diff options
Diffstat (limited to '')
-rw-r--r-- | src/video_core/command_classes/codecs/codec.h | 6 | ||||
-rw-r--r-- | src/video_core/command_classes/codecs/h264.cpp | 3 | ||||
-rw-r--r-- | src/video_core/command_classes/codecs/h264.h | 10 | ||||
-rw-r--r-- | src/video_core/command_classes/codecs/vp9.cpp | 2 | ||||
-rw-r--r-- | src/video_core/command_classes/codecs/vp9.h | 2 | ||||
-rw-r--r-- | src/video_core/command_classes/codecs/vp9_types.h | 5 | ||||
-rw-r--r-- | src/video_core/command_classes/nvdec.h | 4 | ||||
-rw-r--r-- | src/video_core/dma_pusher.cpp | 80 | ||||
-rw-r--r-- | src/video_core/dma_pusher.h | 49 | ||||
-rw-r--r-- | src/video_core/gpu.cpp | 48 | ||||
-rw-r--r-- | src/video_core/gpu.h | 25 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_device.cpp | 9 | ||||
-rw-r--r-- | src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp | 2 | ||||
-rw-r--r-- | src/video_core/texture_cache/surface_params.cpp | 1 |
14 files changed, 175 insertions, 71 deletions
diff --git a/src/video_core/command_classes/codecs/codec.h b/src/video_core/command_classes/codecs/codec.h index cb67094f6..5bbe6a332 100644 --- a/src/video_core/command_classes/codecs/codec.h +++ b/src/video_core/command_classes/codecs/codec.h @@ -42,11 +42,11 @@ public: void Decode(); /// Returns most recently decoded frame - AVFrame* GetCurrentFrame(); - const AVFrame* GetCurrentFrame() const; + [[nodiscard]] AVFrame* GetCurrentFrame(); + [[nodiscard]] const AVFrame* GetCurrentFrame() const; /// Returns the value of current_codec - NvdecCommon::VideoCodec GetCurrentCodec() const; + [[nodiscard]] NvdecCommon::VideoCodec GetCurrentCodec() const; private: bool initialized{}; diff --git a/src/video_core/command_classes/codecs/h264.cpp b/src/video_core/command_classes/codecs/h264.cpp index 549a40f52..33e063e20 100644 --- a/src/video_core/command_classes/codecs/h264.cpp +++ b/src/video_core/command_classes/codecs/h264.cpp @@ -43,7 +43,8 @@ H264::H264(GPU& gpu_) : gpu(gpu_) {} H264::~H264() = default; -std::vector<u8>& H264::ComposeFrameHeader(NvdecCommon::NvdecRegisters& state, bool is_first_frame) { +const std::vector<u8>& H264::ComposeFrameHeader(NvdecCommon::NvdecRegisters& state, + bool is_first_frame) { H264DecoderContext context{}; gpu.MemoryManager().ReadBlock(state.picture_info_offset, &context, sizeof(H264DecoderContext)); diff --git a/src/video_core/command_classes/codecs/h264.h b/src/video_core/command_classes/codecs/h264.h index f2292fd2f..273449495 100644 --- a/src/video_core/command_classes/codecs/h264.h +++ b/src/video_core/command_classes/codecs/h264.h @@ -51,14 +51,14 @@ public: void WriteScalingList(const std::vector<u8>& list, s32 start, s32 count); /// Return the bitstream as a vector. - std::vector<u8>& GetByteArray(); - const std::vector<u8>& GetByteArray() const; + [[nodiscard]] std::vector<u8>& GetByteArray(); + [[nodiscard]] const std::vector<u8>& GetByteArray() const; private: void WriteBits(s32 value, s32 bit_count); void WriteExpGolombCodedInt(s32 value); void WriteExpGolombCodedUInt(u32 value); - s32 GetFreeBufferBits(); + [[nodiscard]] s32 GetFreeBufferBits(); void Flush(); s32 buffer_size{8}; @@ -74,8 +74,8 @@ public: ~H264(); /// Compose the H264 header of the frame for FFmpeg decoding - std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state, - bool is_first_frame = false); + [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state, + bool is_first_frame = false); private: struct H264ParameterSet { diff --git a/src/video_core/command_classes/codecs/vp9.cpp b/src/video_core/command_classes/codecs/vp9.cpp index 42520f856..ab44fdc9e 100644 --- a/src/video_core/command_classes/codecs/vp9.cpp +++ b/src/video_core/command_classes/codecs/vp9.cpp @@ -854,7 +854,7 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() { return uncomp_writer; } -std::vector<u8>& VP9::ComposeFrameHeader(NvdecCommon::NvdecRegisters& state) { +const std::vector<u8>& VP9::ComposeFrameHeader(NvdecCommon::NvdecRegisters& state) { std::vector<u8> bitstream; { Vp9FrameContainer curr_frame = GetCurrentFrame(state); diff --git a/src/video_core/command_classes/codecs/vp9.h b/src/video_core/command_classes/codecs/vp9.h index 05c9682fa..e2504512c 100644 --- a/src/video_core/command_classes/codecs/vp9.h +++ b/src/video_core/command_classes/codecs/vp9.h @@ -119,7 +119,7 @@ public: /// Composes the VP9 frame from the GPU state information. Based on the official VP9 spec /// documentation - std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state); + [[nodiscard]] const std::vector<u8>& ComposeFrameHeader(NvdecCommon::NvdecRegisters& state); /// Returns true if the most recent frame was a hidden frame. [[nodiscard]] bool WasFrameHidden() const { diff --git a/src/video_core/command_classes/codecs/vp9_types.h b/src/video_core/command_classes/codecs/vp9_types.h index a50acf6e8..4f0b05d22 100644 --- a/src/video_core/command_classes/codecs/vp9_types.h +++ b/src/video_core/command_classes/codecs/vp9_types.h @@ -231,9 +231,8 @@ struct PictureInfo { u32 surface_params{}; INSERT_PADDING_WORDS(3); - Vp9PictureInfo Convert() const { - - return Vp9PictureInfo{ + [[nodiscard]] Vp9PictureInfo Convert() const { + return { .is_key_frame = (vp9_flags & FrameFlags::IsKeyFrame) != 0, .intra_only = (vp9_flags & FrameFlags::IntraOnly) != 0, .last_frame_was_key = (vp9_flags & FrameFlags::LastFrameIsKeyFrame) != 0, diff --git a/src/video_core/command_classes/nvdec.h b/src/video_core/command_classes/nvdec.h index af14f9857..eec4443f9 100644 --- a/src/video_core/command_classes/nvdec.h +++ b/src/video_core/command_classes/nvdec.h @@ -26,8 +26,8 @@ public: void ProcessMethod(Method method, const std::vector<u32>& arguments); /// Return most recently decoded frame - AVFrame* GetFrame(); - const AVFrame* GetFrame() const; + [[nodiscard]] AVFrame* GetFrame(); + [[nodiscard]] const AVFrame* GetFrame() const; private: /// Invoke codec to decode a frame diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index f2f96ac33..105b85a92 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include "common/cityhash.h" #include "common/microprofile.h" #include "core/core.h" #include "core/memory.h" @@ -12,6 +13,20 @@ namespace Tegra { +void CommandList::RefreshIntegrityChecks(GPU& gpu) { + command_list_hashes.resize(command_lists.size()); + + for (std::size_t index = 0; index < command_lists.size(); ++index) { + const CommandListHeader command_list_header = command_lists[index]; + std::vector<CommandHeader> command_headers(command_list_header.size); + gpu.MemoryManager().ReadBlockUnsafe(command_list_header.addr, command_headers.data(), + command_list_header.size * sizeof(u32)); + command_list_hashes[index] = + Common::CityHash64(reinterpret_cast<char*>(command_headers.data()), + command_list_header.size * sizeof(u32)); + } +} + DmaPusher::DmaPusher(Core::System& system, GPU& gpu) : gpu{gpu}, system{system} {} DmaPusher::~DmaPusher() = default; @@ -45,32 +60,51 @@ bool