From 55c4488ab2127adbc4a24e7e7753f79fc06d96aa Mon Sep 17 00:00:00 2001
From: Ben Vanik <ben.vanik@gmail.com>
Date: Sun, 4 Jan 2015 14:23:28 -0800
Subject: [PATCH] Fast-pathing MDI's that have only one draw.

---
 src/xenia/gpu/gl4/draw_batcher.cc      | 33 ++++++++++++++++++++------
 src/xenia/kernel/xboxkrnl_threading.cc |  2 +-
 2 files changed, 27 insertions(+), 8 deletions(-)

diff --git a/src/xenia/gpu/gl4/draw_batcher.cc b/src/xenia/gpu/gl4/draw_batcher.cc
index c5d0bdb39..304992a80 100644
--- a/src/xenia/gpu/gl4/draw_batcher.cc
+++ b/src/xenia/gpu/gl4/draw_batcher.cc
@@ -335,14 +335,33 @@ bool DrawBatcher::Flush(FlushMode mode) {
             batch_state_.command_stride, vertex_buffer_count);
       }
     } else {
-      if (batch_state_.indexed) {
-        glMultiDrawElementsIndirect(prim_type, batch_state_.index_type,
-                                    indirect_offset, batch_state_.draw_count,
-                                    batch_state_.command_stride);
+      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 {
-        glMultiDrawArraysIndirect(prim_type, indirect_offset,
-                                  batch_state_.draw_count,
-                                  batch_state_.command_stride);
+        // Full multi-draw.
+        if (batch_state_.indexed) {
+          glMultiDrawElementsIndirect(prim_type, batch_state_.index_type,
+                                      indirect_offset, batch_state_.draw_count,
+                                      batch_state_.command_stride);
+        } else {
+          glMultiDrawArraysIndirect(prim_type, indirect_offset,
+                                    batch_state_.draw_count,
+                                    batch_state_.command_stride);
+        }
       }
     }
 
diff --git a/src/xenia/kernel/xboxkrnl_threading.cc b/src/xenia/kernel/xboxkrnl_threading.cc
index e63ca2ef0..4c404aa1b 100644
--- a/src/xenia/kernel/xboxkrnl_threading.cc
+++ b/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);