Skip to content

Commit

Permalink
[d3d9] Upload DYNAMIC+SYSMEM vertex and index data for each draw
Browse files Browse the repository at this point in the history
  • Loading branch information
@K0bin
K0bin committed Jan 23, 2024
1 parent 13b63a0 commit f067c06
Show file tree
Hide file tree
Showing 6 changed files with 212 additions and 19 deletions.
4 changes: 4 additions & 0 deletions src/d3d9/d3d9_common_buffer.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -211,6 +211,10 @@ namespace dxvk {
: DxvkCsThread::SynchronizeAll;
}

bool IsSysmemDynamic() const {
return m_desc.Pool == D3DPOOL_SYSTEMMEM && (m_desc.Usage & D3DUSAGE_DYNAMIC) != 0;
}

private:

Rc<DxvkBuffer> CreateBuffer() const;
Expand Down
193 changes: 183 additions & 10 deletions src/d3d9/d3d9_device.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2615,7 +2615,22 @@ namespace dxvk {
if (unlikely(!PrimitiveCount))
return S_OK;

PrepareDraw(PrimitiveType, true);
bool dynamicSysmemVBOs = true;
for (uint32_t i = 0; i < caps::MaxStreams && dynamicSysmemVBOs; i++) {
auto* vbo = GetCommonBuffer(m_state.vertexBuffers[i].vertexBuffer);
dynamicSysmemVBOs &= vbo == nullptr || vbo->IsSysmemDynamic();
}

PrepareDraw(PrimitiveType, !dynamicSysmemVBOs, false);

if (unlikely(dynamicSysmemVBOs)) {
// If this draw only uses D3DPOOL_SYSTEMMEM + D3DUSAGE_DYNAMIC buffers,
// we only upload the bits of data that this specific draw actually uses.
// Similar to DrawPrimitiveUp
uint32_t vertexCount = GetVertexCount(PrimitiveType, PrimitiveCount);
UploadDynamicSysmemBuffers(StartVertex, vertexCount, 0, 0);
StartVertex = 0;
}

EmitCs([this,
cPrimType = PrimitiveType,
Expand Down Expand Up @@ -2652,7 +2667,34 @@ namespace dxvk {
if (unlikely(!PrimitiveCount))
return S_OK;

PrepareDraw(PrimitiveType, true);
bool dynamicSysmemVBOs = true;
for (uint32_t i = 0; i < caps::MaxStreams && dynamicSysmemVBOs; i++) {
auto* vbo = GetCommonBuffer(m_state.vertexBuffers[i].vertexBuffer);
dynamicSysmemVBOs &= vbo == nullptr || vbo->IsSysmemDynamic();
}
D3D9CommonBuffer* ibo = GetCommonBuffer(m_state.indices);
bool dynamicSysmemIBO = ibo->IsSysmemDynamic();

PrepareDraw(PrimitiveType, !dynamicSysmemVBOs, !dynamicSysmemIBO);

if (unlikely(dynamicSysmemVBOs || dynamicSysmemIBO)) {
uint32_t vertexCount = GetVertexCount(PrimitiveType, PrimitiveCount);
UploadDynamicSysmemBuffers(
BaseVertexIndex + MinVertexIndex,
dynamicSysmemVBOs ? NumVertices : 0,
StartIndex,
dynamicSysmemIBO ? vertexCount : 0
);
if (dynamicSysmemVBOs) {
// If this draw only uses D3DPOOL_SYSTEMMEM + D3DUSAGE_DYNAMIC buffers,
// we only upload the bits of data that this specific draw actually uses.
// Similar to DrawPrimitiveUp
BaseVertexIndex = -MinVertexIndex;
}
if (dynamicSysmemIBO) {
StartIndex = 0;
}
}

EmitCs([this,
cPrimType = PrimitiveType,
Expand Down Expand Up @@ -2688,7 +2730,7 @@ namespace dxvk {
if (unlikely(!PrimitiveCount))
return S_OK;

PrepareDraw(PrimitiveType, false);
PrepareDraw(PrimitiveType, false, false);

uint32_t vertexCount = GetVertexCount(PrimitiveType, PrimitiveCount);

Expand Down Expand Up @@ -2740,7 +2782,7 @@ namespace dxvk {
if (unlikely(!PrimitiveCount))
return S_OK;

PrepareDraw(PrimitiveType, false);
PrepareDraw(PrimitiveType, false, false);

uint32_t vertexCount = GetVertexCount(PrimitiveType, PrimitiveCount);

