aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/renderer_opengl/util_shaders.cpp
diff options
context:
space:
mode:
authorameerj <52414509+ameerj@users.noreply.github.com>2021-07-04 22:48:41 -0400
committerameerj <52414509+ameerj@users.noreply.github.com>2021-07-31 21:36:26 -0400
commit5665d055476fa793192523c3cb6fe06369d58674 (patch)
tree45a3ed50774c2106dcbeee9d38415f057d64a6fe /src/video_core/renderer_opengl/util_shaders.cpp
parent15c0c213b1efb63f1d6f4900409fca8c8984e973 (diff)
astc_decoder: Optimize the use EncodingData
This buffer was a list of EncodingData structures sorted by their bit length, with some duplication from the cpu decoder implementation. We can take advantage of its sorted property to optimize its usage in the shader. Thanks to wwylele for the optimization idea.
Diffstat (limited to 'src/video_core/renderer_opengl/util_shaders.cpp')
-rw-r--r--src/video_core/renderer_opengl/util_shaders.cpp13
1 files changed, 4 insertions, 9 deletions
diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp
index 37a4d1d9d..a2b264700 100644
--- a/src/video_core/renderer_opengl/util_shaders.cpp
+++ b/src/video_core/renderer_opengl/util_shaders.cpp
@@ -60,19 +60,15 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
swizzle_table_buffer.Create();
- astc_buffer.Create();
glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
- glNamedBufferStorage(astc_buffer.handle, sizeof(ASTC_ENCODINGS_VALUES), &ASTC_ENCODINGS_VALUES,
- 0);
}
UtilShaders::~UtilShaders() = default;
void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
std::span<const VideoCommon::SwizzleParameters> swizzles) {
- static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
- static constexpr GLuint BINDING_INPUT_BUFFER = 1;
- static constexpr GLuint BINDING_ENC_BUFFER = 2;
+ static constexpr GLuint BINDING_INPUT_BUFFER = 0;
+ static constexpr GLuint BINDING_SWIZZLE_BUFFER = 1;
static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;
const Extent2D tile_size{
@@ -81,7 +77,6 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
};
program_manager.BindComputeProgram(astc_decoder_program.handle);
glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
- glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle);
glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
glUniform2ui(1, tile_size.width, tile_size.height);
@@ -103,11 +98,11 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
glUniform1ui(6, params.block_height);
glUniform1ui(7, params.block_height_mask);
- glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
- GL_WRITE_ONLY, GL_RGBA8);
// ASTC texture data
glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset);
+ glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
+ GL_WRITE_ONLY, GL_RGBA8);
glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
}