1
0
Fork 0
forked from suyu/suyu

vk_rasterizer: Use VK_EXT_extended_dynamic_state

This commit is contained in:
ReinUsesLisp 2020-06-22 03:03:27 -03:00
parent a6db8e5f4d
commit c94b398f14
5 changed files with 356 additions and 46 deletions

View file

@ -354,11 +354,27 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa
color_blend_ci.pAttachments = cb_attachments.data();
std::memset(color_blend_ci.blendConstants, 0, sizeof(color_blend_ci.blendConstants));
static constexpr std::array dynamic_states = {
std::vector dynamic_states = {
VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR,
VK_DYNAMIC_STATE_DEPTH_BIAS, VK_DYNAMIC_STATE_BLEND_CONSTANTS,
VK_DYNAMIC_STATE_DEPTH_BOUNDS, VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
VK_DYNAMIC_STATE_STENCIL_WRITE_MASK, VK_DYNAMIC_STATE_STENCIL_REFERENCE};
VK_DYNAMIC_STATE_STENCIL_WRITE_MASK, VK_DYNAMIC_STATE_STENCIL_REFERENCE,
};
if (device.IsExtExtendedDynamicStateSupported()) {
static constexpr std::array extended = {
VK_DYNAMIC_STATE_CULL_MODE_EXT,
VK_DYNAMIC_STATE_FRONT_FACE_EXT,
VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY_EXT,
VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE_EXT,
VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE_EXT,
VK_DYNAMIC_STATE_DEPTH_COMPARE_OP_EXT,
VK_DYNAMIC_STATE_DEPTH_BOUNDS_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE_EXT,
VK_DYNAMIC_STATE_STENCIL_OP_EXT,
};
dynamic_states.insert(dynamic_states.end(), extended.begin(), extended.end());
}
VkPipelineDynamicStateCreateInfo dynamic_state_ci;
dynamic_state_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO;

