prifak/yuzu
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

video_core: Implement maxwell3d draw texture method

Browse Source
This commit is contained in:
Feng Chen
2022-12-29 15:51:16 +08:00
parent b22892a306
commit 944d2e4154
14 changed files with 127 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
src/video_core/engines/draw_manager.cpp
View File

@@ -51,6 +51,10 @@ void DrawManager::ProcessMethodCall(u32 method, u32 argument) {
LOG_WARNING(HW_GPU, "(STUBBED) called");
break;
}
case MAXWELL3D_REG_INDEX(draw_texture.src_y0): {
DrawTexture();
break;
}
default:
break;
}
@@ -179,6 +183,33 @@ void DrawManager::DrawIndexSmall(u32 argument) {
ProcessDraw(true, 1);
}
void DrawManager::DrawTexture() {
const auto& regs{maxwell3d->regs};
draw_texture_state.dst_x0 = static_cast<float>(regs.draw_texture.dst_x0) / 4096.f;
draw_texture_state.dst_y0 = static_cast<float>(regs.draw_texture.dst_y0) / 4096.f;
const auto dst_width = static_cast<float>(regs.draw_texture.dst_width) / 4096.f;
const auto dst_height = static_cast<float>(regs.draw_texture.dst_height) / 4096.f;
const bool lower_left{regs.window_origin.mode !=
Maxwell3D::Regs::WindowOrigin::Mode::UpperLeft};
if (lower_left) {
draw_texture_state.dst_y0 -= dst_height;
}
draw_texture_state.dst_x1 = draw_texture_state.dst_x0 + dst_width;
draw_texture_state.dst_y1 = draw_texture_state.dst_y0 + dst_height;
draw_texture_state.src_x0 = static_cast<float>(regs.draw_texture.src_x0) / 4096.f;
draw_texture_state.src_y0 = static_cast<float>(regs.draw_texture.src_y0) / 4096.f;
draw_texture_state.src_x1 =
(static_cast<float>(regs.draw_texture.dx_du) / 4294967295.f) * dst_width +
draw_texture_state.src_x0;
draw_texture_state.src_y1 =
(static_cast<float>(regs.draw_texture.dy_dv) / 4294967295.f) * dst_height +
draw_texture_state.src_y0;
draw_texture_state.src_sampler = regs.draw_texture.src_sampler;
draw_texture_state.src_texture = regs.draw_texture.src_texture;
maxwell3d->rasterizer->DrawTexture();
}
void DrawManager::UpdateTopology() {
const auto& regs{maxwell3d->regs};
switch (regs.primitive_topology_control) {

20
src/video_core/engines/draw_manager.h
View File

@@ -32,6 +32,19 @@ public:
std::vector<u8> inline_index_draw_indexes;
};
struct DrawTextureState {
f32 dst_x0;
f32 dst_y0;
f32 dst_x1;
f32 dst_y1;
f32 src_x0;
f32 src_y0;
f32 src_x1;
f32 src_y1;
u32 src_sampler;
u32 src_texture;
};
struct IndirectParams {
bool is_indexed;
bool include_count;
@@ -64,6 +77,10 @@ public:
return draw_state;
}
const DrawTextureState& GetDrawTextureState() const {
return draw_texture_state;
}
IndirectParams& GetIndirectParams() {
return indirect_state;
}
@@ -81,6 +98,8 @@ private:
void DrawIndexSmall(u32 argument);
void DrawTexture();
void UpdateTopology();
void ProcessDraw(bool draw_indexed, u32 instance_count);
@@ -89,6 +108,7 @@ private:
Maxwell3D* maxwell3d{};
State draw_state{};
DrawTextureState draw_texture_state{};
IndirectParams indirect_state{};
};
} // namespace Tegra::Engines

16
src/video_core/engines/maxwell_3d.h
View File

@@ -1599,6 +1599,20 @@ public:
};
static_assert(sizeof(TIRModulationCoeff) == 0x4);
struct DrawTexture {
s32 dst_x0;
s32 dst_y0;
s32 dst_width;
s32 dst_height;
s64 dx_du;
s64 dy_dv;
u32 src_sampler;
u32 src_texture;
s32 src_x0;
s32 src_y0;
};
static_assert(sizeof(DrawTexture) == 0x30);
struct ReduceColorThreshold {
union {
BitField<0, 8, u32> all_hit_once;
@@ -2751,7 +2765,7 @@ public:
u32 reserved_sw_method2; ///< 0x102C
std::array<TIRModulationCoeff, 5> tir_modulation_coeff; ///< 0x1030
std::array<u32, 15> spare_nop; ///< 0x1044
INSERT_PADDING_BYTES_NOINIT(0x30);
DrawTexture draw_texture; ///< 0x1080
std::array<u32, 7> reserved_sw_method3_to_7; ///< 0x10B0
ReduceColorThreshold reduce_color_thresholds_unorm8; ///< 0x10CC
std::array<u32, 4> reserved_sw_method10_to_13; ///< 0x10D0

3
src/video_core/rasterizer_interface.h
View File

@@ -46,6 +46,9 @@ public:
/// Dispatches an indirect draw invocation
virtual void DrawIndirect() {}
/// Dispatches an draw texture invocation
virtual void DrawTexture() = 0;
/// Clear the current framebuffer
virtual void Clear(u32 layer_count) = 0;

1
src/video_core/renderer_null/null_rasterizer.cpp
View File

@@ -21,6 +21,7 @@ RasterizerNull::RasterizerNull(Core::Memory::Memory& cpu_memory_, Tegra::GPU& gp
RasterizerNull::~RasterizerNull() = default;
void RasterizerNull::Draw(bool is_indexed, u32 instance_count) {}
void RasterizerNull::DrawTexture() {}
void RasterizerNull::Clear(u32 layer_count) {}
void RasterizerNull::DispatchCompute() {}
void RasterizerNull::ResetCounter(VideoCore::QueryType type) {}

