diff options
| -rw-r--r-- | src/video_core/engines/sw_blitter/blitter.cpp | 49 |
1 files changed, 35 insertions, 14 deletions
diff --git a/src/video_core/engines/sw_blitter/blitter.cpp b/src/video_core/engines/sw_blitter/blitter.cpp index e08204f223..5f066254e9 100644 --- a/src/video_core/engines/sw_blitter/blitter.cpp +++ b/src/video_core/engines/sw_blitter/blitter.cpp @@ -6,6 +6,7 @@ #include <vector> #include "common/alignment.h" +#include "common/literals.h" #include "common/scratch_buffer.h" #include "video_core/engines/sw_blitter/blitter.h" #include "video_core/engines/sw_blitter/converter.h" @@ -24,6 +25,7 @@ using VideoCore::Surface::PixelFormatFromRenderTargetFormat; namespace Tegra::Engines::Blitter { using namespace Texture; +using namespace Common::Literals; namespace { @@ -174,6 +176,22 @@ struct SoftwareBlitEngine::BlitEngineImpl { Common::ScratchBuffer<u8> dst_buffer; Common::ScratchBuffer<f32> intermediate_src; Common::ScratchBuffer<f32> intermediate_dst; + + // Buffers grown for one huge surface (64-256 MiB) are freed after the blit rather than kept + // for the rest of the session; the usual sizes stay. + void ReleaseLargeBuffers() { + constexpr size_t LargeCapacity = 16_MiB; + for (auto* const buffer : {&tmp_buffer, &src_buffer, &dst_buffer}) { + if (buffer->capacity() > LargeCapacity) { + buffer->release(); + } + } + for (auto* const buffer : {&intermediate_src, &intermediate_dst}) { + if (buffer->capacity() * sizeof(f32) > LargeCapacity) { + buffer->release(); + } + } + } ConverterFactory converter_factory; }; @@ -268,20 +286,23 @@ bool SoftwareBlitEngine::Blit(Fermi2D::Surface& src, Fermi2D::Surface& dst, const size_t dst_size = get_surface_size(dst, dst_bytes_per_pixel); const SubSurface dst_part = CoveredRows(dst, dst_bytes_per_pixel, config.dst_x0, config.dst_y0, dst_extent_x, dst_extent_y, dst_size); - Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::SafeReadWrite> - tmp_buffer2(memory_manager, dst.Address() + dst_part.offset, dst_part.size, - &impl->tmp_buffer); - - if (dst.linear == Fermi2D::MemoryLayout::BlockLinear) { - SwizzleSubrect(tmp_buffer2, impl->dst_buffer, dst_bytes_per_pixel, dst.width, - dst_part.height, dst.depth, config.dst_x0, dst_part.origin_y, dst_extent_x, - dst_extent_y, dst.block_height, dst.block_depth, - dst_extent_x * dst_bytes_per_pixel); - } else { - ProcessPitchLinear<true>(impl->dst_buffer, tmp_buffer2, dst_extent_x, dst_extent_y, - dst.pitch, config.dst_x0, dst_part.origin_y, - static_cast<size_t>(dst_bytes_per_pixel)); - } + { + Tegra::Memory::GpuGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::SafeReadWrite> + tmp_buffer2(memory_manager, dst.Address() + dst_part.offset, dst_part.size, + &impl->tmp_buffer); + + if (dst.linear == Fermi2D::MemoryLayout::BlockLinear) { + SwizzleSubrect(tmp_buffer2, impl->dst_buffer, dst_bytes_per_pixel, dst.width, + dst_part.height, dst.depth, config.dst_x0, dst_part.origin_y, + dst_extent_x, dst_extent_y, dst.block_height, dst.block_depth, + dst_extent_x * dst_bytes_per_pixel); + } else { + ProcessPitchLinear<true>(impl->dst_buffer, tmp_buffer2, dst_extent_x, dst_extent_y, + dst.pitch, config.dst_x0, dst_part.origin_y, + static_cast<size_t>(dst_bytes_per_pixel)); + } + } // tmp_buffer2 writes the destination back here + impl->ReleaseLargeBuffers(); return true; } |