summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/video_core/engines/maxwell_3d.cpp12
-rw-r--r--src/video_core/engines/maxwell_3d.h3
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer.cpp2
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer_cache.cpp18
-rw-r--r--src/video_core/renderer_opengl/gl_rasterizer_cache.h24
-rw-r--r--src/video_core/renderer_opengl/gl_shader_decompiler.cpp2
-rw-r--r--src/video_core/renderer_opengl/gl_shader_gen.cpp2
-rw-r--r--src/video_core/renderer_opengl/gl_shader_manager.cpp7
-rw-r--r--src/video_core/renderer_opengl/gl_shader_manager.h5
9 files changed, 54 insertions, 21 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index a46ed4bd7..68f91cc75 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -222,6 +222,18 @@ void Maxwell3D::DrawArrays() {
debug_context->OnEvent(Tegra::DebugContext::Event::FinishedPrimitiveBatch, nullptr);
}
+ // Both instance configuration registers can not be set at the same time.
+ ASSERT_MSG(!regs.draw.instance_next || !regs.draw.instance_cont,
+ "Illegal combination of instancing parameters");
+
+ if (regs.draw.instance_next) {
+ // Increment the current instance *before* drawing.
+ state.current_instance += 1;
+ } else if (!regs.draw.instance_cont) {
+ // Reset the current instance to 0.
+ state.current_instance = 0;
+ }
+
const bool is_indexed{regs.index_array.count && !regs.vertex_buffer.count};
rasterizer.AccelerateDrawBatch(is_indexed);
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 1b30ce018..771eb5abc 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -638,6 +638,8 @@ public:
union {
u32 vertex_begin_gl;
BitField<0, 16, PrimitiveTopology> topology;
+ BitField<26, 1, u32> instance_next;
+ BitField<27, 1, u32> instance_cont;
};
} draw;
@@ -830,6 +832,7 @@ public:
};
std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages;
+ u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering.
};
State state{};
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 9d1549fe9..93eadde7a 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -124,7 +124,7 @@ std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
glBindVertexBuffer(index, stream_buffer.GetHandle(), vertex_buffer_offset,
vertex_array.stride);
- ASSERT_MSG(vertex_array.divisor == 0, "Vertex buffer divisor unimplemented");
+ ASSERT_MSG(vertex_array.divisor == 0, "Instanced vertex arrays are not supported");
}
// Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index b6947b97b..ac4db82cf 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -142,14 +142,16 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form
{GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // RG32UI
{GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // R32UI
+ // Depth formats
+ {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F
+ {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT, ComponentType::UNorm,
+ false}, // Z16
+
// DepthStencil formats
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, ComponentType::UNorm,
false}, // Z24S8
{GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8, ComponentType::UNorm,
- false}, // S8Z24
- {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F
- {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT, ComponentType::UNorm,
- false}, // Z16
+ false}, // S8Z24
{GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, GL_FLOAT_32_UNSIGNED_INT_24_8_REV,
ComponentType::Float, false}, // Z32FS8
}};
@@ -283,10 +285,10 @@ static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPU
MortonCopy<true, PixelFormat::RG8S>,
MortonCopy<true, PixelFormat::RG32UI>,
MortonCopy<true, PixelFormat::R32UI>,
- MortonCopy<true, PixelFormat::Z24S8>,
- MortonCopy<true, PixelFormat::S8Z24>,
MortonCopy<true, PixelFormat::Z32F>,
MortonCopy<true, PixelFormat::Z16>,
+ MortonCopy<true, PixelFormat::Z24S8>,
+ MortonCopy<true, PixelFormat::S8Z24>,
MortonCopy<true, PixelFormat::Z32FS8>,
// clang-format on
};
@@ -339,10 +341,10 @@ static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPU
MortonCopy<false, PixelFormat::RG8S>,
MortonCopy<false, PixelFormat::RG32UI>,
MortonCopy<false, PixelFormat::R32UI>,
- MortonCopy<false, PixelFormat::Z24S8>,
- MortonCopy<false, PixelFormat::S8Z24>,
MortonCopy<false, PixelFormat::Z32F>,
MortonCopy<false, PixelFormat::Z16>,
+ MortonCopy<false, PixelFormat::Z24S8>,
+ MortonCopy<false, PixelFormat::S8Z24>,
MortonCopy<false, PixelFormat::Z32FS8>,
// clang-format on
};
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
index 55cf3782c..beec01746 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -68,11 +68,15 @@ struct SurfaceParams {
MaxColorFormat,
+ // Depth formats
+ Z32F = 42,
+ Z16 = 43,
+
+ MaxDepthFormat,
+
// DepthStencil formats
- Z24S8 = 42,
- S8Z24 = 43,
- Z32F = 44,
- Z16 = 45,
+ Z24S8 = 44,
+ S8Z24 = 45,
Z32FS8 = 46,
MaxDepthStencilFormat,
@@ -153,10 +157,10 @@ struct SurfaceParams {
1, // RG8S
1, // RG32UI
1, // R32UI
- 1, // Z24S8
- 1, // S8Z24
1, // Z32F
1, // Z16
+ 1, // Z24S8
+ 1, // S8Z24
1, // Z32FS8
}};
@@ -211,10 +215,10 @@ struct SurfaceParams {
16, // RG8S
64, // RG32UI
32, // R32UI
- 32, // Z24S8
- 32, // S8Z24
32, // Z32F
16, // Z16
+ 32, // Z24S8
+ 32, // S8Z24
64, // Z32FS8
}};
@@ -587,6 +591,10 @@ struct SurfaceParams {
return SurfaceType::ColorTexture;
}
+ if (static_cast<size_t>(pixel_format) < static_cast<size_t>(PixelFormat::MaxDepthFormat)) {
+ return SurfaceType::Depth;
+ }
+
if (static_cast<size_t>(pixel_format) <
static_cast<size_t>(PixelFormat::MaxDepthStencilFormat)) {
return SurfaceType::DepthStencil;
diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
index e0dfdbb9f..07006b55e 100644
--- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp
@@ -541,7 +541,7 @@ private:
// vertex shader, and what's the value of the fourth element when inside a Tess Eval
// shader.
