From 3e1cbb7d14a17425dc1b2abc5f71b20050a2afb1 Mon Sep 17 00:00:00 2001 From: James Rowe Date: Fri, 17 Nov 2017 10:27:51 -0700 Subject: [PATCH] OpenGL Cache: Split CachedSurface Breaks CachedSurface into two classes, the parameters used to create or find a cached surface, and the actual cached surface. This also adds a few helper methods for getting surfaces from cache --- .../renderer_opengl/gl_rasterizer_cache.cpp | 781 ++++++++++++------ .../renderer_opengl/gl_rasterizer_cache.h | 156 +++- src/video_core/renderer_opengl/gl_state.h | 2 +- 3 files changed, 649 insertions(+), 290 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index e37da1d8e..756a76ea6 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -342,6 +342,231 @@ static bool BlitTextures(GLuint src_tex, const MathUtil::Rectangle& src_rec return true; } +SurfaceParams SurfaceParams::FromInterval(SurfaceInterval interval) const { + SurfaceParams params = *this; + + const u32 stride_tiled_bytes = BytesInPixels(stride * (is_tiled ? 8 : 1)); + PAddr aligned_start = + addr + Common::AlignDown(boost::icl::first(interval) - addr, stride_tiled_bytes); + PAddr aligned_end = + addr + Common::AlignUp(boost::icl::last_next(interval) - addr, stride_tiled_bytes); + + if (aligned_end - aligned_start > stride_tiled_bytes) { + params.addr = aligned_start; + params.height = (aligned_end - aligned_start) / BytesInPixels(stride); + } else { + // 1 row + ASSERT(aligned_end - aligned_start == stride_tiled_bytes); + const u32 tiled_alignment = BytesInPixels(is_tiled ? 8 * 8 : 1); + aligned_start = + addr + Common::AlignDown(boost::icl::first(interval) - addr, tiled_alignment); + aligned_end = + addr + Common::AlignUp(boost::icl::last_next(interval) - addr, tiled_alignment); + params.addr = aligned_start; + params.width = PixelsInBytes(aligned_end - aligned_start) / (is_tiled ? 8 : 1); + params.height = is_tiled ? 8 : 1; + } + params.UpdateParams(); + + return params; +} + +SurfaceInterval SurfaceParams::GetSubRectInterval(MathUtil::Rectangle unscaled_rect) const { + if (unscaled_rect.GetHeight() == 0 || unscaled_rect.GetWidth() == 0) { + return {}; + } + + if (unscaled_rect.bottom > unscaled_rect.top) { + std::swap(unscaled_rect.top, unscaled_rect.bottom); + } + + if (is_tiled) { + unscaled_rect.left = Common::AlignDown(unscaled_rect.left, 8) * 8; + unscaled_rect.bottom = Common::AlignDown(unscaled_rect.bottom, 8) / 8; + unscaled_rect.right = Common::AlignUp(unscaled_rect.right, 8) * 8; + unscaled_rect.top = Common::AlignUp(unscaled_rect.top, 8) / 8; + } + + const u32 stride_tiled = (!is_tiled ? stride : stride * 8); + + const u32 pixel_offset = + stride_tiled * (!is_tiled ? unscaled_rect.bottom : (height / 8) - unscaled_rect.top) + + unscaled_rect.left; + + const u32 pixels = (unscaled_rect.GetHeight() - 1) * stride_tiled + unscaled_rect.GetWidth(); + + return {addr + BytesInPixels(pixel_offset), addr + BytesInPixels(pixel_offset + pixels)}; +} + +MathUtil::Rectangle SurfaceParams::GetSubRect(const SurfaceParams& sub_surface) const { + const u32 begin_pixel_index = PixelsInBytes(sub_surface.addr - addr); + + if (is_tiled) { + const int x0 = (begin_pixel_index % (stride * 8)) / 8; + const int y0 = (begin_pixel_index / (stride * 8)) * 8; + return MathUtil::Rectangle(x0, height - y0, x0 + sub_surface.width, + height - (y0 + sub_surface.height)); // Top to bottom + } + + const int x0 = begin_pixel_index % stride; + const int y0 = begin_pixel_index / stride; + return MathUtil::Rectangle(x0, y0 + sub_surface.height, x0 + sub_surface.width, + y0); // Bottom to top +} + +MathUtil::Rectangle SurfaceParams::GetScaledSubRect(const SurfaceParams& sub_surface) const { + auto rect = GetSubRect(sub_surface); + rect.left = rect.left * res_scale; + rect.right = rect.right * res_scale; + rect.top = rect.top * res_scale; + rect.bottom = rect.bottom * res_scale; + return rect; +} + +bool SurfaceParams::ExactMatch(const SurfaceParams& other_surface) const { + return (other_surface.addr == addr && other_surface.width == width && + other_surface.height == height && other_surface.stride == stride && + other_surface.pixel_format == pixel_format && other_surface.is_tiled == is_tiled); +} + +bool SurfaceParams::CanSubRect(const SurfaceParams& sub_surface) const { + if (sub_surface.addr < addr || sub_surface.end > end || sub_surface.stride != stride || + sub_surface.pixel_format != pixel_format || sub_surface.is_tiled != is_tiled || + (sub_surface.addr - addr) * 8 % GetFormatBpp() != 0) + return false; + + auto rect = GetSubRect(sub_surface); + + if (rect.left + sub_surface.width > stride) { + return false; + } + + if (is_tiled) { + return PixelsInBytes(sub_surface.addr - addr) % 64 == 0 && sub_surface.height % 8 == 0 && + sub_surface.width % 8 == 0; + } + + return true; +} + +bool SurfaceParams::CanExpand(const