DmaPusher::Step() { return false; } - const CommandList& command_list{dma_pushbuffer.front()}; - ASSERT_OR_EXECUTE(!command_list.empty(), { - // Somehow the command_list is empty, in order to avoid a crash - // We ignore it and assume its size is 0. - dma_pushbuffer.pop(); - dma_pushbuffer_subindex = 0; - return true; - }); - const CommandListHeader command_list_header{command_list[dma_pushbuffer_subindex++]}; - const GPUVAddr dma_get = command_list_header.addr; - - if (dma_pushbuffer_subindex >= command_list.size()) { - // We've gone through the current list, remove it from the queue - dma_pushbuffer.pop(); - dma_pushbuffer_subindex = 0; - } + CommandList& command_list{dma_pushbuffer.front()}; - if (command_list_header.size == 0) { - return true; - } + ASSERT_OR_EXECUTE( + command_list.command_lists.size() || command_list.prefetch_command_list.size(), { + // Somehow the command_list is empty, in order to avoid a crash + // We ignore it and assume its size is 0. + dma_pushbuffer.pop(); + dma_pushbuffer_subindex = 0; + return true; + }); - // Push buffer non-empty, read a word - command_headers.resize(command_list_header.size); - gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), - command_list_header.size * sizeof(u32)); + if (command_list.prefetch_command_list.size()) { + // Prefetched command list from nvdrv, used for things like synchronization + command_headers = std::move(command_list.prefetch_command_list); + dma_pushbuffer.pop(); + } else { + const CommandListHeader command_list_header{ + command_list.command_lists[dma_pushbuffer_subindex]}; + const u64 next_hash = command_list.command_list_hashes[dma_pushbuffer_subindex++]; + const GPUVAddr dma_get = command_list_header.addr; + + if (dma_pushbuffer_subindex >= command_list.command_lists.size()) { + // We've gone through the current list, remove it from the queue + dma_pushbuffer.pop(); + dma_pushbuffer_subindex = 0; + } + if (command_list_header.size == 0) { + return true; + } + + // Push buffer non-empty, read a word + command_headers.resize(command_list_header.size); + gpu.MemoryManager().ReadBlockUnsafe(dma_get, command_headers.data(), + command_list_header.size * sizeof(u32)); + + // Integrity check + const u64 new_hash = Common::CityHash64(reinterpret_cast<char*>(command_headers.data()), + command_list_header.size * sizeof(u32)); + if (new_hash != next_hash) { + LOG_CRITICAL(HW_GPU, "CommandList at addr=0x{:X} is corrupt, skipping!", dma_get); + dma_pushbuffer.pop(); + return true; + } + } for (std::size_t index = 0; index < command_headers.size();) { const CommandHeader& command_header = command_headers[index]; diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h index efa90d170..8496ba2da 100644 --- a/src/video_core/dma_pusher.h +++ b/src/video_core/dma_pusher.h @@ -27,6 +27,31 @@ enum class SubmissionMode : u32 { IncreaseOnce = 5 }; +// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence +// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4. +// So the values you see in docs might be multiplied by 4. +enum class BufferMethods : u32 { + BindObject = 0x0, + Nop = 0x2, + SemaphoreAddressHigh = 0x4, + SemaphoreAddressLow = 0x5, + SemaphoreSequence = 0x6, + SemaphoreTrigger = 0x7, + NotifyIntr = 0x8, + WrcacheFlush = 0x9, + Unk28 = 0xA, + UnkCacheFlush = 0xB, + RefCnt = 0x14, + SemaphoreAcquire = 0x1A, + SemaphoreRelease = 0x1B, + FenceValue = 0x1C, + FenceAction = 0x1D, + WaitForInterrupt = 0x1E, + Unk7c = 0x1F, + Yield = 