ASSERT(stage == Maxwell3D::Regs::ShaderStage::Vertex);
- return "vec4(0, 0, uintBitsToFloat(gl_InstanceID), uintBitsToFloat(gl_VertexID))";
+ return "vec4(0, 0, uintBitsToFloat(instance_id.x), uintBitsToFloat(gl_VertexID))";
default:
const u32 index{static_cast<u32>(attribute) -
static_cast<u32>(Attribute::Index::Attribute_0)};
diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp
index 129c777d1..57e0e1726 100644
--- a/src/video_core/renderer_opengl/gl_shader_gen.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp
@@ -38,6 +38,7 @@ out vec4 position;
layout (std140) uniform vs_config {
vec4 viewport_flip;
+ uvec4 instance_id;
};
void main() {
@@ -90,6 +91,7 @@ out vec4 color;
layout (std140) uniform fs_config {
vec4 viewport_flip;
+ uvec4 instance_id;
};
void main() {
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.cpp b/src/video_core/renderer_opengl/gl_shader_manager.cpp
index 415d42fda..f0886caac 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_manager.cpp
@@ -37,11 +37,16 @@ void SetShaderUniformBlockBindings(GLuint shader) {
} // namespace Impl
void MaxwellUniformData::SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage) {
- const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+ const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+ const auto& regs = gpu.regs;
+ const auto& state = gpu.state;
// TODO(bunnei): Support more than one viewport
viewport_flip[0] = regs.viewport_transform[0].scale_x < 0.0 ? -1.0f : 1.0f;
viewport_flip[1] = regs.viewport_transform[0].scale_y < 0.0 ? -1.0f : 1.0f;
+
+ // We only assign the instance to the first component of the vector, the rest is just padding.
+ instance_id[0] = state.current_instance;
}
} // namespace GLShader
diff --git a/src/video_core/renderer_opengl/gl_shader_manager.h b/src/video_core/renderer_opengl/gl_shader_manager.h
index 716933a0b..75fa73605 100644
--- a/src/video_core/renderer_opengl/gl_shader_manager.h
+++ b/src/video_core/renderer_opengl/gl_shader_manager.h
@@ -24,14 +24,15 @@ void SetShaderUniformBlockBindings(GLuint shader);
} // namespace Impl
/// Uniform structure for the Uniform Buffer Object, all vectors must be 16-byte aligned
-// NOTE: Always keep a vec4 at the end. The GL spec is not clear wether the alignment at
+// NOTE: Always keep a vec4 at the end. The GL spec is not clear whether the alignment at
// the end of a uniform block is included in UNIFORM_BLOCK_DATA_SIZE or not.
// Not following that rule will cause problems on some AMD drivers.
struct MaxwellUniformData {
void SetFromRegs(const Maxwell3D::State::ShaderStageInfo& shader_stage);
alignas(16) GLvec4 viewport_flip;
+ alignas(16) GLuvec4 instance_id;
};
-static_assert(sizeof(MaxwellUniformData) == 16, "MaxwellUniformData structure size is incorrect");
+static_assert(sizeof(MaxwellUniformData) == 32, "MaxwellUniformData structure size is incorrect");
static_assert(sizeof(MaxwellUniformData) < 16384,
"MaxwellUniformData structure must be less than 16kb as per the OpenGL spec");
generated by cgit v1.2.3 (git 2.25.1) at 2024-09-15 21:06:32 +0200