Expand Down Expand Up @@ -2827,7 +2869,7 @@ namespace dxvk {
D3D9CommonBuffer* dst = static_cast<D3D9VertexBuffer*>(pDestBuffer)->GetCommonBuffer();
D3D9VertexDecl* decl = static_cast<D3D9VertexDecl*> (pVertexDecl);

PrepareDraw(D3DPT_FORCE_DWORD, true);
PrepareDraw(D3DPT_FORCE_DWORD, true, true);

if (decl == nullptr) {
DWORD FVF = dst->Desc()->FVF;
Expand All @@ -2842,7 +2884,7 @@ namespace dxvk {
decl = iter->second.ptr();
}

uint32_t offset = DestIndex * decl->GetSize();
uint32_t offset = DestIndex * decl->GetSize(0);

auto slice = dst->GetBufferSlice<D3D9_COMMON_BUFFER_TYPE_REAL>();
slice = slice.subSlice(offset, slice.length() - offset);
Expand Down Expand Up @@ -2889,7 +2931,7 @@ namespace dxvk {
}

if (dst->GetMapMode() == D3D9_COMMON_BUFFER_MAP_MODE_BUFFER) {
uint32_t copySize = VertexCount * decl->GetSize();
uint32_t copySize = VertexCount * decl->GetSize(0);

EmitCs([
cSrcBuffer = dst->GetBuffer<D3D9_COMMON_BUFFER_TYPE_REAL>(),
Expand Down Expand Up @@ -5091,6 +5133,124 @@ namespace dxvk {
return D3D_OK;
}



void D3D9DeviceEx::UploadDynamicSysmemBuffers(
UINT FirstVertexIndex,
UINT NumVertices,
UINT FirstIndex,
UINT NumIndices
) {
// The UP buffer allocator will invalidate,
// so we can only use 1 UP buffer slice per draw.
// First we calculate the size of that UP buffer slice
// and store all sizes and offsets into it.

uint32_t upBufferSize = 0;
std::array<uint32_t, caps::MaxStreams> vboUPBufferOffsets;
std::array<uint32_t, caps::MaxStreams> vboUPBufferSizes;
for (uint32_t i = 0; i < caps::MaxStreams && NumVertices != 0; i++) {
vboUPBufferOffsets[i] = upBufferSize;

auto* vbo = GetCommonBuffer(m_state.vertexBuffers[i].vertexBuffer);
if (likely(vbo == nullptr)) {
vboUPBufferSizes[i] = 0;
continue;
}
const uint32_t vertexStride = m_state.vertexDecl->GetSize(i);
uint32_t offset = FirstVertexIndex * vertexStride;
const uint32_t vertexBufferSize = vbo->Desc()->Size;
if (offset < vertexBufferSize) {
const uint32_t vertexDataSize = std::min(NumVertices * vertexStride, vertexBufferSize - offset);
vboUPBufferSizes[i] = vertexDataSize;
upBufferSize += vertexDataSize;
}
}

uint32_t iboUPBufferSize = 0;
uint32_t iboUPBufferOffset = 0;
if (NumIndices != 0) {
auto* ibo = GetCommonBuffer(m_state.indices);
if (likely(ibo != nullptr)) {
uint32_t indexStride = ibo->Desc()->Format == D3D9Format::INDEX16 ? 2 : 4;
VkIndexType indexType = DecodeIndexType(ibo->Desc()->Format);
uint32_t offset = indexStride * FirstIndex;
uint32_t indexBufferSize = ibo->Desc()->Size;
if (offset < indexBufferSize) {
iboUPBufferSize = std::min(NumIndices * indexStride, indexBufferSize - offset);
iboUPBufferOffset = upBufferSize;
upBufferSize += iboUPBufferSize;
}
}
}

if (upBufferSize == 0) {
return;
}

auto upSlice = AllocUPBuffer(upBufferSize);

