diff options
Diffstat (limited to 'src/video_core')
-rw-r--r-- | src/video_core/engines/maxwell_3d.cpp | 91 | ||||
-rw-r--r-- | src/video_core/engines/maxwell_3d.h | 16 | ||||
-rw-r--r-- | src/video_core/engines/shader_bytecode.h | 4 | ||||
-rw-r--r-- | src/video_core/gpu.cpp | 18 | ||||
-rw-r--r-- | src/video_core/gpu.h | 2 | ||||
-rw-r--r-- | src/video_core/memory_manager.cpp | 14 | ||||
-rw-r--r-- | src/video_core/memory_manager.h | 7 | ||||
-rw-r--r-- | src/video_core/shader/decode/arithmetic_integer.cpp | 4 | ||||
-rw-r--r-- | src/video_core/shader/decode/bfi.cpp | 2 |
9 files changed, 92 insertions, 66 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 7cea146f0..0b3e8749b 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -9,6 +9,7 @@ #include "core/core_timing.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/engines/shader_type.h" +#include "video_core/gpu.h" #include "video_core/memory_manager.h" #include "video_core/rasterizer_interface.h" #include "video_core/textures/texture.h" @@ -519,61 +520,63 @@ void Maxwell3D::ProcessFirmwareCall4() { regs.reg_array[0xd00] = 1; } -void Maxwell3D::ProcessQueryGet() { +void Maxwell3D::StampQueryResult(u64 payload, bool long_query) { + struct LongQueryResult { + u64_le value; + u64_le timestamp; + }; + static_assert(sizeof(LongQueryResult) == 16, "LongQueryResult has wrong size"); const GPUVAddr sequence_address{regs.query.QueryAddress()}; - // Since the sequence address is given as a GPU VAddr, we have to convert it to an application - // VAddr before writing. + if (long_query) { + // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast + // GPU, this command may actually take a while to complete in real hardware due to GPU + // wait queues. + LongQueryResult query_result{payload, system.GPU().GetTicks()}; + memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); + } else { + memory_manager.Write<u32>(sequence_address, static_cast<u32>(payload)); + } +} +void Maxwell3D::ProcessQueryGet() { // TODO(Subv): Support the other query units. ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop, "Units other than CROP are unimplemented"); - u64 result = 0; - - // TODO(Subv): Support the other query variables - switch (regs.query.query_get.select) { - case Regs::QuerySelect::Zero: - // This seems to actually write the query sequence to the query address. - result = regs.query.query_sequence; + switch (regs.query.query_get.operation) { + case Regs::QueryOperation::Release: { + const u64 result = regs.query.query_sequence; + StampQueryResult(result, regs.query.query_get.short_query == 0); break; - default: - result = 1; - UNIMPLEMENTED_MSG("Unimplemented query select type {}", - static_cast<u32>(regs.query.query_get.select.Value())); } - - // TODO(Subv): Research and implement how query sync conditions work. - - struct LongQueryResult { - u64_le value; - u64_le timestamp; - }; - static_assert(sizeof(LongQueryResult) == 16, "LongQueryResult has wrong size"); - - switch (regs.query.query_get.mode) { - case Regs::QueryMode::Write: - case Regs::QueryMode::Write2: { - u32 sequence = regs.query.query_sequence; - if (regs.query.query_get.short_query) { - // Write the current query sequence to the sequence address. - // TODO(Subv): Find out what happens if you use a long query type but mark it as a short - // query. - memory_manager.Write<u32>(sequence_address, sequence); - } else { - // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast - // GPU, this command may actually take a while to complete in real hardware due to GPU - // wait queues. - LongQueryResult query_result{}; - query_result.value = result; - // TODO(Subv): Generate a real GPU timestamp and write it here instead of CoreTiming - query_result.timestamp = system.CoreTiming().GetTicks(); - memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); + case Regs::QueryOperation::Acquire: { + // Todo(Blinkhawk): Under this operation, the GPU waits for the CPU + // to write a value that matches the current payload. + UNIMPLEMENTED_MSG("Unimplemented query operation ACQUIRE"); + break; + } + case Regs::QueryOperation::Counter: { + u64 result{}; + switch (regs.query.query_get.select) { + case Regs::QuerySelect::Zero: + result = 0; + break; + default: + result = 1; + UNIMPLEMENTED_MSG("Unimplemented query select type {}", + static_cast<u32>(regs.query.query_get.select.Value())); } + StampQueryResult(result, regs.query.query_get.short_query == 0); + break; + } + case Regs::QueryOperation::Trap: { + UNIMPLEMENTED_MSG("Unimplemented query operation TRAP"); + break; + } + default: { + UNIMPLEMENTED_MSG("Unknown query operation"); break; } - default: - UNIMPLEMENTED_MSG("Query mode {} not implemented", - static_cast<u32>(regs.query.query_get.mode.Value())); } } diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h index 7b1912a66..0a2af54e5 100644 --- a/src/video_core/engines/maxwell_3d.h +++ b/src/video_core/engines/maxwell_3d.h @@ -71,12 +71,11 @@ public: static constexpr std::size_t MaxConstBuffers = 18; static constexpr std::size_t MaxConstBufferSize = 0x10000; - enum class QueryMode : u32 { - Write = 0, - Sync = 1, - // TODO(Subv): It is currently unknown what the difference between method 2 and method 0 - // is. - Write2 = 2, + enum class QueryOperation : u32 { + Release = 0, + Acquire = 1, + Counter = 2, + Trap = 3, }; enum class QueryUnit : u32 { @@ -1081,7 +1080,7 @@ public: u32 query_sequence; union { u32 raw; - BitField<0, 2, QueryMode> mode; + BitField<0, 2, QueryOperation> operation; BitField<4, 1, u32> fence; BitField<12, 4, QueryUnit> unit; BitField<16, 1, QuerySyncCondition> sync_cond; @@ -1413,6 +1412,9 @@ private: /// Handles a write to the QUERY_GET register. void ProcessQueryGet(); + // Writes the query result accordingly + void StampQueryResult(u64 payload, bool long_query); + // Handles Conditional Rendering void ProcessQueryCondition(); diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 402869fde..c9bc83cd7 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -1677,11 +1677,11 @@ union Instruction { } xmad; union { - BitField<20, 14, u64> offset; + BitField<20, 14, u64> shifted_offset; BitField<34, 5, u64> index; u64 GetOffset() const { - return offset * 4; + return shifted_offset * 4; } } cbuf34; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 062ca83b8..7d7137109 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -6,6 +6,7 @@ #include "common/microprofile.h" #include "core/core.h" #include "core/core_timing.h" +#include "core/core_timing_util.h" #include "core/memory.h" #include "video_core/engines/fermi_2d.h" #include "video_core/engines/kepler_compute.h" @@ -23,7 +24,7 @@ MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192)); GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async) : system{system}, renderer{renderer}, is_async{is_async} { auto& rasterizer{renderer.Rasterizer()}; - memory_manager = std::make_unique<Tegra::MemoryManager>(system); + memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer); dma_pusher = std::make_unique<Tegra::DmaPusher>(*this); maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager); fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer); @@ -122,6 +123,19 @@ bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) { return true; } +u64 GPU::GetTicks() const { + // This values were reversed engineered by fincs from NVN + // The gpu clock is reported in units of 385/625 nanoseconds + constexpr u64 gpu_ticks_num = 384; + constexpr u64 gpu_ticks_den = 625; + + const u64 cpu_ticks = system.CoreTiming().GetTicks(); + const u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count(); + const u64 nanoseconds_num = nanoseconds / gpu_ticks_den; + const u64 nanoseconds_rem = nanoseconds % gpu_ticks_den; + return nanoseconds_num * gpu_ticks_num + (nanoseconds_rem * gpu_ticks_num) / gpu_ticks_den; +} + void GPU::FlushCommands() { renderer.Rasterizer().FlushCommands(); } @@ -340,7 +354,7 @@ void GPU::ProcessSemaphoreTriggerMethod() { block.sequence = regs.semaphore_sequence; // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of // CoreTiming - block.timestamp = system.CoreTiming().GetTicks(); + block.timestamp = GetTicks(); memory_manager->WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block, sizeof(block)); } else { diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index b648317bb..07727210c 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -192,6 +192,8 @@ public: bool CancelSyncptInterrupt(u32 syncpoint_id, u32 value); + u64 GetTicks() const; + std::unique_lock<std::mutex> LockSync() { return std::unique_lock{sync_mutex}; } diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp index f1d50be3e..11848fbce 100644 --- a/src/video_core/memory_manager.cpp +++ b/src/video_core/memory_manager.cpp @@ -9,12 +9,13 @@ #include "core/hle/kernel/process.h" #include "core/hle/kernel/vm_manager.h" #include "core/memory.h" -#include "video_core/gpu.h" #include "video_core/memory_manager.h" +#include "video_core/rasterizer_interface.h" namespace Tegra { -MemoryManager::MemoryManager(Core::System& system) : system{system} { +MemoryManager::MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer) + : rasterizer{rasterizer}, system{system} { std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr); std::fill(page_table.attributes.begin(), page_table.attributes.end(), Common::PageType::Unmapped); @@ -83,8 +84,7 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) { const auto cpu_addr = GpuToCpuAddress(gpu_addr); ASSERT(cpu_addr); - system.GPU().FlushAndInvalidateRegion(cache_addr, aligned_size); - + rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size); UnmapRange(gpu_addr, aligned_size); ASSERT(system.CurrentProcess() ->VMManager() @@ -242,7 +242,7 @@ void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, const std::s switch (page_table.attributes[page_index]) { case Common::PageType::Memory: { const u8* src_ptr{page_table.pointers[page_index] + page_offset}; - system.GPU().FlushRegion(ToCacheAddr(src_ptr), copy_amount); + rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount); std::memcpy(dest_buffer, src_ptr, copy_amount); break; } @@ -292,7 +292,7 @@ void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, const switch (page_table.attributes[page_index]) { case Common::PageType::Memory: { u8* dest_ptr{page_table.pointers[page_index] + page_offset}; - system.GPU().InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount); + rasterizer.InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount); std::memcpy(dest_ptr, src_buffer, copy_amount); break; } @@ -340,7 +340,7 @@ void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, const std:: switch (page_table.attributes[page_index]) { case Common::PageType::Memory: { const u8* src_ptr{page_table.pointers[page_index] + page_offset}; - system.GPU().FlushRegion(ToCacheAddr(src_ptr), copy_amount); + rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount); WriteBlock(dest_addr, src_ptr, copy_amount); break; } diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h index 393447eb4..aea010087 100644 --- a/src/video_core/memory_manager.h +++ b/src/video_core/memory_manager.h @@ -10,6 +10,10 @@ #include "common/common_types.h" #include "common/page_table.h" +namespace VideoCore { +class RasterizerInterface; +} + namespace Core { class System; } @@ -47,7 +51,7 @@ struct VirtualMemoryArea { class MemoryManager final { public: - explicit MemoryManager(Core::System& system); + explicit MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer); ~MemoryManager(); GPUVAddr AllocateSpace(u64 size, u64 align); @@ -172,6 +176,7 @@ private: Common::PageTable page_table{page_bits}; VMAMap vma_map; + VideoCore::RasterizerInterface& rasterizer; Core::System& system; }; diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp index e60875cc4..21366869d 100644 --- a/src/video_core/shader/decode/arithmetic_integer.cpp +++ b/src/video_core/shader/decode/arithmetic_integer.cpp @@ -166,13 +166,13 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) { const auto [op_rhs, test] = [&]() -> std::pair<Node, Node> { switch (opcode->get().GetId()) { case OpCode::Id::ICMP_CR: - return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset), + return {GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()), GetRegister(instr.gpr39)}; case OpCode::Id::ICMP_R: return {GetRegister(instr.gpr20), GetRegister(instr.gpr39)}; case OpCode::Id::ICMP_RC: return {GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset)}; + GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; case OpCode::Id::ICMP_IMM: return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)}; default: diff --git a/src/video_core/shader/decode/bfi.cpp b/src/video_core/shader/decode/bfi.cpp index f992bbe2a..70d1c055b 100644 --- a/src/video_core/shader/decode/bfi.cpp +++ b/src/video_core/shader/decode/bfi.cpp @@ -21,7 +21,7 @@ u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) { switch (opcode->get().GetId()) { case OpCode::Id::BFI_RC: return {GetRegister(instr.gpr39), - GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset)}; + GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset())}; case OpCode::Id::BFI_IMM_R: return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)}; default: |