From 894cc9d876a70947aecc7a1a3f9ef869e8088f42 Mon Sep 17 00:00:00 2001 From: Feng Chen Date: Wed, 17 Nov 2021 12:21:17 +0800 Subject: Fix image update/download error when width too small --- .../renderer_opengl/gl_texture_cache.cpp | 27 ++++++++++++++-------- src/video_core/renderer_opengl/gl_texture_cache.h | 1 + 2 files changed, 18 insertions(+), 10 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 2f7d98d8b..5cfb6bb8a 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -317,13 +317,12 @@ void AttachTexture(GLuint fbo, GLenum attachment, const ImageView* image_view) { } } -OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_format) { +OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_format, + GLsizei gl_num_levels) { const GLenum target = ImageTarget(info); const GLsizei width = info.size.width; const GLsizei height = info.size.height; const GLsizei depth = info.size.depth; - const int max_host_mip_levels = std::bit_width(info.size.width); - const GLsizei num_levels = std::min(info.resources.levels, max_host_mip_levels); const GLsizei num_layers = info.resources.layers; const GLsizei num_samples = info.num_samples; @@ -335,10 +334,10 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form } switch (target) { case GL_TEXTURE_1D_ARRAY: - glTextureStorage2D(handle, num_levels, gl_internal_format, width, num_layers); + glTextureStorage2D(handle, gl_num_levels, gl_internal_format, width, num_layers); break; case GL_TEXTURE_2D_ARRAY: - glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, num_layers); + glTextureStorage3D(handle, gl_num_levels, gl_internal_format, width, height, num_layers); break; case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: { // TODO: Where should 'fixedsamplelocations' come from? @@ -348,10 +347,10 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form break; } case GL_TEXTURE_RECTANGLE: - glTextureStorage2D(handle, num_levels, gl_internal_format, width, height); + glTextureStorage2D(handle, gl_num_levels, gl_internal_format, width, height); break; case GL_TEXTURE_3D: - glTextureStorage3D(handle, num_levels, gl_internal_format, width, height, depth); + glTextureStorage3D(handle, gl_num_levels, gl_internal_format, width, height, depth); break; case GL_TEXTURE_BUFFER: UNREACHABLE(); @@ -686,7 +685,9 @@ Image::Image(TextureCacheRuntime& runtime_, const VideoCommon::ImageInfo& info_, gl_format = tuple.format; gl_type = tuple.type; } - texture = MakeImage(info, gl_internal_format); + const int max_host_mip_levels = std::bit_width(info.size.width); + gl_num_levels = std::min(info.resources.levels, max_host_mip_levels); + texture = MakeImage(info, gl_internal_format, gl_num_levels); current_texture = texture.handle; if (runtime->device.HasDebuggingToolAttached()) { const std::string name = VideoCommon::Name(*this); @@ -714,6 +715,9 @@ void Image::UploadMemory(const ImageBufferMap& map, u32 current_image_height = std::numeric_limits::max(); for (const VideoCommon::BufferImageCopy& copy : copies) { + if (copy.image_subresource.base_level >= gl_num_levels) { + continue; + } if (current_row_length != copy.buffer_row_length) { current_row_length = copy.buffer_row_length; glPixelStorei(GL_UNPACK_ROW_LENGTH, current_row_length); @@ -743,6 +747,9 @@ void Image::DownloadMemory(ImageBufferMap& map, u32 current_image_height = std::numeric_limits::max(); for (const VideoCommon::BufferImageCopy& copy : copies) { + if (copy.image_subresource.base_level >= gl_num_levels) { + continue; + } if (current_row_length != copy.buffer_row_length) { current_row_length = copy.buffer_row_length; glPixelStorei(GL_PACK_ROW_LENGTH, current_row_length); @@ -782,7 +789,7 @@ GLuint Image::StorageHandle() noexcept { } store_view.Create(); glTextureView(store_view.handle, ImageTarget(info), current_texture, GL_RGBA8, 0, - info.resources.levels, 0, info.resources.layers); + gl_num_levels, 0, info.resources.layers); return store_view.handle; default: return current_texture; @@ -946,7 +953,7 @@ void Image::Scale(bool up_scale) { auto dst_info = info; dst_info.size.width = scaled_width; dst_info.size.height = scaled_height; - upscaled_backup = MakeImage(dst_info, gl_internal_format); + upscaled_backup = MakeImage(dst_info, gl_internal_format, gl_num_levels); } const u32 src_width = up_scale ? original_width : scaled_width; const u32 src_height = up_scale ? original_height : scaled_height; diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 1bb762568..30037a6a2 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -219,6 +219,7 @@ private: GLenum gl_internal_format = GL_NONE; GLenum gl_format = GL_NONE; GLenum gl_type = GL_NONE; + GLsizei gl_num_levels{}; TextureCacheRuntime* runtime{}; GLuint current_texture{}; }; -- cgit v1.2.3 From 6dd6dc046c23957d3d8e040a92893acf31ca7980 Mon Sep 17 00:00:00 2001 From: Morph <39850852+Morph1984@users.noreply.github.com> Date: Wed, 17 Nov 2021 15:05:07 -0500 Subject: renderer_opengl: Implement S8_UINT stencil format --- .../renderer_opengl/gl_texture_cache.cpp | 26 ++++++++++++++++++---- src/video_core/renderer_opengl/gl_texture_cache.h | 4 ++-- src/video_core/renderer_opengl/maxwell_to_gl.h | 1 + 3 files changed, 25 insertions(+), 6 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 2f7d98d8b..d46ebd3ea 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -148,6 +148,8 @@ GLenum AttachmentType(PixelFormat format) { switch (const SurfaceType type = VideoCore::Surface::GetFormatType(format); type) { case SurfaceType::Depth: return GL_DEPTH_ATTACHMENT; + case SurfaceType::Stencil: + return GL_STENCIL_ATTACHMENT; case SurfaceType::DepthStencil: return GL_DEPTH_STENCIL_ATTACHMENT; default: @@ -897,6 +899,8 @@ void Image::Scale(bool up_scale) { return GL_COLOR_ATTACHMENT0; case SurfaceType::Depth: return GL_DEPTH_ATTACHMENT; + case SurfaceType::Stencil: + return GL_STENCIL_ATTACHMENT; case SurfaceType::DepthStencil: return GL_DEPTH_STENCIL_ATTACHMENT; default: @@ -910,8 +914,10 @@ void Image::Scale(bool up_scale) { return GL_COLOR_BUFFER_BIT; case SurfaceType::Depth: return GL_DEPTH_BUFFER_BIT; + case SurfaceType::Stencil: + return GL_STENCIL_BUFFER_BIT; case SurfaceType::DepthStencil: - return GL_STENCIL_BUFFER_BIT | GL_DEPTH_BUFFER_BIT; + return GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; default: UNREACHABLE(); return GL_COLOR_BUFFER_BIT; @@ -923,8 +929,10 @@ void Image::Scale(bool up_scale) { return 0; case SurfaceType::Depth: return 1; - case SurfaceType::DepthStencil: + case SurfaceType::Stencil: return 2; + case SurfaceType::DepthStencil: + return 3; default: UNREACHABLE(); return 0; @@ -1254,10 +1262,20 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::spanformat) == SurfaceType::DepthStencil) { + switch (GetFormatType(image_view->format)) { + case SurfaceType::Depth: + buffer_bits |= GL_DEPTH_BUFFER_BIT; + break; + case SurfaceType::Stencil: + buffer_bits |= GL_STENCIL_BUFFER_BIT; + break; + case SurfaceType::DepthStencil: buffer_bits |= GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT; - } else { + break; + default: + UNREACHABLE(); buffer_bits |= GL_DEPTH_BUFFER_BIT; + break; } const GLenum attachment = AttachmentType(image_view->format); AttachTexture(handle, attachment, image_view); diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 1bb762568..16224e6b3 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -162,8 +162,8 @@ private: std::array null_image_views{}; - std::array rescale_draw_fbos; - std::array rescale_read_fbos; + std::array rescale_draw_fbos; + std::array rescale_read_fbos; const Settings::ResolutionScalingInfo& resolution; }; diff --git a/src/video_core/renderer_opengl/maxwell_to_gl.h b/src/video_core/renderer_opengl/maxwell_to_gl.h index 39158aa3e..daba42ed9 100644 --- a/src/video_core/renderer_opengl/maxwell_to_gl.h +++ b/src/video_core/renderer_opengl/maxwell_to_gl.h @@ -108,6 +108,7 @@ constexpr std::array FORMAT_TAB {GL_RGB9_E5, GL_RGB, GL_UNSIGNED_INT_5_9_9_9_REV}, // E5B9G9R9_FLOAT {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT}, // D32_FLOAT {GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT, GL_UNSIGNED_SHORT}, // D16_UNORM + {GL_STENCIL_INDEX8, GL_STENCIL, GL_UNSIGNED_BYTE}, // S8_UINT {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // D24_UNORM_S8_UINT {GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8}, // S8_UINT_D24_UNORM {GL_DEPTH32F_STENCIL8, GL_DEPTH_STENCIL, -- cgit v1.2.3 From 