Skip to content

Commit

Permalink
[D3D12] Replace unused shared memory view with a null view
Browse files Browse the repository at this point in the history 
Fixes the PIX validation warning about missing resource states on every guest draw. Also potentially prevents drivers from making assumptions about the shared memory buffer based on the bindings, though no such cases are currently known.
  • Loading branch information
@Triang3l
Triang3l committed Oct 23, 2022
1 parent 5fde7c6 commit 4add1f6
Show file tree
Hide file tree
Showing 2 changed files with 141 additions and 56 deletions.
155 changes: 109 additions & 46 deletions src/xenia/gpu/d3d12/d3d12_command_processor.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1000,6 +1000,36 @@ bool D3D12CommandProcessor::SetupContext() {
parameter.Descriptor.RegisterSpace = 0;
parameter.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
}
// Shared memory SRV and UAV.
D3D12_DESCRIPTOR_RANGE root_shared_memory_view_ranges[2];
{
auto& parameter =
root_parameters_bindless[kRootParameter_Bindless_SharedMemory];
parameter.ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
parameter.DescriptorTable.NumDescriptorRanges =
uint32_t(xe::countof(root_shared_memory_view_ranges));
parameter.DescriptorTable.pDescriptorRanges =
root_shared_memory_view_ranges;
parameter.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
{
auto& range = root_shared_memory_view_ranges[0];
range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
range.NumDescriptors = 1;
range.BaseShaderRegister =
UINT(DxbcShaderTranslator::SRVMainRegister::kSharedMemory);
range.RegisterSpace = UINT(DxbcShaderTranslator::SRVSpace::kMain);
range.OffsetInDescriptorsFromTableStart = 0;
}
{
auto& range = root_shared_memory_view_ranges[1];
range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
range.NumDescriptors = 1;
range.BaseShaderRegister =
UINT(DxbcShaderTranslator::UAVRegister::kSharedMemory);
range.RegisterSpace = 0;
range.OffsetInDescriptorsFromTableStart = 1;
}
}
// Sampler heap.
D3D12_DESCRIPTOR_RANGE root_bindless_sampler_range;
{
Expand All @@ -1019,7 +1049,7 @@ bool D3D12CommandProcessor::SetupContext() {
root_bindless_sampler_range.OffsetInDescriptorsFromTableStart = 0;
}
// View heap.
D3D12_DESCRIPTOR_RANGE root_bindless_view_ranges[6];
D3D12_DESCRIPTOR_RANGE root_bindless_view_ranges[4];
{
auto& parameter =
root_parameters_bindless[kRootParameter_Bindless_ViewHeap];
Expand All @@ -1028,34 +1058,6 @@ bool D3D12CommandProcessor::SetupContext() {
parameter.DescriptorTable.NumDescriptorRanges = 0;
parameter.DescriptorTable.pDescriptorRanges = root_bindless_view_ranges;
parameter.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
// Shared memory SRV.
{
assert_true(parameter.DescriptorTable.NumDescriptorRanges <
xe::countof(root_bindless_view_ranges));
auto& range = root_bindless_view_ranges[parameter.DescriptorTable
.NumDescriptorRanges++];
range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
range.NumDescriptors = 1;
range.BaseShaderRegister =
UINT(DxbcShaderTranslator::SRVMainRegister::kSharedMemory);
range.RegisterSpace = UINT(DxbcShaderTranslator::SRVSpace::kMain);
range.OffsetInDescriptorsFromTableStart =
UINT(SystemBindlessView::kSharedMemoryRawSRV);
}
// Shared memory UAV.
{
assert_true(parameter.DescriptorTable.NumDescriptorRanges <
xe::countof(root_bindless_view_ranges));
auto& range = root_bindless_view_ranges[parameter.DescriptorTable
.NumDescriptorRanges++];
range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
range.NumDescriptors = 1;
range.BaseShaderRegister =
UINT(DxbcShaderTranslator::UAVRegister::kSharedMemory);
range.RegisterSpace = 0;
range.OffsetInDescriptorsFromTableStart =
UINT(SystemBindlessView::kSharedMemoryRawUAV);
}
// EDRAM.
if (render_target_cache_->GetPath() ==
RenderTargetCache::Path::kPixelShaderInterlock) {
Expand Down Expand Up @@ -1418,6 +1420,20 @@ bool D3D12CommandProcessor::SetupContext() {
if (bindless_resources_used_) {
// Create the system bindless descriptors once all resources are
// initialized.
// kNullRawSRV.
ui::d3d12::util::CreateBufferRawSRV(
device,
provider.OffsetViewDescriptor(
view_bindless_heap_cpu_start_,
uint32_t(SystemBindlessView::kNullRawSRV)),
nullptr, 0);
// kNullRawUAV.
ui::d3d12::util::CreateBufferRawUAV(
device,
provider.OffsetViewDescriptor(
view_bindless_heap_cpu_start_,
uint32_t(SystemBindlessView::kNullRawUAV)),
nullptr, 0);
// kNullTexture2DArray.
D3D12_SHADER_RESOURCE_VIEW_DESC null_srv_desc;
null_srv_desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
Expand Down Expand Up @@ -2272,7 +2288,8 @@ bool D3D12CommandProcessor::IssueDraw(xenos::PrimitiveType primitive_type,
used_texture_mask, normalized_depth_control, normalized_color_mask);

