N-archive/citra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Merge pull request #5905 from SachinVin/intel

Browse source 
Remove CopyImageSubData; and enable TextureView reinterpreter only on nvidia
This commit is contained in:
bunnei 2022-01-04 17:41:14 -08:00 committed by GitHub
commit 83913e68f3
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 12 additions and 33 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
src/video_core/renderer_opengl/gl_format_reinterpreter.cpp
View file

@ -282,10 +282,11 @@ void main() {
state.draw.shader_program = cur_program; state.draw.shader_program = cur_program;
state.Apply(); state.Apply();
// OES_texture_view doesn't seem to support D24S8 views, at least on adreno // Nvidia seem to be the only one to support D24S8 views, at least on windows
// so instead it will do an intermediate copy before running through the shader // so for everyone else it will do an intermediate copy before running through the shader
if (GLAD_GL_ARB_texture_view) { std::string_view vendor{reinterpret_cast<const char*>(glGetString(GL_VENDOR))};
texture_view_func = glTextureView; if (vendor.find("NVIDIA") != vendor.npos) {
use_texture_view = true;
} else { } else {
LOG_INFO(Render_OpenGL, LOG_INFO(Render_OpenGL,
"Texture views are unsupported, reinterpretation will do intermediate copy"); "Texture views are unsupported, reinterpretation will do intermediate copy");
@ -302,11 +303,10 @@ void main() {
OpenGLState state; OpenGLState state;
state.texture_units[0].texture_2d = src_tex; state.texture_units[0].texture_2d = src_tex;
if (texture_view_func) { if (use_texture_view) {
temp_tex.Create(); temp_tex.Create();
glActiveTexture(GL_TEXTURE1); glActiveTexture(GL_TEXTURE1);
texture_view_func(temp_tex.handle, GL_TEXTURE_2D, src_tex, GL_DEPTH24_STENCIL8, 0, 1, 0, glTextureView(temp_tex.handle, GL_TEXTURE_2D, src_tex, GL_DEPTH24_STENCIL8, 0, 1, 0, 1);
1);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_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) { } else if (src_rect.top > temp_rect.top || src_rect.right > temp_rect.right) {
@ -331,7 +331,7 @@ void main() {
state.Apply(); state.Apply();
glActiveTexture(GL_TEXTURE1); glActiveTexture(GL_TEXTURE1);
if (!texture_view_func) { if (!use_texture_view) {
glCopyImageSubData(src_tex, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0, 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, temp_tex.handle, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0,
src_rect.GetWidth(), src_rect.GetHeight(), 1); src_rect.GetWidth(), src_rect.GetHeight(), 1);
@ -348,13 +348,13 @@ void main() {
glUniform2i(src_offset_loc, src_rect.left, src_rect.bottom); glUniform2i(src_offset_loc, src_rect.left, src_rect.bottom);
glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
if (texture_view_func) { if (use_texture_view) {
temp_tex.Release(); temp_tex.Release();
} }
} }
private: private:
decltype(glTextureView) texture_view_func = nullptr; bool use_texture_view{};
OGLProgram program{}; OGLProgram program{};
GLint dst_size_loc{-1}, src_size_loc{-1}, src_offset_loc{-1}; GLint dst_size_loc{-1}, src_size_loc{-1}, src_offset_loc{-1};
OGLVertexArray vao{}; OGLVertexArray vao{};
@ -362,30 +362,9 @@ private:
Common::Rectangle<u32> temp_rect{0, 0, 0, 0}; 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() {
// Older Nvidia GPUs don't seem to properly support using glCopyImageSubData to copy D24S8 to if ((GLAD_GL_ARB_stencil_texturing && GLAD_GL_ARB_texture_storage && GLAD_GL_ARB_copy_image) ||
// RGBA8. This is a heuristic check that relies on the newer drivers returning something similar GLES) {
// to `3.3.0 NVIDIA 471.41`, and older, buggy ones returning just `3.3.0`.
std::string_view version{reinterpret_cast<const char*>(glGetString(GL_VERSION))};
if (version.find("NVIDIA") != version.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 &&
GLAD_GL_ARB_copy_image) ||
GLES) {
reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8}, reinterpreters.emplace(PixelFormatPair{PixelFormat::RGBA8, PixelFormat::D24S8},
std::make_unique<ShaderD24S8toRGBA8>()); std::make_unique<ShaderD24S8toRGBA8>());
LOG_INFO(Render_OpenGL, "Using shader for D24S8 to RGBA8 reinterpretation"); LOG_INFO(Render_OpenGL, "Using shader for D24S8 to RGBA8 reinterpretation");