Merge pull request #9423 from vonchenplus/vulkan_quad_strip

video_core: Implement all vulkan topology
This commit is contained in:
liamwhite 2022-12-28 20:59:23 -05:00 committed by GitHub
commit 9fdacb5e3a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 244 additions and 124 deletions

View File

@ -666,8 +666,9 @@ void BufferCache<P>::BindHostGeometryBuffers(bool is_indexed) {
BindHostIndexBuffer(); BindHostIndexBuffer();
} else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) { } else if constexpr (!HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
const auto& draw_state = maxwell3d->draw_manager->GetDrawState(); const auto& draw_state = maxwell3d->draw_manager->GetDrawState();
if (draw_state.topology == Maxwell::PrimitiveTopology::Quads) { if (draw_state.topology == Maxwell::PrimitiveTopology::Quads ||
runtime.BindQuadArrayIndexBuffer(draw_state.vertex_buffer.first, draw_state.topology == Maxwell::PrimitiveTopology::QuadStrip) {
runtime.BindQuadIndexBuffer(draw_state.topology, draw_state.vertex_buffer.first,
draw_state.vertex_buffer.count); draw_state.vertex_buffer.count);
} }
} }

View File

@ -16,6 +16,7 @@ layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
layout (push_constant) uniform PushConstants { layout (push_constant) uniform PushConstants {
uint base_vertex; uint base_vertex;
int index_shift; // 0: uint8, 1: uint16, 2: uint32 int index_shift; // 0: uint8, 1: uint16, 2: uint32
int is_strip; // 0: quads 1: quadstrip
}; };
void main() { void main() {
@ -28,9 +29,10 @@ void main() {
int flipped_shift = 2 - index_shift; int flipped_shift = 2 - index_shift;
int mask = (1 << flipped_shift) - 1; int mask = (1 << flipped_shift) - 1;
const int quad_swizzle[6] = int[](0, 1, 2, 0, 2, 3); const int quads_swizzle[6] = int[](0, 1, 2, 0, 2, 3);
const int quad_strip_swizzle[6] = int[](0, 3, 1, 0, 2, 3);
for (uint vertex = 0; vertex < 6; ++vertex) { for (uint vertex = 0; vertex < 6; ++vertex) {
int offset = primitive * 4 + quad_swizzle[vertex]; int offset = (is_strip == 0 ? primitive * 4 + quads_swizzle[vertex] : primitive * 2 + quad_strip_swizzle[vertex]);
int int_offset = offset >> flipped_shift; int int_offset = offset >> flipped_shift;
int bit_offset = (offset & mask) * index_size; int bit_offset = (offset & mask) * index_size;
uint packed_input = input_indexes[int_offset]; uint packed_input = input_indexes[int_offset];

