diff options
Diffstat (limited to '')
| -rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer_cache.cpp | 139 |
1 files changed, 55 insertions, 84 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index d6048f639..61d670dcb 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -28,6 +28,7 @@ #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_opengl/gl_rasterizer_cache.h" #include "video_core/renderer_opengl/gl_state.h" +#include "video_core/textures/astc.h" #include "video_core/textures/decoders.h" #include "video_core/utils.h" #include "video_core/video_core.h" @@ -50,18 +51,22 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form {GL_RGB5_A1, GL_RGBA, GL_UNSIGNED_SHORT_1_5_5_5_REV, false}, // A1B5G5R5 {GL_R8, GL_RED, GL_UNSIGNED_BYTE, false}, // R8 {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT, false}, // RGBA16F + {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, false}, // R11FG11FB10F {GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT1 {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT23 {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT45 {GL_COMPRESSED_RED_RGTC1, GL_RED, GL_UNSIGNED_INT_8_8_8_8, true}, // DXN1 + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, false}, // ASTC_2D_4X4 }}; static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) { const SurfaceType type = SurfaceParams::GetFormatType(pixel_format); if (type == SurfaceType::ColorTexture) { ASSERT(static_cast<size_t>(pixel_format) < tex_format_tuples.size()); - // For now only UNORM components are supported, or RGBA16F which is type FLOAT - ASSERT(component_type == ComponentType::UNorm || pixel_format == PixelFormat::RGBA16F); + // For now only UNORM components are supported, or either R11FG11FB10F or RGBA16F which are + // type FLOAT + ASSERT(component_type == ComponentType::UNorm || pixel_format == PixelFormat::RGBA16F || + pixel_format == PixelFormat::R11FG11FB10F); return tex_format_tuples[static_cast<unsigned int>(pixel_format)]; } else if (type == SurfaceType::Depth || type == SurfaceType::DepthStencil) { // TODO(Subv): Implement depth formats @@ -83,6 +88,23 @@ static u16 GetResolutionScaleFactor() { : Settings::values.resolution_factor); } +static void ConvertASTCToRGBA8(std::vector<u8>& data, PixelFormat format, u32 width, u32 height) { + u32 block_width{}; + u32 block_height{}; + + switch (format) { + case PixelFormat::ASTC_2D_4X4: + block_width = 4; + block_height = 4; + break; + default: + NGLOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast<u32>(format)); + UNREACHABLE(); + } + + data = Tegra::Texture::ASTC::Decompress(data, width, height, block_width, block_height); +} + template <bool morton_to_gl, PixelFormat format> void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, Tegra::GPUVAddr base, Tegra::GPUVAddr start, Tegra::GPUVAddr end) { @@ -94,6 +116,12 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, Tegra:: auto data = Tegra::Texture::UnswizzleTexture( *gpu.memory_manager->GpuToCpuAddress(base), SurfaceParams::TextureFormatFromPixelFormat(format), stride, height, block_height); + + if (SurfaceParams::IsFormatASTC(format)) { + // ASTC formats are converted to RGBA8 in software, as most PC GPUs do not support this + ConvertASTCToRGBA8(data, format, stride, height); + } + std::memcpy(gl_buffer, data.data(), data.size()); } else { // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should check @@ -110,11 +138,12 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr, Tegra: Tegra::GPUVAddr), SurfaceParams::MaxPixelFormat> morton_to_gl_fns = { - MortonCopy<true, PixelFormat::ABGR8>, MortonCopy<true, PixelFormat::B5G6R5>, - MortonCopy<true, PixelFormat::A2B10G10R10>, MortonCopy<true, PixelFormat::A1B5G5R5>, - MortonCopy<true, PixelFormat::R8>, MortonCopy<true, PixelFormat::RGBA16F>, - MortonCopy<true, PixelFormat::DXT1>, MortonCopy<true, PixelFormat::DXT23>, - MortonCopy<true, PixelFormat::DXT45>, MortonCopy<true, PixelFormat::DXN1>, + MortonCopy<true, PixelFormat::ABGR8>, MortonCopy<true, PixelFormat::B5G6R5>, + MortonCopy<true, PixelFormat::A2B10G10R10>, MortonCopy<true, PixelFormat::A1B5G5R5>, + MortonCopy<true, PixelFormat::R8>, MortonCopy<true, PixelFormat::RGBA16F>, + MortonCopy<true, PixelFormat::R11FG11FB10F>, MortonCopy<true, PixelFormat::DXT1>, + MortonCopy<true, PixelFormat::DXT23>, MortonCopy<true, PixelFormat::DXT45>, + MortonCopy<true, PixelFormat::DXN1>, MortonCopy<true, PixelFormat::ASTC_2D_4X4>, }; static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr, Tegra::GPUVAddr, @@ -127,11 +156,13 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr, Tegra: MortonCopy<false, PixelFormat::A1B5G5R5>, MortonCopy<false, PixelFormat::R8>, MortonCopy<false, PixelFormat::RGBA16F>, + MortonCopy<false, PixelFormat::R11FG11FB10F>, // TODO(Subv): Swizzling the DXT1/DXT23/DXT45/DXN1 formats is not yet supported nullptr, nullptr, nullptr, nullptr, + MortonCopy<false, PixelFormat::ABGR8>, }; // Allocate an uninitialized texture of appropriate size and format for the surface @@ -164,60 +195,10 @@ static void AllocateSurfaceTexture(GLuint texture, const FormatTuple& format_tup static bool BlitTextures(GLuint src_tex, const MathUtil::Rectangle<u32>& src_rect, GLuint dst_tex, const MathUtil::Rectangle<u32>& dst_rect, SurfaceType type, GLuint read_fb_handle, GLuint draw_fb_handle) { - OpenGLState state = OpenGLState::GetCurState(); - - OpenGLState prev_state = state; - SCOPE_EXIT({ prev_state.Apply(); }); - - // Make sure textures aren't bound to texture units, since going to bind them to framebuffer - // components - state.ResetTexture(src_tex); - state.ResetTexture(dst_tex); - - state.draw.read_framebuffer = read_fb_handle; - state.draw.draw_framebuffer = draw_fb_handle; - state.Apply(); - - u32 buffers = 0; - - if (type == SurfaceType::ColorTexture) { - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, src_tex, - 0); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, - 0); - - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, dst_tex, - 0); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, - 0); - - buffers = GL_COLOR_BUFFER_BIT; - } else if (type == SurfaceType::Depth) { - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, src_tex, 0); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); - - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, dst_tex, 0); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); - - buffers = GL_DEPTH_BUFFER_BIT; - } else if (type == SurfaceType::DepthStencil) { - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, - src_tex, 0); - - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, - dst_tex, 0); - - buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; - } - - glBlitFramebuffer(src_rect.left, src_rect.bottom, src_rect.right, src_rect.top, dst_rect.left, - dst_rect.bottom, dst_rect.right, dst_rect.top, buffers, - buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST); + glCopyImageSubData(src_tex, GL_TEXTURE_2D, 0, src_rect.left, src_rect.bottom, 0, dst_tex, + GL_TEXTURE_2D, 0, dst_rect.left, dst_rect.bottom, 0, src_rect.GetWidth(), + src_rect.GetHeight(), 0); return true; } @@ -594,7 +575,7 @@ void CachedSurface::UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint glCompressedTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, static_cast<GLsizei>(rect.GetWidth() * GetCompresssionFactor()), static_cast<GLsizei>(rect.GetHeight() * GetCompresssionFactor()), 0, - size, &gl_buffer[buffer_offset]); + static_cast<GLsizei>(size), &gl_buffer[buffer_offset]); } else { glTexSubImage2D(GL_TEXTURE_2D, 0, x0, y0, static_cast<GLsizei>(rect.GetWidth()), static_cast<GLsizei>(rect.GetHeight()), tuple.format, tuple.type, @@ -933,9 +914,6 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, ScaleMatc // Use GetSurfaceSubRect instead ASSERT(params.width == params.stride); - ASSERT(!params.is_tiled || - (params.GetActualWidth() % 8 == 0 && params.GetActualHeight() % 8 == 0)); - // Check for an exact match in existing surfaces Surface surface = FindMatch<MatchFlags::Exact | MatchFlags::Invalid>(surface_cache, params, match_res_scale); @@ -1078,8 +1056,11 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu params.addr = config.tic.Address(); params.is_tiled = config.tic.IsTiled(); params.pixel_format = SurfaceParams::PixelFormatFromTextureFormat(config.tic.format); - params.width = config.tic.Width() / params.GetCompresssionFactor(); - params.height = config.tic.Height() / params.GetCompresssionFactor(); + + params.width = Common::AlignUp(config.tic.Width(), params.GetCompresssionFactor()) / + params.GetCompresssionFactor(); + params.height = Common::AlignUp(config.tic.Height(), params.GetCompresssionFactor()) / + params.GetCompresssionFactor(); // TODO(Subv): Different types per component are not supported. ASSERT(config.tic.r_type.Value() == config.tic.g_type.Value() && @@ -1090,6 +1071,13 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu if (config.tic.IsTiled()) { params.block_height = config.tic.BlockHeight(); + + // TODO(bunnei): The below align up is a hack. This is here because some compressed textures + // are not a multiple of their own compression factor, and so this accounts for that. This + // could potentially result in an extra row of 4px being decoded if a texture is not a + // multiple of 4. + params.width = Common::AlignUp(params.width, 4); + params.height = Common::AlignUp(params.height, 4); } else { // Use the texture-provided stride value if the texture isn't tiled. params.stride = static_cast<u32>(params.PixelsInBytes(config.tic.Pitch())); @@ -1097,23 +1085,6 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu params.UpdateParams(); - if (config.tic.Width() % 8 != 0 || config.tic.Height() % 8 != 0 || - params.stride != params.width) { - Surface src_surface; - MathUtil::Rectangle<u32> rect; - std::tie(src_surface, rect) = GetSurfaceSubRect(params, ScaleMatch::Ignore, true); - - params.res_scale = src_surface->res_scale; - Surface tmp_surface = CreateSurface(params); - BlitTextures(src_surface->texture.handle, rect, tmp_surface->texture.handle, - tmp_surface->GetScaledRect(), - SurfaceParams::GetFormatType(params.pixel_format), read_framebuffer.handle, - draw_framebuffer.handle); - - remove_surfaces.emplace(tmp_surface); - return tmp_surface; - } - return GetSurface(params, ScaleMatch::Ignore, true); } @@ -1288,7 +1259,7 @@ void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, Tegra::GPUVA const auto interval = *it & validate_interval; // Look for a valid surface to copy from - SurfaceParams params = surface->FromInterval(interval); + SurfaceParams params = *surface; Surface copy_surface = FindMatch<MatchFlags::Copy>(surface_cache, params, ScaleMatch::Ignore, interval); |