From cbd488e15740a50f568b1894d9ee8f303fc6694a Mon Sep 17 00:00:00 2001
From: rogerman <rogerman@users.noreply.github.com>
Date: Wed, 25 Jul 2018 15:18:36 -0700
Subject: [PATCH] SoftRasterizer: Force the number of rendering threads to be a
 power-of-two since any additional threads that are not power-of-two will be
 wasted.

---
 desmume/src/frontend/cocoa/cocoa_GPU.mm | 31 +------------------------
 desmume/src/rasterize.cpp               | 19 ++++++++++++++-
 2 files changed, 19 insertions(+), 31 deletions(-)

diff --git a/desmume/src/frontend/cocoa/cocoa_GPU.mm b/desmume/src/frontend/cocoa/cocoa_GPU.mm
index 8aa520031..291533b3f 100644
--- a/desmume/src/frontend/cocoa/cocoa_GPU.mm
+++ b/desmume/src/frontend/cocoa/cocoa_GPU.mm
@@ -448,36 +448,7 @@ public:
 	if (numberThreads == 0)
 	{
 		isCPUCoreCountAuto = YES;
-		
-		if (numberCores >= 96)
-		{
-			numberCores = 128;
-		}
-		else if (numberCores >= 48)
-		{
-			numberCores = 64;
-		}
-		else if (numberCores >= 24)
-		{
-			numberCores = 32;
-		}
-		else if (numberCores >= 16)
-		{
-			numberCores = 16;
-		}
-		else if (numberCores >= 8)
-		{
-			numberCores = 8;
-		}
-		else if (numberCores >= 4)
-		{
-			numberCores = 4;
-		}
-		else if (numberCores >= 2)
-		{
-			numberCores = 2;
-		}
-		else
+		if (numberCores < 2)
 		{
 			numberCores = 1;
 		}
diff --git a/desmume/src/rasterize.cpp b/desmume/src/rasterize.cpp
index 0617a0ac0..130c58637 100644
--- a/desmume/src/rasterize.cpp
+++ b/desmume/src/rasterize.cpp
@@ -1444,7 +1444,24 @@ SoftRasterizerRenderer::SoftRasterizerRenderer()
 	
 	_HACK_viewer_rasterizerUnit.SetSLI(0, 1, false);
 	
-	_threadCount = CommonSettings.num_cores;
+	const size_t coreCount = CommonSettings.num_cores;
+	_threadCount = coreCount;
+	
+	// SoftRasterizer works best when the number of threads is a power-of-two.
+	// Ensure that the thread count is set to the previous power-of-two if the
+	// core count isn't already a power-of-two.
+	_threadCount--;
+	_threadCount |= (_threadCount >> 1);
+	_threadCount |= (_threadCount >> 2);
+	_threadCount |= (_threadCount >> 4);
+	_threadCount |= (_threadCount >> 8);
+	_threadCount |= (_threadCount >> 16);
+	_threadCount++;
+	
+	if (_threadCount != coreCount)
+	{
+		_threadCount >>= 1;
+	}
 	
 	if (_threadCount > SOFTRASTERIZER_MAX_THREADS)
 	{