From 6b333d862bf220e912364dfff2e337449a5e0e0a Mon Sep 17 00:00:00 2001 From: bunnei Date: Wed, 17 Oct 2018 20:32:29 -0400 Subject: [PATCH] fermi_2d: Add support for more accurate surface copies. --- src/video_core/engines/fermi_2d.cpp | 9 ++++++--- src/video_core/renderer_opengl/gl_rasterizer.cpp | 6 ++++++ 2 files changed, 12 insertions(+), 3 deletions(-) diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index 597b279b95..74e44c7fe5 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -47,9 +47,12 @@ void Fermi2D::HandleSurfaceCopy() { u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format); if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst)) { - // TODO(bunnei): The below implementation currently will not get hit, as - // AccelerateSurfaceCopy tries to always copy and will always return success. This should be - // changed once we properly support flushing. + rasterizer.FlushRegion(source_cpu, src_bytes_per_pixel * regs.src.width * regs.src.height); + // We have to invalidate the destination region to evict any outdated surfaces from the + // cache. We do this before actually writing the new data because the destination address + // might contain a dirty surface that will have to be written back to memory. + rasterizer.InvalidateRegion(dest_cpu, + dst_bytes_per_pixel * regs.dst.width * regs.dst.height); if (regs.src.linear == regs.dst.linear) { // If the input layout and the output layout are the same, just perform a raw copy. diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 4682530333..3daccf82f0 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -659,6 +659,12 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) { bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src, const Tegra::Engines::Fermi2D::Regs::Surface& dst) { MICROPROFILE_SCOPE(OpenGL_Blits); + + if (Settings::values.use_accurate_gpu_emulation) { + // Skip the accelerated copy and perform a slow but more accurate copy + return false; + } + res_cache.FermiCopySurface(src, dst); return true; }