Implement fast_cmyk_to_rgb without calling cmyk_to_rgb.

Same algorithm, just implemented in fixed point with a 1 place cache
and checks for trivial black/white rather than floating point.

1 files changed, 130 insertions, 9 deletions

diff --git a/fitz/res_colorspace.c b/fitz/res_colorspace.c
index 3ae8648e..7b36ac1e 100644
--- a/fitz/res_colorspace.c
+++ b/fitz/res_colorspace.c
@@ -321,18 +321,139 @@ static void fast_cmyk_to_rgb(fz_context *ctx, fz_pixmap *dst, fz_pixmap *src)
 	unsigned char *s = src->samples;
 	unsigned char *d = dst->samples;
 	int n = src->w * src->h;
+	unsigned int C,M,Y,K,r,g,b;
+
+	C = 0;
+	M = 0;
+	Y = 0;
+	K = 0;
+	r = 255;
+	g = 255;
+	b = 255;
+
 	while (n--)
 	{
 #ifdef SLOWCMYK
-		float cmyk[4], rgb[3];
-		cmyk[0] = s[0] / 255.0f;
-		cmyk[1] = s[1] / 255.0f;
-		cmyk[2] = s[2] / 255.0f;
-		cmyk[3] = s[3] / 255.0f;
-		cmyk_to_rgb(ctx, NULL, cmyk, rgb);
-		d[0] = rgb[0] * 255;
-		d[1] = rgb[1] * 255;
-		d[2] = rgb[2] * 255;
+		unsigned int c = s[0];
+		unsigned int m = s[1];
+		unsigned int y = s[2];
+		unsigned int k = s[3];
+		unsigned int cm, c1m, cm1, c1m1, c1m1y, c1m1y1, c1my, c1my1, cm1y, cm1y1, cmy, cmy1;
+		unsigned int x0, x1;
+
+		if (c == C && m == M && y == Y && k == K)
+		{
+			/* Nothing to do */
+		}
+		else if (k == 0 && c == 0 && m == 0 && y == 0)
+		{
+			r = g = b = 255;
+		}
+		else if (k == 255)
+		{
+			r = g = b = 0;
+		}
+		else
+		{
+			c += c>>7;
+			m += m>>7;
+			y += y>>7;
+			k += k>>7;
+			y >>= 1; /* Ditch 1 bit of Y to avoid overflow */
+			cm = c * m;
+			c1m = (m<<8) - cm;
+			cm1 = (c<<8) - cm;
+			c1m1 = ((256 - m)<<8) - cm1;
+			c1m1y  = c1m1 * y;
+			c1m1y1 = (c1m1<<7) - c1m1y;
+			c1my  = c1m * y;
+			c1my1 = (c1m<<7) - c1my;
+			cm1y  = cm1 * y;
+			cm1y1 = (cm1<<7) - cm1y;
+			cmy  = cm * y;
+			cmy1 = (cm<<7) - cmy;
+
+			/* this is a matrix multiplication, unrolled for performance */
+			x1 = c1m1y1 * k;	/* 0 0 0 1 */
+			x0 = (c1m1y1<<8) - x1;	/* 0 0 0 0 */
+			x1 = x1>>8;		/* From 23 fractional bits to 15 */
+			r = g = b = x0;
+			r += 35 * x1; /* 0.1373 */
+			g += 31 * x1; /* 0.1216 */
+			b += 32 * x1; /* 0.1255 */
+
+			x1 = c1m1y * k;		/* 0 0 1 1 */
+			x0 = (c1m1y<<8) - x1;	/* 0 0 1 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			r += 28 * x1; /* 0.1098 */
+			g += 26 * x1; /* 0.1020 */
+			r += x0;
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			g += 243 * x0; /* 0.9490 */
+
+			x1 = c1my1 * k;		/* 0 1 0 1 */
+			x0 = (c1my1<<8) - x1;	/* 0 1 0 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			r += 36 * x1;  /* 0.1412 */
+			r += 237 * x0; /* 0.9255 */
+			b += 141 * x0; /* 0.5490 */
+
+			x1 = c1my * k;		/* 0 1 1 1 */
+			x0 = (c1my<<8) - x1;	/* 0 1 1 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			r += 34 * x1;  /* 0.1333 */
+			r += 238 * x0; /* 0.9294 */
+			g += 28 * x0;  /* 0.1098 */
+			b += 36 * x0;  /* 0.1412 */
+
+			x1 = cm1y1 * k;		/* 1 0 0 1 */
+			x0 = (cm1y1<<8) - x1;	/* 1 0 0 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			g += 15 * x1;  /* 0.0588 */
+			b += 36 * x1;  /* 0.1412 */
+			g += 174 * x0; /* 0.6784 */
+			b += 240 * x0; /* 0.9373 */
+
+			x1 = cm1y * k;		/* 1 0 1 1 */
+			x0 = (cm1y<<8) - x1;	/* 1 0 1 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			g += 19 * x1;  /* 0.0745 */
+			g += 167 * x0; /* 0.6510 */
+			b += 80 * x0;  /* 0.3137 */
+
+			x1 = cmy1 * k;		/* 1 1 0 1 */
+			x0 = (cmy1<<8) - x1;	/* 1 1 0 0 */
+			x1 >>= 8;		/* From 23 fractional bits to 15 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			b += 2 * x1;   /* 0.0078 */
+			r += 46 * x0;  /* 0.1804 */
+			g += 49 * x0;  /* 0.1922 */
+			b += 147 * x0; /* 0.5725 */
+
+			x0 = cmy * (256-k);	/* 1 1 1 0 */
+			x0 >>= 8;		/* From 23 fractional bits to 15 */
+			r += 54 * x0; /* 0.2118 */
+			g += 54 * x0; /* 0.2119 */
+			b += 57 * x0; /* 0.2235 */
+
+			r -= (r>>8);
+			g -= (g>>8);
+			b -= (b>>8);
+			r = r>>23;
+			g = g>>23;
+			b = b>>23;
+			C = c;
+			M = m;
+			Y = y;
+			K = k;
+		}
+		d[0] = r;
+		d[1] = g;
+		d[2] = b;
 #else
 		d[0] = 255 - (unsigned char)fz_mini(s[0] + s[3], 255);
 		d[1] = 255 - (unsigned char)fz_mini(s[1] + s[3], 255);