sneedmca/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu synced 2024-11-04 22:37:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Merge pull request #1295 from bunnei/accurate-copies

Browse source 
gl_rasterizer_cache: Improve accuracy of caching and copies.
This commit is contained in:
bunnei 2018-09-11 23:12:15 -04:00 committed by GitHub
commit 522a11a11f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 12 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
View file

@ -53,8 +53,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format)); params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format));
params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format)); params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format));
params.unaligned_height = config.tic.Height(); params.unaligned_height = config.tic.Height();
params.cache_width = Common::AlignUp(params.width, 8);
params.cache_height = Common::AlignUp(params.height, 8);
params.target = SurfaceTargetFromTextureType(config.tic.texture_type); params.target = SurfaceTargetFromTextureType(config.tic.texture_type);
switch (params.target) { switch (params.target) {
@ -89,8 +87,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.width = config.width; params.width = config.width;
params.height = config.height; params.height = config.height;
params.unaligned_height = config.height; params.unaligned_height = config.height;
params.cache_width = Common::AlignUp(params.width, 8);
params.cache_height = Common::AlignUp(params.height, 8);
params.target = SurfaceTarget::Texture2D; params.target = SurfaceTarget::Texture2D;
params.depth = 1; params.depth = 1;
params.size_in_bytes = params.SizeInBytes(); params.size_in_bytes = params.SizeInBytes();
@ -110,8 +106,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) {
params.width = zeta_width; params.width = zeta_width;
params.height = zeta_height; params.height = zeta_height;
params.unaligned_height = zeta_height; params.unaligned_height = zeta_height;
params.cache_width = Common::AlignUp(params.width, 8);
params.cache_height = Common::AlignUp(params.height, 8);
params.target = SurfaceTarget::Texture2D; params.target = SurfaceTarget::Texture2D;
params.depth = 1; params.depth = 1;
params.size_in_bytes = params.SizeInBytes(); params.size_in_bytes = params.SizeInBytes();
@ -814,16 +808,20 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& surface,
// Get a new surface with the new parameters, and blit the previous surface to it // Get a new surface with the new parameters, and blit the previous surface to it
Surface new_surface{GetUncachedSurface(new_params)}; Surface new_surface{GetUncachedSurface(new_params)};
// If format is unchanged, we can do a faster blit without reinterpreting pixel data if (params.pixel_format == new_params.pixel_format ||
if (params.pixel_format == new_params.pixel_format) { !Settings::values.use_accurate_framebuffers) {
// If the format is the same, just do a framebuffer blit. This is significantly faster than
// using PBOs. The is also likely less accurate, as textures will be converted rather than
// reinterpreted.
BlitTextures(surface->Texture().handle, params.GetRect(), new_surface->Texture().handle, BlitTextures(surface->Texture().handle, params.GetRect(), new_surface->Texture().handle,
params.GetRect(), params.type, read_framebuffer.handle, params.GetRect(), params.type, read_framebuffer.handle,
draw_framebuffer.handle); draw_framebuffer.handle);
return new_surface; } else {
} // When use_accurate_framebuffers setting is enabled, perform a more accurate surface copy,
// where pixels are reinterpreted as a new format (without conversion). This code path uses
// OpenGL PBOs and is quite slow.
// When using accurate framebuffers, always copy old data to new surface, regardless of format
if (Settings::values.use_accurate_framebuffers) {
auto source_format = GetFormatTuple(params.pixel_format, params.component_type); auto source_format = GetFormatTuple(params.pixel_format, params.component_type);
auto dest_format = GetFormatTuple(new_params.pixel_format, new_params.component_type); auto dest_format = GetFormatTuple(new_params.pixel_format, new_params.component_type);

8
src/video_core/renderer_opengl/gl_rasterizer_cache.h
View file

@ -680,8 +680,8 @@ struct SurfaceParams {
/// Checks if surfaces are compatible for caching /// Checks if surfaces are compatible for caching
bool IsCompatibleSurface(const SurfaceParams& other) const { bool IsCompatibleSurface(const SurfaceParams& other) const {
return std::tie(pixel_format, type, cache_width, cache_height) == return std::tie(pixel_format, type, width, height) ==
std::tie(other.pixel_format, other.type, other.cache_width, other.cache_height); std::tie(other.pixel_format, other.type, other.width, other.height);
} }
VAddr addr; VAddr addr;
@ -696,10 +696,6 @@ struct SurfaceParams {
u32 unaligned_height; u32 unaligned_height;
size_t size_in_bytes; size_t size_in_bytes;
SurfaceTarget target; SurfaceTarget target;
// Parameters used for caching only
u32 cache_width;
u32 cache_height;
}; };
}; // namespace OpenGL }; // namespace OpenGL