Skip to content

WIP: Fence #73

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

Open
3Shain wants to merge 7 commits into
base: main
Choose a base branch
from
Open

WIP: Fence #73

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
193561b
refactor(dxmt): remove PBF based synchronization
3Shain Dec 15, 2025
a4d29a3
refactor(dxmt, d3d11): collect pre/post-raster access information
3Shain Dec 16, 2025
22b42a4
refactor(dxmt): allow present encoder to wait/update multiple fences
3Shain Dec 16, 2025
d0ac37b
feat(dxmt): implement fence-based synchronization
3Shain Dec 16, 2025
6a3d9fe
feat(dxmt): simplify fences to wait if indirect dependency can be proved
3Shain Dec 16, 2025
cc00468
feat(dxmt): track buffer access at sub-allocaiton level
3Shain Dec 17, 2025
a9300cf
feat(dxmt): re-implement encoder reorder & coalesc, base on fence inf…
3Shain Dec 17, 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
16 changes: 8 additions & 8 deletions src/d3d11/d3d11_context_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1563,7 +1563,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([IndexType, IndexBufferOffset, Primitive, ArgBuffer = bindable->buffer(),
AlignedByteOffsetForArgs](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
enc.bumpVisibilityResultOffset();
auto [index_buffer, index_sub_offset] = enc.currentIndexBuffer();
auto &cmd = enc.encodeRenderCommand<wmtcmd_render_draw_indexed_indirect>();
Expand Down Expand Up @@ -1598,7 +1598,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
}
if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([Primitive, ArgBuffer = bindable->buffer(), AlignedByteOffsetForArgs](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
enc.bumpVisibilityResultOffset();
auto &cmd = enc.encodeRenderCommand<wmtcmd_render_draw_indirect>();
cmd.type = WMTRenderCommandDrawIndirect;
Expand All @@ -1616,7 +1616,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
auto max_object_threadgroups = max_object_threadgroups_;
if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([=, topo = state_.InputAssembler.Topology, ArgBuffer = bindable->buffer()](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto dispatch_arg = enc.allocateTempBuffer1(sizeof(DXMT_DISPATCH_ARGUMENTS), 4);

auto [vertex_per_warp, vertex_increment_per_wrap] = get_gs_vertex_count(topo);
Expand Down Expand Up @@ -1647,7 +1647,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p

if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([=, topo = state_.InputAssembler.Topology, ArgBuffer = bindable->buffer()](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto dispatch_arg = enc.allocateTempBuffer1(sizeof(DXMT_DISPATCH_ARGUMENTS), 4);

auto [vertex_per_warp, vertex_increment_per_wrap] = get_gs_vertex_count(topo);
Expand Down Expand Up @@ -1679,7 +1679,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
auto max_object_threadgroups = max_object_threadgroups_;
if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([=, topo = state_.InputAssembler.Topology, ArgBuffer = bindable->buffer()](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto dispatch_arg = enc.allocateTempBuffer1(sizeof(DXMT_DISPATCH_ARGUMENTS), 4);

auto PatchPerGroup = 32 / enc.tess_threads_per_patch;
Expand Down Expand Up @@ -1713,7 +1713,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p

if (auto bindable = reinterpret_cast<D3D11ResourceCommon *>(pBufferForArgs)) {
EmitOP([=, topo = state_.InputAssembler.Topology, ArgBuffer = bindable->buffer()](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto [buffer, buffer_offset] = enc.access<true>(ArgBuffer, AlignedByteOffsetForArgs, 20, DXMT_ENCODER_RESOURCE_ACESS_READ);
auto dispatch_arg = enc.allocateTempBuffer1(sizeof(DXMT_DISPATCH_ARGUMENTS), 4);