SurfaceParams& expanded_surface) const { + if (pixel_format == PixelFormat::Invalid || pixel_format != expanded_surface.pixel_format || + is_tiled != expanded_surface.is_tiled || addr > expanded_surface.end || + expanded_surface.addr > end || stride != expanded_surface.stride) + return false; + + const u32 byte_offset = + std::max(expanded_surface.addr, addr) - std::min(expanded_surface.addr, addr); + + const int x0 = byte_offset % BytesInPixels(stride); + const int y0 = byte_offset / BytesInPixels(stride); + + return x0 == 0 && (!is_tiled || y0 % 8 == 0); +} + +bool SurfaceParams::CanTexCopy(const SurfaceParams& texcopy_params) const { + if (pixel_format == PixelFormat::Invalid || addr > texcopy_params.addr || + end < texcopy_params.end || ((texcopy_params.addr - addr) * 8) % GetFormatBpp() != 0 || + (texcopy_params.width * 8) % GetFormatBpp() != 0 || + (texcopy_params.stride * 8) % GetFormatBpp() != 0) + return false; + + const u32 begin_pixel_index = PixelsInBytes(texcopy_params.addr - addr); + const int x0 = begin_pixel_index % stride; + const int y0 = begin_pixel_index / stride; + + if (!is_tiled) + return ((texcopy_params.height == 1 || PixelsInBytes(texcopy_params.stride) == stride) && + x0 + PixelsInBytes(texcopy_params.width) <= stride); + + return (PixelsInBytes(texcopy_params.addr - addr) % 64 == 0 && + PixelsInBytes(texcopy_params.width) % 64 == 0 && + (texcopy_params.height == 1 || PixelsInBytes(texcopy_params.stride) == stride * 8) && + x0 + PixelsInBytes(texcopy_params.width / 8) <= stride); +} + +bool CachedSurface::CanFill(const SurfaceParams& dest_surface, + SurfaceInterval fill_interval) const { + if (type == SurfaceType::Fill && IsRegionValid(fill_interval) && + boost::icl::first(fill_interval) >= addr && + boost::icl::last_next(fill_interval) <= end && // dest_surface is within our fill range + dest_surface.FromInterval(fill_interval).GetInterval() == + fill_interval) { // make sure interval is a rectangle in dest surface + if (fill_size * 8 != dest_surface.GetFormatBpp()) { + // Check if bits repeat for our fill_size + const u32 dest_bytes_per_pixel = std::max(dest_surface.GetFormatBpp() / 8, 1u); + std::vector fill_test(fill_size * dest_bytes_per_pixel); + + for (u32 i = 0; i < dest_bytes_per_pixel; ++i) + std::memcpy(&fill_test[i * fill_size], &fill_data[0], fill_size); + + for (u32 i = 0; i < fill_size; ++i) + if (std::memcmp(&fill_test[dest_bytes_per_pixel * i], &fill_test[0], + dest_bytes_per_pixel) != 0) + return false; + + if (dest_surface.GetFormatBpp() == 4 && (fill_test[0] & 0xF) != (fill_test[0] >> 4)) + return false; + } + return true; + } + return false; +} + +bool CachedSurface::CanCopy(const SurfaceParams& dest_surface, + SurfaceInterval copy_interval) const { + SurfaceParams subrect_params = dest_surface.FromInterval(copy_interval); + ASSERT(subrect_params.GetInterval() == copy_interval); + if (CanSubRect(subrect_params)) + return true; + + if (CanFill(dest_surface, copy_interval)) + return true; + + return false; +} + +SurfaceInterval SurfaceParams::GetCopyableInterval(const Surface& src_surface) const { + SurfaceInterval result{}; + const auto valid_regions = + SurfaceRegions(GetInterval() & src_surface->GetInterval()) - src_surface->invalid_regions; + for (auto& valid_interval : valid_regions) { + const SurfaceInterval aligned_interval{ + addr + Common::AlignUp(boost::icl::first(valid_interval) - addr, + BytesInPixels(is_tiled ? 8 * 8 : 1)), + addr + Common::AlignDown(boost::icl::last_next(valid_interval) - addr, + BytesInPixels(is_tiled ? 8 * 8 : 1))}; + + if (BytesInPixels(is_tiled ? 8 * 8 : 1) > boost::icl::length(valid_interval) || + boost::icl::length(aligned_interval) == 0) { + continue; + } + + // Get the rectangle within aligned_interval + const u32 stride_bytes = BytesInPixels(stride) * (is_tiled ? 8 : 1); + SurfaceInterval rect_interval{ + addr + Common::AlignUp(boost::icl::first(aligned_interval) - addr, stride_bytes), + addr + Common::AlignDown(boost::icl::last_next(aligned_interval) - addr, stride_bytes), + }; + if (boost::icl::first(rect_interval) > boost::icl::last_next(rect_interval)) { + // 1 row + rect_interval = aligned_interval; + } else if (boost::icl::length(rect_interval) == 0) { + // 2 rows that do not make a rectangle, return the larger one + const SurfaceInterval row1{boost::icl::first(aligned_interval), + boost::icl::first(rect_interval)}; + const SurfaceInterval row2{boost::icl::first(rect_interval), + boost::icl::last_next(aligned_interval)}; + rect_interval = (boost::icl::length(row1) > boost::icl::length(row2)) ? row1 : row2; + } + + if (boost::icl::length(rect_interval) > boost::icl::length(result)) { + result = rect_interval; + } + } + return result; +} + bool RasterizerCacheOpenGL::TryBlitSurfaces(CachedSurface* src_surface, const MathUtil::Rectangle& src_rect, CachedSurface* dst_surface, @@ -381,201 +606,61 @@ static void AllocateSurfaceTexture(GLuint texture, const FormatTuple& format_tup cur_state.Apply(); } -MICROPROFILE_DEFINE(OpenGL_SurfaceUpload, "OpenGL", "Surface Upload", MP_RGB(128, 64, 192)); -CachedSurface* RasterizerCacheOpenGL::GetSurface(const CachedSurface& params, bool match_res_scale, - bool load_if_create) { - using PixelFormat = CachedSurface::PixelFormat; - using SurfaceType = CachedSurface::SurfaceType; +MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192)); +void CachedSurface::LoadGLBuffer(PAddr load_start, PAddr load_end) { + ASSERT(type != SurfaceType::Fill); - if (params.addr == 0) { - return nullptr; + const u8* const texture_src_data = Memory::GetPhysicalPointer(addr); + if (texture_src_data == nullptr) + return; + + if (gl_buffer == nullptr) { + gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format); + gl_buffer.reset(new u8[gl_buffer_size]); } - u32 params_size = - params.width * params.height * CachedSurface::GetFormatBpp(params.pixel_format) / 8; + // TODO: Should probably be done in ::Memory:: and check for other regions too + if (load_start < Memory::VRAM_VADDR_END && load_end > Memory::VRAM_VADDR_END) + load_end = Memory::VRAM_VADDR_END; - // Check for an exact match in existing surfaces - CachedSurface* best_exact_surface = nullptr; - float exact_surface_goodness = -1.f; + if (load_start < Memory::VRAM_VADDR && load_end > Memory::VRAM_VADDR) + load_start = Memory::VRAM_VADDR; - auto surface_interval = - boost::icl::interval::right_open(params.addr, params.addr + params_size); - auto range = surface_cache.equal_range(surface_interval); - for (auto it = range.first; it != range.second; ++it) { - for (auto it2 = it->second.begin(); it2 != it->second.end(); ++it2) { - CachedSurface* surface = it2->get(); + MICROPROFILE_SCOPE(OpenGL_SurfaceLoad); - // Check if the request matches the surface exactly - if (params.addr == surface->addr && params.width == surface->width && - params.height == surface->height && params.pixel_format == surface->pixel_format) { - // Make sure optional param-matching criteria are fulfilled - bool tiling_match = (params.is_tiled == surface->is_tiled); - bool res_scale_match = (params.res_scale_width == surface->res_scale_width && - params.res_scale_height == surface->res_scale_height); - if (!match_res_scale || res_scale_match) { - // Prioritize same-tiling and highest resolution surfaces - float match_goodness = - (float)tiling_match + surface->res_scale_width * surface->res_scale_height; - if (match_goodness > exact_surface_goodness || surface->dirty) { - exact_surface_goodness = match_goodness; - best_exact_surface = surface; - } - } - } - } - } + ASSERT(load_start >= addr && load_end <= end); + const u32 start_offset = load_start - addr; - // Return the best exact surface if found - if (best_exact_surface != nullptr) { - return best_exact_surface; - } - - // No matching surfaces found, so create a new one - u8* texture_src_data = Memory::GetPhysicalPointer(params.addr); - if (texture_src_data == nullptr) { - return nullptr; - } - - MICROPROFILE_SCOPE(OpenGL_SurfaceUpload); - - // Stride only applies to linear images. - ASSERT(params.pixel_stride == 0 || !params.is_tiled); - - std::shared_ptr new_surface = std::make_shared(); - - new_surface->addr = params.addr; - new_surface->size = params_size; - - new_surface->texture.Create(); - new_surface->width = params.width; - new_surface->height = params.height; - new_surface->pixel_stride = params.pixel_stride; - new_surface->res_scale_width = params.res_scale_width; - new_surface->res_scale_height = params.res_scale_height; - - new_surface->is_tiled = params.is_tiled; - new_surface->pixel_format = params.pixel_format; - new_surface->dirty = false; - - if (!load_if_create) { - // Don't load any data; just allocate the surface's texture - AllocateSurfaceTexture(new_surface->texture.handle, new_surface->pixel_format, - new_surface->GetScaledWidth(), new_surface->GetScaledHeight()); + if (!is_tiled) { + ASSERT(type == SurfaceType::Color); + std::memcpy(&gl_buffer[start_offset], texture_src_data + start_offset, + load_end - load_start); } else { - // TODO: Consider attempting subrect match in existing surfaces and direct blit here instead - // of memory upload below if that's a common scenario in some game + if (type == SurfaceType::Texture) { + Pica::Texture::TextureInfo tex_info{}; + tex_info.width = width; + tex_info.height = height; + tex_info.format = static_cast(pixel_format); + tex_info.SetDefaultStride(); + tex_info.physical_address = addr; - Memory::RasterizerFlushRegion(params.addr, params_size); + const auto load_interval = SurfaceInterval(load_start, load_end); + const auto rect = GetSubRect(FromInterval(load_interval)); + ASSERT(FromInterval(load_interval).GetInterval() == load_interval); - // Load