// Update constant buffers, descriptors and root parameters.
if (!UpdateBindings(vertex_shader, pixel_shader, root_signature)) {
if (!UpdateBindings(vertex_shader, pixel_shader, root_signature,
memexport_used)) {
return false;
}
// Must not call anything that can change the descriptor heap from now on!
Expand Down Expand Up @@ -2890,6 +2907,7 @@ bool D3D12CommandProcessor::BeginSubmission(bool is_guest_command) {
cbuffer_binding_float_pixel_.up_to_date = false;
cbuffer_binding_bool_loop_.up_to_date = false;
cbuffer_binding_fetch_.up_to_date = false;
current_shared_memory_binding_is_uav_.reset();
if (bindless_resources_used_) {
cbuffer_binding_descriptor_indices_vertex_.up_to_date = false;
cbuffer_binding_descriptor_indices_pixel_.up_to_date = false;
Expand Down Expand Up @@ -3649,9 +3667,10 @@ void D3D12CommandProcessor::UpdateSystemConstantValues(
cbuffer_binding_system_.up_to_date &= !dirty;
}

bool D3D12CommandProcessor::UpdateBindings(
const D3D12Shader* vertex_shader, const D3D12Shader* pixel_shader,
ID3D12RootSignature* root_signature) {
bool D3D12CommandProcessor::UpdateBindings(const D3D12Shader* vertex_shader,
const D3D12Shader* pixel_shader,
ID3D12RootSignature* root_signature,
bool shared_memory_is_uav) {
const ui::d3d12::D3D12Provider& provider = GetD3D12Provider();
ID3D12Device* device = provider.GetDevice();
const RegisterFile& regs = *register_file_;
Expand Down Expand Up @@ -3688,6 +3707,9 @@ bool D3D12CommandProcessor::UpdateBindings(
uint32_t root_parameter_bool_loop_constants =
bindless_resources_used_ ? kRootParameter_Bindless_BoolLoopConstants
: kRootParameter_Bindful_BoolLoopConstants;
uint32_t root_parameter_shared_memory_and_bindful_edram =
bindless_resources_used_ ? kRootParameter_Bindless_SharedMemory
: kRootParameter_Bindful_SharedMemoryAndEdram;

//
// Update root constant buffers that are common for bindful and bindless.
Expand Down Expand Up @@ -3852,6 +3874,13 @@ bool D3D12CommandProcessor::UpdateBindings(
// Update descriptors.
//

if (!current_shared_memory_binding_is_uav_.has_value() ||
current_shared_memory_binding_is_uav_.value() != shared_memory_is_uav) {
current_shared_memory_binding_is_uav_ = shared_memory_is_uav;
current_graphics_root_up_to_date_ &=
~(1u << root_parameter_shared_memory_and_bindful_edram);
}

