diff options
Diffstat (limited to 'src/video_core')
37 files changed, 950 insertions, 738 deletions
diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index f0f450edb..8be7bd594 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -289,8 +289,11 @@ std::pair<typename P::Buffer*, u32> BufferCache<P>::ObtainBuffer(GPUVAddr gpu_ad MarkWrittenBuffer(buffer_id, *cpu_addr, size); break; case ObtainBufferOperation::DiscardWrite: { - IntervalType interval{*cpu_addr, size}; + VAddr cpu_addr_start = Common::AlignDown(*cpu_addr, 64); + VAddr cpu_addr_end = Common::AlignUp(*cpu_addr + size, 64); + IntervalType interval{cpu_addr_start, cpu_addr_end}; ClearDownload(interval); + common_ranges.subtract(interval); break; } default: @@ -1159,6 +1162,11 @@ void BufferCache<P>::UpdateDrawIndirect() { .size = static_cast<u32>(size), .buffer_id = FindBuffer(*cpu_addr, static_cast<u32>(size)), }; + VAddr cpu_addr_start = Common::AlignDown(*cpu_addr, 64); + VAddr cpu_addr_end = Common::AlignUp(*cpu_addr + size, 64); + IntervalType interval{cpu_addr_start, cpu_addr_end}; + ClearDownload(interval); + common_ranges.subtract(interval); }; if (current_draw_indirect->include_count) { update(current_draw_indirect->count_start_address, sizeof(u32), diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp index 9f1b340a9..58ce0d8c2 100644 --- a/src/video_core/dma_pusher.cpp +++ b/src/video_core/dma_pusher.cpp @@ -14,6 +14,7 @@ namespace Tegra { constexpr u32 MacroRegistersStart = 0xE00; +constexpr u32 ComputeInline = 0x6D; DmaPusher::DmaPusher(Core::System& system_, GPU& gpu_, MemoryManager& memory_manager_, Control::ChannelState& channel_state_) @@ -83,12 +84,35 @@ bool DmaPusher::Step() { dma_state.dma_get, command_list_header.size * sizeof(u32)); } } - Core::Memory::GpuGuestMemory<Tegra::CommandHeader, - Core::Memory::GuestMemoryFlags::UnsafeRead> - headers(memory_manager, dma_state.dma_get, command_list_header.size, &command_headers); - ProcessCommands(headers); + const auto safe_process = [&] { + Core::Memory::GpuGuestMemory<Tegra::CommandHeader, + Core::Memory::GuestMemoryFlags::SafeRead> + headers(memory_manager, dma_state.dma_get, command_list_header.size, + &command_headers); + ProcessCommands(headers); + }; + const auto unsafe_process = [&] { + Core::Memory::GpuGuestMemory<Tegra::CommandHeader, + Core::Memory::GuestMemoryFlags::UnsafeRead> + headers(memory_manager, dma_state.dma_get, command_list_header.size, + &command_headers); + ProcessCommands(headers); + }; + if (Settings::IsGPULevelHigh()) { + if (dma_state.method >= MacroRegistersStart) { + unsafe_process(); + return true; + } + if (subchannel_type[dma_state.subchannel] == Engines::EngineTypes::KeplerCompute && + dma_state.method == ComputeInline) { + unsafe_process(); + return true; + } + safe_process(); + return true; + } + unsafe_process(); } - return true; } diff --git a/src/video_core/dma_pusher.h b/src/video_core/dma_pusher.h index 8a2784cdc..c9fab2d90 100644 --- a/src/video_core/dma_pusher.h +++ b/src/video_core/dma_pusher.h @@ -130,8 +130,10 @@ public: void DispatchCalls(); - void BindSubchannel(Engines::EngineInterface* engine, u32 subchannel_id) { + void BindSubchannel(Engines::EngineInterface* engine, u32 subchannel_id, + Engines::EngineTypes engine_type) { subchannels[subchannel_id] = engine; + subchannel_type[subchannel_id] = engine_type; } void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); @@ -170,6 +172,7 @@ private: const bool ib_enable{true}; ///< IB mode enabled std::array<Engines::EngineInterface*, max_subchannels> subchannels{}; + std::array<Engines::EngineTypes, max_subchannels> subchannel_type; GPU& gpu; Core::System& system; diff --git a/src/video_core/engines/engine_interface.h b/src/video_core/engines/engine_interface.h index 392322358..54631ee6c 100644 --- a/src/video_core/engines/engine_interface.h +++ b/src/video_core/engines/engine_interface.h @@ -11,6 +11,14 @@ namespace Tegra::Engines { +enum class EngineTypes : u32 { + KeplerCompute, + Maxwell3D, + Fermi2D, + MaxwellDMA, + KeplerMemory, +}; + class EngineInterface { public: virtual ~EngineInterface() = default; diff --git a/src/video_core/engines/engine_upload.h b/src/video_core/engines/engine_upload.h index 7242d2529..21bf8aeb4 100644 --- a/src/video_core/engines/engine_upload.h +++ b/src/video_core/engines/engine_upload.h @@ -69,6 +69,14 @@ public: /// Binds a rasterizer to this engine. void BindRasterizer(VideoCore::RasterizerInterface* rasterizer); + GPUVAddr ExecTargetAddress() const { + return regs.dest.Address(); + } + + u32 GetUploadSize() const { + return copy_size; + } + private: void ProcessData(std::span<const u8> read_buffer); diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index a38d9528a..cd61ab222 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -43,16 +43,33 @@ void KeplerCompute::CallMethod(u32 method, u32 method_argument, bool is_last_cal switch (method) { case KEPLER_COMPUTE_REG_INDEX(exec_upload): { + UploadInfo info{.upload_address = upload_address, + .exec_address = upload_state.ExecTargetAddress(), + .copy_size = upload_state.GetUploadSize()}; + uploads.push_back(info); upload_state.ProcessExec(regs.exec_upload.linear != 0); break; } case KEPLER_COMPUTE_REG_INDEX(data_upload): { + upload_address = current_dma_segment; upload_state.ProcessData(method_argument, is_last_call); break; } - case KEPLER_COMPUTE_REG_INDEX(launch): + case KEPLER_COMPUTE_REG_INDEX(launch): { + const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address(); + + for (auto& data : uploads) { + const GPUVAddr offset = data.exec_address - launch_desc_loc; + if (offset / sizeof(u32) == LAUNCH_REG_INDEX(grid_dim_x) && + memory_manager.IsMemoryDirty(data.upload_address, data.copy_size)) { + indirect_compute = {data.upload_address}; + } + } + uploads.clear(); ProcessLaunch(); + indirect_compute = std::nullopt; break; + } default: break; } @@ -62,6 +79,7 @@ void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amoun u32 methods_pending) { switch (method) { case KEPLER_COMPUTE_REG_INDEX(data_upload): + upload_address = current_dma_segment; upload_state.ProcessData(base_start, amount); return; default: diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h index 2092e685f..735e05fb4 100644 --- a/src/video_core/engines/kepler_compute.h +++ b/src/video_core/engines/kepler_compute.h @@ -5,6 +5,7 @@ #include <array> #include <cstddef> +#include <optional> #include <vector> #include "common/bit_field.h" #include "common/common_funcs.h" @@ -36,6 +37,9 @@ namespace Tegra::Engines { #define KEPLER_COMPUTE_REG_INDEX(field_name) \ (offsetof(Tegra::Engines::KeplerCompute::Regs, field_name) / sizeof(u32)) +#define LAUNCH_REG_INDEX(field_name) \ + (offsetof(Tegra::Engines::KeplerCompute::LaunchParams, field_name) / sizeof(u32)) + class KeplerCompute final : public EngineInterface { public: explicit KeplerCompute(Core::System& system, MemoryManager& memory_manager); @@ -201,6 +205,10 @@ public: void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) override; + std::optional<GPUVAddr> GetIndirectComputeAddress() const { + return indirect_compute; + } + private: void ProcessLaunch(); @@ -216,6 +224,15 @@ private: MemoryManager& memory_manager; VideoCore::RasterizerInterface* rasterizer = nullptr; Upload::State upload_state; + GPUVAddr upload_address; + + struct UploadInfo { + GPUVAddr upload_address; + GPUVAddr exec_address; + u32 copy_size; + }; + std::vector<UploadInfo> uploads; + std::optional<GPUVAddr> indirect_compute{}; }; #define ASSERT_REG_POSITION(field_name, position) \ diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index c3696096d..06e349e43 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -257,6 +257,7 @@ u32 Maxwell3D::GetMaxCurrentVertices() { const u32 address_size = static_cast<u32>(gpu_addr_end - gpu_addr_begin); num_vertices = std::max( num_vertices, address_size / std::max(attribute.SizeInBytes(), array.stride.Value())); + break; } return num_vertices; } @@ -269,10 +270,13 @@ size_t Maxwell3D::EstimateIndexBufferSize() { std::numeric_limits<u32>::max()}; const size_t byte_size = regs.index_buffer.FormatSizeInBytes(); const size_t log2_byte_size = Common::Log2Ceil64(byte_size); + const size_t cap{GetMaxCurrentVertices() * 3 * byte_size}; + const size_t lower_cap = + std::min<size_t>(static_cast<size_t>(end_address - start_address), cap); return std::min<size_t>( memory_manager.GetMemoryLayoutSize(start_address, byte_size * max_sizes[log2_byte_size]) / byte_size, - static_cast<size_t>(end_address - start_address)); + lower_cap); } u32 Maxwell3D::ProcessShadowRam(u32 method, u32 argument) { diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index cd8e24b0b..da8eab7ee 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -5,6 +5,7 @@ #include "common/assert.h" #include "common/logging/log.h" #include "common/microprofile.h" +#include "common/polyfill_ranges.h" #include "common/settings.h" #include "core/core.h" #include "core/memory.h" diff --git a/src/video_core/engines/puller.cpp b/src/video_core/engines/puller.cpp index 7718a09b3..6de2543b7 100644 --- a/src/video_core/engines/puller.cpp +++ b/src/video_core/engines/puller.cpp @@ -34,19 +34,24 @@ void Puller::ProcessBindMethod(const MethodCall& method_call) { bound_engines[method_call.subchannel] = engine_id; switch (engine_id) { case EngineID::FERMI_TWOD_A: - dma_pusher.BindSubchannel(channel_state.fermi_2d.get(), method_call.subchannel); + dma_pusher.BindSubchannel(channel_state.fermi_2d.get(), method_call.subchannel, + EngineTypes::Fermi2D); break; case EngineID::MAXWELL_B: - dma_pusher.BindSubchannel(channel_state.maxwell_3d.get(), method_call.subchannel); + dma_pusher.BindSubchannel(channel_state.maxwell_3d.get(), method_call.subchannel, + EngineTypes::Maxwell3D); break; case EngineID::KEPLER_COMPUTE_B: - dma_pusher.BindSubchannel(channel_state.kepler_compute.get(), method_call.subchannel); + dma_pusher.BindSubchannel(channel_state.kepler_compute.get(), method_call.subchannel, + EngineTypes::KeplerCompute); break; case EngineID::MAXWELL_DMA_COPY_A: - dma_pusher.BindSubchannel(channel_state.maxwell_dma.get(), method_call.subchannel); + dma_pusher.BindSubchannel(channel_state.maxwell_dma.get(), method_call.subchannel, + EngineTypes::MaxwellDMA); break; case EngineID::KEPLER_INLINE_TO_MEMORY_B: - dma_pusher.BindSubchannel(channel_state.kepler_memory.get(), method_call.subchannel); + dma_pusher.BindSubchannel(channel_state.kepler_memory.get(), method_call.subchannel, + EngineTypes::KeplerMemory); break; default: UNIMPLEMENTED_MSG("Unimplemented engine {:04X}", engine_id); diff --git a/src/video_core/host1x/codecs/codec.cpp b/src/video_core/host1x/codecs/codec.cpp index 220cce28a..8d7da50fc 100644 --- a/src/video_core/host1x/codecs/codec.cpp +++ b/src/video_core/host1x/codecs/codec.cpp @@ -319,6 +319,7 @@ void Codec::Decode() { LOG_WARNING(Service_NVDRV, "Zero width or height in frame"); return; } + bool is_interlaced = initial_frame->interlaced_frame != 0; if (av_codec_ctx->hw_device_ctx) { final_frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter}; ASSERT_MSG(final_frame, "av_frame_alloc final_frame failed"); @@ -334,7 +335,7 @@ void Codec::Decode() { UNIMPLEMENTED_MSG("Unexpected video format: {}", final_frame->format); return; } - if (!final_frame->interlaced_frame) { + if (!is_interlaced) { av_frames.push(std::move(final_frame)); } else { if (!filters_initialized) { diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index e61d9af80..c4d459077 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -50,6 +50,7 @@ set(SHADER_FILES vulkan_blit_depth_stencil.frag vulkan_color_clear.frag vulkan_color_clear.vert + vulkan_depthstencil_clear.frag vulkan_fidelityfx_fsr_easu_fp16.comp vulkan_fidelityfx_fsr_easu_fp32.comp vulkan_fidelityfx_fsr_rcas_fp16.comp diff --git a/src/video_core/host_shaders/astc_decoder.comp b/src/video_core/host_shaders/astc_decoder.comp index bf2693559..5ff17cd0c 100644 --- a/src/video_core/host_shaders/astc_decoder.comp +++ b/src/video_core/host_shaders/astc_decoder.comp @@ -33,26 +33,14 @@ UNIFORM(6) uint block_height_mask; END_PUSH_CONSTANTS struct EncodingData { - uint encoding; - uint num_bits; - uint bit_value; - uint quint_trit_value; + uint data; }; -struct TexelWeightParams { - uvec2 size; - uint max_weight; - bool dual_plane; - bool error_state; - bool void_extent_ldr; - bool void_extent_hdr; -}; - -layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 { +layout(binding = BINDING_INPUT_BUFFER, std430) readonly restrict buffer InputBufferU32 { uvec4 astc_data[]; }; -layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image; +layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly restrict image2DArray dest_image; const uint GOB_SIZE_X_SHIFT = 6; const uint GOB_SIZE_Y_SHIFT = 3; @@ -60,64 +48,21 @@ const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT; const uint BYTES_PER_BLOCK_LOG2 = 4; -const int JUST_BITS = 0; -const int QUINT = 1; -const int TRIT = 2; +const uint JUST_BITS = 0u; +const uint QUINT = 1u; +const uint TRIT = 2u; // ASTC Encodings data, sorted in ascending order based on their BitLength value // (see GetBitLength() function) -EncodingData encoding_values[22] = EncodingData[]( - EncodingData(JUST_BITS, 0, 0, 0), EncodingData(JUST_BITS, 1, 0, 0), EncodingData(TRIT, 0, 0, 0), - EncodingData(JUST_BITS, 2, 0, 0), EncodingData(QUINT, 0, 0, 0), EncodingData(TRIT, 1, 0, 0), - EncodingData(JUST_BITS, 3, 0, 0), EncodingData(QUINT, 1, 0, 0), EncodingData(TRIT, 2, 0, 0), - EncodingData(JUST_BITS, 4, 0, 0), EncodingData(QUINT, 2, 0, 0), EncodingData(TRIT, 3, 0, 0), - EncodingData(JUST_BITS, 