From ef3a0ae64a9e5c7a7a27cc123439b220f626cb29 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Mon, 20 Apr 2020 02:16:56 -0400 Subject: DMAPusher: Propagate multimethod writes into the engines. --- src/video_core/gpu.cpp | 47 +++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 43 insertions(+), 4 deletions(-) (limited to 'src/video_core/gpu.cpp') diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 3b7572d61a..54e7876a64 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -209,16 +209,31 @@ void GPU::CallMethod(const MethodCall& method_call) { ASSERT(method_call.subchannel < bound_engines.size()); - if (ExecuteMethodOnEngine(method_call)) { + if (ExecuteMethodOnEngine(method_call.method)) { CallEngineMethod(method_call); } else { CallPullerMethod(method_call); } } -bool GPU::ExecuteMethodOnEngine(const MethodCall& method_call) { - const auto method = static_cast(method_call.method); - return method >= BufferMethods::NonPullerMethods; +void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, + subchannel); + + ASSERT(subchannel < bound_engines.size()); + + if (ExecuteMethodOnEngine(method)) { + CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); + } else { + for (std::size_t i = 0; i < amount; i++) { + CallPullerMethod({method, base_start[i], subchannel, methods_pending - static_cast(i)}); + } + } +} + +bool GPU::ExecuteMethodOnEngine(u32 method) { + const auto buffer_method = static_cast(method); + return buffer_method >= BufferMethods::NonPullerMethods; } void GPU::CallPullerMethod(const MethodCall& method_call) { @@ -298,6 +313,30 @@ void GPU::CallEngineMethod(const MethodCall& method_call) { } } +void GPU::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending) { + const EngineID engine = bound_engines[subchannel]; + + switch (engine) { + case EngineID::FERMI_TWOD_A: + fermi_2d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_B: + maxwell_3d->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_COMPUTE_B: + kepler_compute->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::MAXWELL_DMA_COPY_A: + maxwell_dma->CallMultiMethod(method, base_start, amount, methods_pending); + break; + case EngineID::KEPLER_INLINE_TO_MEMORY_B: + kepler_memory->CallMultiMethod(method, base_start, amount, methods_pending); + break; + default: + UNIMPLEMENTED_MSG("Unimplemented engine"); + } +} + void GPU::ProcessBindMethod(const MethodCall& method_call) { // Bind the current subchannel to the desired engine id. LOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", method_call.subchannel, -- cgit v1.2.3 From c8f4549d43deda62570a45eabff8986af2130eaf Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Mon, 20 Apr 2020 13:20:52 -0400 Subject: GPU: Add Fast GPU Time Option. --- src/core/settings.h | 1 + src/video_core/gpu.cpp | 6 +++++- src/yuzu/configuration/config.cpp | 2 ++ src/yuzu/configuration/configure_graphics_advanced.cpp | 2 ++ src/yuzu/configuration/configure_graphics_advanced.ui | 7 +++++++ src/yuzu_cmd/config.cpp | 2 ++ src/yuzu_tester/config.cpp | 2 ++ 7 files changed, 21 insertions(+), 1 deletion(-) (limited to 'src/video_core/gpu.cpp') diff --git a/src/core/settings.h b/src/core/settings.h index 7d09253f58..163900f0be 100644 --- a/src/core/settings.h +++ b/src/core/settings.h @@ -446,6 +446,7 @@ struct Values { bool use_asynchronous_gpu_emulation; bool use_vsync; bool force_30fps_mode; + bool use_fast_gpu_time; float bg_red; float bg_green; diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index 54e7876a64..d2bd3cc9d5 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -9,6 +9,7 @@ #include "core/core_timing_util.h" #include "core/frontend/emu_window.h" #include "core/memory.h" +#include "core/settings.h" #include "video_core/engines/fermi_2d.h" #include "video_core/engines/kepler_compute.h" #include "video_core/engines/kepler_memory.h" @@ -154,7 +155,10 @@ u64 GPU::GetTicks() const { constexpr u64 gpu_ticks_den = 625; const u64 cpu_ticks = system.CoreTiming().GetTicks(); - const u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count(); + u64 nanoseconds = Core::Timing::CyclesToNs(cpu_ticks).count(); + if (Settings::values.use_fast_gpu_time) { + nanoseconds /= 256; + } 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; diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp index 196a3a1168..c7c11b9dde 100644 --- a/src/yuzu/configuration/config.cpp +++ b/src/yuzu/configuration/config.cpp @@ -644,6 +644,7 @@ void