From e0ea2f5f6e7c177cf0944b32acb7e24234953951 Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Thu, 18 Oct 2018 15:04:33 -0400 Subject: [PATCH] Fixed Layered Textures Loading and Cubemaps --- .../renderer_opengl/gl_rasterizer_cache.cpp | 135 +++++++++--------- .../renderer_opengl/gl_rasterizer_cache.h | 40 ++++++ src/video_core/textures/decoders.cpp | 6 +- 3 files changed, 109 insertions(+), 72 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 9c89253836..591ec7998b 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -78,6 +78,29 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { } } +std::size_t SurfaceParams::InnerMemorySize(bool layer_only) const { + const u32 compression_factor{GetCompressionFactor(pixel_format)}; + const u32 bytes_per_pixel{GetBytesPerPixel(pixel_format)}; + u32 m_depth = (layer_only ? 1U : depth); + u32 m_width = std::max(1U, width / compression_factor); + u32 m_height = std::max(1U, height / compression_factor); + std::size_t size = Tegra::Texture::CalculateSize(is_tiled, bytes_per_pixel, m_width, m_height, + m_depth, block_height, block_depth); + u32 m_block_height = block_height; + u32 m_block_depth = block_depth; + std::size_t block_size_bytes = 512 * block_height * block_depth; // 512 is GOB size + for (u32 i = 1; i < max_mip_level; i++) { + m_width = std::max(1U, m_width / 2); + m_height = std::max(1U, m_height / 2); + m_depth = std::max(1U, m_depth / 2); + m_block_height = std::max(1U, m_block_height / 2); + m_block_depth = std::max(1U, m_block_depth / 2); + size += Tegra::Texture::CalculateSize(is_tiled, bytes_per_pixel, m_width, m_height, m_depth, + m_block_height, m_block_depth); + } + return is_tiled ? Common::AlignUp(size, block_size_bytes) : size; +} + /*static*/ SurfaceParams SurfaceParams::CreateForTexture( const Tegra::Texture::FullTextureInfo& config, const GLShader::SamplerEntry& entry) { SurfaceParams params{}; @@ -124,6 +147,7 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { break; } + params.is_layered = SurfaceTargetIsLayered(params.target); params.max_mip_level = config.tic.max_mip_level + 1; params.rt = {}; @@ -150,6 +174,7 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { params.target = SurfaceTarget::Texture2D; params.depth = 1; params.max_mip_level = 0; + params.is_layered = false; // Render target specific parameters, not used for caching params.rt.index = static_cast(index); @@ -182,6 +207,7 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) { params.target = SurfaceTarget::Texture2D; params.depth = 1; params.max_mip_level = 0; + params.is_layered = false; params.rt = {}; params.InitCacheParameters(zeta_address); @@ -361,10 +387,11 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 d } } -static constexpr std::array - morton_to_gl_fns = { - // clang-format off +using GLConversionArray = std::array; + +static constexpr GLConversionArray morton_to_gl_fns = { + // clang-format off MortonCopy, MortonCopy, MortonCopy, @@ -418,13 +445,11 @@ static constexpr std::array, MortonCopy, MortonCopy, - // clang-format on + // clang-format on }; -static constexpr std::array - gl_to_morton_fns = { - // clang-format off +static constexpr GLConversionArray gl_to_morton_fns = { + // clang-format off MortonCopy, MortonCopy, MortonCopy, @@ -479,9 +504,35 @@ static constexpr std::array, MortonCopy, MortonCopy, - // clang-format on + // clang-format on }; +void SwizzleFunc(const GLConversionArray& functions, const SurfaceParams& params, + std::vector& gl_buffer) { + u32 depth = params.depth; + if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { + // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. + depth = 1U; + } + if (params.is_layered) { + u64 offset = 0; + u64 offset_gl = 0; + u64 layer_size = params.LayerMemorySize(); + u64 gl_size = params.LayerSizeGL(); + for (u32 i = 0; i < depth; i++) { + functions[static_cast(params.pixel_format)]( + params.width, params.block_height, params.height, params.block_depth, 1, + gl_buffer.data() + offset_gl, gl_size, params.addr + offset); + offset += layer_size; + offset_gl += gl_size; + } + } else { + functions[static_cast(params.pixel_format)]( + params.width, params.block_height, params.height, params.block_depth, depth, + gl_buffer.data(), gl_buffer.size(), params.addr); + } +} + static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface, GLuint read_fb_handle, GLuint draw_fb_handle, GLenum src_attachment = 0, GLenum dst_attachment = 0, std::size_t cubemap_face = 0) { @@ -881,21 +932,10 @@ void CachedSurface::LoadGLBuffer() { gl_buffer.resize(params.size_in_bytes_gl); if (params.is_tiled) { - u32 depth = params.depth; - u32 block_depth = params.block_depth; - ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast(params.target)); - if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { - // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. - depth = 1U; - block_depth = 1U; - } - - morton_to_gl_fns[static_cast(params.pixel_format)]( - params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(), - gl_buffer.size(), params.addr); + SwizzleFunc(morton_to_gl_fns, params, gl_buffer); } else { const auto texture_src_data{Memory::GetPointer(params.addr)}; const auto texture_src_data_end{texture_src_data + params.size_in_bytes_gl}; @@ -929,19 +969,10 @@ void CachedSurface::FlushGLBuffer() { const u8* const texture_src_data = Memory::GetPointer(params.addr); ASSERT(texture_src_data); if (params.is_tiled) { - u32 depth = params.depth; - u32 block_depth = params.block_depth; - ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast(params.target)); - if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { - // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. - depth = 1U; - } - gl_to_morton_fns[static_cast(params.pixel_format)]( - params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(), - gl_buffer.size(), GetAddr()); + SwizzleFunc(gl_to_morton_fns, params, gl_buffer); } else { std::memcpy(Memory::GetPointer(GetAddr()), gl_buffer.data(), GetSizeInBytes()); } @@ -1179,7 +1210,7 @@ void RasterizerCacheOpenGL::AccurateCopySurface(const Surface& src_surface, const Surface& dst_surface) { const auto& src_params{src_surface->GetSurfaceParams()}; const auto& dst_params{dst_surface->GetSurfaceParams()}; - FlushRegion(src_params.addr, dst_params.size_in_bytes); + FlushRegion(src_params.addr, dst_params.MemorySize()); LoadSurface(dst_surface); } @@ -1221,44 +1252,10 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, CopySurface(old_surface, new_surface, copy_pbo.handle); } break; + case SurfaceParams::SurfaceTarget::TextureCubemap: case SurfaceParams::SurfaceTarget::Texture3D: AccurateCopySurface(old_surface, new_surface); break; - case SurfaceParams::SurfaceTarget::TextureCubemap: { - if (old_params.rt.array_mode != 1) { - // TODO(bunnei): This is used by Breath of the Wild, I'm not sure how to implement this - // yet (array rendering used as a cubemap texture). - LOG_CRITICAL(HW_GPU, "Unhandled rendertarget array_mode {}", old_params.rt.array_mode); - UNREACHABLE(); - return new_surface; - } - - // This seems to be used for render-to-cubemap texture - ASSERT_MSG(old_params.target == SurfaceParams::SurfaceTarget::Texture2D, "Unexpected"); - ASSERT_MSG(old_params.pixel_format == new_params.pixel_format, "Unexpected"); - ASSERT_MSG(old_params.rt.base_layer == 0, "Unimplemented"); - - // TODO(bunnei): Verify the below - this stride seems to be in 32-bit words, not pixels. - // Tested with Splatoon 2, Super Mario Odyssey, and Breath of the Wild. - const std::size_t byte_stride{old_params.rt.layer_stride * sizeof(u32)}; - - for (std::size_t index = 0; index < new_params.depth; ++index) { - Surface face_surface{TryGetReservedSurface(old_params)}; - ASSERT_MSG(face_surface, "Unexpected"); - - if (is_blit) { - BlitSurface(face_surface, new_surface, read_framebuffer.handle, - draw_framebuffer.handle, face_surface->GetSurfaceParams().rt.index, - new_params.rt.index, index); - } else { - CopySurface(face_surface, new_surface, copy_pbo.handle, - face_surface->GetSurfaceParams().rt.index, new_params.rt.index, index); - } - - old_params.addr += byte_stride; - } - break; - } default: LOG_CRITICAL(Render_OpenGL, "Unimplemented surface target={}", static_cast(new_params.target)); @@ -1266,7 +1263,7 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface, } return new_surface; -} +} // namespace OpenGL Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(VAddr addr) const { return TryGet(addr); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index 0dd0d90a38..50a7ab47d9 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -168,6 +168,23 @@ struct SurfaceParams { } } + static bool SurfaceTargetIsLayered(SurfaceTarget target) { + switch (target) { + case SurfaceTarget::Texture1D: + case SurfaceTarget::Texture2D: + case SurfaceTarget::Texture3D: + return false; + case SurfaceTarget::Texture1DArray: + case SurfaceTarget::Texture2DArray: + case SurfaceTarget::TextureCubemap: + return true; + default: + LOG_CRITICAL(HW_GPU, "Unimplemented surface_target={}", static_cast(target)); + UNREACHABLE(); + return false; + } + } + /** * Gets the compression factor for the specified PixelFormat. This applies to just the * "compressed width" and "compressed height", not the overall compression factor of a @@ -742,6 +759,25 @@ struct SurfaceParams { return size_in_bytes_gl / 6; } + /// Returns the exact size of memory occupied by the texture in VRAM, including mipmaps. + std::size_t MemorySize() const { + std::size_t size = InnerMemorySize(is_layered); + if (is_layered) + return size * depth; + return size; + } + + /// Returns the exact size of the memory occupied by a layer in a texture in VRAM, including + /// mipmaps. + std::size_t LayerMemorySize() const { + return InnerMemorySize(true); + } + + /// Returns the size of a layer of this surface in OpenGL. + std::size_t LayerSizeGL() const { + return SizeInBytesRaw(true) / depth; + } + /// Creates SurfaceParams from a texture configuration static SurfaceParams CreateForTexture(const Tegra::Texture::FullTextureInfo& config, const GLShader::SamplerEntry& entry); @@ -782,6 +818,7 @@ struct SurfaceParams { u32 unaligned_height; SurfaceTarget target; u32 max_mip_level; + bool is_layered; // Parameters used for caching VAddr addr; @@ -797,6 +834,9 @@ struct SurfaceParams { u32 layer_stride; u32 base_layer; } rt; + +private: + std::size_t InnerMemorySize(bool layer_only = false) const; }; }; // namespace OpenGL diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index f1b40e7f55..56c61b60c9 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -320,13 +320,13 @@ std::vector DecodeTexture(const std::vector& texture_data, TextureFormat std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height, u32 block_depth) { if (tiled) { - const u32 gobs_in_x = 64 / bytes_per_pixel; + const u32 gobs_in_x = 64; const u32 gobs_in_y = 8; const u32 gobs_in_z = 1; - const u32 aligned_width = Common::AlignUp(width, gobs_in_x); + const u32 aligned_width = Common::AlignUp(width * bytes_per_pixel, gobs_in_x); const u32 aligned_height = Common::AlignUp(height, gobs_in_y * block_height); const u32 aligned_depth = Common::AlignUp(depth, gobs_in_z * block_depth); - return aligned_width * aligned_height * aligned_depth * bytes_per_pixel; + return aligned_width * aligned_height * aligned_depth; } else { return width * height * depth * bytes_per_pixel; }