View file

@ -186,13 +186,22 @@ bool HasToPreserveDepthContents(bool is_clear, const Maxwell& regs) {
scissor.max_y < regs.zeta_height;
}
template <std::size_t N>
std::array<VkDeviceSize, N> ExpandStrides(const std::array<u16, N>& strides) {
std::array<VkDeviceSize, N> expanded;
std::copy(strides.begin(), strides.end(), expanded.begin());
return expanded;
}
} // Anonymous namespace
class BufferBindings final {
public:
void AddVertexBinding(VkBuffer buffer, VkDeviceSize offset) {
void AddVertexBinding(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size, u32 stride) {
vertex.buffers[vertex.num_buffers] = buffer;
vertex.offsets[vertex.num_buffers] = offset;
vertex.sizes[vertex.num_buffers] = size;
vertex.strides[vertex.num_buffers] = static_cast<u16>(stride);
++vertex.num_buffers;
}
@ -202,76 +211,76 @@ public:
index.type = type;
}
void Bind(VKScheduler& scheduler) const {
void Bind(const VKDevice& device, VKScheduler& scheduler) const {
// Use this large switch case to avoid dispatching more memory in the record lambda than
// what we need. It looks horrible, but it's the best we can do on standard C++.
switch (vertex.num_buffers) {
case 0:
return BindStatic<0>(scheduler);
return BindStatic<0>(device, scheduler);
case 1:
return BindStatic<1>(scheduler);
return BindStatic<1>(device, scheduler);
case 2:
return BindStatic<2>(scheduler);
return BindStatic<2>(device, scheduler);
case 3:
return BindStatic<3>(scheduler);
return BindStatic<3>(device, scheduler);
case 4:
return BindStatic<4>(scheduler);
return BindStatic<4>(device, scheduler);
case 5:
return BindStatic<5>(scheduler);
return BindStatic<5>(device, scheduler);
case 6:
return BindStatic<6>(scheduler);
return BindStatic<6>(device, scheduler);
case 7:
return BindStatic<7>(scheduler);
return BindStatic<7>(device, scheduler);
case 8:
return BindStatic<8>(scheduler);
return BindStatic<8>(device, scheduler);
case 9:
return BindStatic<9>(scheduler);
return BindStatic<9>(device, scheduler);
case 10:
return BindStatic<10>(scheduler);
return BindStatic<10>(device, scheduler);
case 11:
return BindStatic<11>(scheduler);
return BindStatic<11>(device, scheduler);
case 12:
return BindStatic<12>(scheduler);
return BindStatic<12>(device, scheduler);
case 13:
return BindStatic<13>(scheduler);
return BindStatic<13>(device, scheduler);
case 14:
return BindStatic<14>(scheduler);
return BindStatic<14>(device, scheduler);
case 15:
return BindStatic<15>(scheduler);
return BindStatic<15>(device, scheduler);
case 16:
return BindStatic<16>(scheduler);
return BindStatic<16>(device, scheduler);
case 17:
return BindStatic<17>(scheduler);
return BindStatic<17>(device, scheduler);
case 18:
return BindStatic<18>(scheduler);
return BindStatic<18>(device, scheduler);
case 19:
return BindStatic<19>(scheduler);
return BindStatic<19>(device, scheduler);
case 20:
return BindStatic<20>(scheduler);
return BindStatic<20>(device, scheduler);
case 21:
return BindStatic<21>(scheduler);
return BindStatic<21>(device, scheduler);
case 22:
return BindStatic<22>(scheduler);
return BindStatic<22>(device, scheduler);
case 23:
return BindStatic<23>(scheduler);
return BindStatic<23>(device, scheduler);
case 24:
return BindStatic<24>(scheduler);
return BindStatic<24>(device, scheduler);
case 25:
return BindStatic<25>(scheduler);
return BindStatic<25>(device, scheduler);
case 26:
return BindStatic<26>(scheduler);
return BindStatic<26>(device, scheduler);
case 27:
return BindStatic<27>(scheduler);
return BindStatic<27>(device, scheduler);
case 28:
return BindStatic<28>(scheduler);
return BindStatic<28>(device, scheduler);
case 29:
return BindStatic<29>(scheduler);
return BindStatic<29>(device, scheduler);
case 30:
return BindStatic<30>(scheduler);
return BindStatic<30>(device, scheduler);
case 31:
return BindStatic<31>(scheduler);
return BindStatic<31>(device, scheduler);
case 32:
return BindStatic<32>(scheduler);