// Now copy the actual data and bind it.
for (uint32_t i = 0; i < caps::MaxStreams && NumVertices != 0; i++) {
if (unlikely(vboUPBufferSizes[i] == 0)) {
EmitCs([
cStream = i
](DxvkContext* ctx) {
ctx->bindVertexBuffer(cStream, DxvkBufferSlice(), 0);
});
m_flags.set(D3D9DeviceFlag::DirtyVertexBuffers);
continue;
}

auto* vbo = GetCommonBuffer(m_state.vertexBuffers[i].vertexBuffer);

const uint32_t vertexStride = m_state.vertexDecl->GetSize(i);
uint32_t offset = FirstVertexIndex * vertexStride;

uint8_t* data = reinterpret_cast<uint8_t*>(upSlice.mapPtr) + vboUPBufferOffsets[i];
uint8_t* src = reinterpret_cast<uint8_t*>(vbo->GetMappedSlice().mapPtr) + offset;
std::memcpy(data, src, vboUPBufferSizes[i]);

auto vboSlice = upSlice.slice.subSlice(vboUPBufferOffsets[i], vboUPBufferSizes[i]);
EmitCs([
cStream = i,
cBufferSlice = std::move(vboSlice),
cStride = vertexStride
](DxvkContext* ctx) mutable {
ctx->bindVertexBuffer(cStream, std::move(cBufferSlice), cStride);
});
m_flags.set(D3D9DeviceFlag::DirtyVertexBuffers);
}

if (NumIndices != 0) {
if (unlikely(iboUPBufferSize == 0)) {
EmitCs([
cIndexType = VK_INDEX_TYPE_UINT32
](DxvkContext* ctx) {
ctx->bindIndexBuffer(DxvkBufferSlice(), cIndexType);
});
m_flags.set(D3D9DeviceFlag::DirtyIndexBuffer);
} else {
auto* ibo = GetCommonBuffer(m_state.indices);
uint32_t indexStride = ibo->Desc()->Format == D3D9Format::INDEX16 ? 2 : 4;
VkIndexType indexType = DecodeIndexType(ibo->Desc()->Format);
uint32_t offset = indexStride * FirstIndex;
uint8_t* data = reinterpret_cast<uint8_t*>(upSlice.mapPtr) + iboUPBufferOffset;
uint8_t* src = reinterpret_cast<uint8_t*>(ibo->GetMappedSlice().mapPtr) + offset;
std::memcpy(data, src, iboUPBufferSize);

auto iboSlice = upSlice.slice.subSlice(iboUPBufferOffset, iboUPBufferSize);
EmitCs([
cBufferSlice = std::move(iboSlice),
cIndexType = indexType
](DxvkContext* ctx) mutable {
ctx->bindIndexBuffer(std::move(cBufferSlice), cIndexType);
});
m_flags.set(D3D9DeviceFlag::DirtyIndexBuffer);
}
}
}


void D3D9DeviceEx::EmitCsChunk(DxvkCsChunkRef&& chunk) {
m_csSeqNum = m_csThread.dispatchChunk(std::move(chunk));
Expand Down Expand Up @@ -6454,7 +6614,7 @@ namespace dxvk {
}


void D3D9DeviceEx::PrepareDraw(D3DPRIMITIVETYPE PrimitiveType, bool UploadBuffers) {
void D3D9DeviceEx::PrepareDraw(D3DPRIMITIVETYPE PrimitiveType, bool UploadVBOs, bool UploadIBO) {
if (unlikely(m_activeHazardsRT != 0 || m_activeHazardsDS != 0))
MarkRenderHazards();

Expand All @@ -6467,7 +6627,7 @@ namespace dxvk {

for (uint32_t i = 0; i < caps::MaxStreams; i++) {
auto* vbo = GetCommonBuffer(m_state.vertexBuffers[i].vertexBuffer);
if (vbo != nullptr && vbo->NeedsUpload() && UploadBuffers)
if (vbo != nullptr && vbo->NeedsUpload() && UploadVBOs)
FlushBuffer(vbo);
}

Expand All @@ -6483,7 +6643,7 @@ namespace dxvk {
GenerateTextureMips(texturesToGen);

auto* ibo = GetCommonBuffer(m_state.indices);
if (ibo != nullptr && ibo->NeedsUpload() && UploadBuffers)
if (ibo != nullptr && ibo->NeedsUpload() && UploadIBO)
FlushBuffer(ibo);

UpdateFog();
Expand Down Expand Up @@ -6616,6 +6776,19 @@ namespace dxvk {
}

BindSpecConstants();

if (unlikely(m_flags.test(D3D9DeviceFlag::DirtyVertexBuffers) && UploadVBOs)) {
for (uint32_t i = 0; i < caps::MaxStreams; i++) {
const D3D9VBO& vbo = m_state.vertexBuffers[i];
BindVertexBuffer(i, vbo.vertexBuffer.ptr(), vbo.offset, vbo.stride);
}
m_flags.clr(D3D9DeviceFlag::DirtyVertexBuffers);
}

if (unlikely(m_flags.test(D3D9DeviceFlag::DirtyIndexBuffer) && UploadIBO)) {
BindIndices();
m_flags.clr(D3D9DeviceFlag::DirtyIndexBuffer);
}
}