5, 0, 0), EncodingData(QUINT, 3, 0, 0), EncodingData(TRIT, 4, 0, 0), - EncodingData(JUST_BITS, 6, 0, 0), EncodingData(QUINT, 4, 0, 0), EncodingData(TRIT, 5, 0, 0), - EncodingData(JUST_BITS, 7, 0, 0), EncodingData(QUINT, 5, 0, 0), EncodingData(TRIT, 6, 0, 0), - EncodingData(JUST_BITS, 8, 0, 0) -); - -// The following constants are expanded variants of the Replicate() -// function calls corresponding to the following arguments: -// value: index into the generated table -// num_bits: the after "REPLICATE" in the table name. i.e. 4 is num_bits in REPLICATE_4. -// to_bit: the integer after "TO_" -const uint REPLICATE_BIT_TO_7_TABLE[2] = uint[](0, 127); -const uint REPLICATE_1_BIT_TO_9_TABLE[2] = uint[](0, 511); - -const uint REPLICATE_1_BIT_TO_8_TABLE[2] = uint[](0, 255); -const uint REPLICATE_2_BIT_TO_8_TABLE[4] = uint[](0, 85, 170, 255); -const uint REPLICATE_3_BIT_TO_8_TABLE[8] = uint[](0, 36, 73, 109, 146, 182, 219, 255); -const uint REPLICATE_4_BIT_TO_8_TABLE[16] = - uint[](0, 17, 34, 51, 68, 85, 102, 119, 136, 153, 170, 187, 204, 221, 238, 255); -const uint REPLICATE_5_BIT_TO_8_TABLE[32] = - uint[](0, 8, 16, 24, 33, 41, 49, 57, 66, 74, 82, 90, 99, 107, 115, 123, 132, 140, 148, 156, 165, - 173, 181, 189, 198, 206, 214, 222, 231, 239, 247, 255); -const uint REPLICATE_1_BIT_TO_6_TABLE[2] = uint[](0, 63); -const uint REPLICATE_2_BIT_TO_6_TABLE[4] = uint[](0, 21, 42, 63); -const uint REPLICATE_3_BIT_TO_6_TABLE[8] = uint[](0, 9, 18, 27, 36, 45, 54, 63); -const uint REPLICATE_4_BIT_TO_6_TABLE[16] = - uint[](0, 4, 8, 12, 17, 21, 25, 29, 34, 38, 42, 46, 51, 55, 59, 63); -const uint REPLICATE_5_BIT_TO_6_TABLE[32] = - uint[](0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 33, 35, 37, 39, 41, 43, 45, - 47, 49, 51, 53, 55, 57, 59, 61, 63); -const uint REPLICATE_6_BIT_TO_8_TABLE[64] = - uint[](0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 65, 69, 73, 77, 81, 85, 89, - 93, 97, 101, 105, 109, 113, 117, 121, 125, 130, 134, 138, 142, 146, 150, 154, 158, 162, - 166, 170, 174, 178, 182, 186, 190, 195, 199, 203, 207, 211, 215, 219, 223, 227, 231, 235, - 239, 243, 247, 251, 255); -const uint REPLICATE_7_BIT_TO_8_TABLE[128] = - uint[](0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, 38, 40, 42, 44, - 46, 48, 50, 52, 54, 56, 58, 60, 62, 64, 66, 68, 70, 72, 74, 76, 78, 80, 82, 84, 86, 88, - 90, 92, 94, 96, 98, 100, 102, 104, 106, 108, 110, 112, 114, 116, 118, 120, 122, 124, 126, - 129, 131, 133, 135, 137, 139, 141, 143, 145, 147, 149, 151, 153, 155, 157, 159, 161, 163, - 165, 167, 169, 171, 173, 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, - 201, 203, 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, 235, - 237, 239, 241, 243, 245, 247, 249, 251, 253, 255); +const uint encoding_values[22] = uint[]( + (JUST_BITS), (JUST_BITS | (1u << 8u)), (TRIT), (JUST_BITS | (2u << 8u)), + (QUINT), (TRIT | (1u << 8u)), (JUST_BITS | (3u << 8u)), (QUINT | (1u << 8u)), + (TRIT | (2u << 8u)), (JUST_BITS | (4u << 8u)), (QUINT | (2u << 8u)), (TRIT | (3u << 8u)), + (JUST_BITS | (5u << 8u)), (QUINT | (3u << 8u)), (TRIT | (4u << 8u)), (JUST_BITS | (6u << 8u)), + (QUINT | (4u << 8u)), (TRIT | (5u << 8u)), (JUST_BITS | (7u << 8u)), (QUINT | (5u << 8u)), + (TRIT | (6u << 8u)), (JUST_BITS | (8u << 8u))); // Input ASTC texture globals -uint current_index = 0; -int bitsread = 0; int total_bitsread = 0; uvec4 local_buff; @@ -125,50 +70,60 @@ uvec4 local_buff; uvec4 color_endpoint_data; int color_bitsread = 0; -// Four values, two endpoints, four maximum partitions -uint color_values[32]; -int colvals_index = 0; - -// Weight data globals -uvec4 texel_weight_data; -int texel_bitsread = 0; +// Global "vector" to be pushed into when decoding +// At most will require BLOCK_WIDTH x BLOCK_HEIGHT in single plane mode +// At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode +// So the maximum would be 144 (12 x 12) elements, x 2 for two planes +#define DIVCEIL(number, divisor) (number + divisor - 1) / divisor +#define ARRAY_NUM_ELEMENTS 144 +#define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4) +uint result_vector[ARRAY_NUM_ELEMENTS * 2]; -bool texel_flag = false; - -// Global "vectors" to be pushed into when decoding -EncodingData result_vector[144]; int result_index = 0; +uint result_vector_max_index; +bool result_limit_reached = false; -EncodingData texel_vector[144]; -int texel_vector_index = 0; +// EncodingData helpers +uint Encoding(EncodingData val) { + return bitfieldExtract(val.data, 0, 8); +} +uint NumBits(EncodingData val) { + return bitfieldExtract(val.data, 8, 8); +} +uint BitValue(EncodingData val) { + return bitfieldExtract(val.data, 16, 8); +} +uint QuintTritValue(EncodingData val) { + return bitfieldExtract(val.data, 24, 8); +} -uint unquantized_texel_weights[2][144]; +void Encoding(inout EncodingData val, uint v) { + val.data = bitfieldInsert(val.data, v, 0, 8); +} +void NumBits(inout EncodingData val, uint v) { + val.data = bitfieldInsert(val.data, v, 8, 8); +} +void BitValue(inout EncodingData val, uint v) { + val.data = bitfieldInsert(val.data, v, 16, 8); +} +void QuintTritValue(inout EncodingData val, uint v) { + val.data = bitfieldInsert(val.data, v, 24, 8); +} -uint SwizzleOffset(uvec2 pos) { - uint x = pos.x; - uint y = pos.y; - return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 + - (y % 2) * 16 + (x % 16); +EncodingData CreateEncodingData(uint encoding, uint num_bits, uint bit_val, uint quint_trit_val) { + return EncodingData(((encoding) << 0u) | ((num_bits) << 8u) | + ((bit_val) << 16u) | ((quint_trit_val) << 24u)); } -// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)] -// is the same as [(num_bits - 1):0] and repeats all the way down. -uint Replicate(uint val, uint num_bits, uint to_bit) { - const uint v = val & uint((1 << num_bits) - 1); - uint res = v; - uint reslen = num_bits; - while (reslen < to_bit) { - uint comp = 0; - if (num_bits > to_bit - reslen) { - uint newshift = to_bit - reslen; - comp = num_bits - newshift; - num_bits = newshift; - } - res = uint(res << num_bits); - res = uint(res | (v >> comp)); - reslen += num_bits; + +void ResultEmplaceBack(EncodingData val) { + if (result_index >= result_vector_max_index) { + // Alert callers to avoid decoding more than needed by this phase + result_limit_reached = true; + return; } - return res; + result_vector[result_index] = val.data; + ++result_index; } uvec4 ReplicateByteTo16(uvec4 value) { @@ -176,64 +131,40 @@ uvec4 ReplicateByteTo16(uvec4 value) { } uint ReplicateBitTo7(uint value) { - return REPLICATE_BIT_TO_7_TABLE[value]; + return value * 127; } uint ReplicateBitTo9(uint value) { - return REPLICATE_1_BIT_TO_9_TABLE[value]; + return value * 511; } -uint FastReplicate(uint value, uint num_bits, uint to_bit) { - if (num_bits == 0) { +uint ReplicateBits(uint value, uint num_bits, uint to_bit) { + if (value == 0 || num_bits == 0) { return 0; } - if (num_bits == to_bit) { + if (num_bits >= to_bit) { return value; } - if (to_bit == 6) { - switch (num_bits) { - case 1: - return REPLICATE_1_BIT_TO_6_TABLE[value]; - case 2: - return REPLICATE_2_BIT_TO_6_TABLE[value]; - case 3: - return REPLICATE_3_BIT_TO_6_TABLE[value]; - case 4: - return REPLICATE_4_BIT_TO_6_TABLE[value]; - case 5: - return REPLICATE_5_BIT_TO_6_TABLE[value]; - default: - break; - } - } else { /* if (to_bit == 8) */ - switch (num_bits) { - case 1: - return REPLICATE_1_BIT_TO_8_TABLE[value]; - case 2: - return REPLICATE_2_BIT_TO_8_TABLE[value]; - case 3: - return REPLICATE_3_BIT_TO_8_TABLE[value]; - case 4: - return REPLICATE_4_BIT_TO_8_TABLE[value]; - case 5: - return REPLICATE_5_BIT_TO_8_TABLE[value]; - case 6: - return REPLICATE_6_BIT_TO_8_TABLE[value]; - case 7: - return REPLICATE_7_BIT_TO_8_TABLE[value]; - default: - break; - } + const uint v = value & uint((1 << num_bits) - 1); + uint res = v; + uint reslen = num_bits; + while (reslen < to_bit) { + const uint num_dst_bits_to_shift_up = min(num_bits, to_bit - reslen); + const uint num_src_bits_to_shift_down = num_bits - num_dst_bits_to_shift_up; + + res <<= num_dst_bits_to_shift_up; + res |= (v >> num_src_bits_to_shift_down); + reslen += num_bits; } - return Replicate(value, num_bits, to_bit); + return res; } uint FastReplicateTo8(uint value, uint num_bits) { - return FastReplicate(value, num_bits, 8); + return ReplicateBits(value, num_bits, 8); } uint FastReplicateTo6(uint value, uint num_bits) { - return FastReplicate(value, num_bits, 6); + return ReplicateBits(value, num_bits, 6); } uint Div3Floor(uint v) { @@ -266,15 +197,15 @@ uint Hash52(uint p) { return p; } -uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) { - if (small_block) { +uint Select2DPartition(uint seed, uint x, uint y, uint partition_count) { + if ((block_dims.y * block_dims.x) < 32) { x <<= 1; y <<= 1; } seed += (partition_count - 1) * 1024; - uint rnum = Hash52(uint(seed)); + const uint rnum = Hash52(uint(seed)); uint seed1 = uint(rnum & 0xF); uint seed2 = uint((rnum >> 4) & 0xF); uint seed3 = uint((rnum >> 8) & 0xF); @@ -342,53 +273,52 @@ uint ExtractBits(uvec4 payload, int offset, int bits) { if (bits <= 0) { return 0; } - int last_offset = offset + bits - 1; - int shifted_offset = offset >> 5; + if (bits > 32) { + return 0; + } + const int last_offset = offset + bits - 1; + const int shifted_offset = offset >> 5; if ((last_offset >> 5) == shifted_offset) { return bitfieldExtract(payload[shifted_offset], offset & 31, bits); } - int first_bits = 32 - (offset & 31); - int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits)); - int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits)); + const int first_bits = 32 - (offset & 31); + const int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits)); + const int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits)); return result_first | (result_second << first_bits); } uint StreamBits(uint num_bits) { - int int_bits = int(num_bits); - uint ret = ExtractBits(local_buff, total_bitsread, int_bits); + const int int_bits = int(num_bits); + const uint ret = ExtractBits(local_buff, total_bitsread, int_bits); total_bitsread += int_bits; return ret; } +void SkipBits(uint num_bits) { + const int int_bits = int(num_bits); + total_bitsread += int_bits; +} + uint StreamColorBits(uint num_bits) { - uint ret = 0; - int int_bits = int(num_bits); - if (texel_flag) { - ret = ExtractBits(texel_weight_data, texel_bitsread, int_bits); - texel_bitsread += int_bits; - } else { - ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits); - color_bitsread += int_bits; - } + const int int_bits = int(num_bits); + const uint ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits); + color_bitsread += int_bits; return ret; } -void ResultEmplaceBack(EncodingData val) { - if (texel_flag) { - texel_vector[texel_vector_index] = val; - ++texel_vector_index; - } else { - result_vector[result_index] = val; - ++result_index; - } +EncodingData GetEncodingFromVector(uint index) { + const uint data = result_vector[index]; + return EncodingData(data); } // Returns the number of bits required to encode n_vals values. uint GetBitLength(uint n_vals, uint encoding_index) { - uint total_bits = encoding_values[encoding_index].num_bits * n_vals; - if (encoding_values[encoding_index].encoding == TRIT) { + const EncodingData encoding_value = EncodingData(encoding_values[encoding_index]); + const uint encoding = Encoding(encoding_value); + uint total_bits = NumBits(encoding_value) * n_vals; + if (encoding == TRIT) { total_bits += Div5Ceil(n_vals * 8); - } else if (encoding_values[encoding_index].encoding == QUINT) { + } else if (encoding == QUINT) { total_bits += Div3Ceil(n_vals * 7); } return total_bits; @@ -403,7 +333,7 @@ uint GetNumWeightValues(uvec2 size, bool dual_plane) { } uint GetPackedBitSize(uvec2 size, bool dual_plane, uint max_weight) { - uint n_vals = GetNumWeightValues(size, dual_plane); + const uint n_vals = GetNumWeightValues(size, dual_plane); return GetBitLength(n_vals, max_weight); } @@ -412,87 +342,74 @@ uint BitsBracket(uint bits, uint pos) { } uint BitsOp(uint bits, uint start, uint end) { - if (start == end) { - return BitsBracket(bits, start); - } else if (start > end) { - uint t = start; - start = end; - end = t; - } - - uint mask = (1 << (end - start + 1)) - 1; + const uint mask = (1 << (end - start + 1)) - 1; return ((bits >> start) & mask); } void DecodeQuintBlock(uint num_bits) { - uint m[3]; - uint q[3]; - uint Q; + uvec3 m; + uvec4 qQ; m[0] = StreamColorBits(num_bits); - Q = StreamColorBits(3); + qQ.w = StreamColorBits(3); m[1] = StreamColorBits(num_bits); - Q |= StreamColorBits(2) << 3; + qQ.w |= StreamColorBits(2) << 3; m[2] = StreamColorBits(num_bits); - Q |= StreamColorBits(2) << 5; - if (BitsOp(Q, 1, 2) == 3 && BitsOp(Q, 5, 6) == 0) { - q[0] = 4; - q[1] = 4; - q[2] = (BitsBracket(Q, 0) << 2) | ((BitsBracket(Q, 4) & ~BitsBracket(Q, 0)) << 1) | - (BitsBracket(Q, 3) & ~BitsBracket(Q, 0)); + qQ.w |= StreamColorBits(2) << 5; + if (BitsOp(qQ.w, 1, 2) == 3 && BitsOp(qQ.w, 5, 6) == 0) { + qQ.x = 4; + qQ.y = 4; + qQ.z = (BitsBracket(qQ.w, 0) << 2) | ((BitsBracket(qQ.w, 4) & ~BitsBracket(qQ.w, 0)) << 1) | + (BitsBracket(qQ.w, 3) & ~BitsBracket(qQ.w, 0)); } else { uint C = 0; - if (BitsOp(Q, 1, 2) == 3) { - q[2] = 4; - C = (BitsOp(Q, 3, 4) << 3) | ((~BitsOp(Q, 5, 6) & 3) << 1) | BitsBracket(Q, 0); + if (BitsOp(qQ.w, 1, 2) == 3) { + qQ.z = 4; + C = (BitsOp(qQ.w, 3, 4) << 3) | ((~BitsOp(qQ.w, 5, 6) & 3) << 1) | BitsBracket(qQ.w, 0); } else { - q[2] = BitsOp(Q, 5, 6); - C = BitsOp(Q, 0, 4); + qQ.z = BitsOp(qQ.w, 5, 6); + C = BitsOp(qQ.w, 0, 4); } if (BitsOp(C, 0, 2) == 5) { - q[1] = 4; - q[0] = BitsOp(C, 