3 files changed, 20 insertions, 4 deletions

diff --git a/src/video_core/engines/maxwell_3d.cpp b/src/video_core/engines/maxwell_3d.cpp
index c8c92757a..5d516cdb3 100644
--- a/src/video_core/engines/maxwell_3d.cpp
+++ b/src/video_core/engines/maxwell_3d.cpp
@@ -329,6 +329,10 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         ProcessMacroBind(method_call.argument);
         break;
     }
+    case MAXWELL3D_REG_INDEX(firmware[4]): {
+        ProcessFirmwareCall4();
+        break;
+    }
     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]):
@@ -419,6 +423,14 @@ void Maxwell3D::ProcessMacroBind(u32 data) {
     macro_positions[regs.macros.entry++] = data;
 }
 
+void Maxwell3D::ProcessFirmwareCall4() {
+    LOG_WARNING(HW_GPU, "(STUBBED) called");
+
+    // Firmware call 4 is a blob that changes some registers depending on its parameters.
+    // These registers don't affect emulation and so are stubbed by setting 0xd00 to 1.
+    regs.reg_array[0xd00] = 1;
+}
+
 void Maxwell3D::ProcessQueryGet() {
     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
diff --git a/src/video_core/engines/maxwell_3d.h b/src/video_core/engines/maxwell_3d.h
index f67a5389f..e5ec90717 100644
--- a/src/video_core/engines/maxwell_3d.h
+++ b/src/video_core/engines/maxwell_3d.h
@@ -1089,7 +1089,9 @@ public:
                     INSERT_PADDING_WORDS(14);
                 } shader_config[MaxShaderProgram];
 
-                INSERT_PADDING_WORDS(0x80);
+                INSERT_PADDING_WORDS(0x60);
+
+                u32 firmware[0x20];
 
                 struct {
                     u32 cb_size;
@@ -1319,6 +1321,9 @@ private:
     /// Handles writes to the macro bind register.
     void ProcessMacroBind(u32 data);
 
+    /// Handles firmware blob 4
+    void ProcessFirmwareCall4();
+
     /// Handles a write to the CLEAR_BUFFERS register.
     void ProcessClearBuffers();
 
@@ -1431,6 +1436,7 @@ ASSERT_REG_POSITION(vertex_array[0], 0x700);
 ASSERT_REG_POSITION(independent_blend, 0x780);
 ASSERT_REG_POSITION(vertex_array_limit[0], 0x7C0);
 ASSERT_REG_POSITION(shader_config[0], 0x800);
+ASSERT_REG_POSITION(firmware, 0x8C0);
 ASSERT_REG_POSITION(const_buffer, 0x8E0);
 ASSERT_REG_POSITION(cb_bind[0], 0x904);
 ASSERT_REG_POSITION(tex_cb_index, 0x982);
diff --git a/src/video_core/macro_interpreter.cpp b/src/video_core/macro_interpreter.cpp
index 4e1cb98db..62afc0d11 100644
--- a/src/video_core/macro_interpreter.cpp
+++ b/src/video_core/macro_interpreter.cpp
@@ -131,9 +131,7 @@ bool MacroInterpreter::Step(u32 offset, bool is_delay_slot) {
 
     // An instruction with the Exit flag will not actually
     // cause an exit if it's executed inside a delay slot.
-    // TODO(Blinkhawk): Reversed to always exit. The behavior explained above requires further
-    // testing on the MME code.
-    if (opcode.is_exit) {
+    if (opcode.is_exit && !is_delay_slot) {
         // Exit has a delay slot, execute the next instruction
         Step(offset, true);
         return false;