4 files changed, 74 insertions, 27 deletions

diff --git a/src/video_core/command_processor.cpp b/src/video_core/command_processor.cpp
index 1d578582e..d4cdb4ab2 100644
--- a/src/video_core/command_processor.cpp
+++ b/src/video_core/command_processor.cpp
@@ -24,12 +24,37 @@ namespace Tegra {
 
 enum class BufferMethods {
     BindObject = 0,
+    SetGraphMacroCode = 0x45,
+    SetGraphMacroCodeArg = 0x46,
+    SetGraphMacroEntry = 0x47,
     CountBufferMethods = 0x100,
 };
 
-void GPU::WriteReg(u32 method, u32 subchannel, u32 value) {
-    LOG_WARNING(HW_GPU, "Processing method %08X on subchannel %u value %08X", method, subchannel,
-                value);
+void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params) {
+    LOG_WARNING(HW_GPU, "Processing method %08X on subchannel %u value %08X remaining params %u",
+                method, subchannel, value, remaining_params);
+
+    if (method == static_cast<u32>(BufferMethods::SetGraphMacroEntry)) {
+        // Prepare to upload a new macro, reset the upload counter.
+        LOG_DEBUG(HW_GPU, "Uploading GPU macro %08X", value);
+        current_macro_entry = value;
+        current_macro_code.clear();
+        return;
+    }
+
+    if (method == static_cast<u32>(BufferMethods::SetGraphMacroCodeArg)) {
+        // Append a new code word to the current macro.
+        current_macro_code.push_back(value);
+
+        // There are no more params remaining, submit the code to the 3D engine.
+        if (remaining_params == 0) {
+            maxwell_3d->SubmitMacroCode(current_macro_entry, std::move(current_macro_code));
+            current_macro_entry = InvalidGraphMacroEntry;
+            current_macro_code.clear();
+        }
+
+        return;
+    }
 
     if (method == static_cast<u32>(BufferMethods::BindObject)) {
         // Bind the current subchannel to the desired engine id.
@@ -54,7 +79,7 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value) {
         fermi_2d->WriteReg(method, value);
         break;
     case EngineID::MAXWELL_B:
-        maxwell_3d->WriteReg(method, value);
+        maxwell_3d->WriteReg(method, value, remaining_params);
         break;
     case EngineID::MAXWELL_COMPUTE_B:
         maxwell_compute->WriteReg(method, value);
@@ -78,7 +103,8 @@ void GPU::ProcessCommandList(GPUVAddr address, u32 size) {
         case SubmissionMode::Increasing: {
             // Increase the method value with each argument.
             for (unsigned i = 0; i < header.arg_count; ++i) {
-                WriteReg(header.method + i, header.subchannel, Memory::Read32(current_addr));
+                WriteReg(header.method + i, header.subchannel, Memory::Read32(current_addr),
+                         header.arg_count - i - 1);
                 current_addr += sizeof(u32);
             }
             break;
@@ -87,31 +113,31 @@ void GPU::ProcessCommandList(GPUVAddr address, u32 size) {
         case SubmissionMode::NonIncreasing: {
             // Use the same method value for all arguments.
             for (unsigned i = 0; i < header.arg_count; ++i) {
-                WriteReg(header.method, header.subchannel, Memory::Read32(current_addr));
+                WriteReg(header.method, header.subchannel, Memory::Read32(current_addr),
+                         header.arg_count - i - 1);
                 current_addr += sizeof(u32);
             }
             break;
         }
         case SubmissionMode::IncreaseOnce: {
             ASSERT(header.arg_count.Value() >= 1);
+
             // Use the original method for the first argument and then the next method for all other
             // arguments.
+            WriteReg(header.method, header.subchannel, Memory::Read32(current_addr),
+                     header.arg_count - 1);
+            current_addr += sizeof(u32);
 
-            // Process this command as a method call instead of a register write. Gather
-            // all the parameters first and then pass them at once to the CallMethod function.
-            std::vector<u32> parameters(header.arg_count);
-
-            for (unsigned i = 0; i < header.arg_count; ++i) {
-                parameters[i] = Memory::Read32(current_addr);
+            for (unsigned i = 1; i < header.arg_count; ++i) {
+                WriteReg(header.method + 1, header.subchannel, Memory::Read32(current_addr),
+                         header.arg_count - i - 1);
                 current_addr += sizeof(u32);
             }
-
-            CallMethod(header.method, header.subchannel, parameters);
             break;
         }
         case SubmissionMode::Inline: {
             // The register value is stored in the bits 16-28 as an immediate
-            WriteReg(header.method, header.subchannel, header.inline_data);
+            WriteReg(header.method, header.subchannel, header.inline_data, 0);
             break;
         }
         default:
diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index 67b1b4e7f..49a138c1d 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -17,14 +17,21 @@ const std::unordered_map<u32, Maxwell3D::MethodInfo> Maxwell3D::method_handlers
 