0857f82913d0bcf2de4721233f74cd40ecddcdae Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sat, 20 Nov 2021 06:15:29 +0100 Subject: TextureCache: Implement buffer copies on Vulkan. --- .../renderer_opengl/gl_texture_cache.cpp | 4 +- src/video_core/renderer_opengl/gl_texture_cache.h | 7 +- .../renderer_vulkan/vk_texture_cache.cpp | 174 +++++++++++++++++++++ src/video_core/renderer_vulkan/vk_texture_cache.h | 11 +- src/video_core/texture_cache/texture_cache.h | 4 +- src/video_core/texture_cache/texture_cache_base.h | 2 - 6 files changed, 193 insertions(+), 9 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index 6956535e5..e70bbec81 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -526,8 +526,8 @@ void TextureCacheRuntime::CopyImage(Image& dst_image, Image& src_image, } } -void TextureCacheRuntime::ConvertImage(Image& dst, Image& src, - std::span copies) { +void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, + std::span copies) { LOG_DEBUG(Render_OpenGL, "Converting {} to {}", src.info.format, dst.info.format); format_conversion_pass.ConvertImage(dst, src, copies); } diff --git a/src/video_core/renderer_opengl/gl_texture_cache.h b/src/video_core/renderer_opengl/gl_texture_cache.h index 578f8d523..ad5157d66 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.h +++ b/src/video_core/renderer_opengl/gl_texture_cache.h @@ -84,9 +84,13 @@ public: u64 GetDeviceLocalMemory() const; + bool ShouldReinterpret([[maybe_unused]] Image& dst, [[maybe_unused]] Image& src) { + return true; + } + void CopyImage(Image& dst, Image& src, std::span copies); - void ConvertImage(Image& dst, Image& src, std::span copies); + void ReinterpretImage(Image& dst, Image& src, std::span copies); void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled) { UNIMPLEMENTED(); @@ -338,7 +342,6 @@ struct TextureCacheParams { static constexpr bool FRAMEBUFFER_BLITS = true; static constexpr bool HAS_EMULATED_COPIES = true; static constexpr bool HAS_DEVICE_MEMORY_INFO = true; - static constexpr bool HAS_PIXEL_FORMAT_CONVERSIONS = true; using Runtime = OpenGL::TextureCacheRuntime; using Image = OpenGL::Image; diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index af1a11059..02215cfc2 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -308,6 +308,19 @@ constexpr VkBorderColor ConvertBorderColor(const std::array& color) { }; } +[[nodiscard]] VkBufferImageCopy MakeBufferImageCopy(const VideoCommon::ImageCopy& copy, bool is_src, + VkImageAspectFlags aspect_mask) noexcept { + return VkBufferImageCopy{ + .bufferOffset = 0, + .bufferRowLength = 0, + .bufferImageHeight = 0, + .imageSubresource = MakeImageSubresourceLayers( + is_src ? copy.src_subresource : copy.dst_subresource, aspect_mask), + .imageOffset = MakeOffset3D(is_src ? copy.src_offset : copy.dst_offset), + .imageExtent = MakeExtent3D(copy.extent), + }; +} + [[maybe_unused]] [[nodiscard]] std::vector TransformBufferCopies( std::span copies, size_t buffer_offset) { std::vector result(copies.size()); @@ -754,6 +767,167 @@ StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size) { return staging_buffer_pool.Request(size, MemoryUsage::Download); } +bool TextureCacheRuntime::ShouldReinterpret(Image& dst, Image& src) { + if (VideoCore::Surface::GetFormatType(dst.info.format) == + VideoCore::Surface::SurfaceType::DepthStencil) { + return !device.IsExtShaderStencilExportSupported(); + } + return false; +} + +[[nodiscard]] size_t NextPow2(size_t value) { + return static_cast(1ULL << ((8U * sizeof(size_t)) - std::countl_zero(value - 1U))); +} + +VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) { + const auto level = (8 * sizeof(size_t)) - std::countl_zero(needed_size - 1ULL); + if (buffer_commits[level]) { + return *buffers[level]; + } + const auto new_size = NextPow2(needed_size); + VkBufferUsageFlags flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | + VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | + VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT; + buffers[level] = device.GetLogical().CreateBuffer({ + .