Add optimized function written in C that depends on integers being exactly 32-bit.

7 files changed, 102 insertions, 7 deletions

diff --git a/Makefile b/Makefile
index cb98ed38..2623963a 100644
--- a/Makefile
+++ b/Makefile
@@ -75,7 +75,7 @@ FITZ_OBJ=$(FITZ_SRC:fitz/%.c=$(OBJDIR)/%.o)
 $(FITZ_OBJ): $(FITZ_HDR)
 
 DRAW_SRC=$(addprefix draw/, $(ARCH_SRC) \
-	blendmodes.c glyphcache.c imagedraw.c imagescale.c imageunpack.c meshdraw.c \
+	archport.c blendmodes.c glyphcache.c imagedraw.c imagescale.c imageunpack.c meshdraw.c \
 	pathfill.c pathscan.c pathstroke.c porterduff.c )
 DRAW_OBJ=$(DRAW_SRC:draw/%.c=$(OBJDIR)/%.o)
 $(DRAW_OBJ): $(FITZ_HDR)
diff --git a/draw/archarm.c b/draw/archarm.c
index 609a1b90..c5f9d13c 100644
--- a/draw/archarm.c
+++ b/draw/archarm.c
@@ -98,7 +98,7 @@ path_w4i1o4arm(byte * restrict argb, byte * restrict src, byte cov, int len, byt
 }
 
 void
-fz_accelerate(void)
+fz_acceleratearch(void)
 {
 	fz_path_w4i1o4 = path_w4i1o4arm;
 }
diff --git a/draw/archport.c b/draw/archport.c
new file mode 100644
index 00000000..2b45b7c0
--- /dev/null
+++ b/draw/archport.c
@@ -0,0 +1,94 @@
+#include "fitz.h"
+
+/* This C implementation was a prototype of the algorithm used
+ * in the ARM code in draw/archarm.c. It is conceivable that on some
+ * architectures/compilers this may be preferable to the vanilla
+ * version below. */
+
+static void
+path_w4i1o4_32bit(unsigned char * restrict argb,
+	unsigned char * restrict src, unsigned char cov, int len,
+	unsigned char * restrict dst)
+{
+	/* COLOR * coverage + DST * (256-coverage) = (COLOR - DST)*coverage + DST*256 */
+	unsigned int *dst32 = (unsigned int *)(void *)dst;
+	int alpha = argb[0];
+	unsigned int rb = argb[1] | argb[3] << 16;
+	unsigned int ag = 255 | argb[2] << 16;
+	const int MASK = 0xFF00FF00;
+
+	/* sanity test */
+	if (sizeof(int) != 4 || sizeof(unsigned int) != 4)
+		abort();
+
+	if (alpha != 255)
+	{
+		alpha += alpha>>7; /* alpha is now in the 0...256 range */
+		while (len--)
+		{
+			unsigned int ca, drb, dag, crb, cag;
+			cov += *src; *src++ = 0;
+			ca = cov+(cov>>7); /* ca is in 0...256 range */
+			ca = (ca*alpha)>>8; /* ca is is in 0...256 range */
+			dag = *dst32++;
+			if (ca != 0)
+			{
+				drb = dag & MASK;
+				dag = (dag<<8) & MASK;
+				crb = rb - (drb>>8);
+				cag = ag - (dag>>8);
+				drb += crb * ca;
+				dag += cag * ca;
+				drb = drb & MASK;
+				dag = dag & MASK;
+				dag = drb | (dag>>8);
+				dst32[-1] = dag;
+			}
+		}
+	}
+	else
+	{
+		while (len--)
+		{
+			unsigned int ca, drb, dag, crb, cag;
+			cov += *src; *src++ = 0;
+			ca = cov+(cov>>7); /* ca is in 0...256 range */
+			dag = *dst32++;
+			if (ca == 0)
+				continue;
+			if (ca == 255)
+			{
+				dag = (rb<<8) | ag;
+			}
+			else
+			{
+				drb = dag & MASK;
+				dag = (dag<<8) & MASK;
+				crb = rb - (drb>>8);
+				cag = ag - (dag>>8);
+				drb += crb * ca;
+				dag += cag * ca;
+				drb = drb & MASK;
+				dag = dag & MASK;
+				dag = drb | (dag>>8);
+			}
+			dst32[-1] = dag;
+		}
+	}
+}
+
+void fz_accelerate(void)
+{
+	if (sizeof(int) == 4)
+	{
+		fz_path_w4i1o4 = path_w4i1o4_32bit;
+	}
+
+	if (sizeof(int) == 8)
+	{
+	}
+
+#ifdef HAVE_CPUDEP
+	fz_acceleratearch();
+#endif
+}
diff --git a/draw/archx86.c b/draw/archx86.c
index b75fb28f..0c313f33 100644
--- a/draw/archx86.c
+++ b/draw/archx86.c
@@ -220,7 +220,7 @@ static void img_4o4mmx(FZ_PSRC, FZ_PDST, FZ_PCTM)
 
 #if defined (ARCH_X86) || defined(ARCH_X86_64)
 void
-fz_accelerate(void)
+fz_acceleratearch(void)
 {
 #  ifdef HAVE_MMX
 	if (fz_cpuflags & HAVE_MMX)
diff --git a/fitz/base_cpudep.c b/fitz/base_cpudep.c
index bc9bee91..980bc9d8 100644
--- a/fitz/base_cpudep.c
+++ b/fitz/base_cpudep.c
@@ -12,10 +12,6 @@ unsigned fz_cpuflags = 0;
 
 #ifndef HAVE_CPUDEP
 
-void fz_accelerate(void)
-{
-}
-
 void fz_cpudetect(void)
 {
 }
diff --git a/fitz/fitz_draw.h b/fitz/fitz_draw.h
index 0e53401f..cd155712 100644
--- a/fitz/fitz_draw.h
+++ b/fitz/fitz_draw.h
@@ -452,6 +452,7 @@ void fz_dashpath(fz_gel *gel, fz_path *path, fz_strokestate *stroke, fz_matrix c
  */
 
 extern void fz_accelerate(void);
+extern void fz_acceleratearch(void);
 
 extern void (*fz_duff_non)(unsigned char*,int,int,unsigned char*,int,int,int);
 extern void (*fz_duff_nimon)(unsigned char*,int,int,unsigned char*,int,int,unsigned char*,int,int,int);
diff --git a/win32/draw/draw.vcproj b/win32/draw/draw.vcproj
index af1280dc..45954e31 100644
--- a/win32/draw/draw.vcproj
+++ b/win32/draw/draw.vcproj
@@ -147,6 +147,10 @@
 			Name="draw"

 			>

 			<File

+				RelativePath="..\..\draw\archport.c"

+				>

+			</File>

+			<File

 				RelativePath="..\..\draw\archx86.c"

 				>

 			</File>