auto PatchPerGroup = 32 / enc.tess_threads_per_patch;
Expand Down Expand Up @@ -4679,7 +4679,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
auto &so_slot0 = state_.StreamOutput.Targets[0];
if (so_slot0.Offset == 0xFFFFFFFF) {
EmitST([slot0 = so_slot0.Buffer->buffer()](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(slot0, 0, slot0->length(), DXMT_ENCODER_RESOURCE_ACESS_WRITE);
auto [buffer, buffer_offset] = enc.access<true>(slot0, 0, slot0->length(), DXMT_ENCODER_RESOURCE_ACESS_WRITE);
auto &cmd = enc.encodeRenderCommand<wmtcmd_render_setbuffer>();
cmd.type = WMTRenderCommandSetVertexBuffer;
cmd.buffer = buffer->buffer();;
Expand All @@ -4690,7 +4690,7 @@ template <typename ContextInternalState> class MTLD3D11DeviceContextImplBase : p
});
} else {
EmitST([slot0 = so_slot0.Buffer->buffer(), offset = so_slot0.Offset](ArgumentEncodingContext &enc) {
auto [buffer, buffer_offset] = enc.access(slot0, 0, slot0->length(), DXMT_ENCODER_RESOURCE_ACESS_WRITE);
auto [buffer, buffer_offset] = enc.access<true>(slot0, 0, slot0->length(), DXMT_ENCODER_RESOURCE_ACESS_WRITE);
auto &cmd = enc.encodeRenderCommand<wmtcmd_render_setbuffer>();
cmd.type = WMTRenderCommandSetVertexBuffer;
cmd.buffer = buffer->buffer();;
Expand Down
10 changes: 5 additions & 5 deletions src/dxmt/dxmt_buffer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,14 +22,17 @@ BufferAllocation::BufferAllocation(WMT::Device device, const WMTBufferInfo &info
suballocation_count_ = DXMT_PAGE_SIZE / suballocation_size_;
info_.length = DXMT_PAGE_SIZE;
}
fenceTrackers.reserve(suballocation_count_);
for (auto i = 0u; i < suballocation_count_; i++) {
fenceTrackers.push_back({});
}
if (flags_.test(BufferAllocationFlag::CpuPlaced)) {
placed_buffer = wsi::aligned_malloc(info_.length, DXMT_PAGE_SIZE);
info_.memory.set(placed_buffer);
}
obj_ = device.newBuffer(info_);
gpuAddress_ = info_.gpu_address;
mappedMemory_ = info_.memory.get_accessible_or_null();
depkey = EncoderDepSet::generateNewKey(global_buffer_seq.fetch_add(1));
};

BufferAllocation::~BufferAllocation() {
Expand Down Expand Up @@ -129,10 +132,7 @@ Buffer::createView(BufferViewDescriptor const &descriptor) {

Rc<BufferAllocation>
Buffer::allocate(Flags<BufferAllocationFlag> flags) {
WMTResourceOptions options = WMTResourceStorageModeShared;
if (flags.test(BufferAllocationFlag::GpuReadonly)) {
options |= WMTResourceHazardTrackingModeUntracked;
}
WMTResourceOptions options = WMTResourceHazardTrackingModeUntracked;
if (flags.test(BufferAllocationFlag::CpuWriteCombined)) {
options |= WMTResourceOptionCPUCacheModeWriteCombined;
}
Expand Down
6 changes: 5 additions & 1 deletion src/dxmt/dxmt_buffer.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,6 +88,10 @@ class BufferAllocation : public Allocation {
return current_suballocation_ * stride;
}

uint32_t currentSuballocation() {
return current_suballocation_;
}

void
updateContents(uint64_t offset, const void *data, uint64_t length, uint32_t suballocation = 0) noexcept {
if (likely(mappedMemory_ != nullptr && !flags_.test(BufferAllocationFlag::GpuManaged))) {
Expand All @@ -98,7 +102,7 @@ class BufferAllocation : public Allocation {
}

DXMT_RESOURCE_RESIDENCY_STATE residencyState;
EncoderDepKey depkey;
std::vector<GenericAccessTracker> fenceTrackers;

private:
BufferAllocation(WMT::Device device, const WMTBufferInfo &info, Flags<BufferAllocationFlag> flags);
Expand Down
Loading
Loading