return BindStatic<32>(device, scheduler);
}
UNREACHABLE();
}
@ -282,6 +291,8 @@ private:
std::size_t num_buffers = 0;
std::array<VkBuffer, Maxwell::NumVertexArrays> buffers;
std::array<VkDeviceSize, Maxwell::NumVertexArrays> offsets;
std::array<VkDeviceSize, Maxwell::NumVertexArrays> sizes;
std::array<u16, Maxwell::NumVertexArrays> strides;
} vertex;
struct {
@ -291,15 +302,23 @@ private:
} index;
template <std::size_t N>
void BindStatic(VKScheduler& scheduler) const {
void BindStatic(const VKDevice& device, VKScheduler& scheduler) const {
if (device.IsExtExtendedDynamicStateSupported()) {
if (index.buffer) {
BindStatic<N, true>(scheduler);
BindStatic<N, true, true>(scheduler);
} else {
BindStatic<N, false>(scheduler);
BindStatic<N, false, true>(scheduler);
}
} else {
if (index.buffer) {
BindStatic<N, true, false>(scheduler);
} else {
BindStatic<N, false, false>(scheduler);
}
}
}
template <std::size_t N, bool is_indexed>
template <std::size_t N, bool is_indexed, bool has_extended_dynamic_state>
void BindStatic(VKScheduler& scheduler) const {
static_assert(N <= Maxwell::NumVertexArrays);
if constexpr (N == 0) {
@ -311,6 +330,31 @@ private:
std::copy(vertex.buffers.begin(), vertex.buffers.begin() + N, buffers.begin());
std::copy(vertex.offsets.begin(), vertex.offsets.begin() + N, offsets.begin());
if constexpr (has_extended_dynamic_state) {
// With extended dynamic states we can specify the length and stride of a vertex buffer
std::array<VkDeviceSize, N> sizes;
std::array<u16, N> strides;
std::copy(vertex.sizes.begin(), vertex.sizes.begin() + N, sizes.begin());
std::copy(vertex.strides.begin(), vertex.strides.begin() + N, strides.begin());
if constexpr (is_indexed) {
scheduler.Record(
[buffers, offsets, sizes, strides, index = index](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type);
cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(),
offsets.data(), sizes.data(),
ExpandStrides(strides).data());
});
} else {
scheduler.Record([buffers, offsets, sizes, strides](vk::CommandBuffer cmdbuf) {
cmdbuf.BindVertexBuffers2EXT(0, static_cast<u32>(N), buffers.data(),
offsets.data(), sizes.data(),
ExpandStrides(strides).data());
});
}
return;
}
if constexpr (is_indexed) {
// Indexed draw
scheduler.Record([buffers, offsets, index = index](vk::CommandBuffer cmdbuf) {
@ -402,7 +446,7 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
UpdateDynamicStates();
buffer_bindings.Bind(scheduler);
buffer_bindings.Bind(device, scheduler);
BeginTransformFeedback();
@ -893,6 +937,17 @@ void RasterizerVulkan::UpdateDynamicStates() {
UpdateBlendConstants(regs);
UpdateDepthBounds(regs);
UpdateStencilFaces(regs);
if (device.IsExtExtendedDynamicStateSupported()) {
UpdateCullMode(regs);
UpdateDepthBoundsTestEnable(regs);
UpdateDepthTestEnable(regs);
UpdateDepthWriteEnable(regs);
UpdateDepthCompareOp(regs);
UpdateFrontFace(regs);
UpdatePrimitiveTopology(regs);
UpdateStencilOp(regs);
UpdateStencilTestEnable(regs);
}
}
void RasterizerVulkan::BeginTransformFeedback() {
@ -952,13 +1007,13 @@ void RasterizerVulkan::SetupVertexArrays(BufferBindings& buffer_bindings) {
const GPUVAddr end{regs.vertex_array_limit[index].LimitAddress()};
ASSERT(end >= start);
const std::size_t size{end - start};
const std::size_t size = end - start;
if (size == 0) {
buffer_bindings.AddVertexBinding(DefaultBuffer(), 0);
buffer_bindings.AddVertexBinding(DefaultBuffer(), 0, DEFAULT_BUFFER_SIZE, 0);
continue;
}
const auto info = buffer_cache.UploadMemory(start, size);
buffer_bindings.AddVertexBinding(info.handle, info.offset);
buffer_bindings.AddVertexBinding(info.handle, info.offset, size, vertex_array.stride);
}
}
@ -1310,6 +1365,117 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
}
}
void RasterizerVulkan::UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchCullMode()) {
return;
}
scheduler.Record(
[enabled = regs.cull_test_enabled, cull_face = regs.cull_face](vk::CommandBuffer cmdbuf) {
