From daedf9d158ba6d9265ac1b75bf015889880a856c Mon Sep 17 00:00:00 2001
From: "alexey.lysiuk" <alexey.lysiuk@gmail.com>
Date: Fri, 20 Dec 2019 11:04:50 +0200
Subject: [PATCH] - improved normalNx scaling performance by ~10%

Actual boost heavily depends on platform's memory architecture, made it a bit more cache friendly in general
---
 src/gamedata/textures/hires/hqresize.cpp | 46 +++++++++++-------------
 1 file changed, 21 insertions(+), 25 deletions(-)

diff --git a/src/gamedata/textures/hires/hqresize.cpp b/src/gamedata/textures/hires/hqresize.cpp
index 8e93689dd..f67d4a137 100644
--- a/src/gamedata/textures/hires/hqresize.cpp
+++ b/src/gamedata/textures/hires/hqresize.cpp
@@ -236,29 +236,7 @@ static unsigned char *scaleNxHelper( void (*scaleNxFunction) ( uint32_t* , uint3
 	return newBuffer;
 }
 
-static void normalNx ( uint32_t* inputBuffer, uint32_t* outputBuffer, int inWidth, int inHeight, int size )
-{
-	const int width = size * inWidth;
-	const int height = size * inHeight;
-
-	for ( int i = 0; i < inWidth; ++i )
-	{
-		for ( int j = 0; j < inHeight; ++j )
-		{
-			const uint32_t E = inputBuffer[ i     +inWidth*j    ];
-			for ( int k = 0; k < size; k++ )
-			{
-				for ( int l = 0; l < size; l++ )
-				{
-					outputBuffer[size*i+k + width*(size*j+l)] = E;
-				}
-			}
-		}
-	}
-}
-
-static unsigned char *normalNxHelper( void (normalNxFunction) ( uint32_t* , uint32_t* , int , int, int),
-							  const int N,
+static unsigned char *normalNx(const int N,
 							  unsigned char *inputBuffer,
 							  const int inWidth,
 							  const int inHeight,
@@ -269,7 +247,25 @@ static unsigned char *normalNxHelper( void (normalNxFunction) ( uint32_t* , uint
 	outHeight = N *inHeight;
 	unsigned char * newBuffer = new unsigned char[outWidth*outHeight*4];
 
-	normalNxFunction ( reinterpret_cast<uint32_t*> ( inputBuffer ), reinterpret_cast<uint32_t*> ( newBuffer ), inWidth, inHeight, N );
+	uint32_t *const inBuffer = reinterpret_cast<uint32_t *>(inputBuffer);
+	uint32_t *const outBuffer = reinterpret_cast<uint32_t *>(newBuffer);
+
+	for (int y = 0; y < inHeight; ++y)
+	{
+		const int inRowPos = inWidth * y;
+		const int outRowPos = outWidth * N * y;
+
+		for (int x = 0; x < inWidth; ++x)
+		{
+			std::fill_n(&outBuffer[outRowPos + N * x], N, inBuffer[inRowPos + x]);
+		}
+
+		for (int c = 1; c < N; ++c)
+		{
+			std::copy_n(&outBuffer[outRowPos], outWidth, &outBuffer[outRowPos + outWidth * c]);
+		}
+	}
+
 	delete[] inputBuffer;
 	return newBuffer;
 }
@@ -495,7 +491,7 @@ void FTexture::CreateUpsampledTextureBuffer(FTextureBuffer &texbuffer, bool hasA
 		else if (type == 5)
 			texbuffer.mBuffer = xbrzHelper(xbrzOldScale, mult, texbuffer.mBuffer, inWidth, inHeight, texbuffer.mWidth, texbuffer.mHeight);
 		else if (type == 6)
-			texbuffer.mBuffer = normalNxHelper(&normalNx, mult, texbuffer.mBuffer, inWidth, inHeight, texbuffer.mWidth, texbuffer.mHeight);
+			texbuffer.mBuffer = normalNx(mult, texbuffer.mBuffer, inWidth, inHeight, texbuffer.mWidth, texbuffer.mHeight);
 		else
 			return;
 	}