 Maxwell3D::Maxwell3D(MemoryManager& memory_manager) : memory_manager(memory_manager) {}
 
-void Maxwell3D::AttemptMethodCall(u32 method, const std::vector<u32>& parameters) {
+void Maxwell3D::SubmitMacroCode(u32 entry, std::vector<u32> code) {
+    uploaded_macros[entry * 2 + MacroRegistersStart] = std::move(code);
+}
+
+void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters) {
     // TODO(Subv): Write an interpreter for the macros uploaded via registers 0x45 and 0x47
+
+    // The requested macro must have been uploaded already.
+    ASSERT_MSG(uploaded_macros.find(method) != uploaded_macros.end(), "Macro %08X was not uploaded",
+               method);
+
     auto itr = method_handlers.find(method);
     ASSERT_MSG(itr != method_handlers.end(), "Unhandled method call %08X", method);
 
-    // Only execute the macro handler once we've been fed the expected number of parameters.
-    if (itr->second.arguments != parameters.size())
-        return;
+    ASSERT(itr->second.arguments == parameters.size());
 
     (this->*itr->second.handler)(parameters);
 
@@ -33,7 +40,7 @@ void Maxwell3D::AttemptMethodCall(u32 method, const std::vector<u32>& parameters
     macro_params.clear();
 }
 
-void Maxwell3D::WriteReg(u32 method, u32 value) {
+void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Maxwell3D register, increase the size of the Regs structure");
 
@@ -56,8 +63,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value) {
 
         macro_params.push_back(value);
 
-        // Try to call the macro with the current number of parameters.
-        AttemptMethodCall(executing_macro, macro_params);
+        // Call the macro when there are no more parameters in the command buffer
+        if (remaining_params == 0) {
+            CallMacroMethod(executing_macro, macro_params);
+        }
         return;
     }
 
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index 75a1c05bc..05820a21e 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -21,7 +21,10 @@ public:
     ~Maxwell3D() = default;
 
     /// Write the value to the register identified by method.
-    void WriteReg(u32 method, u32 value);
+    void WriteReg(u32 method, u32 value, u32 remaining_params);
+
+    /// Uploads the code for a GPU macro program associated with the specified entry.
+    void SubmitMacroCode(u32 entry, std::vector<u32> code);
 
     /// Register structure of the Maxwell3D engine.
     /// TODO(Subv): This structure will need to be made bigger as more registers are discovered.
@@ -198,18 +201,19 @@ public:
 private:
     MemoryManager& memory_manager;
 
+    std::unordered_map<u32, std::vector<u32>> uploaded_macros;
+
     /// Macro method that is currently being executed / being fed parameters.
     u32 executing_macro = 0;
     /// Parameters that have been submitted to the macro call so far.
     std::vector<u32> macro_params;
 
     /**
-     * Attempts a method call to this engine. Will return without doing anything if the number of
-     * parameters doesn't match what is expected for the method.
+     * Call a macro on this engine.
      * @param method Method to call
      * @param parameters Arguments to the method call
      */
-    void AttemptMethodCall(u32 method, const std::vector<u32>& parameters);
+    void CallMacroMethod(u32 method, const std::vector<u32>& parameters);
 
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index ba7781756..d2e4ff52d 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -6,6 +6,7 @@
 
 #include <memory>
 #include <unordered_map>
+#include <vector>
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/engines/maxwell_3d.h"
@@ -38,8 +39,10 @@ public:
     std::unique_ptr<MemoryManager> memory_manager;
 
 private:
+    static constexpr u32 InvalidGraphMacroEntry = 0xFFFFFFFF;
+
     /// Writes a single register in the engine bound to the specified subchannel
-    void WriteReg(u32 method, u32 subchannel, u32 value);
+    void WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params);
 
     /// Mapping of command subchannels to their bound engine ids.
     std::unordered_map<u32, EngineID> bound_engines;
@@ -50,6 +53,11 @@ private:
     std::unique_ptr<Engines::Fermi2D> fermi_2d;
     /// Compute engine
     std::unique_ptr<Engines::MaxwellCompute> maxwell_compute;
+
+    /// Entry of the macro that is currently being uploaded
+    u32 current_macro_entry = InvalidGraphMacroEntry;
+    /// Code being uploaded for the current macro
+    std::vector<u32> current_macro_code;
 };
 
 } // namespace Tegra