Merge pull request #3742 from FernandoS27/command-list

Optimize GPU Command Lists and Introduce Fast GPU Time Option

10 files changed, 117 insertions, 0 deletions

diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp
index bace6affb..8a47614d2 100644
--- a/src/video_core/engines/fermi_2d.cpp
+++ b/src/video_core/engines/fermi_2d.cpp
@@ -28,6 +28,12 @@ void Fermi2D::CallMethod(const GPU::MethodCall& method_call) {
     }
 }
 
+void Fermi2D::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<u32>(i)});
+    }
+}
+
 static std::pair<u32, u32> DelimitLine(u32 src_1, u32 src_2, u32 dst_1, u32 dst_2, u32 src_line) {
     const u32 line_a = src_2 - src_1;
     const u32 line_b = dst_2 - dst_1;
diff --git a/src/video_core/engines/fermi_2d.h b/src/video_core/engines/fermi_2d.h
index dba342c70..939a5966d 100644
--- a/src/video_core/engines/fermi_2d.h
+++ b/src/video_core/engines/fermi_2d.h
@@ -39,6 +39,9 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
     enum class Origin : u32 {
         Center = 0,
         Corner = 1,
diff --git a/src/video_core/engines/kepler_compute.cpp b/src/video_core/engines/kepler_compute.cpp
index 368c75a66..00a12175f 100644
--- a/src/video_core/engines/kepler_compute.cpp
+++ b/src/video_core/engines/kepler_compute.cpp
@@ -51,6 +51,13 @@ void KeplerCompute::CallMethod(const GPU::MethodCall& method_call) {
     }
 }
 
+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<u32>(i)});
+    }
+}
+
 Texture::FullTextureInfo KeplerCompute::GetTexture(std::size_t offset) const {
     const std::bitset<8> cbuf_mask = launch_description.const_buffer_enable_mask.Value();
     ASSERT(cbuf_mask[regs.tex_cb_index]);
diff --git a/src/video_core/engines/kepler_compute.h b/src/video_core/engines/kepler_compute.h
index eeb79c56f..fe55fdfd0 100644
--- a/src/video_core/engines/kepler_compute.h
+++ b/src/video_core/engines/kepler_compute.h
@@ -202,6 +202,9 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
     Texture::FullTextureInfo GetTexture(std::size_t offset) const;
 
     /// Given a texture handle, returns the TSC and TIC entries.
diff --git a/src/video_core/engines/kepler_memory.cpp b/src/video_core/engines/kepler_memory.cpp
index 597872e43..586ff15dc 100644
--- a/src/video_core/engines/kepler_memory.cpp
+++ b/src/video_core/engines/kepler_memory.cpp
@@ -41,4 +41,11 @@ void KeplerMemory::CallMethod(const GPU::MethodCall& method_call) {
     }
 }
 
+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<u32>(i)});
+    }
+}
+
 } // namespace Tegra::Engines
diff --git a/src/video_core/engines/kepler_memory.h b/src/video_core/engines/kepler_memory.h
index 396fb6e86..bb26fb030 100644
--- a/src/video_core/engines/kepler_memory.h
+++ b/src/video_core/engines/kepler_memory.h
@@ -40,6 +40,9 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
     struct Regs {
         static constexpr size_t NUM_REGS = 0x7F;
 
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 2824ed707..39e3b66a2 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -280,6 +280,58 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
     }
 }
 
+void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
+                                u32 methods_pending) {
+    // Methods after 0xE00 are special, they're actually triggers for some microcode that was
+    // uploaded to the GPU during initialization.
+    if (method >= MacroRegistersStart) {
+        // We're trying to execute a macro
+        if (executing_macro == 0) {
+            // A macro call must begin by writing the macro method's register, not its argument.
+            ASSERT_MSG((method % 2) == 0,
+                       "Can't start macro execution by writing to the ARGS register");
+            executing_macro = method;
+        }
+
+        for (std::size_t i = 0; i < amount; i++) {
+            macro_params.push_back(base_start[i]);
+        }
+
+        // Call the macro when there are no more parameters in the command buffer
+        if (amount == methods_pending) {
+            CallMacroMethod(executing_macro, macro_params.size(), macro_params.data());
+            macro_params.clear();
+        }
+        return;
+    }
+    switch (method) {
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[1]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[2]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[3]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[4]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[5]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[6]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[7]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[8]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[9]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[10]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[11]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[12]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[13]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[14]):
+    case MAXWELL3D_REG_INDEX(const_buffer.cb_data[15]): {
+        ProcessCBMultiData(method, base_start, amount);
+        break;
+    }
+    default: {
+        for (std::size_t i = 0; i < amount; i++) {
+            CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+        }
+    }
+    }
+}
+
 void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) {
     if (mme_draw.current_mode == MMEDrawMode::Undefined) {
         if (mme_draw.gl_begin_consume) {
@@ -570,6 +622,28 @@ void Maxwell3D::StartCBData(u32 method) {
     ProcessCBData(regs.const_buffer.cb_data[cb_data_state.id]);
 }
 
+void Maxwell3D::ProcessCBMultiData(u32 method, const u32* start_base, u32 amount) {
+    if (cb_data_state.current != method) {
+        if (cb_data_state.current != null_cb_data) {
+            FinishCBData();
+        }
+        constexpr u32 first_cb_data = MAXWELL3D_REG_INDEX(const_buffer.cb_data[0]);
+        cb_data_state.start_pos = regs.const_buffer.cb_pos;
+        cb_data_state.id = method - first_cb_data;
+        cb_data_state.current = method;
+        cb_data_state.counter = 0;
+    }
+    const std::size_t id = cb_data_state.id;
+    const std::size_t size = amount;
+    std::size_t i = 0;
+    for (; i < size; i++) {
+        cb_data_state.buffer[id][cb_data_state.counter] = start_base[i];
+        cb_data_state.counter++;
+    }
+    // Increment the current buffer position.
+    regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4 * amount;
+}
+
 void Maxwell3D::FinishCBData() {
     // Write the input value to the current const buffer at the current position.
     const GPUVAddr buffer_address = regs.const_buffer.BufferAddress();
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 7bbc6600b..3dfba8197 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -1359,6 +1359,9 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
     /// Write the value to the register identified by method.
     void CallMethodFromMME(const GPU::MethodCall& method_call);
 
@@ -1512,6 +1515,7 @@ private:
     /// Handles a write to the CB_DATA[i] register.
     void StartCBData(u32 method);
     void ProcessCBData(u32 value);
+    void ProcessCBMultiData(u32 method, const u32* start_base, u32 amount);
     void FinishCBData();
 
     /// Handles a write to the CB_BIND register.
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index 3bfed6ab8..6630005b0 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -36,6 +36,13 @@ 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) {
+    for (std::size_t i = 0; i < amount; i++) {
+        CallMethod({method, base_start[i], 0, methods_pending - static_cast<u32>(i)});
+    }
+}
+
 void MaxwellDMA::HandleCopy() {
     LOG_TRACE(HW_GPU, "Requested a DMA copy");
 
diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h
index 4f40d1d1f..c43ed8194 100644
--- a/src/video_core/engines/maxwell_dma.h
+++ b/src/video_core/engines/maxwell_dma.h
@@ -35,6 +35,9 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    /// Write multiple values to the register identified by method.
+    void CallMultiMethod(u32 method, const u32* base_start, u32 amount, u32 methods_pending);
+
     struct Regs {
         static constexpr std::size_t NUM_REGS = 0x1D6;