->device_info.device_generated_commands_properties_nv;
if ((signature->state_template.stride != signature->desc.ByteStride && max_command_count > 1) ||
(arg_buffer_offset & (props->minIndirectCommandsBufferOffsetAlignment - 1)) ||
(count_buffer && (count_buffer_offset & (props->minSequencesCountBufferOffsetAlignment - 1))) ||
patch_args.debug_tag ||
require_ibo_update)
{
if (FAILED(hr = d3d12_command_allocator_allocate_scratch_memory(list->allocator,
VKD3D_SCRATCH_POOL_KIND_DEVICE_STORAGE,
sizeof(uint32_t),
list->device->device_info.device_generated_commands_properties_nv.minSequencesCountBufferOffsetAlignment,
~0u, &count_allocation)))
require_patch = true;
}
if (require_patch)
{
if (FAILED(hr = d3d12_command_signature_allocate_stream_memory_for_list(
list, signature, max_command_count, &stream_allocation)))
{
WARN("Failed to allocate count memory.\n");
WARN("Failed to allocate stream memory.\n");
return;
}
}
patch_args.template_va = signature->state_template.buffer_va;
patch_args.api_buffer_va = d3d12_resource_get_va(arg_buffer, arg_buffer_offset);
patch_args.device_generated_commands_va = stream_allocation.va;
patch_args.indirect_count_va = count_buffer ? d3d12_resource_get_va(count_buffer, count_buffer_offset) : 0;
patch_args.dst_indirect_count_va = count_buffer ? count_allocation.va : 0;
patch_args.api_buffer_word_stride = signature->desc.ByteStride / sizeof(uint32_t);
patch_args.device_generated_commands_word_stride = signature->state_template.stride / sizeof(uint32_t);
patch_args.debug_tag = 0; /* Modify to non-zero value as desired when debugging. */
if (patch_args.debug_tag != 0)
{
/* Makes log easier to understand since a sorted log will appear in-order. */
static uint32_t vkd3d_implicit_instance_count;
patch_args.implicit_instance = vkd3d_atomic_uint32_increment(
&vkd3d_implicit_instance_count, vkd3d_memory_order_relaxed) - 1;
}
barrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
barrier.pNext = NULL;
barrier.srcAccessMask = 0;
barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
if (!list->execute_indirect.has_observed_transition_to_indirect)
{
/* Fast path, throw the template resolve to the init command buffer. */
d3d12_command_allocator_allocate_init_command_buffer(list->allocator, list);
vk_patch_cmd_buffer = list->vk_init_commands;
if (!list->execute_indirect.has_emitted_indirect_to_compute_barrier)
if (count_buffer)
{
if (FAILED(hr = d3d12_command_allocator_allocate_scratch_memory(list->allocator,
VKD3D_SCRATCH_POOL_KIND_DEVICE_STORAGE,
sizeof(uint32_t),
props->minSequencesCountBufferOffsetAlignment,
~0u, &count_allocation)))
{
WARN("Failed to allocate count memory.\n");
return;
}
}
patch_args.template_va = signature->state_template.buffer_va;
patch_args.api_buffer_va = d3d12_resource_get_va(arg_buffer, arg_buffer_offset);
patch_args.device_generated_commands_va = stream_allocation.va;
patch_args.indirect_count_va = count_buffer ? d3d12_resource_get_va(count_buffer, count_buffer_offset) : 0;
patch_args.dst_indirect_count_va = count_buffer ? count_allocation.va : 0;
patch_args.api_buffer_word_stride = signature->desc.ByteStride / sizeof(uint32_t);
patch_args.device_generated_commands_word_stride = signature->state_template.stride / sizeof(uint32_t);
if (patch_args.debug_tag != 0)
{
/* Makes log easier to understand since a sorted log will appear in-order. */
static uint32_t vkd3d_implicit_instance_count;
patch_args.implicit_instance = vkd3d_atomic_uint32_increment(
&vkd3d_implicit_instance_count, vkd3d_memory_order_relaxed) - 1;
}
barrier.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER;
barrier.pNext = NULL;
barrier.srcAccessMask = 0;
barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
if (!list->execute_indirect.has_observed_transition_to_indirect)
{
/* Fast path, throw the template resolve to the init command buffer. */
d3d12_command_allocator_allocate_init_command_buffer(list->allocator, list);
vk_patch_cmd_buffer = list->vk_init_commands;
if (!list->execute_indirect.has_emitted_indirect_to_compute_barrier)
{
VK_CALL(vkCmdPipelineBarrier(vk_patch_cmd_buffer, VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 1, &barrier, 0, NULL, 0, NULL));
list->execute_indirect.has_emitted_indirect_to_compute_barrier = true;
}
}
else
{
vk_patch_cmd_buffer = list->vk_command_buffer;
d3d12_command_list_end_current_render_pass(list, true);
VK_CALL(vkCmdPipelineBarrier(vk_patch_cmd_buffer, VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 1, &barrier, 0, NULL, 0, NULL));
list->execute_indirect.has_emitted_indirect_to_compute_barrier = true;
d3d12_command_list_invalidate_current_pipeline(list, true);
}
}
else
{
vk_patch_cmd_buffer = list->vk_command_buffer;
d3d12_command_list_end_current_render_pass(list, true);
VK_CALL(vkCmdPipelineBarrier(vk_patch_cmd_buffer, VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 1, &barrier, 0, NULL, 0, NULL));
d3d12_command_list_invalidate_current_pipeline(list, true);
}
VK_CALL(vkCmdPushConstants(vk_patch_cmd_buffer, signature->state_template.pipeline.vk_pipeline_layout,
VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(patch_args), &patch_args));
VK_CALL(vkCmdBindPipeline(vk_patch_cmd_buffer, VK_PIPELINE_BIND_POINT_COMPUTE,
signature->state_template.pipeline.vk_pipeline));
VK_CALL(vkCmdPushConstants(vk_patch_cmd_buffer, signature->state_template.pipeline.vk_pipeline_layout,
VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(patch_args), &patch_args));
VK_CALL(vkCmdBindPipeline(vk_patch_cmd_buffer, VK_PIPELINE_BIND_POINT_COMPUTE,
signature->state_template.pipeline.vk_pipeline));
/* One workgroup processes the patching for one draw. We could potentially use indirect dispatch
* to restrict the patching work to just the indirect count, but meh, just more barriers.