View File

@ -301,6 +301,8 @@ VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const Device& device,
return VK_PRIMITIVE_TOPOLOGY_POINT_LIST; return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
case Maxwell::PrimitiveTopology::Lines: case Maxwell::PrimitiveTopology::Lines:
return VK_PRIMITIVE_TOPOLOGY_LINE_LIST; return VK_PRIMITIVE_TOPOLOGY_LINE_LIST;
case Maxwell::PrimitiveTopology::LineLoop:
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
case Maxwell::PrimitiveTopology::LineStrip: case Maxwell::PrimitiveTopology::LineStrip:
return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP; return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
case Maxwell::PrimitiveTopology::Triangles: case Maxwell::PrimitiveTopology::Triangles:
@ -309,15 +311,28 @@ VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const Device& device,
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP; return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP;
case Maxwell::PrimitiveTopology::TriangleFan: case Maxwell::PrimitiveTopology::TriangleFan:
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN; return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
case Maxwell::PrimitiveTopology::LinesAdjacency:
return VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY;
case Maxwell::PrimitiveTopology::LineStripAdjacency:
return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY;
case Maxwell::PrimitiveTopology::TrianglesAdjacency:
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY;
case Maxwell::PrimitiveTopology::TriangleStripAdjacency:
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY;
case Maxwell::PrimitiveTopology::Quads: case Maxwell::PrimitiveTopology::Quads:
// TODO(Rodrigo): Use VK_PRIMITIVE_TOPOLOGY_QUAD_LIST_EXT whenever it releases case Maxwell::PrimitiveTopology::QuadStrip:
// TODO: Use VK_PRIMITIVE_TOPOLOGY_QUAD_LIST_EXT/VK_PRIMITIVE_TOPOLOGY_QUAD_STRIP_EXT
// whenever it releases
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST; return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
case Maxwell::PrimitiveTopology::Patches: case Maxwell::PrimitiveTopology::Patches:
return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST; return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
default: case Maxwell::PrimitiveTopology::Polygon:
LOG_WARNING(Render_Vulkan, "Draw mode is Polygon with a polygon mode of lines should be a "
"single body and not a bunch of triangles.");
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
}
UNIMPLEMENTED_MSG("Unimplemented topology={}", topology); UNIMPLEMENTED_MSG("Unimplemented topology={}", topology);
return {}; return {};
}
} }
VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type, VkFormat VertexFormat(const Device& device, Maxwell::VertexAttribute::Type type,