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .size = new_size, + .usage = flags, + .sharingMode = VK_SHARING_MODE_EXCLUSIVE, + .queueFamilyIndexCount = 0, + .pQueueFamilyIndices = nullptr, + }); + buffer_commits[level] = std::make_unique( + memory_allocator.Commit(buffers[level], MemoryUsage::DeviceLocal)); + return *buffers[level]; +} + +void TextureCacheRuntime::ReinterpretImage(Image& dst, Image& src, + std::span copies) { + std::vector vk_in_copies(copies.size()); + std::vector vk_out_copies(copies.size()); + const VkImageAspectFlags src_aspect_mask = src.AspectMask(); + const VkImageAspectFlags dst_aspect_mask = dst.AspectMask(); + + std::ranges::transform(copies, vk_in_copies.begin(), [src_aspect_mask](const auto& copy) { + return MakeBufferImageCopy(copy, true, src_aspect_mask); + }); + std::ranges::transform(copies, vk_out_copies.begin(), [dst_aspect_mask](const auto& copy) { + return MakeBufferImageCopy(copy, false, dst_aspect_mask); + }); + const u32 img_bpp = BytesPerBlock(src.info.format); + size_t total_size = 0; + for (const auto& copy : copies) { + total_size += copy.extent.width * copy.extent.height * copy.extent.depth * img_bpp; + } + const VkBuffer copy_buffer = GetTemporaryBuffer(total_size); + const VkImage dst_image = dst.Handle(); + const VkImage src_image = src.Handle(); + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([dst_image, src_image, copy_buffer, src_aspect_mask, dst_aspect_mask, + vk_in_copies, vk_out_copies](vk::CommandBuffer cmdbuf) { + RangedBarrierRange dst_range; + RangedBarrierRange src_range; + for (const VkBufferImageCopy& copy : vk_in_copies) { + src_range.AddLayers(copy.imageSubresource); + } + for (const VkBufferImageCopy& copy : vk_out_copies) { + dst_range.AddLayers(copy.imageSubresource); + } + static constexpr VkMemoryBarrier READ_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + }; + static constexpr VkMemoryBarrier WRITE_BARRIER{ + .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT, + }; + const std::array pre_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = src_range.SubresourceRange(src_aspect_mask), + }, + }; + const std::array middle_in_barrier{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = 0, + .dstAccessMask = 0, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = src_image, + .subresourceRange = src_range.SubresourceRange(src_aspect_mask), + }, + }; + const std::array middle_out_barrier{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_GENERAL, + .newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = dst_range.SubresourceRange(dst_aspect_mask), + }, + }; + const std::array post_barriers{ + VkImageMemoryBarrier{ + .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, + .pNext = nullptr, + .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT, + .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | + VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | + VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | + VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | + VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT, + .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, + .newLayout = VK_IMAGE_LAYOUT_GENERAL, + .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED, + .image = dst_image, + .subresourceRange = dst_range.SubresourceRange(dst_aspect_mask), + }, + }; + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, {}, {}, pre_barriers); + + cmdbuf.CopyImageToBuffer(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, copy_buffer, + vk_in_copies); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, WRITE_BARRIER, nullptr, middle_in_barrier); + + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, + 0, READ_BARRIER, {}, middle_out_barrier); + cmdbuf.CopyBufferToImage(copy_buffer, dst_image, VK_IMAGE_LAYOUT_GENERAL, vk_out_copies); + cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, + 0, {}, {}, post_barriers); + }); +} + void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst, ImageView& src, const Region2D& dst_region, const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter, diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h index f5f8f9a74..44e9dcee4 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.h +++ b/src/video_core/renderer_vulkan/vk_texture_cache.h @@ -61,6 +61,10 @@ public: void CopyImage(Image& dst, Image& src, std::span copies); + bool ShouldReinterpret(Image& dst, Image& src); + + void