1
src/video_core/renderer_null/null_rasterizer.h
View File

@@ -31,6 +31,7 @@ public:
~RasterizerNull() override;
void Draw(bool is_indexed, u32 instance_count) override;
void DrawTexture() override;
void Clear(u32 layer_count) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;

1
src/video_core/renderer_opengl/gl_device.cpp
View File

@@ -166,6 +166,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
has_shader_int64 = HasExtension(extensions, "GL_ARB_gpu_shader_int64");
has_amd_shader_half_float = GLAD_GL_AMD_gpu_shader_half_float;
has_sparse_texture_2 = GLAD_GL_ARB_sparse_texture2;
has_draw_texture = GLAD_GL_NV_draw_texture;
warp_size_potentially_larger_than_guest = !is_nvidia && !is_intel;
need_fastmath_off = is_nvidia;
can_report_memory = GLAD_GL_NVX_gpu_memory_info;

5
src/video_core/renderer_opengl/gl_device.h
View File

@@ -146,6 +146,10 @@ public:
return has_sparse_texture_2;
}
bool HasDrawTexture() const {
return has_draw_texture;
}
bool IsWarpSizePotentiallyLargerThanGuest() const {
return warp_size_potentially_larger_than_guest;
}
@@ -216,6 +220,7 @@ private:
bool has_shader_int64{};
bool has_amd_shader_half_float{};
bool has_sparse_texture_2{};
bool has_draw_texture{};
bool warp_size_potentially_larger_than_guest{};
bool need_fastmath_off{};
bool has_cbuf_ftou_bug{};

34
src/video_core/renderer_opengl/gl_rasterizer.cpp
View File

@@ -318,6 +318,40 @@ void RasterizerOpenGL::DrawIndirect() {
buffer_cache.SetDrawIndirect(nullptr);
}
void RasterizerOpenGL::DrawTexture() {
MICROPROFILE_SCOPE(OpenGL_Drawing);
SCOPE_EXIT({ gpu.TickWork(); });
query_cache.UpdateCounters();
texture_cache.SynchronizeGraphicsDescriptors();
texture_cache.UpdateRenderTargets(false);
state_tracker.BindFramebuffer(texture_cache.GetFramebuffer()->Handle());
SyncState();
const auto& draw_texture_state = maxwell3d->draw_manager->GetDrawTextureState();
const auto& sampler = texture_cache.GetGraphicsSampler(draw_texture_state.src_sampler);
const auto& texture = texture_cache.GetImageView(draw_texture_state.src_texture);
if (device.HasDrawTexture()) {
glDrawTextureNV(texture.DefaultHandle(), sampler->Handle(),
static_cast<float>(draw_texture_state.dst_x0),
static_cast<float>(draw_texture_state.dst_y0),
static_cast<float>(draw_texture_state.dst_x1),
static_cast<float>(draw_texture_state.dst_y1), 0,
static_cast<float>(draw_texture_state.src_x0) / texture.size.width,
static_cast<float>(draw_texture_state.src_y0) / texture.size.height,
static_cast<float>(draw_texture_state.src_x1) / texture.size.width,
static_cast<float>(draw_texture_state.src_y1) / texture.size.height);
} else {
UNIMPLEMENTED();
}
++num_queued_commands;
}
void RasterizerOpenGL::DispatchCompute() {
ComputePipeline* const pipeline{shader_cache.CurrentComputePipeline()};
if (!pipeline) {

1
src/video_core/renderer_opengl/gl_rasterizer.h
View File

@@ -70,6 +70,7 @@ public:
void Draw(bool is_indexed, u32 instance_count) override;
void DrawIndirect() override;
void DrawTexture() override;
void Clear(u32 layer_count) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;

4
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -265,6 +265,10 @@ void RasterizerVulkan::DrawIndirect() {
buffer_cache.SetDrawIndirect(nullptr);
}
void RasterizerVulkan::DrawTexture() {
}
void RasterizerVulkan::Clear(u32 layer_count) {
MICROPROFILE_SCOPE(Vulkan_Clearing);

1
src/video_core/renderer_vulkan/vk_rasterizer.h
View File

@@ -66,6 +66,7 @@ public:
void Draw(bool is_indexed, u32 instance_count) override;
void DrawIndirect() override;
void DrawTexture() override;
void Clear(u32 layer_count) override;
void DispatchCompute() override;
void ResetCounter(VideoCore::QueryType type) override;

7
src/video_core/texture_cache/texture_cache.h
View File

@@ -148,6 +148,13 @@ typename P::ImageView& TextureCache<P>::GetImageView(ImageViewId id) noexcept {
return slot_image_views[id];
}
template <class P>
typename P::ImageView& TextureCache<P>::GetImageView(u32 index) noexcept {
const auto image_view_id = VisitImageView(channel_state->graphics_image_table,
channel_state->graphics_image_view_ids, index);
return slot_image_views[image_view_id];
}
template <class P>
void TextureCache<P>::MarkModification(ImageId id) noexcept {
MarkModification(slot_images[id]);

3
src/video_core/texture_cache/texture_cache_base.h
View File

@@ -129,6 +129,9 @@ public:
/// Return a reference to the given image view id
[[nodiscard]] ImageView& GetImageView(ImageViewId id) noexcept;
/// Get the imageview from the graphics descriptor table in the specified index
[[nodiscard]] ImageView& GetImageView(u32 index) noexcept;
/// Mark an image as modified from the GPU
void MarkModification(ImageId id) noexcept;