restore previous twiddle table for non-square tex

core/rend/TexCache.cpp

@@ -23,23 +23,60 @@ u32 palette32_ram[1024];
 u32 pal_hash_256[4];
 u32 pal_hash_16[64];
 
-u32 detwiddle[1024];
+u32 detwiddle[2][10][1024];
+//input : address in the yyyyyxxxxx format
+//output : address in the xyxyxyxy format
+//U : x resolution , V : y resolution
+//twiddle works on 64b words
 
-void BuildTwiddleTable()
+
+static u32 twiddle_slow(u32 x,u32 y,u32 x_sz,u32 y_sz)
 {
-    for (u32 j = 0; j < ARRAY_SIZE(detwiddle); j++)
-    {
-        u32 detwiddled = 0;
-        for (int i = 0; i < 10; i++)
-        {
-            u32 shift = 1 << i;
-            detwiddled |= ((j & shift) << i);
-        }
-    	detwiddle[j] = detwiddled;
-    }
+	u32 rv=0;//low 2 bits are directly passed  -> needs some misc stuff to work.However
+			 //Pvr internally maps the 64b banks "as if" they were twiddled :p
+
+	u32 sh=0;
+	x_sz>>=1;
+	y_sz>>=1;
+	while(x_sz!=0 || y_sz!=0)
+	{
+		if (y_sz)
+		{
+			u32 temp=y&1;
+			rv|=temp<<sh;
+
+			y_sz>>=1;
+			y>>=1;
+			sh++;
+		}
+		if (x_sz)
+		{
+			u32 temp=x&1;
+			rv|=temp<<sh;
+
+			x_sz>>=1;
+			x>>=1;
+			sh++;
+		}
+	}	
+	return rv;
 }
 
-static OnLoad btt(&BuildTwiddleTable);
+static void BuildTwiddleTables()
+{
+	for (u32 s = 0; s < 10; s++)
+	{
+		u32 x_sz = 1024;
+		u32 y_sz = 2 << s;
+		for (u32 i = 0; i < x_sz; i++)
+		{
+			detwiddle[0][s][i] = twiddle_slow(i, 0, x_sz, y_sz);
+			detwiddle[1][s][i] = twiddle_slow(0, i, y_sz, x_sz);
+		}
+	}
+}
+
+static OnLoad btt(&BuildTwiddleTables);
 
 void palette_update()
 {
@@ -1099,3 +1136,4 @@ void dump_screenshot(u8 *buffer, u32 width, u32 height, bool alpha, u32 rowPitch
 
 }
 #endif
+

core/rend/TexCache.h

@@ -17,7 +17,7 @@ extern u32 pal_hash_256[4];
 extern u32 pal_hash_16[64];
 extern bool KillTex;
 
-extern u32 detwiddle[1024];
+extern u32 detwiddle[2][10][1024];
 
 template<class pixel_type>
 class PixelBuffer
@@ -168,6 +168,8 @@ __forceinline u32 YUV422(s32 Y,s32 Yu,s32 Yv)
 	return PixelPacker::packRGB(clamp(0,255,R),clamp(0,255,G),clamp(0,255,B));
 }
 
+#define twop(x,y,bcx,bcy) (detwiddle[0][bcy][x]+detwiddle[1][bcx][y])
+
 //pixel packers !
 struct pp_565
 {
@@ -517,23 +519,24 @@ void texture_PL(PixelBuffer<pixel_type>* pb,u8* p_in,u32 Width,u32 Height)
 	}
 }
 
-static inline u32 get_tw_texel_position(u32 x, u32 y)
-{
-    return detwiddle[y]  |  detwiddle[x] << 1;
-}
-
 template<class PixelConvertor, class pixel_type>
 void texture_TW(PixelBuffer<pixel_type>* pb,u8* p_in,u32 Width,u32 Height)
 {
 	pb->amove(0,0);
 
-	const u32 divider = PixelConvertor::xpp * PixelConvertor::ypp;
+	const u32 divider=PixelConvertor::xpp*PixelConvertor::ypp;
 
-	for (u32 y = 0; y < Height; y += PixelConvertor::ypp)
+	unsigned long bcx_,bcy_;
+	bcx_=bitscanrev(Width);
+	bcy_=bitscanrev(Height);
+	const u32 bcx=bcx_-1;
+	const u32 bcy=bcy_-1;
+
+	for (u32 y=0;y<Height;y+=PixelConvertor::ypp)
 	{
-		for (u32 x = 0; x < Width; x += PixelConvertor::xpp)
+		for (u32 x=0;x<Width;x+=PixelConvertor::xpp)
 		{
-			u8* p = &p_in[get_tw_texel_position(x, y) / divider * 8];
+			u8* p = &p_in[(twop(x,y,bcx,bcy)/divider)<<3];
 			PixelConvertor::Convert(pb,p);
 
 			pb->rmovex(PixelConvertor::xpp);
@@ -548,14 +551,18 @@ void texture_VQ(PixelBuffer<pixel_type>* pb,u8* p_in,u32 Width,u32 Height)
 	p_in+=256*4*2;
 	pb->amove(0,0);
 
-	Height /= PixelConvertor::ypp;
-	Width /= PixelConvertor::xpp;
+	const u32 divider=PixelConvertor::xpp*PixelConvertor::ypp;
+	unsigned long bcx_,bcy_;
+	bcx_=bitscanrev(Width);
+	bcy_=bitscanrev(Height);
+	const u32 bcx=bcx_-1;
+	const u32 bcy=bcy_-1;
 
-	for (u32 y = 0; y < Height; y++)
+	for (u32 y=0;y<Height;y+=PixelConvertor::ypp)
 	{
-		for (u32 x = 0; x < Width; x++)
+		for (u32 x=0;x<Width;x+=PixelConvertor::xpp)
 		{
-			u8 p = p_in[get_tw_texel_position(x, y)];
+			u8 p = p_in[twop(x,y,bcx,bcy)/divider];
 			PixelConvertor::Convert(pb,&vq_codebook[p*8]);
 
 			pb->rmovex(PixelConvertor::xpp);