0x20, + NonPullerMethods = 0x40, +}; + struct CommandListHeader { union { u64 raw; @@ -49,9 +74,29 @@ union CommandHeader { static_assert(std::is_standard_layout_v<CommandHeader>, "CommandHeader is not standard layout"); static_assert(sizeof(CommandHeader) == sizeof(u32), "CommandHeader has incorrect size!"); +static constexpr CommandHeader BuildCommandHeader(BufferMethods method, u32 arg_count, + SubmissionMode mode) { + CommandHeader result{}; + result.method.Assign(static_cast<u32>(method)); + result.arg_count.Assign(arg_count); + result.mode.Assign(mode); + return result; +} + class GPU; -using CommandList = std::vector<Tegra::CommandListHeader>; +struct CommandList final { + CommandList() = default; + explicit CommandList(std::size_t size) : command_lists(size) {} + explicit CommandList(std::vector<Tegra::CommandHeader>&& prefetch_command_list) + : prefetch_command_list{std::move(prefetch_command_list)} {} + + void RefreshIntegrityChecks(GPU& gpu); + + std::vector<Tegra::CommandListHeader> command_lists; + std::vector<u64> command_list_hashes; + std::vector<Tegra::CommandHeader> prefetch_command_list; +}; /** * The DmaPusher class implements DMA submission to FIFOs, providing an area of memory that the @@ -60,7 +105,7 @@ using CommandList = std::vector<Tegra::CommandListHeader>; * See https://envytools.readthedocs.io/en/latest/hw/fifo/dma-pusher.html#fifo-dma-pusher for * details on this implementation. */ -class DmaPusher { +class DmaPusher final { public: explicit DmaPusher(Core::System& system, GPU& gpu); ~DmaPusher(); diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 171f78183..ebd149c3a 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -194,30 +194,6 @@ void GPU::SyncGuestHost() { void GPU::OnCommandListEnd() { renderer->Rasterizer().ReleaseFences(); } -// Note that, traditionally, methods are treated as 4-byte addressable locations, and hence -// their numbers are written down multiplied by 4 in Docs. Here we are not multiply by 4. -// So the values you see in docs might be multiplied by 4. -enum class BufferMethods { - BindObject = 0x0, - Nop = 0x2, - SemaphoreAddressHigh = 0x4, - SemaphoreAddressLow = 0x5, - SemaphoreSequence = 0x6, - SemaphoreTrigger = 0x7, - NotifyIntr = 0x8, - WrcacheFlush = 0x9, - Unk28 = 0xA, - UnkCacheFlush = 0xB, - RefCnt = 0x14, - SemaphoreAcquire = 0x1A, - SemaphoreRelease = 0x1B, - FenceValue = 0x1C, - FenceAction = 0x1D, - Unk78 = 0x1E, - Unk7c = 0x1F, - Yield = 0x20, - NonPullerMethods = 0x40, -}; enum class GpuSemaphoreOperation { AcquireEqual = 0x1, @@ -277,7 +253,12 @@ void GPU::CallPullerMethod(const MethodCall& method_call) { case BufferMethods::UnkCacheFlush: case BufferMethods::WrcacheFlush: case BufferMethods::FenceValue: + break; case BufferMethods::FenceAction: + ProcessFenceActionMethod(); + break; + case BufferMethods::WaitForInterrupt: + ProcessWaitForInterruptMethod(); break; case BufferMethods::SemaphoreTrigger: { ProcessSemaphoreTriggerMethod(); @@ -391,6 +372,25 @@ void GPU::ProcessBindMethod(const MethodCall& method_call) { } } +void GPU::ProcessFenceActionMethod() { + switch (regs.fence_action.op) { + case FenceOperation::Acquire: + WaitFence(regs.fence_action.syncpoint_id, regs.fence_value); + break; + case FenceOperation::Increment: + IncrementSyncPoint(regs.fence_action.syncpoint_id); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented operation {}", + static_cast<u32>(regs.fence_action.op.Value())); + } +} + +void