* We'll nop out the workgroup early based on direct count, and the number of threads should be trivial either way. */
VK_CALL(vkCmdDispatch(vk_patch_cmd_buffer, max_command_count, 1, 1));
/* One workgroup processes the patching for one draw. We could potentially use indirect dispatch
* to restrict the patching work to just the indirect count, but meh, just more barriers.
* We'll nop out the workgroup early based on direct count, and the number of threads should be trivial either way. */
VK_CALL(vkCmdDispatch(vk_patch_cmd_buffer, max_command_count, 1, 1));
if (vk_patch_cmd_buffer == list->vk_command_buffer)
{
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
VK_CALL(vkCmdPipelineBarrier(vk_patch_cmd_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
0, 1, &barrier, 0, NULL, 0, NULL));
/* The barrier is deferred if we moved the dispatch to init command buffer. */
if (vk_patch_cmd_buffer == list->vk_command_buffer)
{
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
VK_CALL(vkCmdPipelineBarrier(vk_patch_cmd_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT,
0, 1, &barrier, 0, NULL, 0, NULL));
/* The barrier is deferred if we moved the dispatch to init command buffer. */
}
}
if (!d3d12_command_list_begin_render_pass(list))
@ -9769,13 +9807,7 @@ static void d3d12_command_list_execute_indirect_state_template(
return;
}
/* Bind IBO. If we always update the IBO indirectly, do not validate the index buffer here.
* We can render fine even with a NULL IBO bound. */
for (i = 0; i < signature->desc.NumArgumentDescs; i++)
if (signature->desc.pArgumentDescs[i].Type == D3D12_INDIRECT_ARGUMENT_TYPE_INDEX_BUFFER_VIEW)
break;
if (i == signature->desc.NumArgumentDescs &&
if (!require_ibo_update &&
signature->desc.pArgumentDescs[signature->desc.NumArgumentDescs - 1].Type ==
D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED &&
!d3d12_command_list_update_index_buffer(list))
@ -9799,8 +9831,16 @@ static void d3d12_command_list_execute_indirect_state_template(
if (count_buffer)
{
generated.sequencesCountBuffer = count_allocation.buffer;
generated.sequencesCountOffset = count_allocation.offset;
if (require_patch)
{
generated.sequencesCountBuffer = count_allocation.buffer;
generated.sequencesCountOffset = count_allocation.offset;
}
else
{
generated.sequencesCountBuffer = count_buffer->res.vk_buffer;
generated.sequencesCountOffset = count_buffer->mem.offset + count_buffer_offset;
}
}
else
{
@ -9808,8 +9848,19 @@ static void d3d12_command_list_execute_indirect_state_template(
generated.sequencesCountOffset = 0;
}
stream.buffer = stream_allocation.buffer;
stream.offset = stream_allocation.offset;
if (require_patch)
{
stream.buffer = stream_allocation.buffer;
stream.offset = stream_allocation.offset;
}
else
{
stream.buffer = arg_buffer->res.vk_buffer;
stream.offset = arg_buffer->mem.offset + arg_buffer_offset;
}
if (require_patch)
WARN("Template requires patching :(\n");
VK_CALL(vkCmdExecuteGeneratedCommandsNV(list->vk_command_buffer, VK_FALSE, &generated));
@ -13156,6 +13207,7 @@ static HRESULT d3d12_command_signature_init_state_template(struct d3d12_command_
required_stride_alignment = max(required_stride_alignment, required_alignment);
}
stream_stride = max(stream_stride, desc->ByteStride);
stream_stride = align(stream_stride, required_stride_alignment);
if (FAILED(hr = d3d12_command_signature_init_patch_commands_buffer(signature, device, patch_commands, patch_commands_count)))