Config::ReadRendererValues() { Settings::values.use_asynchronous_gpu_emulation = ReadSetting(QStringLiteral("use_asynchronous_gpu_emulation"), false).toBool(); Settings::values.use_vsync = ReadSetting(QStringLiteral("use_vsync"), true).toBool(); + Settings::values.use_fast_gpu_time = ReadSetting(QStringLiteral("use_fast_gpu_time"), true).toBool(); Settings::values.force_30fps_mode = ReadSetting(QStringLiteral("force_30fps_mode"), false).toBool(); @@ -1085,6 +1086,7 @@ void Config::SaveRendererValues() { WriteSetting(QStringLiteral("use_asynchronous_gpu_emulation"), Settings::values.use_asynchronous_gpu_emulation, false); WriteSetting(QStringLiteral("use_vsync"), Settings::values.use_vsync, true); + WriteSetting(QStringLiteral("use_fast_gpu_time"), Settings::values.use_fast_gpu_time, true); WriteSetting(QStringLiteral("force_30fps_mode"), Settings::values.force_30fps_mode, false); // Cast to double because Qt's written float values are not human-readable diff --git a/src/yuzu/configuration/configure_graphics_advanced.cpp b/src/yuzu/configuration/configure_graphics_advanced.cpp index 0a3f47339d..5bb2ae5554 100644 --- a/src/yuzu/configuration/configure_graphics_advanced.cpp +++ b/src/yuzu/configuration/configure_graphics_advanced.cpp @@ -22,6 +22,7 @@ void ConfigureGraphicsAdvanced::SetConfiguration() { ui->gpu_accuracy->setCurrentIndex(static_cast(Settings::values.gpu_accuracy)); ui->use_vsync->setEnabled(runtime_lock); ui->use_vsync->setChecked(Settings::values.use_vsync); + ui->use_fast_gpu_time->setChecked(Settings::values.use_fast_gpu_time); ui->force_30fps_mode->setEnabled(runtime_lock); ui->force_30fps_mode->setChecked(Settings::values.force_30fps_mode); ui->anisotropic_filtering_combobox->setEnabled(runtime_lock); @@ -32,6 +33,7 @@ void ConfigureGraphicsAdvanced::ApplyConfiguration() { auto gpu_accuracy = static_cast(ui->gpu_accuracy->currentIndex()); Settings::values.gpu_accuracy = gpu_accuracy; Settings::values.use_vsync = ui->use_vsync->isChecked(); + Settings::values.use_fast_gpu_time = ui->use_fast_gpu_time->isChecked(); Settings::values.force_30fps_mode = ui->force_30fps_mode->isChecked(); Settings::values.max_anisotropy = ui->anisotropic_filtering_combobox->currentIndex(); } diff --git a/src/yuzu/configuration/configure_graphics_advanced.ui b/src/yuzu/configuration/configure_graphics_advanced.ui index 0c7b383e00..770b80c50d 100644 --- a/src/yuzu/configuration/configure_graphics_advanced.ui +++ b/src/yuzu/configuration/configure_graphics_advanced.ui @@ -69,6 +69,13 @@ + + + + Use Fast GPU Time + + + diff --git a/src/yuzu_cmd/config.cpp b/src/yuzu_cmd/config.cpp index d1ac354bf1..8476a5a168 100644 --- a/src/yuzu_cmd/config.cpp +++ b/src/yuzu_cmd/config.cpp @@ -394,6 +394,8 @@ void Config::ReadValues() { sdl2_config->GetBoolean("Renderer", "use_asynchronous_gpu_emulation", false); Settings::values.use_vsync = static_cast(sdl2_config->GetInteger("Renderer", "use_vsync", 1)); + Settings::values.use_fast_gpu_time = + sdl2_config->GetBoolean("Renderer", "use_fast_gpu_time", true); Settings::values.bg_red = static_cast(sdl2_config->GetReal("Renderer", "bg_red", 0.0)); Settings::values.bg_green = diff --git a/src/yuzu_tester/config.cpp b/src/yuzu_tester/config.cpp index c0325cc3ca..3be58b15da 100644 --- a/src/yuzu_tester/config.cpp +++ b/src/yuzu_tester/config.cpp @@ -130,6 +130,8 @@ void Config::ReadValues() { Settings::values.gpu_accuracy = static_cast(gpu_accuracy_level); Settings::values.use_asynchronous_gpu_emulation = sdl2_config->GetBoolean("Renderer", "use_asynchronous_gpu_emulation", false); + Settings::values.use_fast_gpu_time = + sdl2_config->GetBoolean("Renderer", "use_fast_gpu_time", true); Settings::values.bg_red = static_cast(sdl2_config->GetReal("Renderer", "bg_red", 0.0)); Settings::values.bg_green = -- cgit v1.2.3 From 0cf32d6184603fc1e419c3607dc3c1f900fb6eb0 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Mon, 20 Apr 2020 13:42:14 -0400 Subject: Clang Format. --- src/video_core/engines/kepler_compute.cpp | 3 ++- src/video_core/engines/kepler_memory.cpp | 3 ++- src/video_core/engines/maxwell_dma.cpp | 3 ++- src/video_core/gpu.cpp | 12 +++++++----- src/video_core/gpu.h | 6 ++++-- src/yuzu/configuration/config.cpp | 3 ++- 6 files changed, 19 insertions(+), 11 deletions(-) (limited to 'src/video_core/gpu.cpp') diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp index 894300f57c..00a12175f1 100644 --- a/src/video_core/engines/kepler_compute.cpp +++ b/src/video_core/engines/kepler_compute.cpp @@ -51,7 +51,8 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) { } } -void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { +void KeplerCompute::CallMultiMethod(u32 method, const u32* base_start, u32 amount, + u32 methods_pending) { for (std::size_t i = 0; i < amount; i++) { CallMethod({method, base_start[i], 0, methods_pending - static_cast(i)}); } diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp index e906a11240..586ff15dc9 100644 --- a/src/video_core/engines/kepler_memory.cpp +++ b/src/video_core/engines/kepler_memory.cpp @@ -41,7 +41,8 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) { } } -void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { +void KeplerMemory::CallMultiMethod(u32 method, const u32* base_start, u32 amount, + u32 methods_pending) { for (std::size_t i = 0; i < amount; i++) { CallMethod({method, base_start[i], 0, methods_pending - static_cast(i)}); } diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index 51e606a100..6630005b0d 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -36,7 +36,8 @@ void MaxwellDMA::CallMethod(const GPU::MethodCall& method_call) { #undef MAXWELLDMA_REG_INDEX } -void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending) { +void MaxwellDMA::CallMultiMethod(u32 method, const u32* base_start, u32 amount, + u32 methods_pending) { for (std::size_t i = 0; i < amount; i++) { CallMethod({method, base_start[i], 0, methods_pending - static_cast(i)}); } diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp index d2bd3cc9d5..b87fd873d0 100644 --- a/src/video_core/gpu.cpp +++ b/src/video_core/gpu.cpp @@ -220,9 +220,9 @@ void GPU::CallMethod(const MethodCall& method_call) { } } -void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending) { - LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, - subchannel); +void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { + LOG_TRACE(HW_GPU, "Processing method {:08X} on subchannel {}", method, subchannel); ASSERT(subchannel < bound_engines.size()); @@ -230,7 +230,8 @@ void GPU::CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 CallEngineMultiMethod(method, subchannel, base_start, amount, methods_pending); } else { for (std::size_t i = 0; i < amount; i++) { - CallPullerMethod({method, base_start[i], subchannel, methods_pending - static_cast(i)}); + CallPullerMethod( + {method, base_start[i], subchannel, methods_pending - static_cast(i)}); } } } @@ -317,7 +318,8 @@ void GPU::CallEngineMethod(const MethodCall& method_call) { } } -void GPU::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending) { +void GPU::CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending) { const EngineID engine = bound_engines[subchannel]; switch (engine) { diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h index 4d7e2651c7..dd51c95b7f 100644 --- a/src/video_core/gpu.h +++ b/src/video_core/gpu.h @@ -156,7 +156,8 @@ public: void CallMethod(const MethodCall& method_call); /// Calls a GPU multivalue method. - void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending); + void CallMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); /// Flush all current written commands into the host GPU for execution. void FlushCommands(); @@ -313,7 +314,8 @@ private: void CallEngineMethod(const MethodCall& method_call); /// Calls a GPU engine multivalue method. - void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, u32 methods_pending); + void CallEngineMultiMethod(u32 method, u32 subchannel, const u32* base_start, u32 amount, + u32 methods_pending); /// Determines where the method should be executed. bool ExecuteMethodOnEngine(u32 method); diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp index c7c11b9dde..59c003f380 100644 --- a/src/yuzu/configuration/config.cpp +++ b/src/yuzu/configuration/config.cpp @@ -644,7 +644,8 @@ void Config::ReadRendererValues() { Settings::values.use_asynchronous_gpu_emulation = ReadSetting(QStringLiteral("use_asynchronous_gpu_emulation"), false).toBool(); Settings::values.use_vsync = ReadSetting(QStringLiteral("use_vsync"), true).toBool(); - Settings::values.use_fast_gpu_time = ReadSetting(QStringLiteral("use_fast_gpu_time"), true).toBool(); + Settings::values.use_fast_gpu_time = + ReadSetting(QStringLiteral("use_fast_gpu_time"), true).toBool(); Settings::values.force_30fps_mode = ReadSetting(QStringLiteral("force_30fps_mode"), false).toBool(); -- cgit v1.2.3