GPU::ProcessWaitForInterruptMethod() { + // TODO(bunnei) ImplementMe + LOG_WARNING(HW_GPU, "(STUBBED) called"); +} + void GPU::ProcessSemaphoreTriggerMethod() { const auto semaphoreOperationMask = 0xF; const auto op = diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index b8c613b11..5444b49f3 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -263,6 +263,24 @@ public: return use_nvdec; } + enum class FenceOperation : u32 { + Acquire = 0, + Increment = 1, + }; + + union FenceAction { + u32 raw; + BitField<0, 1, FenceOperation> op; + BitField<8, 24, u32> syncpoint_id; + + static constexpr CommandHeader Build(FenceOperation op, u32 syncpoint_id) { + FenceAction result{}; + result.op.Assign(op); + result.syncpoint_id.Assign(syncpoint_id); + return {result.raw}; + } + }; + struct Regs { static constexpr size_t NUM_REGS = 0x40; @@ -291,10 +309,7 @@ public: u32 semaphore_acquire; u32 semaphore_release; u32 fence_value; - union { - BitField<4, 4, u32> operation; - BitField<8, 8, u32> id; - } fence_action; + FenceAction fence_action; INSERT_UNION_PADDING_WORDS(0xE2); // Puller state @@ -342,6 +357,8 @@ protected: private: void ProcessBindMethod(const MethodCall& method_call); + void ProcessFenceActionMethod(); + void ProcessWaitForInterruptMethod(); void ProcessSemaphoreTriggerMethod(); void ProcessSemaphoreRelease(); void ProcessSemaphoreAcquire(); diff --git a/src/video_core/renderer_vulkan/vk_device.cpp b/src/video_core/renderer_vulkan/vk_device.cpp index e1217ca83..f34ed6735 100644 --- a/src/video_core/renderer_vulkan/vk_device.cpp +++ b/src/video_core/renderer_vulkan/vk_device.cpp @@ -771,13 +771,18 @@ void VKDevice::CollectTelemetryParameters() { VkPhysicalDeviceDriverPropertiesKHR driver{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR, .pNext = nullptr, + .driverID = {}, + .driverName = {}, + .driverInfo = {}, + .conformanceVersion = {}, }; - VkPhysicalDeviceProperties2KHR properties{ + VkPhysicalDeviceProperties2KHR device_properties{ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR, .pNext = &driver, + .properties = {}, }; - physical.GetProperties2KHR(properties); + physical.GetProperties2KHR(device_properties); driver_id = driver.driverID; vendor_name = driver.driverName; diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 696eaeb5f..0e8f9c352 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -159,6 +159,7 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules( .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO, .pNext = nullptr, .flags = 0, + .codeSize = 0, }; std::vector<vk::ShaderModule> modules; @@ -388,6 +389,7 @@ vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpa .logicOp = VK_LOGIC_OP_COPY, .attachmentCount = static_cast<u32>(num_attachments), .pAttachments = cb_attachments.data(), + .blendConstants = {}, }; std::vector dynamic_states{ diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp index e8515321b..13dd16356 100644 --- a/src/video_core/texture_cache/surface_params.cpp +++ b/src/video_core/texture_cache/surface_params.cpp @@ -240,6 +240,7 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface( .is_tiled = is_tiled, .srgb_conversion = config.format == Tegra::RenderTargetFormat::B8G8R8A8_SRGB || config.format == Tegra::RenderTargetFormat::A8B8G8R8_SRGB, + .is_layered = false, .block_width = is_tiled ? std::min(config.BlockWidth(), 5U) : 0U, .block_height = is_tiled ? std::min(config.BlockHeight(), 5U) : 0U, .block_depth = is_tiled ? std::min(config.BlockDepth(), 5U) : 0U, |