View File

@ -51,15 +51,6 @@ size_t BytesPerIndex(VkIndexType index_type) {
} }
} }
template <typename T>
std::array<T, 6> MakeQuadIndices(u32 quad, u32 first) {
std::array<T, 6> indices{0, 1, 2, 0, 2, 3};
for (T& index : indices) {
index = static_cast<T>(first + index + quad * 4);
}
return indices;
}
vk::Buffer CreateBuffer(const Device& device, u64 size) { vk::Buffer CreateBuffer(const Device& device, u64 size) {
VkBufferUsageFlags flags = VkBufferUsageFlags flags =
VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
@ -123,6 +114,187 @@ VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat
return *views.back().handle; return *views.back().handle;
} }
class QuadIndexBuffer {
public:
QuadIndexBuffer(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_)
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
staging_pool{staging_pool_} {}
virtual ~QuadIndexBuffer() = default;
void UpdateBuffer(u32 num_indices_) {
if (num_indices_ <= num_indices) {
return;
}
scheduler.Finish();
num_indices = num_indices_;
index_type = IndexTypeFromNumElements(device, num_indices);
const u32 num_quads = GetQuadsNum(num_indices);
const u32 num_triangle_indices = num_quads * 6;
const u32 num_first_offset_copies = 4;
const size_t bytes_per_index = BytesPerIndex(index_type);
const size_t size_bytes = num_triangle_indices * bytes_per_index * num_first_offset_copies;
buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = size_bytes,
.usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
if (device.HasDebuggingToolAttached()) {
buffer.SetObjectNameEXT("Quad LUT");
}
memory_commit = memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
const StagingBufferRef staging = staging_pool.Request(size_bytes, MemoryUsage::Upload);
u8* staging_data = staging.mapped_span.data();
const size_t quad_size = bytes_per_index * 6;
for (u32 first = 0; first < num_first_offset_copies; ++first) {
for (u32 quad = 0; quad < num_quads; ++quad) {
MakeAndUpdateIndices(staging_data, quad_size, quad, first);
staging_data += quad_size;
}
}
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([src_buffer = staging.buffer, src_offset = staging.offset,
dst_buffer = *buffer, size_bytes](vk::CommandBuffer cmdbuf) {
const VkBufferCopy copy{
.srcOffset = src_offset,
.dstOffset = 0,
.size = size_bytes,
};
const VkBufferMemoryBarrier write_barrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.buffer = dst_buffer,
.offset = 0,
.size = size_bytes,
};
cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, write_barrier);
});
}
void BindBuffer(u32 first) {
const VkIndexType index_type_ = index_type;
const size_t sub_first_offset = static_cast<size_t>(first % 4) * GetQuadsNum(num_indices);
const size_t offset =
(sub_first_offset + GetQuadsNum(first)) * 6ULL * BytesPerIndex(index_type);
scheduler.Record([buffer = *buffer, index_type_, offset](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(buffer, offset, index_type_);
});
}
protected:
virtual u32 GetQuadsNum(u32 num_indices) const = 0;
virtual void MakeAndUpdateIndices(u8* staging_data, size_t quad_size, u32 quad, u32 first) = 0;
const Device& device;
MemoryAllocator& memory_allocator;
Scheduler& scheduler;
StagingBufferPool& staging_pool;
vk::Buffer buffer{};
MemoryCommit memory_commit{};
VkIndexType index_type{};
u32 num_indices = 0;
};
class QuadArrayIndexBuffer : public QuadIndexBuffer {
public:
QuadArrayIndexBuffer(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_)
: QuadIndexBuffer(device_, memory_allocator_, scheduler_, staging_pool_) {}
~QuadArrayIndexBuffer() = default;
private:
u32 GetQuadsNum(u32 num_indices_) const override {
return num_indices_ / 4;
}
template <typename T>
static std::array<T, 6> MakeIndices(u32 quad, u32 first) {
std::array<T, 6> indices{0, 1, 2, 0, 2, 3};
for (T& index : indices) {
index = static_cast<T>(first + index + quad * 4);
}
return indices;
}
void MakeAndUpdateIndices(u8* staging_data, size_t quad_size, u32 quad, u32 first) {
switch (index_type) {
case VK_INDEX_TYPE_UINT8_EXT:
std::memcpy(staging_data, MakeIndices<u8>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT16:
std::memcpy(staging_data, MakeIndices<u16>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT32:
std::memcpy(staging_data, MakeIndices<u32>(quad, first).data(), quad_size);
break;
default:
ASSERT(false);
break;
}
}
};
class QuadStripIndexBuffer : public QuadIndexBuffer {
public:
QuadStripIndexBuffer(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_)
: QuadIndexBuffer(device_, memory_allocator_, scheduler_, staging_pool_) {}
~QuadStripIndexBuffer() = default;
private:
u32 GetQuadsNum(u32 num_indices_) const override {
return num_indices_ >= 4 ? (num_indices_ - 2) / 2 : 0;
}
template <typename T>
static std::array<T, 6> MakeIndices(u32 quad, u32 first) {
std::array<T, 6> indices{0, 3, 1, 0, 2, 3};
for (T& index : indices) {
index = static_cast<T>(first + index + quad * 2);
}
return indices;
}
void MakeAndUpdateIndices(u8* staging_data, size_t quad_size, u32 quad, u32 first) {
switch (index_type) {
case VK_INDEX_TYPE_UINT8_EXT:
std::memcpy(staging_data, MakeIndices<u8>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT16:
std::memcpy(staging_data, MakeIndices<u16>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT32:
std::memcpy(staging_data, MakeIndices<u32>(quad, first).data(), quad_size);
break;
default:
ASSERT(false);
break;
}
}
};
BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_, BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& memory_allocator_,
Scheduler& scheduler_, StagingBufferPool& staging_pool_, Scheduler& scheduler_, StagingBufferPool& staging_pool_,
UpdateDescriptorQueue& update_descriptor_queue_, UpdateDescriptorQueue& update_descriptor_queue_,
@ -130,7 +302,12 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
: device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_}, : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_}, staging_pool{staging_pool_}, update_descriptor_queue{update_descriptor_queue_},
uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue), uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
quad_index_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue) {} quad_index_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue) {
quad_array_index_buffer = std::make_shared<QuadArrayIndexBuffer>(device_, memory_allocator_,
scheduler_, staging_pool_);
quad_strip_index_buffer = std::make_shared<QuadStripIndexBuffer>(device_, memory_allocator_,
scheduler_, staging_pool_);
}
StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) { StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) {
return staging_pool.Request(size, MemoryUsage::Upload); return staging_pool.Request(size, MemoryUsage::Upload);
@ -245,10 +422,11 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat
VkIndexType vk_index_type = MaxwellToVK::IndexFormat(index_format); VkIndexType vk_index_type = MaxwellToVK::IndexFormat(index_format);
VkDeviceSize vk_offset = offset; VkDeviceSize vk_offset = offset;
VkBuffer vk_buffer = buffer; VkBuffer vk_buffer = buffer;
if (topology == PrimitiveTopology::Quads) { if (topology == PrimitiveTopology::Quads || topology == PrimitiveTopology::QuadStrip) {
vk_index_type = VK_INDEX_TYPE_UINT32; vk_index_type = VK_INDEX_TYPE_UINT32;
std::tie(vk_buffer, vk_offset) = std::tie(vk_buffer, vk_offset) =
quad_index_pass.Assemble(index_format, num_indices, base_vertex, buffer, offset); quad_index_pass.Assemble(index_format, num_indices, base_vertex, buffer, offset,
topology == PrimitiveTopology::QuadStrip);
} else if (vk_index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported()) { } else if (vk_index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported()) {
vk_index_type = VK_INDEX_TYPE_UINT16; vk_index_type = VK_INDEX_TYPE_UINT16;
std::tie(vk_buffer, vk_offset) = uint8_pass.Assemble(num_indices, buffer, offset); std::tie(vk_buffer, vk_offset) = uint8_pass.Assemble(num_indices, buffer, offset);
@ -263,7 +441,7 @@ void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat
}); });
} }
void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) { void BufferCacheRuntime::BindQuadIndexBuffer(PrimitiveTopology topology, u32 first, u32 count) {
if (count == 0) { if (count == 0) {
ReserveNullBuffer(); ReserveNullBuffer();
scheduler.Record([this](vk::CommandBuffer cmdbuf) { scheduler.Record([this](vk::CommandBuffer cmdbuf) {
@ -271,16 +449,14 @@ void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) {
}); });
return; return;
} }
ReserveQuadArrayLUT(first + count, true);
// The LUT has the indices 0, 1, 2, and 3 copied as an array if (topology == PrimitiveTopology::Quads) {
// To apply these 'first' offsets we can apply an offset based on the modulus. quad_array_index_buffer->UpdateBuffer(first + count);
const VkIndexType index_type = quad_array_lut_index_type; quad_array_index_buffer->BindBuffer(first);
const size_t sub_first_offset = static_cast<size_t>(first % 4) * (current_num_indices / 4); } else if (topology == PrimitiveTopology::QuadStrip) {
const size_t offset = (sub_first_offset + first / 4) * 6ULL * BytesPerIndex(index_type); quad_strip_index_buffer->UpdateBuffer(first + count);
scheduler.Record([buffer = *quad_array_lut, index_type, offset](vk::CommandBuffer cmdbuf) { quad_strip_index_buffer->BindBuffer(first);
cmdbuf.BindIndexBuffer(buffer, offset, index_type); }
});
} }
void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size,
@ -323,83 +499,6 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer,
}); });
} }
void BufferCacheRuntime::ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle) {
if (num_indices <= current_num_indices) {
return;
}
if (wait_for_idle) {
scheduler.Finish();
}
current_num_indices = num_indices;
quad_array_lut_index_type = IndexTypeFromNumElements(device, num_indices);
const u32 num_quads = num_indices / 4;
const u32 num_triangle_indices = num_quads * 6;
const u32 num_first_offset_copies = 4;
const size_t bytes_per_index = BytesPerIndex(quad_array_lut_index_type);
const size_t size_bytes = num_triangle_indices * bytes_per_index * num_first_offset_copies;
quad_array_lut = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = size_bytes,
.usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
if (device.HasDebuggingToolAttached()) {
quad_array_lut.SetObjectNameEXT("Quad LUT");
}
quad_array_lut_commit = memory_allocator.Commit(quad_array_lut, MemoryUsage::DeviceLocal);
const StagingBufferRef staging = staging_pool.Request(size_bytes, MemoryUsage::Upload);
u8* staging_data = staging.mapped_span.data();
const size_t quad_size = bytes_per_index * 6;
for (u32 first = 0; first < num_first_offset_copies; ++first) {
for (u32 quad = 0; quad < num_quads; ++quad) {
switch (quad_array_lut_index_type) {
case VK_INDEX_TYPE_UINT8_EXT:
std::memcpy(staging_data, MakeQuadIndices<u8>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT16:
std::memcpy(staging_data, MakeQuadIndices<u16>(quad, first).data(), quad_size);
break;
case VK_INDEX_TYPE_UINT32:
std::memcpy(staging_data, MakeQuadIndices<u32>(quad, first).data(), quad_size);
break;
default:
ASSERT(false);
break;
}
staging_data += quad_size;
}
}
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([src_buffer = staging.buffer, src_offset = staging.offset,
dst_buffer = *quad_array_lut, size_bytes](vk::CommandBuffer cmdbuf) {
const VkBufferCopy copy{
.srcOffset = src_offset,
.dstOffset = 0,
.size = size_bytes,
};
const VkBufferMemoryBarrier write_barrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.buffer = dst_buffer,
.offset = 0,
.size = size_bytes,
};
cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT,
0, write_barrier);
});
}
void BufferCacheRuntime::ReserveNullBuffer() { void BufferCacheRuntime::ReserveNullBuffer() {
if (null_buffer) { if (null_buffer) {
return; return;