3, 4); + qQ.y = 4; + qQ.x = BitsOp(C, 3, 4); } else { - q[1] = BitsOp(C, 3, 4); - q[0] = BitsOp(C, 0, 2); + qQ.y = BitsOp(C, 3, 4); + qQ.x = BitsOp(C, 0, 2); } } for (uint i = 0; i < 3; i++) { - EncodingData val; - val.encoding = QUINT; - val.num_bits = num_bits; - val.bit_value = m[i]; - val.quint_trit_value = q[i]; + const EncodingData val = CreateEncodingData(QUINT, num_bits, m[i], qQ[i]); ResultEmplaceBack(val); } } void DecodeTritBlock(uint num_bits) { - uint m[5]; - uint t[5]; - uint T; + uvec4 m; + uvec4 t; + uvec3 Tm5t5; m[0] = StreamColorBits(num_bits); - T = StreamColorBits(2); + Tm5t5.x = StreamColorBits(2); m[1] = StreamColorBits(num_bits); - T |= StreamColorBits(2) << 2; + Tm5t5.x |= StreamColorBits(2) << 2; m[2] = StreamColorBits(num_bits); - T |= StreamColorBits(1) << 4; + Tm5t5.x |= StreamColorBits(1) << 4; m[3] = StreamColorBits(num_bits); - T |= StreamColorBits(2) << 5; - m[4] = StreamColorBits(num_bits); - T |= StreamColorBits(1) << 7; + Tm5t5.x |= StreamColorBits(2) << 5; + Tm5t5.y = StreamColorBits(num_bits); + Tm5t5.x |= StreamColorBits(1) << 7; uint C = 0; - if (BitsOp(T, 2, 4) == 7) { - C = (BitsOp(T, 5, 7) << 2) | BitsOp(T, 0, 1); - t[4] = 2; + if (BitsOp(Tm5t5.x, 2, 4) == 7) { + C = (BitsOp(Tm5t5.x, 5, 7) << 2) | BitsOp(Tm5t5.x, 0, 1); + Tm5t5.z = 2; t[3] = 2; } else { - C = BitsOp(T, 0, 4); - if (BitsOp(T, 5, 6) == 3) { - t[4] = 2; - t[3] = BitsBracket(T, 7); + C = BitsOp(Tm5t5.x, 0, 4); + if (BitsOp(Tm5t5.x, 5, 6) == 3) { + Tm5t5.z = 2; + t[3] = BitsBracket(Tm5t5.x, 7); } else { - t[4] = BitsBracket(T, 7); - t[3] = BitsOp(T, 5, 6); + Tm5t5.z = BitsBracket(Tm5t5.x, 7); + t[3] = BitsOp(Tm5t5.x, 5, 6); } } if (BitsOp(C, 0, 1) == 3) { @@ -508,31 +425,31 @@ void DecodeTritBlock(uint num_bits) { t[1] = BitsOp(C, 2, 3); t[0] = (BitsBracket(C, 1) << 1) | (BitsBracket(C, 0) & ~BitsBracket(C, 1)); } - for (uint i = 0; i < 5; i++) { - EncodingData val; - val.encoding = TRIT; - val.num_bits = num_bits; - val.bit_value = m[i]; - val.quint_trit_value = t[i]; + for (uint i = 0; i < 4; i++) { + const EncodingData val = CreateEncodingData(TRIT, num_bits, m[i], t[i]); ResultEmplaceBack(val); } + const EncodingData val = CreateEncodingData(TRIT, num_bits, Tm5t5.y, Tm5t5.z); + ResultEmplaceBack(val); } void DecodeIntegerSequence(uint max_range, uint num_values) { - EncodingData val = encoding_values[max_range]; + EncodingData val = EncodingData(encoding_values[max_range]); + const uint encoding = Encoding(val); + const uint num_bits = NumBits(val); uint vals_decoded = 0; - while (vals_decoded < num_values) { - switch (val.encoding) { + while (vals_decoded < num_values && !result_limit_reached) { + switch (encoding) { case QUINT: - DecodeQuintBlock(val.num_bits); + DecodeQuintBlock(num_bits); vals_decoded += 3; break; case TRIT: - DecodeTritBlock(val.num_bits); + DecodeTritBlock(num_bits); vals_decoded += 5; break; case JUST_BITS: - val.bit_value = StreamColorBits(val.num_bits); + BitValue(val, StreamColorBits(num_bits)); ResultEmplaceBack(val); vals_decoded++; break; @@ -540,7 +457,7 @@ void DecodeIntegerSequence(uint max_range, uint num_values) { } } -void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) { +void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, out uint color_values[32]) { uint num_values = 0; for (uint i = 0; i < num_partitions; i++) { num_values += ((modes[i] >> 2) + 1) << 1; @@ -549,7 +466,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) { // TODO(ameerj): profile with binary search int range = 0; while (++range < encoding_values.length()) { - uint bit_length = GetBitLength(num_values, range); + const uint bit_length = GetBitLength(num_values, range); if (bit_length > color_data_bits) { break; } @@ -560,48 +477,49 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) { if (out_index >= num_values) { break; } - EncodingData val = result_vector[itr]; - uint bitlen = val.num_bits; - uint bitval = val.bit_value; + const EncodingData val = GetEncodingFromVector(itr); + const uint encoding = Encoding(val); + const uint bitlen = NumBits(val); + const uint bitval = BitValue(val); uint A = 0, B = 0, C = 0, D = 0; A = ReplicateBitTo9((bitval & 1)); - switch (val.encoding) { + switch (encoding) { case JUST_BITS: - color_values[out_index++] = FastReplicateTo8(bitval, bitlen); + color_values[++out_index] = FastReplicateTo8(bitval, bitlen); break; case TRIT: { - D = val.quint_trit_value; + D = QuintTritValue(val); switch (bitlen) { case 1: C = 204; break; case 2: { C = 93; - uint b = (bitval >> 1) & 1; + const uint b = (bitval >> 1) & 1; B = (b << 8) | (b << 4) | (b << 2) | (b << 1); break; } case 3: { C = 44; - uint cb = (bitval >> 1) & 3; + const uint cb = (bitval >> 1) & 3; B = (cb << 7) | (cb << 2) | cb; break; } case 4: { C = 22; - uint dcb = (bitval >> 1) & 7; + const uint dcb = (bitval >> 1) & 7; B = (dcb << 6) | dcb; break; } case 5: { C = 11; - uint edcb = (bitval >> 1) & 0xF; + const uint edcb = (bitval >> 1) & 0xF; B = (edcb << 5) | (edcb >> 2); break; } case 6: { C = 5; - uint fedcb = (bitval >> 1) & 0x1F; + const uint fedcb = (bitval >> 1) & 0x1F; B = (fedcb << 4) | (fedcb >> 4); break; } @@ -609,32 +527,32 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) { break; } case QUINT: { - D = val.quint_trit_value; + D = QuintTritValue(val); switch (bitlen) { case 1: C = 113; break; case 2: { C = 54; - uint b = (bitval >> 1) & 1; + const uint b = (bitval >> 1) & 1; B = (b << 8) | (b << 3) | (b << 2); break; } case 3: { C = 26; - uint cb = (bitval >> 1) & 3; + const uint cb = (bitval >> 1) & 3; B = (cb << 7) | (cb << 1) | (cb >> 1); break; } case 4: { C = 13; - uint dcb = (bitval >> 1) & 7; + const uint dcb = (bitval >> 1) & 7; B = (dcb << 6) | (dcb >> 1); break; } case 5: { C = 6; - uint edcb = (bitval >> 1) & 0xF; + const uint edcb = (bitval >> 1) & 0xF; B = (edcb << 5) | (edcb >> 3); break; } @@ -642,11 +560,11 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) { break; } } - if (val.encoding != JUST_BITS) { + if (encoding != JUST_BITS) { uint T = (D * C) + B; T ^= A; T = (A & 0x80) | (T >> 2); - color_values[out_index++] = T; + color_values[++out_index] = T; } } } @@ -664,139 +582,136 @@ ivec2 BitTransferSigned(int a, int b) { } uvec4 ClampByte(ivec4 color) { - for (uint i = 0; i < 4; ++i) { - color[i] = (color[i] < 0) ? 0 : ((color[i] > 255) ? 255 : color[i]); - } - return uvec4(color); + return uvec4(clamp(color, 0, 255)); } ivec4 BlueContract(int a, int r, int g, int b) { return ivec4(a, (r + b) >> 1, (g + b) >> 1, b); } -void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode) { +void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, uint color_values[32], + inout uint colvals_index) { #define READ_UINT_VALUES(N) \ - uint v[N]; \ + uvec4 V[2]; \ for (uint i = 0; i < N; i++) { \ - v[i] = color_values[colvals_index++]; \ + V[i / 4][i % 4] = color_values[++colvals_index]; \ } - #define READ_INT_VALUES(N) \ - int v[N]; \ + ivec4 V[2]; \ for (uint i = 0; i < N; i++) { \ - v[i] = int(color_values[colvals_index++]); \ + V[i / 4][i % 4] = int(color_values[++colvals_index]); \ } switch (color_endpoint_mode) { case 0: { READ_UINT_VALUES(2) - ep1 = uvec4(0xFF, v[0], v[0], v[0]); - ep2 = uvec4(0xFF, v[1], v[1], v[1]); + ep1 = uvec4(0xFF, V[0].x, V[0].x, V[0].x); + ep2 = uvec4(0xFF, V[0].y, V[0].y, V[0].y); break; } case 1: { READ_UINT_VALUES(2) - uint L0 = (v[0] >> 2) | (v[1] & 0xC0); - uint L1 = min(L0 + (v[1] & 0x3F), 0xFFU); + const uint L0 = (V[0].x >> 2) | (V[0].y & 0xC0); + const uint L1 = min(L0 + (V[0].y & 0x3F), 0xFFU); ep1 = uvec4(0xFF, L0, L0, L0); ep2 = uvec4(0xFF, L1, L1, L1); break; } case 4: { READ_UINT_VALUES(4) - ep1 = uvec4(v[2], v[0], v[0], v[0]); - ep2 = uvec4(v[3], v[1], v[1], v[1]); + ep1 = uvec4(V[0].z, V[0].x, V[0].x, V[0].x); + ep2 = uvec4(V[0].w, V[0].y, V[0].y, V[0].y); break; } case 5: { READ_INT_VALUES(4) - ivec2 transferred = BitTransferSigned(v[1], v[0]); - v[1] = transferred.x; - v[0] = transferred.y; - transferred = BitTransferSigned(v[3], v[2]); - v[3] = transferred.x; - v[2] = transferred.y; - ep1 = ClampByte(ivec4(v[2], v[0], v[0], v[0])); - ep2 = ClampByte(ivec4(v[2] + v[3], v[0] + v[1], v[0] + v[1], v[0] + v[1])); + ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); + V[0].y = transferred.x; + V[0].x = transferred.y; + transferred = BitTransferSigned(V[0].w, V[0].z); + V[0].w = transferred.x; + V[0].z = transferred.y; + ep1 = ClampByte(ivec4(V[0].z, V[0].x, V[0].x, V[0].x)); + ep2 = ClampByte(ivec4(V[0].z + V[0].w, V[0].x + V[0].y, V[0].x + V[0].y, V[0].x + V[0].y)); break; } case 6: { READ_UINT_VALUES(4) - ep1 = uvec4(0xFF, (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8); - ep2 = uvec4(0xFF, v[0], v[1], v[2]); + ep1 = uvec4(0xFF, (V[0].x * V[0].w) >> 8, (V[0].y * V[0].w) >> 8, (V[0].z * V[0].w) >> 8); + ep2 = uvec4(0xFF, V[0].x, V[0].y, V[0].z); break; } case 8: { READ_UINT_VALUES(6) - if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) { - ep1 = uvec4(0xFF, v[0], v[2], v[4]); - ep2 = uvec4(0xFF, v[1], v[3], v[5]); + if ((V[0].y + V[0].w + V[1].y) >= (V[0].x + V[0].z + V[1].x)) { + ep1 = uvec4(0xFF, V[0].x, V[0].z, V[1].x); + ep2 = uvec4(0xFF, V[0].y, V[0].w, V[1].y); } else { - ep1 = uvec4(BlueContract(0xFF, int(v[1]), int(v[3]), int(v[5]))); - ep2 = uvec4(BlueContract(0xFF, int(v[0]), int(v[2]), int(v[4]))); + ep1 = uvec4(BlueContract(0xFF, int(V[0].y), int(V[0].w), int(V[1].y))); + ep2 = uvec4(BlueContract(0xFF, int(V[0].x), int(V[0].z), int(V[1].x))); } break; } case 9: { READ_INT_VALUES(6) - ivec2 transferred = BitTransferSigned(v[1], v[0]); - v[1] = transferred.x; - v[0] = transferred.y; - transferred = BitTransferSigned(v[3], v[2]); - v[3] = transferred.x; - v[2] = transferred.y; - transferred = BitTransferSigned(v[5], v[4]); - v[5] = transferred.x; - v[4] = transferred.y; - if ((v[1] + v[3] + v[5]) >= 0) { - ep1 = ClampByte(ivec4(0xFF, v[0], v[2], v[4])); - ep2 = ClampByte(ivec4(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5])); + ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); + V[0].y = transferred.x; + V[0].x = transferred.y; + transferred = BitTransferSigned(V[0].w, V[0].z); + V[0].w = transferred.x; + V[0].z = transferred.y; + transferred = BitTransferSigned(V[1].y, V[1].x); + V[1].y = transferred.x; + V[1].x = transferred.y; + if ((V[0].y + V[0].w + V[1].y) >= 0) { + ep1 = ClampByte(ivec4(0xFF, V[0].x, V[0].z, V[1].x)); + ep2 = ClampByte(ivec4(0xFF, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); } else { - ep1 = ClampByte(BlueContract(0xFF, v[0] + v[1], v[2] + v[3], v[4] + v[5])); - ep2 = ClampByte(BlueContract(0xFF, v[0], v[2], v[4])); + ep1 = ClampByte(BlueContract(0xFF, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); + ep2 = ClampByte(BlueContract(0xFF, V[0].x, V[0].z, V[1].x)); } break; } case 10: { READ_UINT_VALUES(6) - ep1 = uvec4(v[4], (v[0] * v[3]) >> 8, (v[1] * v[3]) >> 8, (v[2] * v[3]) >> 8); - ep2 = uvec4(v[5], v[0], v[1], v[2]); + ep1 = uvec4(V[1].x, (V[0].x * V[0].w) >> 8, (V[0].y * V[0].w) >> 8, (V[0].z * V[0].w) >> 8); + ep2 = uvec4(V[1].y, V[0].x, V[0].y, V[0].z); break; } case 12: { READ_UINT_VALUES(8) - if ((v[1] + v[3] + v[5]) >= (v[0] + v[2] + v[4])) { - ep1 = uvec4(v[6], v[0], v[2], v[4]); - ep2 = uvec4(v[7], v[1], v[3], v[5]); + if ((V[0].y + V[0].w + V[1].y) >= (V[0].x + V[0].z + V[1].x)) { + ep1 = uvec4(V[1].z, V[0].x, V[0].z, V[1].x); + ep2 = uvec4(V[1].w, V[0].y, V[0].w, V[1].y); } else { - ep1 = uvec4(BlueContract(int(v[7]), int(v[1]), int(v[3]), int(v[5]))); - ep2 = uvec4(BlueContract(int(v[6]), int(v[0]), int(v[2]), int(v[4]))); + ep1 = uvec4(BlueContract(int(V[1].w), int(V[0].y), int(V[0].w), int(V[1].y))); + ep2 = uvec4(BlueContract(int(V[1].z), int(V[0].x), int(V[0].z), int(V[1].x))); } break; } case 13: { READ_INT_VALUES(8) - ivec2 transferred = BitTransferSigned(v[1], v[0]); - v[1] = transferred.x; - v[0] = transferred.y; - transferred = BitTransferSigned(v[3], v[2]); - v[3] = transferred.x; - v[2] = transferred.y; - - transferred = BitTransferSigned(v[5], v[4]); - v[5] = transferred.x; - v[4] = transferred.y; - - transferred = BitTransferSigned(v[7], v[6]); - v[7] = transferred.x; - v[6] = transferred.y; - - if ((v[1] + v[3] + v[5]) >= 0) { - ep1 = ClampByte(ivec4(v[6], v[0], v[2], v[4])); - ep2 = ClampByte(ivec4(v[7] + v[6], v[0] + v[1], v[2] + v[3], v[4] + v[5])); + ivec2 transferred = BitTransferSigned(V[0].y, V[0].x); + V[0].y = transferred.x; + V[0].x = transferred.y; + transferred = BitTransferSigned(V[0].w, V[0].z); + V[0].w = transferred.x; + V[0].z = transferred.y; + + transferred = BitTransferSigned(V[1].y, V[1].x); + V[1].y = transferred.x; + V[1].x = transferred.y; + + transferred = BitTransferSigned(V[1].w, V[1].z); + V[1].w = transferred.x; + V[1].z = transferred.y; + + if ((V[0].y + V[0].w + V[1].y) >= 0) { + ep1 = ClampByte(ivec4(V[1].z, V[0].x, V[0].z, V[1].x)); + ep2 = ClampByte(ivec4(V[1].w + V[1].z, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); } else { - ep1 = ClampByte(BlueContract(v[6] + v[7], v[0] + v[1], v[2] + v[3], v[4] + v[5])); - ep2 = ClampByte(BlueContract(v[6], v[0], v[2], v[4])); + ep1 = ClampByte(BlueContract(V[1].z + V[1].w, V[0].x + V[0].y, V[0].z + V[0].w, V[1].x + V[1].y)); + ep2 = ClampByte(BlueContract(V[1].z, V[0].x, V[0].z, V[1].x)); } break; } @@ -812,36 +727,34 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode) { } uint UnquantizeTexelWeight(EncodingData val) { - uint bitval = val.bit_value; - uint bitlen = val.num_bits; - uint A = ReplicateBitTo7((bitval & 1)); + const uint encoding = Encoding(val); + const uint bitlen = NumBits(val); + const uint bitval = BitValue(val); + const uint A = ReplicateBitTo7((bitval & 1)); uint B = 0, C = 0, D = 0; uint result = 0; - switch (val.encoding) { + const uint bitlen_0_results[5] = {0, 16, 32, 48, 64}; + switch (encoding) { case JUST_BITS: - result = FastReplicateTo6(bitval, bitlen); - break; + return FastReplicateTo6(bitval, bitlen); case TRIT: { - D = val.quint_trit_value; + D = QuintTritValue(val); switch (bitlen) { - case 0: { - uint results[3] = {0, 32, 63}; - result = results[D]; - break; - } + case 0: + return bitlen_0_results[D * 2]; case 1: { C = 50; break; } case 2: { C = 23; - uint b = (bitval >> 1) & 1; + const uint b = (bitval >> 1) & 1; B = (b << 6) | (b << 2) | b; break; } case 3: { C = 11; - uint cb = (bitval >> 1) & 3; + const uint cb = (bitval >> 1) & 3; B = (cb << 5) | cb; break; } @@ -851,20 +764,17 @@ uint UnquantizeTexelWeight(EncodingData val) { break; } case QUINT: { - D = val.quint_trit_value; + D = QuintTritValue(val); switch (bitlen) { - case 0: { - uint results[5] = {0, 16, 32, 47, 63}; - result = results[D]; - break; - } + case 0: + return bitlen_0_results[D]; case 1: { C = 28; break; } case 2: { C = 13; - uint b = (bitval >> 1) & 1; + const uint b = (bitval >> 1) & 1; B = (b << 6) | (b << 1); break; } @@ -872,7 +782,7 @@ uint UnquantizeTexelWeight(EncodingData val) { break; } } - if (val.encoding != JUST_BITS && bitlen > 0) { + if (encoding != JUST_BITS && bitlen > 0) { result = D * C + B; result ^= A; result = (A & 0x20) | (result >> 2); @@ -883,61 +793,77 @@ uint UnquantizeTexelWeight(EncodingData val) { return result; } -void UnquantizeTexelWeights(bool dual_plane, uvec2 size) { - uint weight_idx = 0; - uint unquantized[2][144]; - uint area = size.x * size.y; - for (uint itr = 0; itr < texel_vector_index; itr++) { - unquantized[0][weight_idx] = UnquantizeTexelWeight(texel_vector[itr]); - if (dual_plane) { - ++itr; - unquantized[1][weight_idx] = UnquantizeTexelWeight(texel_vector[itr]); - if (itr == texel_vector_index) { - break; - } - } - if (++weight_idx >= (area)) - break; +void UnquantizeTexelWeights(uvec2 size, bool is_dual_plane) { + const uint num_planes = is_dual_plane ? 2 : 1; + const uint area = size.x * size.y; + const uint loop_count = min(result_index, area * num_planes); + for (uint itr = 0; itr < loop_count; ++itr) { + result_vector[itr] = + UnquantizeTexelWeight(GetEncodingFromVector(itr)); } +} + +uint GetUnquantizedTexelWieght(uint offset_base, uint plane, bool is_dual_plane) { + const uint offset = is_dual_plane ? 2 * offset_base + plane : offset_base; + return result_vector[offset]; +} +uvec4 GetUnquantizedWeightVector(uint t, uint s, uvec2 size, uint plane_index, bool is_dual_plane) { const uint Ds = uint((block_dims.x * 0.5f + 1024) / (block_dims.x - 1)); const uint Dt = uint((block_dims.y * 0.5f + 1024) / (block_dims.y - 1)); - const uint k_plane_scale = dual_plane ? 2 : 1; - for (uint plane = 0; plane < k_plane_scale; plane++) { - for (uint t = 0; t < block_dims.y; t++) { - for (uint s = 0; s < block_dims.x; s++) { - uint cs = Ds * s; - uint ct = Dt * t; - uint gs = (cs * (size.x - 1) + 32) >> 6; - uint gt = (ct * (size.y - 1) + 32) >> 6; - uint js = gs >> 4; - uint fs = gs & 0xF; - uint jt = gt >> 4; - uint ft = gt & 0x0F; - uint w11 = (fs * ft + 8) >> 4; - uint w10 = ft - w11; - uint w01 = fs - w11; - uint w00 = 16 - fs - ft + w11; - uvec4 w = uvec4(w00, w01, w10, w11); - uint v0 = jt * size.x + js; - - uvec4 p = uvec4(0); - if (v0 < area) { - p.x = unquantized[plane][v0]; - } - if ((v0 + 1) < (area)) { - p.y = unquantized[plane][v0 + 1]; - } - if ((v0 + size.x) < (area)) { - p.z = unquantized[plane][(v0 + size.x)]; - } - if ((v0 + size.x + 1) < (area)) { - p.w = unquantized[plane][(v0 + size.x + 1)]; - } - unquantized_texel_weights[plane][t * block_dims.x + s] = (uint(dot(p, w)) + 8) >> 4; - } + const uint area = size.x * size.y; + + const uint cs = Ds * s; + const uint ct = Dt * t; + const uint gs = (cs * (size.x - 1) + 32) >> 6; + const uint gt = (ct * (size.y - 1) + 32) >> 6; + const uint js = gs >> 4; + const uint fs = gs & 0xF; + const uint jt = gt >> 4; + const uint ft = gt & 0x0F; + const uint w11 = (fs * ft + 8) >> 4; + const uint w10 = ft - w11; + const uint w01 = fs - w11; + const uint w00 = 16 - fs - ft + w11; + const uvec4 w = uvec4(w00, w01, w10, w11); + const uint v0 = jt * size.x + js; + + uvec4 p0 = uvec4(0); + uvec4 p1 = uvec4(0); + + if (v0 < area) { + const uint offset_base = v0; + p0.x = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane); + p1.x = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane); + } + if ((v0 + 1) < (area)) { + const uint offset_base = v0 + 1; + p0.y = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane); + p1.y = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane); + } + if ((v0 + size.x) < (area)) { + const uint offset_base = v0 + size.x; + p0.z = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane); + p1.z = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane); + } + if ((v0 + size.x + 1) < (area)) { + const uint offset_base = v0 + size.x + 1; + p0.w = GetUnquantizedTexelWieght(offset_base, 0, is_dual_plane); + p1.w = GetUnquantizedTexelWieght(offset_base, 1, is_dual_plane); + } + + const uint primary_weight = (uint(dot(p0, w)) + 8) >> 4; + + uvec4 weight_vec = uvec4(primary_weight); + + if (is_dual_plane) { + const uint secondary_weight = (uint(dot(p1, w)) + 8) >> 4; + for (uint c = 0; c < 4; c++) { + const bool is_secondary = ((plane_index + 1u) & 3u) == c; + weight_vec[c] = is_secondary ? secondary_weight : primary_weight; } } + return weight_vec; } int FindLayout(uint mode) { @@ -971,80 +897,96 @@ int FindLayout(uint mode) { return 5; } -TexelWeightParams DecodeBlockInfo() { - TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false, false, false, false); - uint mode = StreamBits(11); + +void FillError(ivec3 coord) { + for (uint j = 0; j < block_dims.y; j++) { + for (uint i = 0; i < block_dims.x; i++) { + imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0)); + } + } +} + +void FillVoidExtentLDR(ivec3 coord) { + SkipBits(52); + const uint r_u = StreamBits(16); + const uint g_u = StreamBits(16); + const uint b_u = StreamBits(16); + const uint a_u = StreamBits(16); + const float a = float(a_u) / 65535.0f; + const float r = float(r_u) / 65535.0f; + const float g = float(g_u) / 65535.0f; + const float b = float(b_u) / 65535.0f; + for (uint j = 0; j < block_dims.y; j++) { + for (uint i = 0; i < block_dims.x; i++) { + imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a)); + } + } +} + +bool IsError(uint mode) { if ((mode & 0x1ff) == 0x1fc) { if ((mode & 0x200) != 0) { - params.void_extent_hdr = true; - } else { - params.void_extent_ldr = true; + // params.void_extent_hdr = true; + return true; } if ((mode & 0x400) == 0 || StreamBits(1) == 0) { - params.error_state = true; + return true; } - return params; + return false; } if ((mode & 0xf) == 0) { - params.error_state = true; - return params; + return true; } if ((mode & 3) == 0 && (mode & 0x1c0) == 0x1c0) { - params.error_state = true; - return params; + return true; } + return false; +} + +uvec2 DecodeBlockSize(uint mode) { uint A, B; - uint mode_layout = FindLayout(mode); - switch (mode_layout) { + switch (FindLayout(mode)) { case 0: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x3; - params.size = uvec2(B + 4, A + 2); - break; + return uvec2(B + 4, A + 2); case 1: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x3; - params.size = uvec2(B + 8, A + 2); - break; + return uvec2(B + 8, A + 2); case 2: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x3; - params.size = uvec2(A + 2, B + 8); - break; + return uvec2(A + 2, B + 8); case 3: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x1; - params.size = uvec2(A + 2, B + 6); - break; + return uvec2(A + 2, B + 6); case 4: A = (mode >> 5) & 0x3; B = (mode >> 7) & 0x1; - params.size = uvec2(B + 2, A + 2); - break; + return uvec2(B + 2, A + 2); case 5: A = (mode >> 5) & 0x3; - params.size = uvec2(12, A + 2); - break; + return uvec2(12, A + 2); case 6: A = (mode >> 5) & 0x3; - params.size = uvec2(A + 2, 12); - break; + return uvec2(A + 2, 12); case 7: - params.size = uvec2(6, 10); - break; + return uvec2(6, 10); case 8: - params.size = uvec2(10, 6); - break; + return uvec2(10, 6); case 9: A = (mode >> 5) & 0x3; B = (mode >> 9) & 0x3; - params.size = uvec2(A + 6, B + 6); - break; + return uvec2(A + 6, B + 6); default: - params.error_state = true; - break; + return uvec2(0); } - params.dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0); +} + +uint DecodeMaxWeight(uint mode) { + const uint mode_layout = FindLayout(mode); uint weight_index = (mode & 0x10) != 0 ? 1 : 0; if (mode_layout < 5) { weight_index |= (mode & 0x3) << 1; @@ -1053,64 +995,34 @@ TexelWeightParams DecodeBlockInfo() { } weight_index -= 2; if ((mode_layout != 9) && ((mode & 0x200) != 0)) { - const int max_weights[6] = int[6](7, 8, 9, 10, 11, 12); - params.max_weight = max_weights[weight_index]; - } else { - const int max_weights[6] = int[6](1, 2, 3, 4, 5, 6); - params.max_weight = max_weights[weight_index]; - } - return params; -} - -void FillError(ivec3 coord) { - for (uint j = 0; j < block_dims.y; j++) { - for (uint i = 0; i < block_dims.x; i++) { - imageStore(dest_image, coord + ivec3(i, j, 0), vec4(0.0, 0.0, 0.0, 0.0)); - } - } -} - -void FillVoidExtentLDR(ivec3 coord) { - StreamBits(52); - uint r_u = StreamBits(16); - uint g_u = StreamBits(16); - uint b_u = StreamBits(16); - uint a_u = StreamBits(16); - float a = float(a_u) / 65535.0f; - float r = float(r_u) / 65535.0f; - float g = float(g_u) / 65535.0f; - float b = float(b_u) / 65535.0f; - for (uint j = 0; j < block_dims.y; j++) { - for (uint i = 0; i < block_dims.x; i++) { - imageStore(dest_image, coord + ivec3(i, j, 0), vec4(r, g, b, a)); - } + weight_index += 6; } + return weight_index + 1; } void DecompressBlock(ivec3 coord) { - TexelWeightParams params = DecodeBlockInfo(); - if (params.error_state) { - FillError(coord); - return; - } - if (params.void_extent_hdr) { + uint mode = StreamBits(11); + if (IsError(mode)) { FillError(coord); return; } - if (params.void_extent_ldr) { + if ((mode & 0x1ff) == 0x1fc) { + // params.void_extent_ldr = true; FillVoidExtentLDR(coord); return; } - if ((params.size.x > block_dims.x) || (params.size.y > block_dims.y)) { + const uvec2 size_params = DecodeBlockSize(mode); + if ((size_params.x > block_dims.x) || (size_params.y > block_dims.y)) { FillError(coord); return; } - uint num_partitions = StreamBits(2) + 1; - if (num_partitions > 4 || (num_partitions == 4 && params.dual_plane)) { + const uint num_partitions = StreamBits(2) + 1; + const uint mode_layout = FindLayout(mode); + const bool dual_plane = (mode_layout != 9) && ((mode & 0x400) != 0); + if (num_partitions > 4 || (num_partitions == 4 && dual_plane)) { FillError(coord); return; } - int plane_index = -1; uint partition_index = 1; uvec4 color_endpoint_mode = uvec4(0); uint ced_pointer = 0; @@ -1122,8 +1034,9 @@ void DecompressBlock(ivec3 coord) { partition_index = StreamBits(10); base_cem = StreamBits(6); } - uint base_mode = base_cem & 3; - uint weight_bits = GetPackedBitSize(params.size, params.dual_plane, params.max_weight); + const uint base_mode = base_cem & 3; + const uint max_weight = DecodeMaxWeight(mode); + const uint weight_bits = GetPackedBitSize(size_params, dual_plane, max_weight); uint remaining_bits = 128 - weight_bits - total_bitsread; uint extra_cem_bits = 0; if (base_mode > 0) { @@ -1142,10 +1055,7 @@ void DecompressBlock(ivec3 coord) { } } remaining_bits -= extra_cem_bits; - uint plane_selector_bits = 0; - if (params.dual_plane) { - plane_selector_bits = 2; - } + const uint plane_selector_bits = dual_plane ? 