diff options
Diffstat (limited to '')
| -rw-r--r-- | src/video_core/engines/maxwell_3d.cpp | 24 | ||||
| -rw-r--r-- | src/video_core/engines/puller.cpp | 39 |
2 files changed, 47 insertions, 16 deletions
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp index 3a4646289..950c70dcd 100644 --- a/src/video_core/engines/maxwell_3d.cpp +++ b/src/video_core/engines/maxwell_3d.cpp @@ -242,6 +242,9 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume return; case MAXWELL3D_REG_INDEX(fragment_barrier): return rasterizer->FragmentBarrier(); + case MAXWELL3D_REG_INDEX(invalidate_texture_data_cache): + rasterizer->InvalidateGPUCache(); + return rasterizer->WaitForIdle(); case MAXWELL3D_REG_INDEX(tiled_cache_barrier): return rasterizer->TiledCacheBarrier(); } @@ -472,10 +475,25 @@ void Maxwell3D::ProcessQueryGet() { switch (regs.query.query_get.operation) { case Regs::QueryOperation::Release: - if (regs.query.query_get.fence == 1) { - rasterizer->SignalSemaphore(regs.query.QueryAddress(), regs.query.query_sequence); + if (regs.query.query_get.fence == 1 || regs.query.query_get.short_query != 0) { + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); } else { - StampQueryResult(regs.query.query_sequence, regs.query.query_get.short_query == 0); + struct LongQueryResult { + u64_le value; + u64_le timestamp; + }; + const GPUVAddr sequence_address{regs.query.QueryAddress()}; + const u32 payload = regs.query.query_sequence; + std::function<void()> operation([this, sequence_address, payload] { + LongQueryResult query_result{payload, system.GPU().GetTicks()}; + memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result)); + }); + rasterizer->SignalFence(std::move(operation)); } break; case Regs::QueryOperation::Acquire: diff --git a/src/video_core/engines/puller.cpp b/src/video_core/engines/puller.cpp index 8c17639e4..dd9494efa 100644 --- a/src/video_core/engines/puller.cpp +++ b/src/video_core/engines/puller.cpp @@ -79,12 +79,15 @@ void Puller::ProcessSemaphoreTriggerMethod() { u64 timestamp; }; - Block block{}; - block.sequence = regs.semaphore_sequence; - // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of - // CoreTiming - block.timestamp = gpu.GetTicks(); - memory_manager.WriteBlock(regs.semaphore_address.SemaphoreAddress(), &block, sizeof(block)); + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_sequence; + std::function<void()> operation([this, sequence_address, payload] { + Block block{}; + block.sequence = payload; + block.timestamp = gpu.GetTicks(); + memory_manager.WriteBlock(sequence_address, &block, sizeof(block)); + }); + rasterizer->SignalFence(std::move(operation)); } else { do { const u32 word{memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress())}; @@ -94,6 +97,7 @@ void Puller::ProcessSemaphoreTriggerMethod() { regs.acquire_active = true; regs.acquire_mode = false; if (word != regs.acquire_value) { + rasterizer->ReleaseFences(); std::this_thread::sleep_for(std::chrono::milliseconds(1)); continue; } @@ -101,11 +105,13 @@ void Puller::ProcessSemaphoreTriggerMethod() { regs.acquire_active = true; regs.acquire_mode = true; if (word < regs.acquire_value) { + rasterizer->ReleaseFences(); std::this_thread::sleep_for(std::chrono::milliseconds(1)); continue; } } else if (op == GpuSemaphoreOperation::AcquireMask) { - if (word & regs.semaphore_sequence == 0) { + if (word && regs.semaphore_sequence == 0) { + rasterizer->ReleaseFences(); std::this_thread::sleep_for(std::chrono::milliseconds(1)); continue; } @@ -117,16 +123,23 @@ void Puller::ProcessSemaphoreTriggerMethod() { } void Puller::ProcessSemaphoreRelease() { - rasterizer->SignalSemaphore(regs.semaphore_address.SemaphoreAddress(), regs.semaphore_release); + const GPUVAddr sequence_address{regs.semaphore_address.SemaphoreAddress()}; + const u32 payload = regs.semaphore_release; + std::function<void()> operation([this, sequence_address, payload] { + memory_manager.Write<u32>(sequence_address, payload); + }); + rasterizer->SignalFence(std::move(operation)); } void Puller::ProcessSemaphoreAcquire() { - const u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); + u32 word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); const auto value = regs.semaphore_acquire; - std::this_thread::sleep_for(std::chrono::milliseconds(5)); - if (word != value) { + while (word != value) { regs.acquire_active = true; regs.acquire_value = value; + std::this_thread::sleep_for(std::chrono::milliseconds(1)); + rasterizer->ReleaseFences(); + word = memory_manager.Read<u32>(regs.semaphore_address.SemaphoreAddress()); // TODO(kemathe73) figure out how to do the acquire_timeout regs.acquire_mode = false; regs.acquire_source = false; @@ -147,9 +160,9 @@ void Puller::CallPullerMethod(const MethodCall& method_call) { case BufferMethods::SemaphoreAddressHigh: case BufferMethods::SemaphoreAddressLow: case BufferMethods::SemaphoreSequencePayload: - case BufferMethods::WrcacheFlush: case BufferMethods::SyncpointPayload: break; + case BufferMethods::WrcacheFlush: case BufferMethods::RefCnt: rasterizer->SignalReference(); break; @@ -173,7 +186,7 @@ void Puller::CallPullerMethod(const MethodCall& method_call) { } case BufferMethods::MemOpB: { // Implement this better. - rasterizer->SyncGuestHost(); + rasterizer->InvalidateGPUCache(); break; } case BufferMethods::MemOpC: |