View File

@ -50,6 +50,9 @@ private:
std::vector<BufferView> views; std::vector<BufferView> views;
}; };
class QuadArrayIndexBuffer;
class QuadStripIndexBuffer;
class BufferCacheRuntime { class BufferCacheRuntime {
friend Buffer; friend Buffer;
@ -86,7 +89,7 @@ public:
void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices, void BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format, u32 num_indices,
u32 base_vertex, VkBuffer buffer, u32 offset, u32 size); u32 base_vertex, VkBuffer buffer, u32 offset, u32 size);
void BindQuadArrayIndexBuffer(u32 first, u32 count); void BindQuadIndexBuffer(PrimitiveTopology topology, u32 first, u32 count);
void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride); void BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size, u32 stride);
@ -118,8 +121,6 @@ private:
update_descriptor_queue.AddBuffer(buffer, offset, size); update_descriptor_queue.AddBuffer(buffer, offset, size);
} }
void ReserveQuadArrayLUT(u32 num_indices, bool wait_for_idle);
void ReserveNullBuffer(); void ReserveNullBuffer();
const Device& device; const Device& device;
@ -128,10 +129,8 @@ private:
StagingBufferPool& staging_pool; StagingBufferPool& staging_pool;
UpdateDescriptorQueue& update_descriptor_queue; UpdateDescriptorQueue& update_descriptor_queue;
vk::Buffer quad_array_lut; std::shared_ptr<QuadArrayIndexBuffer> quad_array_index_buffer;
MemoryCommit quad_array_lut_commit; std::shared_ptr<QuadStripIndexBuffer> quad_strip_index_buffer;
VkIndexType quad_array_lut_index_type{};
u32 current_num_indices = 0;
vk::Buffer null_buffer; vk::Buffer null_buffer;
MemoryCommit null_buffer_commit; MemoryCommit null_buffer_commit;