2 : 0; remaining_bits -= plane_selector_bits; if (remaining_bits > 128) { // Bad data, more remaining bits than 4 bytes @@ -1153,17 +1063,17 @@ void DecompressBlock(ivec3 coord) { return; } // Read color data... - uint color_data_bits = remaining_bits; + const uint color_data_bits = remaining_bits; while (remaining_bits > 0) { - int nb = int(min(remaining_bits, 32U)); - uint b = StreamBits(nb); + const int nb = int(min(remaining_bits, 32U)); + const uint b = StreamBits(nb); color_endpoint_data[ced_pointer] = uint(bitfieldExtract(b, 0, nb)); ++ced_pointer; remaining_bits -= nb; } - plane_index = int(StreamBits(plane_selector_bits)); + const uint plane_index = uint(StreamBits(plane_selector_bits)); if (base_mode > 0) { - uint extra_cem = StreamBits(extra_cem_bits); + const uint extra_cem = StreamBits(extra_cem_bits); uint cem = (extra_cem << 6) | base_cem; cem >>= 2; uvec4 C = uvec4(0); @@ -1185,70 +1095,80 @@ void DecompressBlock(ivec3 coord) { color_endpoint_mode[i] |= M[i]; } } else if (num_partitions > 1) { - uint cem = base_cem >> 2; + const uint cem = base_cem >> 2; for (uint i = 0; i < num_partitions; i++) { color_endpoint_mode[i] = cem; } } - DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits); - uvec4 endpoints[4][2]; - for (uint i = 0; i < num_partitions; i++) { - ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]); + uvec4 endpoints0[4]; + uvec4 endpoints1[4]; + { + // This decode phase should at most push 32 elements into the vector + result_vector_max_index = 32; + uint color_values[32]; + uint colvals_index = 0; + DecodeColorValues(color_endpoint_mode, num_partitions, color_data_bits, color_values); + for (uint i = 0; i < num_partitions; i++) { + ComputeEndpoints(endpoints0[i], endpoints1[i], color_endpoint_mode[i], color_values, + colvals_index); + } } + color_endpoint_data = local_buff; + color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx; + const uint clear_byte_start = (weight_bits >> 3) + 1; - texel_weight_data = local_buff; - texel_weight_data = bitfieldReverse(texel_weight_data).wzyx; - uint clear_byte_start = - (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1; - - uint byte_insert = ExtractBits(texel_weight_data, int(clear_byte_start - 1) * 8, 8) & - uint( - ((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1)); - uint vec_index = (clear_byte_start - 1) >> 2; - texel_weight_data[vec_index] = - bitfieldInsert(texel_weight_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8); + const uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) & + uint(((1 << (weight_bits % 8)) - 1)); + const uint vec_index = (clear_byte_start - 1) >> 2; + color_endpoint_data[vec_index] = bitfieldInsert(color_endpoint_data[vec_index], byte_insert, + int((clear_byte_start - 1) % 4) * 8, 8); for (uint i = clear_byte_start; i < 16; ++i) { - uint idx = i >> 2; - texel_weight_data[idx] = bitfieldInsert(texel_weight_data[idx], 0, int(i % 4) * 8, 8); + const uint idx = i >> 2; + color_endpoint_data[idx] = bitfieldInsert(color_endpoint_data[idx], 0, int(i % 4) * 8, 8); } - texel_flag = true; // use texel "vector" and bit stream in integer decoding - DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane)); - UnquantizeTexelWeights(params.dual_plane, params.size); + // Re-init vector variables for next decode phase + result_index = 0; + color_bitsread = 0; + result_limit_reached = false; + // The limit for the Unquantize phase, avoids decoding more data than needed. + result_vector_max_index = size_params.x * size_params.y; + if (dual_plane) { + result_vector_max_index *= 2; + } + DecodeIntegerSequence(max_weight, GetNumWeightValues(size_params, dual_plane)); + + UnquantizeTexelWeights(size_params, dual_plane); for (uint j = 0; j < block_dims.y; j++) { for (uint i = 0; i < block_dims.x; i++) { uint local_partition = 0; if (num_partitions > 1) { - local_partition = Select2DPartition(partition_index, i, j, num_partitions, - (block_dims.y * block_dims.x) < 32); - } - vec4 p; - uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]); - uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]); - uvec4 plane_vec = uvec4(0); - uvec4 weight_vec = uvec4(0); - for (uint c = 0; c < 4; c++) { - if (params.dual_plane && (((plane_index + 1) & 3) == c)) { - plane_vec[c] = 1; - } - weight_vec[c] = unquantized_texel_weights[plane_vec[c]][j * block_dims.x + i]; + local_partition = Select2DPartition(partition_index, i, j, num_partitions); } - vec4 Cf = vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64); - p = (Cf / 65535.0); + const uvec4 C0 = ReplicateByteTo16(endpoints0[local_partition]); + const uvec4 C1 = ReplicateByteTo16(endpoints1[local_partition]); + const uvec4 weight_vec = GetUnquantizedWeightVector(j, i, size_params, plane_index, dual_plane); + const vec4 Cf = + vec4((C0 * (uvec4(64) - weight_vec) + C1 * weight_vec + uvec4(32)) / 64); + const vec4 p = (Cf / 65535.0f); imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar); } } } +uint SwizzleOffset(uvec2 pos) { + const uint x = pos.x; + const uint y = pos.y; + return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + + ((x % 32) / 16) * 32 + (y % 2) * 16 + (x % 16); +} + void main() { uvec3 pos = gl_GlobalInvocationID; pos.x <<= BYTES_PER_BLOCK_LOG2; - - // Read as soon as possible due to its latency const uint swizzle = SwizzleOffset(pos.xy); - const uint block_y = pos.y >> GOB_SIZE_Y_SHIFT; uint offset = 0; @@ -1262,8 +1182,6 @@ void main() { if (any(greaterThanEqual(coord, imageSize(dest_image)))) { return; } - current_index = 0; - bitsread = 0; local_buff = astc_data[offset / 16]; DecompressBlock(coord); } diff --git a/src/video_core/host_shaders/vulkan_depthstencil_clear.frag b/src/video_core/host_shaders/vulkan_depthstencil_clear.frag new file mode 100644 index 000000000..1ac177c7e --- /dev/null +++ b/src/video_core/host_shaders/vulkan_depthstencil_clear.frag @@ -0,0 +1,12 @@ +// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#version 460 core + +layout (push_constant) uniform PushConstants { + vec4 clear_depth; +}; + +void main() { + gl_FragDepth = clear_depth.x; +} diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp index 905505ca1..5d0bb9cc4 100644 --- a/src/video_core/macro/macro.cpp +++ b/src/video_core/macro/macro.cpp @@ -27,14 +27,24 @@ MICROPROFILE_DEFINE(MacroHLE, "GPU", "Execute macro HLE", MP_RGB(128, 192, 192)) namespace Tegra { -static void Dump(u64 hash, std::span<const u32> code) { +static void Dump(u64 hash, std::span<const u32> code, bool decompiled = false) { const auto base_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::DumpDir)}; const auto macro_dir{base_dir / "macros"}; if (!Common::FS::CreateDir(base_dir) || !Common::FS::CreateDir(macro_dir)) { LOG_ERROR(Common_Filesystem, "Failed to create macro dump directories"); return; } - const auto name{macro_dir / fmt::format("{:016x}.macro", hash)}; + auto name{macro_dir / fmt::format("{:016x}.macro", hash)}; + + if (decompiled) { + auto new_name{macro_dir / fmt::format("decompiled_{:016x}.macro", hash)}; + if (Common::FS::Exists(name)) { + (void)Common::FS::RenameFile(name, new_name); + return; + } + name = new_name; + } + std::fstream macro_file(name, std::ios::out | std::ios::binary); if (!macro_file) { LOG_ERROR(Common_Filesystem, "Unable to open or create file at {}", @@ -90,9 +100,6 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { if (!mid_method.has_value()) { cache_info.lle_program = Compile(macro_code->second); cache_info.hash = Common::HashValue(macro_code->second); - if (Settings::values.dump_macros) { - Dump(cache_info.hash, macro_code->second); - } } else { const auto& macro_cached = uploaded_macro_code[mid_method.value()]; const auto rebased_method = method - mid_method.value(); @@ -102,9 +109,6 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { code.size() * sizeof(u32)); cache_info.hash = Common::HashValue(code); cache_info.lle_program = Compile(code); - if (Settings::values.dump_macros) { - Dump(cache_info.hash, code); - } } auto hle_program = hle_macros->GetHLEProgram(cache_info.hash); @@ -117,6 +121,10 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) { MICROPROFILE_SCOPE(MacroHLE); cache_info.hle_program->Execute(parameters, method); } + + if (Settings::values.dump_macros) { + Dump(cache_info.hash, macro_code->second, cache_info.has_hle_program); + } } } diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp index f822fa856..44a771d65 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.cpp @@ -220,7 +220,8 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c ASSERT(num_textures <= MAX_TEXTURES); ASSERT(num_images <= MAX_IMAGES); - const bool assembly_shaders{assembly_programs[0].handle != 0}; + const auto backend = device.GetShaderBackend(); + const bool assembly_shaders{backend == Settings::ShaderBackend::Glasm}; use_storage_buffers = !assembly_shaders || num_storage_buffers <= device.GetMaxGLASMStorageBufferBlocks(); writes_global_memory &= !use_storage_buffers; @@ -230,7 +231,6 @@ GraphicsPipeline::GraphicsPipeline(const Device& device, TextureCache& texture_c GenerateTransformFeedbackState(); } const bool in_parallel = thread_worker != nullptr; - const auto backend = device.GetShaderBackend(); auto func{[this, sources_ = std::move(sources), sources_spirv_ = std::move(sources_spirv), shader_notify, backend, in_parallel, force_context_flush](ShaderContext::Context*) mutable { @@ -559,15 +559,13 @@ void GraphicsPipeline::ConfigureImpl(bool is_indexed) { } void GraphicsPipeline::ConfigureTransformFeedbackImpl() const { - glTransformFeedbackStreamAttribsNV(num_xfb_attribs, xfb_attribs.data(), num_xfb_strides, - xfb_streams.data(), GL_INTERLEAVED_ATTRIBS); + glTransformFeedbackAttribsNV(num_xfb_attribs, xfb_attribs.data(), GL_SEPARATE_ATTRIBS); } void GraphicsPipeline::GenerateTransformFeedbackState() { // TODO(Rodrigo): Inject SKIP_COMPONENTS*_NV when required. An unimplemented message will signal // when this is required. GLint* cursor{xfb_attribs.data()}; - GLint* current_stream{xfb_streams.data()}; for (size_t feedback = 0; feedback < Maxwell::NumTransformFeedbackBuffers; ++feedback) { const auto& layout = key.xfb_state.layouts[feedback]; @@ -575,15 +573,6 @@ void GraphicsPipeline::GenerateTransformFeedbackState() { if (layout.varying_count == 0) { continue; } - *current_stream = static_cast<GLint>(feedback); - if (current_stream != xfb_streams.data()) { - // When stepping one stream, push the expected token - cursor[0] = GL_NEXT_BUFFER_NV; - cursor[1] = 0; - cursor[2] = 0; - cursor += XFB_ENTRY_STRIDE; - } - ++current_stream; const auto& locations = key.xfb_state.varyings[feedback]; std::optional<u32> current_index; @@ -619,7 +608,6 @@ void GraphicsPipeline::GenerateTransformFeedbackState() { } } num_xfb_attribs = static_cast<GLsizei>((cursor - xfb_attribs.data()) / XFB_ENTRY_STRIDE); - num_xfb_strides = static_cast<GLsizei>(current_stream - xfb_streams.data()); } void GraphicsPipeline::WaitForBuild() { diff --git a/src/video_core/renderer_opengl/gl_graphics_pipeline.h b/src/video_core/renderer_opengl/gl_graphics_pipeline.h index 7b3d7eae8..74fc9cc3d 100644 --- a/src/video_core/renderer_opengl/gl_graphics_pipeline.h +++ b/src/video_core/renderer_opengl/gl_graphics_pipeline.h @@ -154,9 +154,7 @@ private: static constexpr std::size_t XFB_ENTRY_STRIDE = 3; GLsizei num_xfb_attribs{}; - GLsizei num_xfb_strides{}; std::array<GLint, 128 * XFB_ENTRY_STRIDE * Maxwell::NumTransformFeedbackBuffers> xfb_attribs{}; - std::array<GLint, Maxwell::NumTransformFeedbackBuffers> xfb_streams{}; std::mutex built_mutex; std::condition_variable built_condvar; diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index aadd6967c..dd03efecd 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -380,6 +380,17 @@ void RasterizerOpenGL::DispatchCompute() { pipeline->SetEngine(kepler_compute, gpu_memory); pipeline->Configure(); const auto& qmd{kepler_compute->launch_description}; + auto indirect_address = kepler_compute->GetIndirectComputeAddress(); + if (indirect_address) { + // DispatchIndirect + static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; + const auto post_op = VideoCommon::ObtainBufferOperation::DiscardWrite; + const auto [buffer, offset] = + buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op); + glBindBuffer(GL_DISPATCH_INDIRECT_BUFFER, buffer->Handle()); + glDispatchComputeIndirect(static_cast<GLintptr>(offset)); + return; + } glDispatchCompute(qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z); ++num_queued_commands; has_written_global_memory |= pipeline->WritesGlobalMemory(); @@ -1335,7 +1346,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, } const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height); static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; - const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing; + const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing + : VideoCommon::ObtainBufferOperation::MarkAsWritten; const auto [buffer, offset] = buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op); @@ -1344,8 +1356,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const std::span copy_span{©, 1}; if constexpr (IS_IMAGE_UPLOAD) { + texture_cache.PrepareImage(image_id, true, false); image->UploadMemory(buffer->Handle(), offset, copy_span); } else { + if (offset % BytesPerBlock(image->info.format)) { + return false; + } texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span, buffer_operand.address, buffer_size); } diff --git a/src/video_core/renderer_opengl/util_shaders.cpp b/src/video_core/renderer_opengl/util_shaders.cpp index 544982d18..c437013e6 100644 --- a/src/video_core/renderer_opengl/util_shaders.cpp +++ b/src/video_core/renderer_opengl/util_shaders.cpp @@ -68,6 +68,7 @@ void UtilShaders::ASTCDecode(Image& image, const StagingBufferMap& map, std::span<const VideoCommon::SwizzleParameters> swizzles) { static constexpr GLuint BINDING_INPUT_BUFFER = 0; static constexpr GLuint BINDING_OUTPUT_IMAGE = 0; + program_manager.LocalMemoryWarmup(); const Extent2D tile_size{ .width = VideoCore::Surface::DefaultBlockWidth(image.info.format), diff --git a/src/video_core/renderer_vulkan/blit_image.cpp b/src/video_core/renderer_vulkan/blit_image.cpp index f74ae972e..1032c9d12 100644 --- a/src/video_core/renderer_vulkan/blit_image.cpp +++ b/src/video_core/renderer_vulkan/blit_image.cpp @@ -16,6 +16,7 @@ #include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h" #include "video_core/host_shaders/vulkan_color_clear_frag_spv.h" #include "video_core/host_shaders/vulkan_color_clear_vert_spv.h" +#include "video_core/host_shaders/vulkan_depthstencil_clear_frag_spv.h" #include "video_core/renderer_vulkan/blit_image.h" #include "video_core/renderer_vulkan/maxwell_to_vk.h" #include "video_core/renderer_vulkan/vk_scheduler.h" @@ -428,6 +429,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_, blit_depth_stencil_frag(BuildShader(device, VULKAN_BLIT_DEPTH_STENCIL_FRAG_SPV)), clear_color_vert(BuildShader(device, VULKAN_COLOR_CLEAR_VERT_SPV)), clear_color_frag(BuildShader(device, VULKAN_COLOR_CLEAR_FRAG_SPV)), + clear_stencil_frag(BuildShader(device, VULKAN_DEPTHSTENCIL_CLEAR_FRAG_SPV)), convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)), convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)), convert_abgr8_to_d24s8_frag(BuildShader(device, CONVERT_ABGR8_TO_D24S8_FRAG_SPV)), @@ -593,6 +595,28 @@ void BlitImageHelper::ClearColor(const Framebuffer* dst_framebuffer, u8 color_ma scheduler.InvalidateState(); } +void BlitImageHelper::ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear, + f32 clear_depth, u8 stencil_mask, u32 stencil_ref, + u32 stencil_compare_mask, const Region2D& dst_region) { + const BlitDepthStencilPipelineKey key{ + .renderpass = dst_framebuffer->RenderPass(), + .depth_clear = depth_clear, + .stencil_mask = stencil_mask, + .stencil_compare_mask = stencil_compare_mask, + .stencil_ref = stencil_ref, + }; + const VkPipeline pipeline = FindOrEmplaceClearStencilPipeline(key); + const VkPipelineLayout layout = *clear_color_pipeline_layout; + scheduler.RequestRenderpass(dst_framebuffer); + scheduler.Record([pipeline, layout, clear_depth, dst_region](vk::CommandBuffer cmdbuf) { + cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline); + BindBlitState(cmdbuf, dst_region); + cmdbuf.PushConstants(layout, VK_SHADER_STAGE_FRAGMENT_BIT, clear_depth); + cmdbuf.Draw(3, 1, 0, 0); + }); + scheduler.InvalidateState(); +} + void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, const ImageView& src_image_view) { const VkPipelineLayout layout = *one_texture_pipeline_layout; @@ -820,6 +844,61 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel return *clear_color_pipelines.back(); } +VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline( + const BlitDepthStencilPipelineKey& key) { + const auto it = std::ranges::find(clear_stencil_keys, key); + if (it != clear_stencil_keys.end()) { + return *clear_stencil_pipelines[std::distance(clear_stencil_keys.begin(), it)]; + } + clear_stencil_keys.push_back(key); + const std::array stages = MakeStages(*clear_color_vert, *clear_stencil_frag); + const auto stencil = VkStencilOpState{ + .failOp = VK_STENCIL_OP_KEEP, + .passOp = VK_STENCIL_OP_REPLACE, + .depthFailOp = VK_STENCIL_OP_KEEP, + .compareOp = VK_COMPARE_OP_ALWAYS, + .compareMask = key.stencil_compare_mask, + .writeMask = key.stencil_mask, + .reference = key.stencil_ref, + }; + const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{ + .