// Get textures and samplers used by the vertex shader, check if the last used
// samplers are compatible and update them.
size_t texture_layout_uid_vertex =
Expand Down Expand Up @@ -4180,12 +4209,14 @@ bool D3D12CommandProcessor::UpdateBindings(
if (write_textures_pixel) {
view_count_partial_update += texture_count_pixel;
}
// All the constants + shared memory SRV and UAV + textures.
// Shared memory SRV and null UAV + null SRV and shared memory UAV +
// textures.
size_t view_count_full_update =
2 + texture_count_vertex + texture_count_pixel;
4 + texture_count_vertex + texture_count_pixel;
if (edram_rov_used) {
// + EDRAM UAV.
++view_count_full_update;
// + EDRAM UAV in two tables (with the shared memory SRV and with the
// shared memory UAV).
view_count_full_update += 2;
}
D3D12_CPU_DESCRIPTOR_HANDLE view_cpu_handle;
D3D12_GPU_DESCRIPTOR_HANDLE view_gpu_handle;
Expand Down Expand Up @@ -4230,10 +4261,25 @@ bool D3D12CommandProcessor::UpdateBindings(
bindful_textures_written_pixel_ = false;
// If updating fully, write the shared memory SRV and UAV descriptors and,
// if needed, the EDRAM descriptor.
gpu_handle_shared_memory_and_edram_ = view_gpu_handle;
// SRV + null UAV + EDRAM.
gpu_handle_shared_memory_srv_and_edram_ = view_gpu_handle;
shared_memory_->WriteRawSRVDescriptor(view_cpu_handle);
view_cpu_handle.ptr += descriptor_size_view;
view_gpu_handle.ptr += descriptor_size_view;
ui::d3d12::util::CreateBufferRawUAV(device, view_cpu_handle, nullptr, 0);
view_cpu_handle.ptr += descriptor_size_view;
view_gpu_handle.ptr += descriptor_size_view;
if (edram_rov_used) {
render_target_cache_->WriteEdramUintPow2UAVDescriptor(view_cpu_handle,
2);
view_cpu_handle.ptr += descriptor_size_view;
view_gpu_handle.ptr += descriptor_size_view;
}
// Null SRV + UAV + EDRAM.
gpu_handle_shared_memory_uav_and_edram_ = view_gpu_handle;
ui::d3d12::util::CreateBufferRawSRV(device, view_cpu_handle, nullptr, 0);
view_cpu_handle.ptr += descriptor_size_view;
view_gpu_handle.ptr += descriptor_size_view;
shared_memory_->WriteRawUAVDescriptor(view_cpu_handle);
view_cpu_handle.ptr += descriptor_size_view;
view_gpu_handle.ptr += descriptor_size_view;
Expand Down Expand Up @@ -4372,6 +4418,31 @@ bool D3D12CommandProcessor::UpdateBindings(
current_graphics_root_up_to_date_ |= 1u
<< root_parameter_bool_loop_constants;
}
if (!(current_graphics_root_up_to_date_ &
(1u << root_parameter_shared_memory_and_bindful_edram))) {
assert_true(current_shared_memory_binding_is_uav_.has_value());
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_shared_memory_and_bindful_edram;
if (bindless_resources_used_) {
gpu_handle_shared_memory_and_bindful_edram =
provider.OffsetViewDescriptor(
view_bindless_heap_gpu_start_,
uint32_t(current_shared_memory_binding_is_uav_.value()
? SystemBindlessView ::
kNullRawSRVAndSharedMemoryRawUAVStart
: SystemBindlessView ::
kSharedMemoryRawSRVAndNullRawUAVStart));
} else {
gpu_handle_shared_memory_and_bindful_edram =
current_shared_memory_binding_is_uav_.value()
? gpu_handle_shared_memory_uav_and_edram_
: gpu_handle_shared_memory_srv_and_edram_;
}
deferred_command_list_.D3DSetGraphicsRootDescriptorTable(
root_parameter_shared_memory_and_bindful_edram,
gpu_handle_shared_memory_and_bindful_edram);
current_graphics_root_up_to_date_ |=
1u << root_parameter_shared_memory_and_bindful_edram;
}
if (bindless_resources_used_) {
if (!(current_graphics_root_up_to_date_ &
(1u << kRootParameter_Bindless_DescriptorIndicesPixel))) {
Expand Down Expand Up @@ -4405,14 +4476,6 @@ bool D3D12CommandProcessor::UpdateBindings(
<< kRootParameter_Bindless_ViewHeap;
}
} else {
if (!(current_graphics_root_up_to_date_ &
(1u << kRootParameter_Bindful_SharedMemoryAndEdram))) {
deferred_command_list_.D3DSetGraphicsRootDescriptorTable(
kRootParameter_Bindful_SharedMemoryAndEdram,
gpu_handle_shared_memory_and_edram_);
current_graphics_root_up_to_date_ |=
1u << kRootParameter_Bindful_SharedMemoryAndEdram;
}
uint32_t extra_index;
extra_index = current_graphics_root_bindful_extras_.textures_pixel;
if (extra_index != RootBindfulExtraParameterIndices::kUnavailable &&
Expand Down
42 changes: 32 additions & 10 deletions src/xenia/gpu/d3d12/d3d12_command_processor.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@
#include <atomic>
#include <deque>
#include <memory>
#include <optional>
#include <string>
#include <unordered_map>
#include <utility>
Expand Down Expand Up @@ -128,11 +129,19 @@ class D3D12CommandProcessor : public CommandProcessor {
uint32_t count, ui::d3d12::util::DescriptorCpuGpuHandlePair* handles_out);
// These are needed often, so they are always allocated.
enum class SystemBindlessView : uint32_t {
kSharedMemoryRawSRV,
// Both may be bound as one root parameter.
kSharedMemoryRawSRVAndNullRawUAVStart,
kSharedMemoryRawSRV = kSharedMemoryRawSRVAndNullRawUAVStart,
kNullRawUAV,