ReinterpretImage(Image& dst, Image& src, std::span copies); + void ConvertImage(Framebuffer* dst, ImageView& dst_view, ImageView& src_view, bool rescaled); bool CanAccelerateImageUpload(Image&) const noexcept { @@ -82,6 +86,8 @@ public: return true; } + [[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size); + const Device& device; VKScheduler& scheduler; MemoryAllocator& memory_allocator; @@ -90,6 +96,10 @@ public: ASTCDecoderPass& astc_decoder_pass; RenderPassCache& render_pass_cache; const Settings::ResolutionScalingInfo& resolution; + + constexpr static size_t indexing_slots = 8 * sizeof(size_t); + std::array buffers{}; + std::array, indexing_slots> buffer_commits{}; }; class Image : public VideoCommon::ImageBase { @@ -316,7 +326,6 @@ struct TextureCacheParams { static constexpr bool FRAMEBUFFER_BLITS = false; static constexpr bool HAS_EMULATED_COPIES = false; static constexpr bool HAS_DEVICE_MEMORY_INFO = true; - static constexpr bool HAS_PIXEL_FORMAT_CONVERSIONS = false; using Runtime = Vulkan::TextureCacheRuntime; using Image = Vulkan::Image; diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 4188f93c5..44a0d42ba 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -1762,8 +1762,8 @@ void TextureCache

::CopyImage(ImageId dst_id, ImageId src_id, std::vector Date: Sat, 20 Nov 2021 14:46:19 +0100 Subject: TextureCache: Refactor and fix linux compiling. --- src/common/bit_util.h | 7 +++++++ src/video_core/renderer_opengl/gl_texture_cache.cpp | 6 ++---- src/video_core/renderer_vulkan/vk_texture_cache.cpp | 7 ++----- 3 files changed, 11 insertions(+), 9 deletions(-) (limited to 'src/video_core/renderer_opengl') diff --git a/src/common/bit_util.h b/src/common/bit_util.h index 64520ca4e..eef8c1c5a 100644 --- a/src/common/bit_util.h +++ b/src/common/bit_util.h @@ -7,6 +7,7 @@ #include #include #include +#include #include "common/common_types.h" @@ -44,4 +45,10 @@ template return static_cast(log2_f + static_cast((value ^ (1ULL << log2_f)) != 0ULL)); } +template +requires std::is_integral_v +[[nodiscard]] T NextPow2(T value) { + return static_cast(1ULL << ((8U * sizeof(T)) - std::countl_zero(value - 1U))); +} + } // namespace Common diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp index e70bbec81..ecb215a7d 100644 --- a/src/video_core/renderer_opengl/gl_texture_cache.cpp +++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp @@ -9,6 +9,7 @@ #include +#include "common/bit_util.h" #include "common/literals.h" #include "common/settings.h" #include "video_core/renderer_opengl/gl_device.h" @@ -397,9 +398,6 @@ OGLTexture MakeImage(const VideoCommon::ImageInfo& info, GLenum gl_internal_form return GL_R32UI; } -[[nodiscard]] u32 NextPow2(u32 value) { - return 1U << (32U - std::countl_zero(value - 1U)); -} } // Anonymous namespace ImageBufferMap::~ImageBufferMap() { @@ -1308,7 +1306,7 @@ void FormatConversionPass::ConvertImage(Image& dst_image, Image& src_image, const u32 copy_size = region.width * region.height * region.depth * img_bpp; if (pbo_size < copy_size) { intermediate_pbo.Create(); - pbo_size = NextPow2(copy_size); + pbo_size = Common::NextPow2(copy_size); glNamedBufferData(intermediate_pbo.handle, pbo_size, nullptr, GL_STREAM_COPY); } // Copy from source to PBO diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp index 02215cfc2..f194110e5 100644 --- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp @@ -8,6 +8,7 @@ #include #include "common/bit_cast.h" +#include "common/bit_util.h" #include "common/settings.h" #include "video_core/engines/fermi_2d.h" @@ -775,16 +776,12 @@ bool TextureCacheRuntime::ShouldReinterpret(Image& dst, Image& src) { return false; } -[[nodiscard]] size_t NextPow2(size_t value) { - return static_cast(1ULL << ((8U * sizeof(size_t)) - std::countl_zero(value - 1U))); -} - VkBuffer TextureCacheRuntime::GetTemporaryBuffer(size_t needed_size) { const auto level = (8 * sizeof(size_t)) - std::countl_zero(needed_size - 1ULL); if (buffer_commits[level]) { return *buffers[level]; } - const auto new_size = NextPow2(needed_size); + const auto new_size = Common::NextPow2(needed_size); VkBufferUsageFlags flags = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT; -- cgit v1.2.3