data from memory to the new surface - OpenGLState cur_state = OpenGLState::GetCurState(); - - GLuint old_tex = cur_state.texture_units[0].texture_2d; - cur_state.texture_units[0].texture_2d = new_surface->texture.handle; - cur_state.Apply(); - glActiveTexture(GL_TEXTURE0); - - if (!new_surface->is_tiled) { - // TODO: Ensure this will always be a color format, not a depth or other format - ASSERT((size_t)new_surface->pixel_format < fb_format_tuples.size()); - const FormatTuple& tuple = fb_format_tuples[(unsigned int)params.pixel_format]; - - glPixelStorei(GL_UNPACK_ROW_LENGTH, (GLint)new_surface->pixel_stride); - glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height, 0, - tuple.format, tuple.type, texture_src_data); - glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); - } else { - SurfaceType type = CachedSurface::GetFormatType(new_surface->pixel_format); - if (type != SurfaceType::Depth && type != SurfaceType::DepthStencil) { - FormatTuple tuple; - if ((size_t)params.pixel_format < fb_format_tuples.size()) { - tuple = fb_format_tuples[(unsigned int)params.pixel_format]; - } else { - // Texture - tuple = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE}; + for (unsigned y = rect.bottom; y < rect.top; ++y) { + for (unsigned x = rect.left; x < rect.right; ++x) { + auto vec4 = + Pica::Texture::LookupTexture(texture_src_data, x, height - 1 - y, tex_info); + const size_t offset = (x + (width * y)) * 4; + std::memcpy(&gl_buffer[offset], vec4.AsArray(), 4); } - - std::vector> tex_buffer(params.width * params.height); - - Pica::Texture::TextureInfo tex_info; - tex_info.width = params.width; - tex_info.height = params.height; - tex_info.format = (Pica::TexturingRegs::TextureFormat)params.pixel_format; - tex_info.SetDefaultStride(); - tex_info.physical_address = params.addr; - - for (unsigned y = 0; y < params.height; ++y) { - for (unsigned x = 0; x < params.width; ++x) { - tex_buffer[x + params.width * y] = Pica::Texture::LookupTexture( - texture_src_data, x, params.height - 1 - y, tex_info); - } - } - - glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height, - 0, GL_RGBA, GL_UNSIGNED_BYTE, tex_buffer.data()); - } else { - // Depth/Stencil formats need special treatment since they aren't sampleable using - // LookupTexture and can't use RGBA format - size_t tuple_idx = (size_t)params.pixel_format - 14; - ASSERT(tuple_idx < depth_format_tuples.size()); - const FormatTuple& tuple = depth_format_tuples[tuple_idx]; - - u32 bytes_per_pixel = CachedSurface::GetFormatBpp(params.pixel_format) / 8; - - // OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type - bool use_4bpp = (params.pixel_format == PixelFormat::D24); - - u32 gl_bytes_per_pixel = use_4bpp ? 4 : bytes_per_pixel; - - std::vector temp_fb_depth_buffer(params.width * params.height * - gl_bytes_per_pixel); - - u8* temp_fb_depth_buffer_ptr = - use_4bpp ? temp_fb_depth_buffer.data() + 1 : temp_fb_depth_buffer.data(); - - MortonCopyPixels(params.pixel_format, params.width, params.height, bytes_per_pixel, - gl_bytes_per_pixel, texture_src_data, temp_fb_depth_buffer_ptr, - true); - - glTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format, params.width, params.height, - 0, tuple.format, tuple.type, temp_fb_depth_buffer.data()); } + } else { + morton_to_gl_fns[static_cast(pixel_format)](stride, height, &gl_buffer[0], addr, + load_start, load_end); } - - // If not 1x scale, blit 1x texture to a new scaled texture and replace texture in surface - if (new_surface->res_scale_width != 1.f || new_surface->res_scale_height != 1.f) { - OGLTexture scaled_texture; - scaled_texture.Create(); - - AllocateSurfaceTexture(scaled_texture.handle, new_surface->pixel_format, - new_surface->GetScaledWidth(), new_surface->GetScaledHeight()); - BlitTextures(new_surface->texture.handle, scaled_texture.handle, - CachedSurface::GetFormatType(new_surface->pixel_format), - MathUtil::Rectangle(0, 0, new_surface->width, new_surface->height), - MathUtil::Rectangle(0, 0, new_surface->GetScaledWidth(), - new_surface->GetScaledHeight())); - - new_surface->texture.Release(); - new_surface->texture.handle = scaled_texture.handle; - scaled_texture.handle = 0; - cur_state.texture_units[0].texture_2d = new_surface->texture.handle; - cur_state.Apply(); - } - - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, 0); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE); - glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE); - - cur_state.texture_units[0].texture_2d = old_tex; - cur_state.Apply(); } - - Memory::RasterizerMarkRegionCached(new_surface->addr, new_surface->size, 1); - surface_cache.add(std::make_pair(boost::icl::interval::right_open( - new_surface->addr, new_surface->addr + new_surface->size), - std::set>({new_surface}))); - return