midou/citra
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

video_core: implement optimized D24S8->RGBA8 reinterpreters

Browse Source
This commit is contained in:
BreadFish64 2020-04-07 17:09:05 -05:00 committed by SachinVin
parent fda121c25c
commit e6b59cb5d8
1 changed files with 168 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

170
src/video_core/renderer_opengl/gl_format_reinterpreter.cpp
View File

@ -220,9 +220,175 @@ private:
GLint d24s8_abgr_viewport_u_id; GLint d24s8_abgr_viewport_u_id;
}; };
class ShaderD24S8toRGBA8 final : public FormatReinterpreterBase {
public:
ShaderD24S8toRGBA8() {
constexpr std::string_view vs_source = R"(
out vec2 dst_coord;
uniform mediump ivec2 dst_size;
const vec2 vertices[4] =
vec2[4](vec2(-1.0, -1.0), vec2(1.0, -1.0), vec2(-1.0, 1.0), vec2(1.0, 1.0));
void main() {
gl_Position = vec4(vertices[gl_VertexID], 0.0, 1.0);
dst_coord = (vertices[gl_VertexID] / 2.0 + 0.5) * vec2(dst_size);
}
)";
constexpr std::string_view fs_source = R"(
in mediump vec2 dst_coord;
out lowp vec4 frag_color;
uniform highp sampler2D depth;
uniform lowp usampler2D stencil;
uniform mediump ivec2 dst_size;
uniform mediump ivec2 src_size;
uniform mediump ivec2 src_offset;
void main() {
mediump ivec2 tex_coord;
if (src_size == dst_size) {
tex_coord = ivec2(dst_coord);
} else {
highp int tex_index = int(dst_coord.y) * dst_size.x + int(dst_coord.x);
mediump int y = tex_index / src_size.x;
tex_coord = ivec2(tex_index - y * src_size.x, y);
}
tex_coord -= src_offset;
highp uint depth_val =
uint(texelFetch(depth, tex_coord, 0).x * (exp2(32.0) - 1.0));
lowp uint stencil_val = texelFetch(stencil, tex_coord, 0).x;
highp uvec4 components =
uvec4(stencil_val, (uvec3(depth_val) >> uvec3(24u, 16u, 8u)) & 0x000000FFu);
frag_color = vec4(components) / (exp2(8.0) - 1.0);
}
)";
program.Create(vs_source.data(), fs_source.data());
dst_size_loc = glGetUniformLocation(program.handle, "dst_size");
src_size_loc = glGetUniformLocation(program.handle, "src_size");
src_offset_loc = glGetUniformLocation(program.handle, "src_offset");
vao.Create();
auto state = OpenGLState::GetCurState();
auto cur_program = state.draw.shader_program;
state.draw.shader_program = program.handle;
state.Apply();
glUniform1i(glGetUniformLocation(program.handle, "stencil"), 1);
state.draw.shader_program = cur_program;
state.Apply();
// OES_texture_view doesn't seem to support D24S8 views, at least on adreno
// so instead it will do an intermediate copy before running through the shader
if (GLAD_GL_ARB_texture_view) {
texture_view_func = glTextureView;
} else {
LOG_INFO(Render_OpenGL,
"Texture views are unsupported, reinterpretation will do intermediate copy");
temp_tex.Create();
}
}
void Reinterpret(GLuint src_tex, const Common::Rectangle<u32>& src_rect, GLuint read_fb_handle,
GLuint dst_tex, const Common::Rectangle<u32>& dst_rect,
GLuint draw_fb_handle) override {
OpenGLState prev_state = OpenGLState::GetCurState();
SCOPE_EXIT({ prev_state.Apply(); });
OpenGLState state;
state.texture_units[0].texture_2d = src_tex;
if (texture_view_func) {
temp_tex.Create();
glActiveTexture(GL_TEXTURE1);
texture_view_func(temp_tex.handle, GL_TEXTURE_2D, src_tex, GL_DEPTH24_STENCIL8, 0, 1, 0,
1);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
} else if (src_rect.top > temp_rect.top || src_rect.right > temp_rect.right) {
temp_tex.Release();
temp_tex.Create();