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .depthTestEnable = VK_FALSE, + .depthWriteEnable = key.depth_clear, + .depthCompareOp = VK_COMPARE_OP_ALWAYS, + .depthBoundsTestEnable = VK_FALSE, + .stencilTestEnable = VK_TRUE, + .front = stencil, + .back = stencil, + .minDepthBounds = 0.0f, + .maxDepthBounds = 0.0f, + }; + clear_stencil_pipelines.push_back(device.GetLogical().CreateGraphicsPipeline({ + .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO, + .pNext = nullptr, + .flags = 0, + .stageCount = static_cast<u32>(stages.size()), + .pStages = stages.data(), + .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO, + .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO, + .pTessellationState = nullptr, + .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO, + .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO, + .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO, + .pDepthStencilState = &depth_stencil_ci, + .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO, + .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO, + .layout = *clear_color_pipeline_layout, + .renderPass = key.renderpass, + .subpass = 0, + .basePipelineHandle = VK_NULL_HANDLE, + .basePipelineIndex = 0, + })); + return *clear_stencil_pipelines.back(); +} + void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth) { if (pipeline) { diff --git a/src/video_core/renderer_vulkan/blit_image.h b/src/video_core/renderer_vulkan/blit_image.h index 2976a7d91..dcfe217aa 100644 --- a/src/video_core/renderer_vulkan/blit_image.h +++ b/src/video_core/renderer_vulkan/blit_image.h @@ -27,6 +27,16 @@ struct BlitImagePipelineKey { Tegra::Engines::Fermi2D::Operation operation; }; +struct BlitDepthStencilPipelineKey { + constexpr auto operator<=>(const BlitDepthStencilPipelineKey&) const noexcept = default; + + VkRenderPass renderpass; + bool depth_clear; + u8 stencil_mask; + u32 stencil_compare_mask; + u32 stencil_ref; +}; + class BlitImageHelper { public: explicit BlitImageHelper(const Device& device, Scheduler& scheduler, @@ -64,6 +74,10 @@ public: void ClearColor(const Framebuffer* dst_framebuffer, u8 color_mask, const std::array<f32, 4>& clear_color, const Region2D& dst_region); + void ClearDepthStencil(const Framebuffer* dst_framebuffer, bool depth_clear, f32 clear_depth, + u8 stencil_mask, u32 stencil_ref, u32 stencil_compare_mask, + const Region2D& dst_region); + private: void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer, const ImageView& src_image_view); @@ -76,6 +90,8 @@ private: [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key); [[nodiscard]] VkPipeline FindOrEmplaceClearColorPipeline(const BlitImagePipelineKey& key); + [[nodiscard]] VkPipeline FindOrEmplaceClearStencilPipeline( + const BlitDepthStencilPipelineKey& key); void ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass, bool is_target_depth); @@ -108,6 +124,7 @@ private: vk::ShaderModule blit_depth_stencil_frag; vk::ShaderModule clear_color_vert; vk::ShaderModule clear_color_frag; + vk::ShaderModule clear_stencil_frag; vk::ShaderModule convert_depth_to_float_frag; vk::ShaderModule convert_float_to_depth_frag; vk::ShaderModule convert_abgr8_to_d24s8_frag; @@ -122,6 +139,8 @@ private: std::vector<vk::Pipeline> blit_depth_stencil_pipelines; std::vector<BlitImagePipelineKey> clear_color_keys; std::vector<vk::Pipeline> clear_color_pipelines; + std::vector<BlitDepthStencilPipelineKey> clear_stencil_keys; + std::vector<vk::Pipeline> clear_stencil_pipelines; vk::Pipeline convert_d32_to_r32_pipeline; vk::Pipeline convert_r32_to_d32_pipeline; vk::Pipeline convert_d16_to_r16_pipeline; diff --git a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp index a8540339d..35bf80ea3 100644 --- a/src/video_core/renderer_vulkan/maxwell_to_vk.cpp +++ b/src/video_core/renderer_vulkan/maxwell_to_vk.cpp @@ -126,7 +126,7 @@ struct FormatTuple { {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable}, // A1R5G5B5_UNORM {VK_FORMAT_A2B10G10R10_UNORM_PACK32, Attachable | Storage}, // A2B10G10R10_UNORM {VK_FORMAT_A2B10G10R10_UINT_PACK32, Attachable | Storage}, // A2B10G10R10_UINT - {VK_FORMAT_A2R10G10B10_UNORM_PACK32, Attachable | Storage}, // A2R10G10B10_UNORM + {VK_FORMAT_A2R10G10B10_UNORM_PACK32, Attachable}, // A2R10G10B10_UNORM {VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable}, // A1B5G5R5_UNORM (flipped with swizzle) {VK_FORMAT_R5G5B5A1_UNORM_PACK16}, // A5B5G5R1_UNORM (specially swizzled) {VK_FORMAT_R8_UNORM, Attachable | Storage}, // R8_UNORM diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp index 454bb66a4..c4c30d807 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp +++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp @@ -66,21 +66,6 @@ std::string BuildCommaSeparatedExtensions( return fmt::format("{}", fmt::join(available_extensions, ",")); } -DebugCallback MakeDebugCallback(const vk::Instance& instance, const vk::InstanceDispatch& dld) { - if (!Settings::values.renderer_debug) { - return DebugCallback{}; - } - const std::optional properties = vk::EnumerateInstanceExtensionProperties(dld); - const auto it = std::ranges::find_if(*properties, [](const auto& prop) { - return std::strcmp(VK_EXT_DEBUG_UTILS_EXTENSION_NAME, prop.extensionName) == 0; - }); - if (it != properties->end()) { - return CreateDebugUtilsCallback(instance); - } else { - return CreateDebugReportCallback(instance); - } -} - } // Anonymous namespace Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dld, @@ -103,7 +88,8 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_, cpu_memory(cpu_memory_), gpu(gpu_), library(OpenLibrary(context.get())), instance(CreateInstance(*library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type, Settings::values.renderer_debug.GetValue())), - debug_callback(MakeDebugCallback(instance, dld)), + debug_messenger(Settings::values.renderer_debug ? CreateDebugUtilsCallback(instance) + : vk::DebugUtilsMessenger{}), surface(CreateSurface(instance, render_window.GetWindowInfo())), device(CreateDevice(instance, dld, *surface)), memory_allocator(device), state_tracker(), scheduler(device, state_tracker), diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h index 89e98425e..590bc1c64 100644 --- a/src/video_core/renderer_vulkan/renderer_vulkan.h +++ b/src/video_core/renderer_vulkan/renderer_vulkan.h @@ -35,8 +35,6 @@ class GPU; namespace Vulkan { -using DebugCallback = std::variant<vk::DebugUtilsMessenger, vk::DebugReportCallback>; - Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dld, VkSurfaceKHR surface); @@ -75,7 +73,7 @@ private: vk::InstanceDispatch dld; vk::Instance instance; - DebugCallback debug_callback; + vk::DebugUtilsMessenger debug_messenger; vk::SurfaceKHR surface; ScreenInfo screen_info; diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp index 60a6ac651..e15865d16 100644 --- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp @@ -529,17 +529,20 @@ void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bi buffer_handles.push_back(handle); } if (device.IsExtExtendedDynamicStateSupported()) { - scheduler.Record([bindings_ = std::move(bindings), + scheduler.Record([this, bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles)](vk::CommandBuffer cmdbuf) { cmdbuf.BindVertexBuffers2EXT(bindings_.min_index, - bindings_.max_index - bindings_.min_index, + std::min(bindings_.max_index - bindings_.min_index, + device.GetMaxVertexInputBindings()), buffer_handles_.data(), bindings_.offsets.data(), bindings_.sizes.data(), bindings_.strides.data()); }); } else { - scheduler.Record([bindings_ = std::move(bindings), + scheduler.Record([this, bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles)](vk::CommandBuffer cmdbuf) { - cmdbuf.BindVertexBuffers(bindings_.min_index, bindings_.max_index - bindings_.min_index, + cmdbuf.BindVertexBuffers(bindings_.min_index, + std::min(bindings_.max_index - bindings_.min_index, + device.GetMaxVertexInputBindings()), buffer_handles_.data(), bindings_.offsets.data()); }); } diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index c1314ca99..a1ec1a100 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -294,10 +294,11 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device texture_cache{texture_cache_}, shader_notify{shader_notify_}, use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()}, use_vulkan_pipeline_cache{Settings::values.use_vulkan_driver_pipeline_cache.GetValue()}, - workers(device.GetDriverID() == VK_DRIVER_ID_QUALCOMM_PROPRIETARY - ? 1 - : (std::max(std::thread::hardware_concurrency(), 2U) - 1), - "VkPipelineBuilder"), +#ifdef ANDROID + workers(1, "VkPipelineBuilder"), +#else + workers(std::max(std::thread::hardware_concurrency(), 2U) - 1, "VkPipelineBuilder"), +#endif serialization_thread(1, "VkPipelineSerialization") { const auto& float_control{device.FloatControlProperties()}; const VkDriverId driver_id{device.GetDriverID()}; @@ -611,9 +612,6 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))}; Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0); - if (Settings::values.dump_shaders) { - env.Dump(hash, key.unique_hashes[index]); - } if (!uses_vertex_a || index != 1) { // Normal path programs[index] = TranslateProgram(pools.inst, pools.block, env, cfg, host_info); @@ -624,6 +622,10 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( programs[index] = MergeDualVertexPrograms(program_va, program_vb, env); } + if (Settings::values.dump_shaders) { + env.Dump(hash, key.unique_hashes[index]); + } + if (programs[index].info.requires_layer_emulation) { layer_source_program = &programs[index]; } @@ -664,6 +666,19 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline( std::move(modules), infos); } catch (const Shader::Exception& exception) { + auto hash = key.Hash(); + size_t env_index{0}; + for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) { + if (key.unique_hashes[index] == 0) { + continue; + } + Shader::Environment& env{*envs[env_index]}; + ++env_index; + + const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))}; + Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0); + env.Dump(hash, key.unique_hashes[index]); + } LOG_ERROR(Render_Vulkan, "{}", exception.what()); return nullptr; } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index aa59889bd..01e76a82c 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -428,15 +428,27 @@ void RasterizerVulkan::Clear(u32 layer_count) { if (aspect_flags == 0) { return; } - scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil, - clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) { - VkClearAttachment attachment; - attachment.aspectMask = aspect_flags; - attachment.colorAttachment = 0; - attachment.clearValue.depthStencil.depth = clear_depth; - attachment.clearValue.depthStencil.stencil = clear_stencil; - cmdbuf.ClearAttachments(attachment, clear_rect); - }); + + if (use_stencil && regs.stencil_front_mask != 0xFF && regs.stencil_front_mask != 0) { + Region2D dst_region = { + Offset2D{.x = clear_rect.rect.offset.x, .y = clear_rect.rect.offset.y}, + Offset2D{.x = clear_rect.rect.offset.x + static_cast<s32>(clear_rect.rect.extent.width), + .y = clear_rect.rect.offset.y + + static_cast<s32>(clear_rect.rect.extent.height)}}; + blit_image.ClearDepthStencil(framebuffer, use_depth, regs.clear_depth, + static_cast<u8>(regs.stencil_front_mask), regs.clear_stencil, + regs.stencil_front_func_mask, dst_region); + } else { + scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil, + clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) { + VkClearAttachment attachment; + attachment.aspectMask = aspect_flags; + attachment.colorAttachment = 0; + attachment.clearValue.depthStencil.depth = clear_depth; + attachment.clearValue.depthStencil.stencil = clear_stencil; + cmdbuf.ClearAttachments(attachment, clear_rect); + }); + } } void RasterizerVulkan::DispatchCompute() { @@ -451,6 +463,20 @@ void RasterizerVulkan::DispatchCompute() { pipeline->Configure(*kepler_compute, *gpu_memory, scheduler, buffer_cache, texture_cache); const auto& qmd{kepler_compute->launch_description}; + auto indirect_address = kepler_compute->GetIndirectComputeAddress(); + if (indirect_address) { + // DispatchIndirect + static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; + const auto post_op = VideoCommon::ObtainBufferOperation::DiscardWrite; + const auto [buffer, offset] = + buffer_cache.ObtainBuffer(*indirect_address, 12, sync_info, post_op); + scheduler.RequestOutsideRenderPassOperationContext(); + scheduler.Record([indirect_buffer = buffer->Handle(), + indirect_offset = offset](vk::CommandBuffer cmdbuf) { + cmdbuf.DispatchIndirect(indirect_buffer, indirect_offset); + }); + return; + } const std::array<u32, 3> dim{qmd.grid_dim_x, qmd.grid_dim_y, qmd.grid_dim_z}; scheduler.RequestOutsideRenderPassOperationContext(); scheduler.Record([dim](vk::CommandBuffer cmdbuf) { cmdbuf.Dispatch(dim[0], dim[1], dim[2]); }); @@ -830,7 +856,8 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, } const u32 buffer_size = static_cast<u32>(buffer_operand.pitch * buffer_operand.height); static constexpr auto sync_info = VideoCommon::ObtainBufferSynchronize::FullSynchronize; - const auto post_op = VideoCommon::ObtainBufferOperation::DoNothing; + const auto post_op = IS_IMAGE_UPLOAD ? VideoCommon::ObtainBufferOperation::DoNothing + : VideoCommon::ObtainBufferOperation::MarkAsWritten; const auto [buffer, offset] = buffer_cache.ObtainBuffer(buffer_operand.address, buffer_size, sync_info, post_op); @@ -839,8 +866,12 @@ bool AccelerateDMA::DmaBufferImageCopy(const Tegra::DMA::ImageCopy& copy_info, const std::span copy_span{©, 1}; if constexpr (IS_IMAGE_UPLOAD) { + texture_cache.PrepareImage(image_id, true, false); image->UploadMemory(buffer->Handle(), offset, copy_span); } else { + if (offset % BytesPerBlock(image->info.format)) { + return false; + } texture_cache.DownloadImageIntoBuffer(image, buffer->Handle(), offset, copy_span, buffer_operand.address, buffer_size); } diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 4457b366f..1bdb0def5 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -719,6 +719,7 @@ typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_ad return nullptr; } const auto& image_map_ids = it->second; + boost::container::small_vector<const ImageBase*, 4> valid_images; for (const ImageMapId map_id : image_map_ids) { const ImageMapView& map = slot_map_views[map_id]; const ImageBase& image = slot_images[map.image_id]; @@ -728,8 +729,20 @@ typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_ad if (image.image_view_ids.empty()) { continue; } - return &slot_image_views[image.image_view_ids.at(0)]; + valid_images.push_back(&image); } + + if (valid_images.size() == 1) [[likely]] { + return &slot_image_views[valid_images[0]->image_view_ids.at(0)]; + } + + if (valid_images.size() > 0) [[unlikely]] { + std::ranges::sort(valid_images, [](const auto* a, const auto* b) { + return a->modification_tick > b->modification_tick; + }); + return &slot_image_views[valid_images[0]->image_view_ids.at(0)]; + } + return nullptr; } diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h index e9ec91265..a40825c9f 100644 --- a/src/video_core/texture_cache/texture_cache_base.h +++ b/src/video_core/texture_cache/texture_cache_base.h @@ -243,6 +243,9 @@ public: /// Create channel state. void CreateChannel(Tegra::Control::ChannelState& channel) final override; + /// Prepare an image to be used + void PrepareImage(ImageId image_id, bool is_modification, bool invalidate); + std::recursive_mutex mutex; private: @@ -387,9 +390,6 @@ private: /// Synchronize image aliases, copying data if needed void SynchronizeAliases(ImageId image_id); - /// Prepare an image to be used - void PrepareImage(ImageId image_id, bool is_modification, bool invalidate); - /// Prepare an image view to be used void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate); diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.cpp b/src/video_core/vulkan_common/vulkan_debug_callback.cpp index 67e8065a4..448df2d3a 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.cpp +++ b/src/video_core/vulkan_common/vulkan_debug_callback.cpp @@ -63,22 +63,6 @@ VkBool32 DebugUtilCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity, return VK_FALSE; } -VkBool32 DebugReportCallback(VkDebugReportFlagsEXT flags, VkDebugReportObjectTypeEXT objectType, - uint64_t object, size_t location, int32_t messageCode, - const char* pLayerPrefix, const char* pMessage, void* pUserData) { - const VkDebugReportFlagBitsEXT severity = static_cast<VkDebugReportFlagBitsEXT>(flags); - const std::string_view message{pMessage}; - if (severity & VK_DEBUG_REPORT_ERROR_BIT_EXT) { - LOG_CRITICAL(Render_Vulkan, "{}", message); - } else if (severity & VK_DEBUG_REPORT_WARNING_BIT_EXT) { - LOG_WARNING(Render_Vulkan, "{}", message); - } else if (severity & VK_DEBUG_REPORT_INFORMATION_BIT_EXT) { - LOG_INFO(Render_Vulkan, "{}", message); - } else if (severity & VK_DEBUG_REPORT_DEBUG_BIT_EXT) { - LOG_DEBUG(Render_Vulkan, "{}", message); - } - return VK_FALSE; -} } // Anonymous namespace vk::DebugUtilsMessenger CreateDebugUtilsCallback(const vk::Instance& instance) { @@ -98,15 +82,4 @@ vk::DebugUtilsMessenger CreateDebugUtilsCallback(const vk::Instance& instance) { }); } -vk::DebugReportCallback CreateDebugReportCallback(const vk::Instance& instance) { - return instance.CreateDebugReportCallback({ - .sType = VK_STRUCTURE_TYPE_DEBUG_REPORT_CALLBACK_CREATE_INFO_EXT, - .pNext = nullptr, - .flags = VK_DEBUG_REPORT_DEBUG_BIT_EXT | VK_DEBUG_REPORT_INFORMATION_BIT_EXT | - VK_DEBUG_REPORT_ERROR_BIT_EXT | VK_DEBUG_REPORT_WARNING_BIT_EXT, - .pfnCallback = DebugReportCallback, - .pUserData = nullptr, - }); -} - } // namespace Vulkan diff --git a/src/video_core/vulkan_common/vulkan_debug_callback.h b/src/video_core/vulkan_common/vulkan_debug_callback.h index a8af7b406..5e940782f 100644 --- a/src/video_core/vulkan_common/vulkan_debug_callback.h +++ b/src/video_core/vulkan_common/vulkan_debug_callback.h @@ -9,6 +9,4 @@ namespace Vulkan { vk::DebugUtilsMessenger CreateDebugUtilsCallback(const vk::Instance& instance); -vk::DebugReportCallback CreateDebugReportCallback(const vk::Instance& instance); - } // namespace Vulkan diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 710929ac5..617417040 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -71,6 +71,11 @@ constexpr std::array R8G8B8_SSCALED{ VK_FORMAT_UNDEFINED, }; +constexpr std::array VK_FORMAT_R32G32B32_SFLOAT{ + VK_FORMAT_R32G32B32A32_SFLOAT, + VK_FORMAT_UNDEFINED, +}; + } // namespace Alternatives enum class NvidiaArchitecture { @@ -103,6 +108,8 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) { return Alternatives::R16G16B16_SSCALED.data(); case VK_FORMAT_R8G8B8_SSCALED: return Alternatives::R8G8B8_SSCALED.data(); + case VK_FORMAT_R32G32B32_SFLOAT: + return Alternatives::VK_FORMAT_R32G32B32_SFLOAT.data(); default: return nullptr; } @@ -130,6 +137,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::Physica VK_FORMAT_A2B10G10R10_UINT_PACK32, VK_FORMAT_A2B10G10R10_UNORM_PACK32, VK_FORMAT_A2B10G10R10_USCALED_PACK32, + VK_FORMAT_A2R10G10B10_UNORM_PACK32, VK_FORMAT_A8B8G8R8_SINT_PACK32, VK_FORMAT_A8B8G8R8_SNORM_PACK32, VK_FORMAT_A8B8G8R8_SRGB_PACK32, @@ -326,6 +334,43 @@ std::vector<const char*> ExtensionListForVulkan( } // Anonymous namespace +void Device::RemoveExtension(bool& extension, const std::string& extension_name) { + extension = false; + loaded_extensions.erase(extension_name); +} + +void Device::RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name) { + if (loaded_extensions.contains(extension_name) && !is_suitable) { + LOG_WARNING(Render_Vulkan, "Removing unsuitable extension {}", extension_name); + this->RemoveExtension(is_suitable, extension_name); + } +} + +template <typename Feature> +void Device::RemoveExtensionFeature(bool& extension, Feature& feature, + const std::string& extension_name) { + // Unload extension. + this->RemoveExtension(extension, extension_name); + + // Save sType and pNext for chain. + VkStructureType sType = feature.sType; + void* pNext = feature.pNext; + + // Clear feature struct and restore chain. + feature = {}; + feature.sType = sType; + feature.pNext = pNext; +} + +template <typename Feature> +void Device::RemoveExtensionFeatureIfUnsuitable(bool is_suitable, Feature& feature, + const std::string& extension_name) { + if (loaded_extensions.contains(extension_name) && !is_suitable) { + LOG_WARNING(Render_Vulkan, "Removing features for unsuitable extension {}", extension_name); + this->RemoveExtensionFeature(is_suitable, feature, extension_name); + } +} + Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR surface, const vk::InstanceDispatch& dld_) : instance{instance_}, dld{dld_}, physical{physical_}, @@ -397,21 +442,20 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR if (is_qualcomm || is_turnip) { LOG_WARNING(Render_Vulkan, "Qualcomm and Turnip drivers have broken VK_EXT_custom_border_color"); - extensions.custom_border_color = false; - loaded_extensions.erase(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); + RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color, + VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); } if (is_qualcomm) { must_emulate_scaled_formats = true; LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken VK_EXT_extended_dynamic_state"); - extensions.extended_dynamic_state = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state, features.extended_dynamic_state, + VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); LOG_WARNING(Render_Vulkan, "Qualcomm drivers have a slow VK_KHR_push_descriptor implementation"); - extensions.push_descriptor = false; - loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); + RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); #if defined(ANDROID) && defined(ARCHITECTURE_arm64) // Patch the driver to enable BCn textures. @@ -440,15 +484,12 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR must_emulate_scaled_formats = true; LOG_WARNING(Render_Vulkan, "ARM drivers have broken VK_EXT_extended_dynamic_state"); - extensions.extended_dynamic_state = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state, features.extended_dynamic_state, + VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); LOG_WARNING(Render_Vulkan, "ARM drivers have broken VK_EXT_extended_dynamic_state2"); - features.extended_dynamic_state2.extendedDynamicState2 = false; - features.extended_dynamic_state2.extendedDynamicState2LogicOp = false; - features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false; - extensions.extended_dynamic_state2 = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state2, features.extended_dynamic_state2, + VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); } if (is_nvidia) { @@ -464,8 +505,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR case NvidiaArchitecture::VoltaOrOlder: if (nv_major_version < 527) { LOG_WARNING(Render_Vulkan, "Volta and older have broken VK_KHR_push_descriptor"); - extensions.push_descriptor = false; - loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); + RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); } break; } @@ -480,8 +520,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR if (version < VK_MAKE_API_VERSION(0, 21, 2, 0)) { LOG_WARNING(Render_Vulkan, "RADV versions older than 21.2 have broken VK_EXT_extended_dynamic_state"); - extensions.extended_dynamic_state = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state, + features.extended_dynamic_state, + VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); } } if (extensions.extended_dynamic_state2 && is_radv) { @@ -490,11 +531,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_WARNING( Render_Vulkan, "RADV versions older than 22.3.1 have broken VK_EXT_extended_dynamic_state2"); - features.extended_dynamic_state2.extendedDynamicState2 = false; - features.extended_dynamic_state2.extendedDynamicState2LogicOp = false; - features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false; - extensions.extended_dynamic_state2 = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state2, + features.extended_dynamic_state2, + VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); } } if (extensions.extended_dynamic_state2 && is_qualcomm) { @@ -504,11 +543,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR // Qualcomm Adreno 7xx drivers do not properly support extended_dynamic_state2. LOG_WARNING(Render_Vulkan, "Qualcomm Adreno 7xx drivers have broken VK_EXT_extended_dynamic_state2"); - features.extended_dynamic_state2.extendedDynamicState2 = false; - features.extended_dynamic_state2.extendedDynamicState2LogicOp = false; - features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false; - extensions.extended_dynamic_state2 = false; - loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); + RemoveExtensionFeature(extensions.extended_dynamic_state2, + features.extended_dynamic_state2, + VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); } } if (extensions.extended_dynamic_state3 && is_radv) { @@ -540,9 +577,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR if (is_rdna2) { LOG_WARNING(Render_Vulkan, "RADV has broken VK_EXT_vertex_input_dynamic_state on RDNA2 hardware"); - features.vertex_input_dynamic_state.vertexInputDynamicState = false; - extensions.vertex_input_dynamic_state = false; - loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.vertex_input_dynamic_state, + features.vertex_input_dynamic_state, + VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); } } if (extensions.vertex_input_dynamic_state && is_qualcomm) { @@ -553,9 +590,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR LOG_WARNING( Render_Vulkan, "Qualcomm Adreno 7xx drivers have broken VK_EXT_vertex_input_dynamic_state"); - features.vertex_input_dynamic_state.vertexInputDynamicState = false; - extensions.vertex_input_dynamic_state = false; - loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.vertex_input_dynamic_state, + features.vertex_input_dynamic_state, + VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); } } @@ -575,8 +612,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR if (!features.shader_float16_int8.shaderFloat16) { LOG_WARNING(Render_Vulkan, "AMD GCN4 and earlier have broken VK_EXT_sampler_filter_minmax"); - extensions.sampler_filter_minmax = false; - loaded_extensions.erase(VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME); + RemoveExtension(extensions.sampler_filter_minmax, + VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME); } } @@ -584,8 +621,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR const u32 version = (properties.properties.driverVersion << 3) >> 3; if (version < VK_MAKE_API_VERSION(27, 20, 100, 0)) { LOG_WARNING(Render_Vulkan, "Intel has broken VK_EXT_vertex_input_dynamic_state"); - extensions.vertex_input_dynamic_state = false; - loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeature(extensions.vertex_input_dynamic_state, + features.vertex_input_dynamic_state, + VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); } } if (features.shader_float16_int8.shaderFloat16 && is_intel_windows) { @@ -612,8 +650,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR // mesa/mesa/-/commit/ff91c5ca42bc80aa411cb3fd8f550aa6fdd16bdc LOG_WARNING(Render_Vulkan, "ANV drivers 22.3.0 to 23.1.0 have broken VK_KHR_push_descriptor"); - extensions.push_descriptor = false; - loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); + RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME); } } if (is_mvk) { @@ -1007,34 +1044,29 @@ bool Device::GetSuitability(bool requires_swapchain) { return suitable; } -void Device::RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name) { - if (loaded_extensions.contains(extension_name) && !is_suitable) { - LOG_WARNING(Render_Vulkan, "Removing unsuitable extension {}", extension_name); - loaded_extensions.erase(extension_name); - } -} - void Device::RemoveUnsuitableExtensions() { // VK_EXT_custom_border_color extensions.custom_border_color = features.custom_border_color.customBorderColors && features.custom_border_color.customBorderColorWithoutFormat; - RemoveExtensionIfUnsuitable(extensions.custom_border_color, - VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.custom_border_color, features.custom_border_color, + VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); // VK_EXT_depth_clip_control extensions.depth_clip_control = features.depth_clip_control.depthClipControl; - RemoveExtensionIfUnsuitable(extensions.depth_clip_control, - VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control, + VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME); // VK_EXT_extended_dynamic_state extensions.extended_dynamic_state = features.extended_dynamic_state.extendedDynamicState; - RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state, - VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state, + features.extended_dynamic_state, + VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME); // VK_EXT_extended_dynamic_state2 extensions.extended_dynamic_state2 = features.extended_dynamic_state2.extendedDynamicState2; - RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state2, - VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state2, + features.extended_dynamic_state2, + VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME); // VK_EXT_extended_dynamic_state3 dynamic_state3_blending = @@ -1048,35 +1080,38 @@ void Device::RemoveUnsuitableExtensions() { extensions.extended_dynamic_state3 = dynamic_state3_blending || dynamic_state3_enables; dynamic_state3_blending = dynamic_state3_blending && extensions.extended_dynamic_state3; dynamic_state3_enables = dynamic_state3_enables && extensions.extended_dynamic_state3; - RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state3, - VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state3, + features.extended_dynamic_state3, + VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME); // VK_EXT_provoking_vertex extensions.provoking_vertex = features.provoking_vertex.provokingVertexLast && features.provoking_vertex.transformFeedbackPreservesProvokingVertex; - RemoveExtensionIfUnsuitable(extensions.provoking_vertex, - VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.provoking_vertex, features.provoking_vertex, + VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME); // VK_KHR_shader_atomic_int64 extensions.shader_atomic_int64 = features.shader_atomic_int64.shaderBufferInt64Atomics && features.shader_atomic_int64.shaderSharedInt64Atomics; - RemoveExtensionIfUnsuitable(extensions.shader_atomic_int64, - VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.shader_atomic_int64, features.shader_atomic_int64, + VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME); // VK_EXT_shader_demote_to_helper_invocation extensions.shader_demote_to_helper_invocation = features.shader_demote_to_helper_invocation.shaderDemoteToHelperInvocation; - RemoveExtensionIfUnsuitable(extensions.shader_demote_to_helper_invocation, - VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.shader_demote_to_helper_invocation, + features.shader_demote_to_helper_invocation, + VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME); // VK_EXT_subgroup_size_control extensions.subgroup_size_control = features.subgroup_size_control.subgroupSizeControl && properties.subgroup_size_control.minSubgroupSize <= GuestWarpSize && properties.subgroup_size_control.maxSubgroupSize >= GuestWarpSize; - RemoveExtensionIfUnsuitable(extensions.subgroup_size_control, - VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.subgroup_size_control, + features.subgroup_size_control, + VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME); // VK_EXT_transform_feedback extensions.transform_feedback = @@ -1086,24 +1121,27 @@ void Device::RemoveUnsuitableExtensions() { properties.transform_feedback.maxTransformFeedbackBuffers > 0 && properties.transform_feedback.transformFeedbackQueries && properties.transform_feedback.transformFeedbackDraw; - RemoveExtensionIfUnsuitable(extensions.transform_feedback, - VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.transform_feedback, features.transform_feedback, + VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME); // VK_EXT_vertex_input_dynamic_state extensions.vertex_input_dynamic_state = features.vertex_input_dynamic_state.vertexInputDynamicState; - RemoveExtensionIfUnsuitable(extensions.vertex_input_dynamic_state, - VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.vertex_input_dynamic_state, + features.vertex_input_dynamic_state, + VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME); // VK_KHR_pipeline_executable_properties if (Settings::values.renderer_shader_feedback.GetValue()) { extensions.pipeline_executable_properties = features.pipeline_executable_properties.pipelineExecutableInfo; - RemoveExtensionIfUnsuitable(extensions.pipeline_executable_properties, - VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.pipeline_executable_properties, + features.pipeline_executable_properties, + VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME); } else { - extensions.pipeline_executable_properties = false; - loaded_extensions.erase(VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME); + RemoveExtensionFeature(extensions.pipeline_executable_properties, + features.pipeline_executable_properties, + VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME); } // VK_KHR_workgroup_memory_explicit_layout @@ -1113,8 +1151,9 @@ void Device::RemoveUnsuitableExtensions() { features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout8BitAccess && features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout16BitAccess && features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayoutScalarBlockLayout; - RemoveExtensionIfUnsuitable(extensions.workgroup_memory_explicit_layout, - VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME); + RemoveExtensionFeatureIfUnsuitable(extensions.workgroup_memory_explicit_layout, + features.workgroup_memory_explicit_layout, + VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME); } void Device::SetupFamilies(VkSurfaceKHR surface) { diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index d8dd41e51..488fdd313 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -639,8 +639,17 @@ private: // Remove extensions which have incomplete feature support. void RemoveUnsuitableExtensions(); + + void RemoveExtension(bool& extension, const std::string& extension_name); void RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name); + template <typename Feature> + void RemoveExtensionFeature(bool& extension, Feature& feature, + const std::string& extension_name); + template <typename Feature> + void RemoveExtensionFeatureIfUnsuitable(bool is_suitable, Feature& feature, + const std::string& extension_name); + /// Sets up queue families. void SetupFamilies(VkSurfaceKHR surface); diff --git a/src/video_core/vulkan_common/vulkan_instance.cpp b/src/video_core/vulkan_common/vulkan_instance.cpp index 72aedb8d8..180657a75 100644 --- a/src/video_core/vulkan_common/vulkan_instance.cpp +++ b/src/video_core/vulkan_common/vulkan_instance.cpp @@ -41,9 +41,6 @@ namespace { bool enable_validation) { std::vector<const char*> extensions; extensions.reserve(6); -#ifdef __APPLE__ - extensions.push_back(VK_KHR_PORTABILITY_ENUMERATION_EXTENSION_NAME); -#endif switch (window_type) { case Core::Frontend::WindowSystemType::Headless: break; @@ -74,11 +71,14 @@ namespace { if (window_type != Core::Frontend::WindowSystemType::Headless) { extensions.push_back(VK_KHR_SURFACE_EXTENSION_NAME); } - if (enable_validation) { - const bool debug_utils = - AreExtensionsSupported(dld, std::array{VK_EXT_DEBUG_UTILS_EXTENSION_NAME}); - extensions.push_back(debug_utils ? VK_EXT_DEBUG_UTILS_EXTENSION_NAME - : VK_EXT_DEBUG_REPORT_EXTENSION_NAME); +#ifdef __APPLE__ + if (AreExtensionsSupported(dld, std::array{VK_KHR_PORTABILITY_ENUMERATION_EXTENSION_NAME})) { + extensions.push_back(VK_KHR_PORTABILITY_ENUMERATION_EXTENSION_NAME); + } +#endif + if (enable_validation && + AreExtensionsSupported(dld, std::array{VK_EXT_DEBUG_UTILS_EXTENSION_NAME})) { + extensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME); } return extensions; } diff --git a/src/video_core/vulkan_common/vulkan_library.cpp b/src/video_core/vulkan_common/vulkan_library.cpp index 47f6f2a03..0130f6a0d 100644 --- a/src/video_core/vulkan_common/vulkan_library.cpp +++ b/src/video_core/vulkan_common/vulkan_library.cpp @@ -19,13 +19,17 @@ std::shared_ptr<Common::DynamicLibrary> OpenLibrary( #else auto library = std::make_shared<Common::DynamicLibrary>(); #ifdef __APPLE__ + const auto libvulkan_filename = + Common::FS::GetBundleDirectory() / "Contents/Frameworks/libvulkan.1.dylib"; + const auto libmoltenvk_filename = + Common::FS::GetBundleDirectory() / "Contents/Frameworks/libMoltenVK.dylib"; + const char* library_paths[] = {std::getenv("LIBVULKAN_PATH"), libvulkan_filename.c_str(), + libmoltenvk_filename.c_str()}; // Check if a path to a specific Vulkan library has been specified. - char* const libvulkan_env = std::getenv("LIBVULKAN_PATH"); - if (!libvulkan_env || !library->Open(libvulkan_env)) { - // Use the libvulkan.dylib from the application bundle. - const auto filename = - Common::FS::GetBundleDirectory() / "Contents/Frameworks/libvulkan.dylib"; - void(library->Open(Common::FS::PathToUTF8String(filename).c_str())); + for (const auto& library_path : library_paths) { + if (library_path && library->Open(library_path)) { + break; + } } #else std::string filename = Common::DynamicLibrary::GetVersionedFilename("vulkan", 1); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index 78e5a248f..c3f388d89 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -92,6 +92,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdCopyImage); X(vkCmdCopyImageToBuffer); X(vkCmdDispatch); + X(vkCmdDispatchIndirect); X(vkCmdDraw); X(vkCmdDrawIndexed); X(vkCmdDrawIndirect); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index c226a2a29..049fa8038 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -203,6 +203,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdCopyImage vkCmdCopyImage{}; PFN_vkCmdCopyImageToBuffer vkCmdCopyImageToBuffer{}; PFN_vkCmdDispatch vkCmdDispatch{}; + PFN_vkCmdDispatchIndirect vkCmdDispatchIndirect{}; PFN_vkCmdDraw vkCmdDraw{}; PFN_vkCmdDrawIndexed vkCmdDrawIndexed{}; PFN_vkCmdDrawIndirect vkCmdDrawIndirect{}; @@ -1209,6 +1210,10 @@ public: dld->vkCmdDispatch(handle, x, y, z); } + void DispatchIndirect(VkBuffer indirect_buffer, VkDeviceSize offset) const noexcept { + dld->vkCmdDispatchIndirect(handle, indirect_buffer, offset); + } + void PipelineBarrier(VkPipelineStageFlags src_stage_mask, VkPipelineStageFlags dst_stage_mask, VkDependencyFlags dependency_flags, Span<VkMemoryBarrier> memory_barriers, Span<VkBufferMemoryBarrier> buffer_barriers, |