new_surface.get(); } CachedSurface* RasterizerCacheOpenGL::GetSurfaceRect(const CachedSurface& params, @@ -826,102 +911,272 @@ CachedSurface* RasterizerCacheOpenGL::TryGetFillSurface(const GPU::Regs::MemoryF return nullptr; } -MICROPROFILE_DEFINE(OpenGL_SurfaceDownload, "OpenGL", "Surface Download", MP_RGB(128, 192, 64)); -void RasterizerCacheOpenGL::FlushSurface(CachedSurface* surface) { - using PixelFormat = CachedSurface::PixelFormat; - using SurfaceType = CachedSurface::SurfaceType; - - if (!surface->dirty) { +MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64)); +void CachedSurface::FlushGLBuffer(PAddr flush_start, PAddr flush_end) { + u8* const dst_buffer = Memory::GetPhysicalPointer(addr); + if (dst_buffer == nullptr) return; + + ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format)); + + // TODO: Should probably be done in ::Memory:: and check for other regions too + // same as loadglbuffer() + if (flush_start < Memory::VRAM_VADDR_END && flush_end > Memory::VRAM_VADDR_END) + flush_end = Memory::VRAM_VADDR_END; + + if (flush_start < Memory::VRAM_VADDR && flush_end > Memory::VRAM_VADDR) + flush_start = Memory::VRAM_VADDR; + + MICROPROFILE_SCOPE(OpenGL_SurfaceFlush); + + ASSERT(flush_start >= addr && flush_end <= end); + const u32 start_offset = flush_start - addr; + const u32 end_offset = flush_end - addr; + + if (type == SurfaceType::Fill) { + const u32 coarse_start_offset = start_offset - (start_offset % fill_size); + const u32 backup_bytes = start_offset % fill_size; + std::array backup_data; + if (backup_bytes) + std::memcpy(&backup_data[0], &dst_buffer[coarse_start_offset], backup_bytes); + + for (u32 offset = coarse_start_offset; offset < end_offset; offset += fill_size) + std::memcpy(&dst_buffer[offset], &fill_data[0], + std::min(fill_size, end_offset - offset)); + + if (backup_bytes) + std::memcpy(&dst_buffer[coarse_start_offset], &backup_data[0], backup_bytes); + } else if (!is_tiled) { + ASSERT(type == SurfaceType::Color); + std::memcpy(dst_buffer + start_offset, &gl_buffer[start_offset], flush_end - flush_start); + } else { + gl_to_morton_fns[static_cast(pixel_format)](stride, height, &gl_buffer[0], addr, + flush_start, flush_end); } +} - MICROPROFILE_SCOPE(OpenGL_SurfaceDownload); - - u8* dst_buffer = Memory::GetPhysicalPointer(surface->addr); - if (dst_buffer == nullptr) { +void CachedSurface::UploadGLTexture(const MathUtil::Rectangle& rect) { + if (type == SurfaceType::Fill) return; + + ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format)); + + // Load data from memory to the surface + GLint x0 = static_cast(rect.left); + GLint y0 = static_cast(rect.bottom); + size_t buffer_offset = (y0 * stride + x0) * GetGLBytesPerPixel(pixel_format); + + const FormatTuple& tuple = GetFormatTuple(pixel_format); + GLuint target_tex = texture.handle; + + // If not 1x scale, create 1x texture that we will blit from to replace texture subrect in + // surface + OGLTexture unscaled_tex; + if (res_scale != 1) { + x0 = 0; + y0 = 0; + + unscaled_tex.Create(); + AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight()); + target_tex = unscaled_tex.handle; } OpenGLState cur_state = OpenGLState::GetCurState(); + GLuint old_tex = cur_state.texture_units[0].texture_2d; - - OGLTexture unscaled_tex; - GLuint texture_to_flush = surface->texture.handle; - - // If not 1x scale, blit scaled texture to a new 1x texture and use that to flush - if (surface->res_scale_width != 1.f || surface->res_scale_height != 1.f) { - unscaled_tex.Create(); - - AllocateSurfaceTexture(unscaled_tex.handle, surface->pixel_format, surface->width, - surface->height); - BlitTextures( - surface->texture.handle, unscaled_tex.handle, - CachedSurface::GetFormatType(surface->pixel_format), - MathUtil::Rectangle(0, 0, surface->GetScaledWidth(), surface->GetScaledHeight()), - MathUtil::Rectangle(0, 0, surface->width, surface->height)); - - texture_to_flush = unscaled_tex.handle; - } - - cur_state.texture_units[0].texture_2d = texture_to_flush; + cur_state.texture_units[0].texture_2d = target_tex; cur_state.Apply(); + + // Ensure no bad interactions with GL_UNPACK_ALIGNMENT + ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0); + glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast(stride)); + glActiveTexture(GL_TEXTURE0); + glTexSubImage2D(GL_TEXTURE_2D, 0, x0, y0, static_cast(rect.GetWidth()), + static_cast(rect.GetHeight()), tuple.format, tuple.type, + &gl_buffer[buffer_offset]); - if (!surface->is_tiled) { - // TODO: Ensure this will always be a color format, not a depth or other format - ASSERT((size_t)surface->pixel_format < fb_format_tuples.size()); - const FormatTuple& tuple = fb_format_tuples[(unsigned int)surface->pixel_format]; - - glPixelStorei(GL_PACK_ROW_LENGTH, (GLint)surface->pixel_stride); - glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, dst_buffer); - glPixelStorei(GL_PACK_ROW_LENGTH, 