From cdddd71d0890a9669ea5003207bf1390db8622ad Mon Sep 17 00:00:00 2001 From: bunnei Date: Tue, 11 Sep 2018 20:12:29 -0400 Subject: [PATCH 1/2] gl_shader_cache: Remove cache_width/cache_height. - This was once an optimization, but we no longer need it with the cache reserve. - This is also inaccurate. --- src/video_core/renderer_opengl/gl_rasterizer_cache.cpp | 6 ------ src/video_core/renderer_opengl/gl_rasterizer_cache.h | 8 ++------ 2 files changed, 2 insertions(+), 12 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 29d61eccd..0d7a38e3e 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -53,8 +53,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) { params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format)); params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format)); 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); switch (params.target) { @@ -89,8 +87,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) { params.width = config.width; params.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.depth = 1; params.size_in_bytes = params.SizeInBytes(); @@ -110,8 +106,6 @@ static VAddr TryGetCpuAddr(Tegra::GPUVAddr gpu_addr) { params.width = zeta_width; params.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.depth = 1; params.size_in_bytes = params.SizeInBytes(); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index e660998d0..57ea8593b 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -680,8 +680,8 @@ struct SurfaceParams { /// Checks if surfaces are compatible for caching bool IsCompatibleSurface(const SurfaceParams& other) const { - return std::tie(pixel_format, type, cache_width, cache_height) == - std::tie(other.pixel_format, other.type, other.cache_width, other.cache_height); + return std::tie(pixel_format, type, width, height) == + std::tie(other.pixel_format, other.type, other.width, other.height); } VAddr addr; @@ -696,10 +696,6 @@ struct SurfaceParams { u32 unaligned_height; size_t size_in_bytes; SurfaceTarget target; - - // Parameters used for caching only - u32 cache_width; - u32 cache_height; }; }; // namespace OpenGL From 7bb226f22da31bcf4aeeaaf7ac07a3ee77347e20 Mon Sep 17 00:00:00 2001 From: bunnei Date: Tue, 11 Sep 2018 20:14:00 -0400 Subject: [PATCH 2/2] gl_rasterizer_cache: Always blit on recreate, regardless of format. - Fixes several rendering issues with Super Mario Odyssey. --- .../renderer_opengl/gl_rasterizer_cache.cpp | 16 ++++++++++------ 1 file changed, 10 insertions(+), 6 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 0d7a38e3e..3b38565f4 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -811,16 +811,20 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& surface, // Get a new surface with the new parameters, and blit the previous surface to it 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, params.GetRect(), params.type, read_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 dest_format = GetFormatTuple(new_params.pixel_format, new_params.component_type);