Fast-pathing MDI's that have only one draw.

src/xenia/gpu/gl4/draw_batcher.cc

@@ -335,6 +335,24 @@ bool DrawBatcher::Flush(FlushMode mode) {
             batch_state_.command_stride, vertex_buffer_count);
       }
     } else {
+      if (batch_state_.draw_count == 1) {
+        // Fast path for one draw. Removes MDI overhead when not required.
+        if (batch_state_.indexed) {
+          auto& cmd = active_draw_.draw_elements_cmd;
+          glDrawElementsInstancedBaseVertexBaseInstance(
+              prim_type, cmd->count, batch_state_.index_type,
+              reinterpret_cast<void*>(
+                  uintptr_t(cmd->first_index) *
+                  (batch_state_.index_type == GL_UNSIGNED_SHORT ? 2 : 4)),
+              cmd->instance_count, cmd->base_vertex, cmd->base_instance);
+        } else {
+          auto& cmd = active_draw_.draw_arrays_cmd;
+          glDrawArraysInstancedBaseInstance(prim_type, cmd->first_index,
+                                            cmd->count, cmd->instance_count,
+                                            cmd->base_instance);
+        }
+      } else {
+        // Full multi-draw.
         if (batch_state_.indexed) {
           glMultiDrawElementsIndirect(prim_type, batch_state_.index_type,
                                       indirect_offset, batch_state_.draw_count,
@@ -345,6 +363,7 @@ bool DrawBatcher::Flush(FlushMode mode) {
                                     batch_state_.command_stride);
         }
       }
+    }
 
     batch_state_.command_range_start = UINTPTR_MAX;
     batch_state_.command_range_length = 0;

src/xenia/kernel/xboxkrnl_threading.cc

@@ -287,7 +287,7 @@ SHIM_CALL KeDelayExecutionThread_shim(PPCContext* ppc_state,
 }
 
 SHIM_CALL NtYieldExecution_shim(PPCContext* ppc_state, KernelState* state) {
-  XELOGD("NtYieldExecution()");
+  //XELOGD("NtYieldExecution()");
   XThread* thread = XThread::GetCurrentThread();
   X_STATUS result = thread->Delay(0, 0, 0);
   SHIM_SET_RETURN_64(0);