0); - } else { - SurfaceType type = CachedSurface::GetFormatType(surface->pixel_format); - if (type != SurfaceType::Depth && type != SurfaceType::DepthStencil) { - ASSERT((size_t)surface->pixel_format < fb_format_tuples.size()); - const FormatTuple& tuple = fb_format_tuples[(unsigned int)surface->pixel_format]; - - u32 bytes_per_pixel = CachedSurface::GetFormatBpp(surface->pixel_format) / 8; - - std::vector temp_gl_buffer(surface->width * surface->height * bytes_per_pixel); - - glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, temp_gl_buffer.data()); - - // Directly copy pixels. Internal OpenGL color formats are consistent so no conversion - // is necessary. - MortonCopyPixels(surface->pixel_format, surface->width, surface->height, - bytes_per_pixel, bytes_per_pixel, dst_buffer, temp_gl_buffer.data(), - false); - } else { - // Depth/Stencil formats need special treatment since they aren't sampleable using - // LookupTexture and can't use RGBA format - size_t tuple_idx = (size_t)surface->pixel_format - 14; - ASSERT(tuple_idx < depth_format_tuples.size()); - const FormatTuple& tuple = depth_format_tuples[tuple_idx]; - - u32 bytes_per_pixel = CachedSurface::GetFormatBpp(surface->pixel_format) / 8; - - // OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type - bool use_4bpp = (surface->pixel_format == PixelFormat::D24); - - u32 gl_bytes_per_pixel = use_4bpp ? 4 : bytes_per_pixel; - - std::vector temp_gl_buffer(surface->width * surface->height * gl_bytes_per_pixel); - - glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, temp_gl_buffer.data()); - - u8* temp_gl_buffer_ptr = use_4bpp ? temp_gl_buffer.data() + 1 : temp_gl_buffer.data(); - - MortonCopyPixels(surface->pixel_format, surface->width, surface->height, - bytes_per_pixel, gl_bytes_per_pixel, dst_buffer, temp_gl_buffer_ptr, - false); - } - } - - surface->dirty = false; + glPixelStorei(GL_UNPACK_ROW_LENGTH, 0); cur_state.texture_units[0].texture_2d = old_tex; cur_state.Apply(); + + if (res_scale != 1) { + auto scaled_rect = rect; + scaled_rect.left *= res_scale; + scaled_rect.top *= res_scale; + scaled_rect.right *= res_scale; + scaled_rect.bottom *= res_scale; + + BlitTextures(unscaled_tex.handle, {0, rect.GetHeight(), rect.GetWidth(), 0}, texture.handle, + scaled_rect, type); + } +} +void CachedSurface::DownloadGLTexture(const MathUtil::Rectangle& rect) { + if (type == SurfaceType::Fill) + return; + + if (gl_buffer == nullptr) { + gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format); + gl_buffer.reset(new u8[gl_buffer_size]); + } + + OpenGLState state = OpenGLState::GetCurState(); + OpenGLState prev_state = state; + SCOPE_EXIT({ prev_state.Apply(); }); + + const FormatTuple& tuple = GetFormatTuple(pixel_format); + + // Ensure no bad interactions with GL_PACK_ALIGNMENT + ASSERT(stride * GetGLBytesPerPixel(pixel_format) % 4 == 0); + glPixelStorei(GL_PACK_ROW_LENGTH, static_cast(stride)); + size_t buffer_offset = (rect.bottom * stride + rect.left) * GetGLBytesPerPixel(pixel_format); + + // If not 1x scale, blit scaled texture to a new 1x texture and use that to flush + OGLTexture unscaled_tex; + if (res_scale != 1) { + auto scaled_rect = rect; + scaled_rect.left *= res_scale; + scaled_rect.top *= res_scale; + scaled_rect.right *= res_scale; + scaled_rect.bottom *= res_scale; + + unscaled_tex.Create(); + AllocateSurfaceTexture(unscaled_tex.handle, tuple, rect.GetWidth(), rect.GetHeight()); + BlitTextures(texture.handle, scaled_rect, unscaled_tex.handle, rect, type); + + state.texture_units[0].texture_2d = unscaled_tex.handle; + state.Apply(); + + glActiveTexture(GL_TEXTURE0); + glGetTexImage(GL_TEXTURE_2D, 0, tuple.format, tuple.type, &gl_buffer[buffer_offset]); + } else { + state.ResetTexture(texture.handle); + state.draw.read_framebuffer = transfer_framebuffers[0].handle; + state.Apply(); + + if (type == SurfaceType::Color || type == SurfaceType::Texture) { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, + texture.handle, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + 0, 0); + } 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, + texture.handle, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); + } else { + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0); + glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, + texture.handle, 0); + } + glReadPixels(static_cast(rect.left), static_cast(rect.bottom), + static_cast(rect.GetWidth()), static_cast(rect.GetHeight()), + tuple.format, tuple.type, &gl_buffer[buffer_offset]); + } + + glPixelStorei(GL_PACK_ROW_LENGTH, 0); +} + +enum MatchFlags { + Invalid = 1, // Flag that can be applied to other match types, invalid matches require + // validation before they can be used + Exact = 1 << 1, // Surfaces perfectly match + SubRect = 1 << 2, // Surface encompasses params + Copy = 1 << 3, // Surface we can copy from + Expand = 1 << 4, // Surface that can expand params + TexCopy = 1 << 5 // Surface that will match a display transfer "texture copy" parameters +}; + +constexpr MatchFlags operator|(MatchFlags lhs, MatchFlags rhs) { + return static_cast(static_cast(lhs) | static_cast(rhs)); +} + +/// Get the best surface match (and its match type) for the given flags +template +Surface FindMatch(const SurfaceCache& surface_cache, const SurfaceParams& params, + ScaleMatch match_scale_type, + boost::optional validate_interval = boost::none) { + Surface match_surface = nullptr; + bool match_valid = false; + u32 match_scale = 0; + SurfaceInterval match_interval{}; + + for (auto& pair : RangeFromInterval(surface_cache, params.GetInterval())) { + for (auto& surface : pair.second) { + const bool res_scale_matched = match_scale_type == ScaleMatch::Exact + ? (params.res_scale == surface->res_scale) + : (params.res_scale <= surface->res_scale); + bool is_valid = + find_flags & MatchFlags::Copy ? true + : // validity will be checked in GetCopyableInterval + surface->IsRegionValid(validate_interval.value_or(params.GetInterval())); + + if (!(find_flags & MatchFlags::Invalid) && !is_valid) + continue; + + auto IsMatch_Helper = [&](auto check_type, auto match_fn) { + if (!(find_flags & check_type)) + return; + + bool matched; + SurfaceInterval surface_interval; + std::tie(matched, surface_interval) = match_fn(); + if (!matched) + return; + + if (!res_scale_matched && match_scale_type != ScaleMatch::Ignore && + surface->type != SurfaceType::Fill) + return; + + // Found a match, update only if this is better than the previous one + auto UpdateMatch = [&] { + match_surface = surface; + match_valid = is_valid; + match_scale = surface->res_scale; + match_interval = surface_interval; + }; + + if (surface->res_scale > match_scale) { + UpdateMatch(); + return; + } else if (surface->res_scale < match_scale) { + return; + } + + if (is_valid && !match_valid) { + UpdateMatch(); + return; + } else if (is_valid != match_valid) { + return; + } + + if (boost::icl::length(surface_interval) > boost::icl::length(match_interval)) { + UpdateMatch(); + } + }; + IsMatch_Helper(std::integral_constant{}, [&] { + return std::make_pair(surface->ExactMatch(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant{}, [&] { + return std::make_pair(surface->CanSubRect(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant{}, [&] { + auto copy_interval = + params.FromInterval(*validate_interval).GetCopyableInterval(surface); + bool matched = boost::icl::length(copy_interval & *validate_interval) != 0 && + surface->CanCopy(params, copy_interval); + return std::make_pair(matched, copy_interval); + }); + IsMatch_Helper(std::integral_constant{}, [&] { + return std::make_pair(surface->CanExpand(params), surface->GetInterval()); + }); + IsMatch_Helper(std::integral_constant{}, [&] { + return std::make_pair(surface->CanTexCopy(params), surface->GetInterval()); + }); + } + } + return match_surface; } void RasterizerCacheOpenGL::FlushRegion(PAddr addr, u32 size, const CachedSurface* skip_surface, diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index 9c2abcd84..92bf7d3be 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -13,6 +13,7 @@ #pragma GCC diagnostic ignored "-Wunused-local-typedefs" #endif #include +#include #ifdef __GNUC__ #pragma GCC diagnostic pop #endif @@ -20,21 +21,37 @@ #include "common/assert.h" #include "common/common_funcs.h" #include "common/common_types.h" +#include "common/math_util.h" #include "core/hw/gpu.h" #include "video_core/regs_framebuffer.h" #include "video_core/regs_texturing.h" #include "video_core/renderer_opengl/gl_resource_manager.h" -namespace MathUtil { -template -struct Rectangle; -} - struct CachedSurface; +using Surface = std::shared_ptr; +using SurfaceSet = std::set; -using SurfaceCache = boost::icl::interval_map>>; +using SurfaceRegions = boost::icl::interval_set; +using SurfaceMap = boost::icl::interval_map; +using SurfaceCache = boost::icl::interval_map; -struct CachedSurface { +using SurfaceInterval = SurfaceCache::interval_type; +static_assert(std::is_same() && + std::is_same(), + "incorrect interval types"); + +using SurfaceRect_Tuple = std::tuple>; +using SurfaceSurfaceRect_Tuple = std::tuple>; + +using PageMap = boost::icl::interval_map; + +enum class ScaleMatch { + Exact, // only accept same res scale + Upscale, // only allow higher scale than params + Ignore // accept every scaled res +}; + +struct SurfaceParams { enum class PixelFormat { // First 5 formats are shared between textures and color buffers RGBA8 = 0, @@ -68,10 +85,11 @@ struct CachedSurface { Texture = 1, Depth = 2, DepthStencil = 3, - Invalid = 4, + Fill = 4, + Invalid = 5 }; - static constexpr unsigned int GetFormatBpp(CachedSurface::PixelFormat