// Both may be bound as one root parameter.
kNullRawSRVAndSharedMemoryRawUAVStart,
kNullRawSRV = kNullRawSRVAndSharedMemoryRawUAVStart,
kSharedMemoryRawUAV,

kSharedMemoryR32UintSRV,
kSharedMemoryR32G32UintSRV,
kSharedMemoryR32G32B32A32UintSRV,
kSharedMemoryRawUAV,
kSharedMemoryR32UintUAV,
kSharedMemoryR32G32UintUAV,
kSharedMemoryR32G32B32A32UintUAV,
Expand Down Expand Up @@ -253,10 +262,10 @@ class D3D12CommandProcessor : public CommandProcessor {
kRootParameter_Bindful_SystemConstants, // +2 = 6 in all.
// Pretty rarely used and rarely changed - flow control constants.
kRootParameter_Bindful_BoolLoopConstants, // +2 = 8 in all.
// Never changed except for when starting a new descriptor heap - shared
// memory byte address buffer, and, if ROV is used for EDRAM, EDRAM R32_UINT
// UAV.
// SRV/UAV descriptor table.
// Changed only when starting a new descriptor heap or when switching
// between shared memory as SRV and UAV - shared memory byte address buffer
// (as SRV and as UAV, either may be null if not used), and, if ROV is used
// for EDRAM, EDRAM R32_UINT UAV.
kRootParameter_Bindful_SharedMemoryAndEdram, // +1 = 9 in all.

kRootParameter_Bindful_Count_Base,
Expand All @@ -280,10 +289,14 @@ class D3D12CommandProcessor : public CommandProcessor {
kRootParameter_Bindless_DescriptorIndicesVertex, // +2 = 6 in VS.
kRootParameter_Bindless_SystemConstants, // +2 = 8 in all.
kRootParameter_Bindless_BoolLoopConstants, // +2 = 10 in all.
// Changed only when switching between shared memory as SRV and UAV - shared
// memory byte address buffer (as SRV and as UAV, either may be null if not
// used).
kRootParameter_Bindless_SharedMemory, // +1 = 11 in all.
// Unbounded sampler descriptor table - changed in case of overflow.
kRootParameter_Bindless_SamplerHeap, // +1 = 11 in all.
kRootParameter_Bindless_SamplerHeap, // +1 = 12 in all.
// Unbounded SRV/UAV descriptor table - never changed.
kRootParameter_Bindless_ViewHeap, // +1 = 12 in all.
kRootParameter_Bindless_ViewHeap, // +1 = 13 in all.

kRootParameter_Bindless_Count,
};
Expand Down Expand Up @@ -362,7 +375,8 @@ class D3D12CommandProcessor : public CommandProcessor {
uint32_t normalized_color_mask);
bool UpdateBindings(const D3D12Shader* vertex_shader,
const D3D12Shader* pixel_shader,
ID3D12RootSignature* root_signature);
ID3D12RootSignature* root_signature,
bool shared_memory_is_uav);

// Returns dword count for one element for a memexport format, or 0 if it's
// not supported by the D3D12 command processor (if it's smaller that 1 dword,
Expand Down Expand Up @@ -622,6 +636,13 @@ class D3D12CommandProcessor : public CommandProcessor {
ConstantBufferBinding cbuffer_binding_descriptor_indices_vertex_;
ConstantBufferBinding cbuffer_binding_descriptor_indices_pixel_;

// Whether the latest shared memory and EDRAM buffer binding contains the
// shared memory UAV rather than the SRV.
// Separate descriptor tables for the SRV and the UAV, even though only one is
// accessed dynamically in the shaders, are used to prevent a validation
// message about missing resource states in PIX.
std::optional<bool> current_shared_memory_binding_is_uav_;

// Pages with the descriptors currently used for handling Xenos draw calls.
uint64_t draw_view_bindful_heap_index_;
uint64_t draw_sampler_bindful_heap_index_;
Expand Down Expand Up @@ -654,7 +675,8 @@ class D3D12CommandProcessor : public CommandProcessor {
std::vector<uint32_t> current_sampler_bindless_indices_pixel_;

// Latest bindful descriptor handles used for handling Xenos draw calls.
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_shared_memory_and_edram_;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_shared_memory_srv_and_edram_;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_shared_memory_uav_and_edram_;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_textures_vertex_;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_textures_pixel_;
D3D12_GPU_DESCRIPTOR_HANDLE gpu_handle_samplers_vertex_;
Expand Down

0 comments on commit 4add1f6

Please sign in to comment.