View File

@ -245,7 +245,7 @@ QuadIndexedPass::QuadIndexedPass(const Device& device_, Scheduler& scheduler_,
UpdateDescriptorQueue& update_descriptor_queue_) UpdateDescriptorQueue& update_descriptor_queue_)
: ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS, : ComputePass(device_, descriptor_pool_, INPUT_OUTPUT_DESCRIPTOR_SET_BINDINGS,
INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO, INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE, INPUT_OUTPUT_BANK_INFO,
COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 2>, VULKAN_QUAD_INDEXED_COMP_SPV), COMPUTE_PUSH_CONSTANT_RANGE<sizeof(u32) * 3>, VULKAN_QUAD_INDEXED_COMP_SPV),
scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_} {} update_descriptor_queue{update_descriptor_queue_} {}
@ -253,7 +253,7 @@ QuadIndexedPass::~QuadIndexedPass() = default;
std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble( std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex, Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex,
VkBuffer src_buffer, u32 src_offset) { VkBuffer src_buffer, u32 src_offset, bool is_strip) {
const u32 index_shift = [index_format] { const u32 index_shift = [index_format] {
switch (index_format) { switch (index_format) {
case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte: case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte:
@ -267,7 +267,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
return 2; return 2;
}(); }();
const u32 input_size = num_vertices << index_shift; const u32 input_size = num_vertices << index_shift;
const u32 num_tri_vertices = (num_vertices / 4) * 6; const u32 num_tri_vertices = (is_strip ? (num_vertices - 2) / 2 : num_vertices / 4) * 6;
const std::size_t staging_size = num_tri_vertices * sizeof(u32); const std::size_t staging_size = num_tri_vertices * sizeof(u32);
const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal); const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
@ -278,8 +278,8 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
const void* const descriptor_data{update_descriptor_queue.UpdateData()}; const void* const descriptor_data{update_descriptor_queue.UpdateData()};
scheduler.RequestOutsideRenderPassOperationContext(); scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex, index_shift,
index_shift](vk::CommandBuffer cmdbuf) { is_strip](vk::CommandBuffer cmdbuf) {
static constexpr u32 DISPATCH_SIZE = 1024; static constexpr u32 DISPATCH_SIZE = 1024;
static constexpr VkMemoryBarrier WRITE_BARRIER{ static constexpr VkMemoryBarrier WRITE_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
@ -287,7 +287,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT, .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_INDEX_READ_BIT, .dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
}; };
const std::array<u32, 2> push_constants{base_vertex, index_shift}; const std::array<u32, 3> push_constants{base_vertex, index_shift, is_strip ? 1u : 0u};
const VkDescriptorSet set = descriptor_allocator.Commit(); const VkDescriptorSet set = descriptor_allocator.Commit();
device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data); device.GetLogical().UpdateDescriptorSet(set, *descriptor_template, descriptor_data);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline); cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);

View File

@ -74,7 +74,7 @@ public:
std::pair<VkBuffer, VkDeviceSize> Assemble( std::pair<VkBuffer, VkDeviceSize> Assemble(
Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices,
u32 base_vertex, VkBuffer src_buffer, u32 src_offset); u32 base_vertex, VkBuffer src_buffer, u32 src_offset, bool is_strip);
private: private:
Scheduler& scheduler; Scheduler& scheduler;

View File

@ -138,12 +138,16 @@ DrawParams MakeDrawParams(const MaxwellDrawState& draw_state, u32 num_instances,
.first_index = is_indexed ? draw_state.index_buffer.first : 0, .first_index = is_indexed ? draw_state.index_buffer.first : 0,
.is_indexed = is_indexed, .is_indexed = is_indexed,
}; };
if (draw_state.topology == Maxwell::PrimitiveTopology::Quads) {
// 6 triangle vertices per quad, base vertex is part of the index // 6 triangle vertices per quad, base vertex is part of the index
// See BindQuadArrayIndexBuffer for more details // See BindQuadIndexBuffer for more details
if (draw_state.topology == Maxwell::PrimitiveTopology::Quads) {
params.num_vertices = (params.num_vertices / 4) * 6; params.num_vertices = (params.num_vertices / 4) * 6;
params.base_vertex = 0; params.base_vertex = 0;
params.is_indexed = true; params.is_indexed = true;
} else if (draw_state.topology == Maxwell::PrimitiveTopology::QuadStrip) {
params.num_vertices = (params.num_vertices - 2) / 2 * 6;
params.base_vertex = 0;
params.is_indexed = true;
} }
return params; return params;
} }