Expand Down
21 changes: 19 additions & 2 deletions src/d3d9/d3d9_device.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,8 @@ namespace dxvk {
DirtyInputLayout,
DirtyViewportScissor,
DirtyMultiSampleState,
DirtyVertexBuffers,
DirtyIndexBuffer,

DirtyFogState,
DirtyFogColor,
Expand Down Expand Up @@ -764,6 +766,21 @@ namespace dxvk {
HRESULT UnlockBuffer(
D3D9CommonBuffer* pResource);

/**
* @brief Uploads data from D3DPOOL_SYSMEM + D3DUSAGE_DYNAMIC buffers and binds the temporary buffers.
*
* @param FirstVertexIndex The first vertex
* @param NumVertices The number of vertices that are accessed. If this is 0, the vertex buffer binding will not be modified.
* @param FirstIndex The first index
* @param NumIndices The number of indices that will be drawn. If this is 0, the index buffer binding will not be modified.
*/
void UploadDynamicSysmemBuffers(
UINT FirstVertexIndex,
UINT NumVertices,
UINT FirstIndex,
UINT NumIndices
);

void SetupFPU();

int64_t DetermineInitialTextureMemory();
Expand Down Expand Up @@ -895,7 +912,7 @@ namespace dxvk {

uint32_t GetInstanceCount() const;

void PrepareDraw(D3DPRIMITIVETYPE PrimitiveType, bool UploadBuffers);
void PrepareDraw(D3DPRIMITIVETYPE PrimitiveType, bool UploadVBOs, bool UploadIBOs);

template <DxsoProgramType ShaderStage>
void BindShader(
Expand Down Expand Up @@ -1063,7 +1080,7 @@ namespace dxvk {
}

inline uint32_t GetUPBufferSize(uint32_t vertexCount, uint32_t stride) {
return (vertexCount - 1) * stride + std::max(m_state.vertexDecl->GetSize(), stride);
return (vertexCount - 1) * stride + std::max(m_state.vertexDecl->GetSize(0), stride);
}

inline void FillUPVertexBuffer(void* buffer, const void* userData, uint32_t dataSize, uint32_t bufferSize) {
Expand Down
2 changes: 1 addition & 1 deletion src/d3d9/d3d9_swvp_emu.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -144,7 +144,7 @@ namespace dxvk {
uint32_t primitiveId = m_module.opLoad(uint_t, primitiveIdPtr);

// The size of any given vertex
uint32_t vertexSize = m_module.constu32(pDecl->GetSize() / sizeof(uint32_t));
uint32_t vertexSize = m_module.constu32(pDecl->GetSize(0) / sizeof(uint32_t));

//The offset of this vertex from the beginning of the buffer
uint32_t thisVertexOffset = m_module.opIMul(uint_t, vertexSize, primitiveId);
Expand Down
4 changes: 2 additions & 2 deletions src/d3d9/d3d9_vertex_declaration.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -354,8 +354,8 @@ namespace dxvk {

void D3D9VertexDecl::Classify() {
for (const auto& element : m_elements) {
if (element.Stream == 0 && element.Type != D3DDECLTYPE_UNUSED)
m_size = std::max(m_size, element.Offset + GetDecltypeSize(D3DDECLTYPE(element.Type)));
if (element.Type != D3DDECLTYPE_UNUSED)
m_sizes[element.Stream] = std::max(m_sizes[element.Stream], element.Offset + GetDecltypeSize(D3DDECLTYPE(element.Type)));

if (element.Usage == D3DDECLUSAGE_COLOR && element.UsageIndex == 0)
m_flags.set(D3D9VertexDeclFlag::HasColor0);
Expand Down
7 changes: 3 additions & 4 deletions src/d3d9/d3d9_vertex_declaration.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -50,8 +50,8 @@ namespace dxvk {
return m_elements;
}

UINT GetSize() const {
return m_size;
UINT GetSize(UINT Stream) const {
return m_sizes[Stream];
}

bool TestFlag(D3D9VertexDeclFlag flag) const {
Expand Down Expand Up @@ -94,8 +94,7 @@ namespace dxvk {

uint32_t m_texcoordMask = 0;

// The size of Stream 0. That's all we care about.
uint32_t m_size = 0;
std::array<uint32_t, caps::MaxStreams> m_sizes = {};

};

Expand Down

0 comments on commit f067c06

Please sign in to comment.