cmdbuf.SetCullModeEXT(enabled ? MaxwellToVK::CullFace(cull_face) : VK_CULL_MODE_NONE);
});
}
void RasterizerVulkan::UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchDepthBoundsTestEnable()) {
return;
}
scheduler.Record([enable = regs.depth_bounds_enable](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthBoundsTestEnableEXT(enable);
});
}
void RasterizerVulkan::UpdateDepthTestEnable(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchDepthTestEnable()) {
return;
}
scheduler.Record([enable = regs.depth_test_enable](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthTestEnableEXT(enable);
});
}
void RasterizerVulkan::UpdateDepthWriteEnable(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchDepthWriteEnable()) {
return;
}
scheduler.Record([enable = regs.depth_write_enabled](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthWriteEnableEXT(enable);
});
}
void RasterizerVulkan::UpdateDepthCompareOp(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchDepthCompareOp()) {
return;
}
scheduler.Record([func = regs.depth_test_func](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthCompareOpEXT(MaxwellToVK::ComparisonOp(func));
});
}
void RasterizerVulkan::UpdateFrontFace(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchFrontFace()) {
return;
}
VkFrontFace front_face = MaxwellToVK::FrontFace(regs.front_face);
if (regs.screen_y_control.triangle_rast_flip != 0) {
front_face = front_face == VK_FRONT_FACE_CLOCKWISE ? VK_FRONT_FACE_COUNTER_CLOCKWISE
: VK_FRONT_FACE_CLOCKWISE;
}
scheduler.Record(
[front_face](vk::CommandBuffer cmdbuf) { cmdbuf.SetFrontFaceEXT(front_face); });
}
void RasterizerVulkan::UpdatePrimitiveTopology(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchPrimitiveTopology()) {
return;
}
const Maxwell::PrimitiveTopology primitive_topology = regs.draw.topology.Value();
scheduler.Record([this, primitive_topology](vk::CommandBuffer cmdbuf) {
cmdbuf.SetPrimitiveTopologyEXT(MaxwellToVK::PrimitiveTopology(device, primitive_topology));
});
}
void RasterizerVulkan::UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchStencilOp()) {
return;
}
const Maxwell::StencilOp fail = regs.stencil_front_op_fail;
const Maxwell::StencilOp zfail = regs.stencil_front_op_zfail;
const Maxwell::StencilOp zpass = regs.stencil_front_op_zpass;
const Maxwell::ComparisonOp compare = regs.stencil_front_func_func;
if (regs.stencil_two_side_enable) {
scheduler.Record([fail, zfail, zpass, compare](vk::CommandBuffer cmdbuf) {
cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_AND_BACK, MaxwellToVK::StencilOp(fail),
MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail),
MaxwellToVK::ComparisonOp(compare));
});
} else {
const Maxwell::StencilOp back_fail = regs.stencil_back_op_fail;
const Maxwell::StencilOp back_zfail = regs.stencil_back_op_zfail;
const Maxwell::StencilOp back_zpass = regs.stencil_back_op_zpass;
const Maxwell::ComparisonOp back_compare = regs.stencil_back_func_func;
scheduler.Record([fail, zfail, zpass, compare, back_fail, back_zfail, back_zpass,
back_compare](vk::CommandBuffer cmdbuf) {
cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_FRONT_BIT, MaxwellToVK::StencilOp(fail),
MaxwellToVK::StencilOp(zpass), MaxwellToVK::StencilOp(zfail),
MaxwellToVK::ComparisonOp(compare));
cmdbuf.SetStencilOpEXT(VK_STENCIL_FACE_BACK_BIT, MaxwellToVK::StencilOp(back_fail),
MaxwellToVK::StencilOp(back_zpass),
MaxwellToVK::StencilOp(back_zfail),
MaxwellToVK::ComparisonOp(back_compare));
});
}
}
void RasterizerVulkan::UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs) {
if (!state_tracker.TouchStencilTestEnable()) {
return;
}
scheduler.Record([enable = regs.stencil_enable](vk::CommandBuffer cmdbuf) {
cmdbuf.SetStencilTestEnableEXT(enable);
});
}
std::size_t RasterizerVulkan::CalculateGraphicsStreamBufferSize(bool is_indexed) const {
std::size_t size = CalculateVertexArraysSize();
if (is_indexed) {