format) { + static constexpr unsigned int GetFormatBpp(PixelFormat format) { constexpr std::array bpp_table = { 32, // RGBA8 24, // RGB8 @@ -93,8 +111,11 @@ struct CachedSurface { 32, // D24S8 }; - ASSERT((unsigned int)format < ARRAY_SIZE(bpp_table)); - return bpp_table[(unsigned int)format]; + ASSERT(static_cast(format) < bpp_table.size()); + return bpp_table[static_cast(format)]; + } + unsigned int GetFormatBpp() const { + return GetFormatBpp(pixel_format); } static PixelFormat PixelFormatFromTextureFormat(Pica::TexturingRegs::TextureFormat format) { @@ -162,31 +183,114 @@ struct CachedSurface { return SurfaceType::Invalid; } + /// Update the params "size", "end" and "type" from the already set "addr", "width", "height" + /// and "pixel_format" + void UpdateParams() { + if (stride == 0) { + stride = width; + } + type = GetFormatType(pixel_format); + size = !is_tiled ? BytesInPixels(stride * (height - 1) + width) + : BytesInPixels(stride * 8 * (height / 8 - 1) + width * 8); + end = addr + size; + } + + SurfaceInterval GetInterval() const { + return SurfaceInterval::right_open(addr, end); + } + + // Returns the outer rectangle containing "interval" + SurfaceParams FromInterval(SurfaceInterval interval) const; + + SurfaceInterval GetSubRectInterval(MathUtil::Rectangle unscaled_rect) const; + + // Returns the region of the biggest valid rectange within interval + SurfaceInterval GetCopyableInterval(const Surface& src_surface) const; + u32 GetScaledWidth() const { - return (u32)(width * res_scale_width); + return width * res_scale; } u32 GetScaledHeight() const { - return (u32)(height * res_scale_height); + return height * res_scale; } - PAddr addr; - u32 size; + MathUtil::Rectangle GetRect() const { + return {0, height, width, 0}; + } - PAddr min_valid; - PAddr max_valid; + MathUtil::Rectangle GetScaledRect() const { + return {0, GetScaledHeight(), GetScaledWidth(), 0}; + } + + u32 PixelsInBytes(u32 size) const { + return size * 8 / GetFormatBpp(pixel_format); + } + + u32 BytesInPixels(u32 pixels) const { + return pixels * GetFormatBpp(pixel_format) / 8; + } + + bool ExactMatch(const SurfaceParams& other_surface) const; + bool CanSubRect(const SurfaceParams& sub_surface) const; + bool CanExpand(const SurfaceParams& expanded_surface) const; + bool CanTexCopy(const SurfaceParams& texcopy_params) const; + + MathUtil::Rectangle GetSubRect(const SurfaceParams& sub_surface) const; + MathUtil::Rectangle GetScaledSubRect(const SurfaceParams& sub_surface) const; + + PAddr addr = 0; + PAddr end = 0; + u32 size = 0; + + u32 width = 0; + u32 height = 0; + u32 stride = 0; + u16 res_scale = 1; + + bool is_tiled = false; + PixelFormat pixel_format = PixelFormat::Invalid; + SurfaceType type = SurfaceType::Invalid; +}; + +struct CachedSurface : SurfaceParams { + bool CanFill(const SurfaceParams& dest_surface, SurfaceInterval fill_interval) const; + bool CanCopy(const SurfaceParams& dest_surface, SurfaceInterval copy_interval) const; + + bool IsRegionValid(SurfaceInterval interval) const { + return (invalid_regions.find(interval) == invalid_regions.end()); + } + + bool IsSurfaceFullyInvalid() const { + return (invalid_regions & GetInterval()) == SurfaceRegions(GetInterval()); + } + + SurfaceRegions invalid_regions; + + u32 fill_size = 0; /// Number of bytes to read from fill_data + std::array fill_data; OGLTexture texture; - u32 width; - u32 height; - /// Stride between lines, in pixels. Only valid for images in linear format. - u32 pixel_stride = 0; - float res_scale_width = 1.f; - float res_scale_height = 1.f; - bool is_tiled; - PixelFormat pixel_format; - bool dirty; + static constexpr unsigned int GetGLBytesPerPixel(PixelFormat format) { + // OpenGL needs 4 bpp alignment for D24 since using GL_UNSIGNED_INT as type + return format == PixelFormat::Invalid + ? 0 + : (format == PixelFormat::D24 || GetFormatType(format) == SurfaceType::Texture) + ? 4 + : SurfaceParams::GetFormatBpp(format) / 8; + } + + std::unique_ptr gl_buffer; + size_t gl_buffer_size = 0; + + // Read/Write data in 3DS memory to/from gl_buffer + void LoadGLBuffer(PAddr load_start, PAddr load_end); + void FlushGLBuffer(PAddr flush_start, PAddr flush_end); + + // Upload/Download data in gl_buffer in/to this surface's texture + void UploadGLTexture(const MathUtil::Rectangle& rect); + void DownloadGLTexture(const MathUtil::Rectangle& rect); }; class RasterizerCacheOpenGL : NonCopyable { diff --git a/src/video_core/renderer_opengl/gl_state.h b/src/video_core/renderer_opengl/gl_state.h index 703aee90e..033d417bc 100644 --- a/src/video_core/renderer_opengl/gl_state.h +++ b/src/video_core/renderer_opengl/gl_state.h @@ -144,7 +144,7 @@ public: OpenGLState(); /// Get the currently active OpenGL state - static OpenGLState& GetCurState() { + static OpenGLState GetCurState() { return cur_state; }