state.texture_units[1].texture_2d = temp_tex.handle;
state.Apply();
glActiveTexture(GL_TEXTURE1);
glTexStorage2D(GL_TEXTURE_2D, 1, GL_DEPTH24_STENCIL8, src_rect.right, src_rect.top);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
temp_rect = src_rect;
}
state.texture_units[1].texture_2d = temp_tex.handle;
state.draw.draw_framebuffer = draw_fb_handle;
state.draw.shader_program = program.handle;
state.draw.vertex_array = vao.handle;
state.viewport = {static_cast<GLint>(dst_rect.left), static_cast<GLint>(dst_rect.bottom),
static_cast<GLsizei>(dst_rect.GetWidth()),
static_cast<GLsizei>(dst_rect.GetHeight())};
state.Apply();
glActiveTexture(GL_TEXTURE1);
if (!texture_view_func) {
glCopyImageSubData(src_tex, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0,
temp_tex.handle, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0,
src_rect.GetWidth(), src_rect.GetHeight(), 1);
}
glTexParameteri(GL_TEXTURE_2D, GL_DEPTH_STENCIL_TEXTURE_MODE, GL_STENCIL_INDEX);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex,
0);
glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
0);
glUniform2i(dst_size_loc, dst_rect.GetWidth(), dst_rect.GetHeight());
glUniform2i(src_size_loc, src_rect.GetWidth(), src_rect.GetHeight());
glUniform2i(src_offset_loc, src_rect.left, src_rect.bottom);
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
if (texture_view_func) {
temp_tex.Release();
}
}
private:
decltype(glTextureView) texture_view_func = nullptr;
OGLProgram program{};
GLint dst_size_loc{-1}, src_size_loc{-1}, src_offset_loc{-1};
OGLVertexArray vao{};
OGLTexture temp_tex{};
Common::Rectangle<u32> temp_rect{0, 0, 0, 0};
};
class CopyImageSubData final : public FormatReinterpreterBase {
void Reinterpret(GLuint src_tex, const Common::Rectangle<u32>& src_rect, GLuint read_fb_handle,
GLuint dst_tex, const Common::Rectangle<u32>& dst_rect,
GLuint draw_fb_handle) override {
glCopyImageSubData(src_tex, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0, dst_tex,
GL_TEXTURE_2D, 0, dst_rect.left, dst_rect.bottom, 0, src_rect.GetWidth(),
src_rect.GetHeight(), 1);
}
};
FormatReinterpreterOpenGL::FormatReinterpreterOpenGL() { FormatReinterpreterOpenGL::FormatReinterpreterOpenGL() {
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8}, std::string_view vendor{reinterpret_cast<const char*>(glGetString(GL_VENDOR))};
std::make_unique<PixelBufferD24S8toABGR>()); if (vendor.find("NVIDIA") != vendor.npos) {
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8},
std::make_unique<CopyImageSubData>());
// Nvidia bends the spec and allows direct copies between color and depth formats
// might as well take advantage of it
LOG_INFO(Render_OpenGL, "Using glCopyImageSubData for D24S8 to RGBA8 reinterpretation");
} else if ((GLAD_GL_ARB_stencil_texturing && GLAD_GL_ARB_texture_storage) || GLES) {
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8},
std::make_unique<ShaderD24S8toRGBA8>());
LOG_INFO(Render_OpenGL, "Using shader for D24S8 to RGBA8 reinterpretation");
} else {
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8},
std::make_unique<PixelBufferD24S8toABGR>());
LOG_INFO(Render_OpenGL, "Using pbo for D24S8 to RGBA8 reinterpretation");
}
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGB5A1, PixelFormat::RGBA4}, reinterpreters.emplace(PixelFormatPair{PixelFormat::RGB5A1, PixelFormat::RGBA4},
std::make_unique<RGBA4toRGB5A1>()); std::make_unique<RGBA4toRGB5A1>());
} }