View file

@ -245,6 +245,16 @@ private:
void UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthBoundsTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthWriteEnable(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateDepthCompareOp(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateFrontFace(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdatePrimitiveTopology(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateStencilOp(Tegra::Engines::Maxwell3D::Regs& regs);
void UpdateStencilTestEnable(Tegra::Engines::Maxwell3D::Regs& regs);
std::size_t CalculateGraphicsStreamBufferSize(bool is_indexed) const;
std::size_t CalculateComputeStreamBufferSize() const;

View file

@ -36,6 +36,15 @@ Flags MakeInvalidationFlags() {
flags[BlendConstants] = true;
flags[DepthBounds] = true;
flags[StencilProperties] = true;
flags[CullMode] = true;
flags[DepthBoundsEnable] = true;
flags[DepthTestEnable] = true;
flags[DepthWriteEnable] = true;
flags[DepthCompareOp] = true;
flags[FrontFace] = true;
flags[PrimitiveTopology] = true;
flags[StencilOp] = true;
flags[StencilTestEnable] = true;
return flags;
}
@ -75,6 +84,57 @@ void SetupDirtyStencilProperties(Tables& tables) {
table[OFF(stencil_back_func_mask)] = StencilProperties;
}
void SetupDirtyCullMode(Tables& tables) {
auto& table = tables[0];
table[OFF(cull_face)] = CullMode;
table[OFF(cull_test_enabled)] = CullMode;
}
void SetupDirtyDepthBoundsEnable(Tables& tables) {
tables[0][OFF(depth_bounds_enable)] = DepthBoundsEnable;
}
void SetupDirtyDepthTestEnable(Tables& tables) {
tables[0][OFF(depth_test_enable)] = DepthTestEnable;
}
void SetupDirtyDepthWriteEnable(Tables& tables) {
tables[0][OFF(depth_write_enabled)] = DepthWriteEnable;
}
void SetupDirtyDepthCompareOp(Tables& tables) {
tables[0][OFF(depth_test_func)] = DepthCompareOp;
}
void SetupDirtyFrontFace(Tables& tables) {
auto& table = tables[0];
table[OFF(front_face)] = FrontFace;
table[OFF(screen_y_control)] = FrontFace;
}
void SetupDirtyPrimitiveTopology(Tables& tables) {
tables[0][OFF(draw.topology)] = PrimitiveTopology;
}
void SetupDirtyStencilOp(Tables& tables) {
auto& table = tables[0];
table[OFF(stencil_front_op_fail)] = StencilOp;
table[OFF(stencil_front_op_zfail)] = StencilOp;
table[OFF(stencil_front_op_zpass)] = StencilOp;
table[OFF(stencil_front_func_func)] = StencilOp;
table[OFF(stencil_back_op_fail)] = StencilOp;
table[OFF(stencil_back_op_zfail)] = StencilOp;
table[OFF(stencil_back_op_zpass)] = StencilOp;
table[OFF(stencil_back_func_func)] = StencilOp;
// Table 0 is used by StencilProperties
tables[1][OFF(stencil_two_side_enable)] = StencilOp;
}
void SetupDirtyStencilTestEnable(Tables& tables) {
tables[0][OFF(stencil_enable)] = StencilTestEnable;
}
} // Anonymous namespace
StateTracker::StateTracker(Core::System& system)
@ -90,6 +150,14 @@ void StateTracker::Initialize() {
SetupDirtyBlendConstants(tables);
SetupDirtyDepthBounds(tables);
SetupDirtyStencilProperties(tables);
SetupDirtyCullMode(tables);
SetupDirtyDepthBoundsEnable(tables);
SetupDirtyDepthTestEnable(tables);
SetupDirtyDepthWriteEnable(tables);
SetupDirtyDepthCompareOp(tables);
SetupDirtyFrontFace(tables);
SetupDirtyPrimitiveTopology(tables);
SetupDirtyStencilOp(tables);
}
void StateTracker::InvalidateCommandBufferState() {

View file

@ -26,6 +26,16 @@ enum : u8 {
DepthBounds,
StencilProperties,
CullMode,
DepthBoundsEnable,
DepthTestEnable,
DepthWriteEnable,
DepthCompareOp,
FrontFace,
PrimitiveTopology,
StencilOp,
StencilTestEnable,
Last
};
static_assert(Last <= std::numeric_limits<u8>::max());
@ -64,6 +74,46 @@ public:
return Exchange(Dirty::StencilProperties, false);
}
bool TouchCullMode() {
return Exchange(Dirty::CullMode, false);
}
bool TouchDepthBoundsTestEnable() {
return Exchange(Dirty::DepthBoundsEnable, false);
}
bool TouchDepthTestEnable() {
return Exchange(Dirty::DepthTestEnable, false);
}
bool TouchDepthBoundsEnable() {
return Exchange(Dirty::DepthBoundsEnable, false);
}
bool TouchDepthWriteEnable() {
return Exchange(Dirty::DepthWriteEnable, false);
}
bool TouchDepthCompareOp() {
return Exchange(Dirty::DepthCompareOp, false);
}
bool TouchFrontFace() {
return Exchange(Dirty::FrontFace, false);
}
bool TouchPrimitiveTopology() {
return Exchange(Dirty::PrimitiveTopology, false);
}
bool TouchStencilOp() {
return Exchange(Dirty::StencilOp, false);
}
bool TouchStencilTestEnable() {
return Exchange(Dirty::StencilTestEnable, false);
}
private:
bool Exchange(std::size_t id, bool new_value) const noexcept {
auto& flags = system.GPU().Maxwell3D().dirty.flags;