gka fixes. use truetypes in fontfile.

7 files changed, 155 insertions, 67 deletions

diff --git a/TODO b/TODO
index 3e324907..2cf21011 100644
--- a/TODO
+++ b/TODO
@@ -1,3 +1,14 @@
+immediate plan:
+	1 - pdf_resolve remake
+	2 - refcount resources
+	3 - put image load/scale into rastfuncs
+	4 - altivec optimize
+	5 - gtk+pdf
+	6 - page labels + dests + outline + annots
+	7 - global font/cmap cache
+
+---
+
 colorspace conversions (v2)
 	- cal*
 	- iccbased
diff --git a/base/cpudep.c b/base/cpudep.c
index 0301213d..c7af38b9 100644
--- a/base/cpudep.c
+++ b/base/cpudep.c
@@ -32,8 +32,8 @@ static void sse(void)
 static void sse2(void)
 { __asm__ ("andpd %xmm0, %xmm0\n\t"); }
 
-static void sse3(void)
-{ __asm__ ("haddps %%xmm0, %%xmm0\n\t" : : : "%xmm0"); }
+/* static void sse3(void) */
+/* { __asm__ ("haddps %%xmm0, %%xmm0\n\t" : : : "%xmm0"); } */
 
 #ifdef ARCH_X86_64
 static void amd64(void)
@@ -47,7 +47,7 @@ static const featuretest features[] = {
 	{ mmxext, HAVE_MMXEXT, "mmxext" },
 	{ sse, HAVE_SSE, "sse" },
 	{ sse2, HAVE_SSE2, "sse2" },
-	{ sse3, HAVE_SSE3, "sse3" },
+/*	{ sse3, HAVE_SSE3, "sse3" }, */
 #ifdef ARCH_X86_64
 	{ amd64, HAVE_AMD64, "amd64" }
 #endif
@@ -105,33 +105,45 @@ sigillhandler(int sig)
 	siglongjmp(jmpbuf, 1);
 }
 
+static int
+enabled(char *env, const char *ext)
+{
+	int len;
+	char *s;
+	if (!env)
+		return 1;
+	len = strlen(ext);
+	while ((s = strstr(env, ext)))
+	{
+		s += len;
+		if (*s == ' ' || *s == ',' || *s == '\0')
+			return 1;
+	}
+	return 0;
+}
+
 static void
 dumpflags(void)
 {
 	unsigned f = fz_cpuflags;
 	int i, n;
 
-	fputs("detected cpu features: ", stdout);
+	fputs("detected cpu features:", stdout);
 	n = 0;
 	for (i = 0; i < sizeof(features) / sizeof(featuretest); i++)
 	{
 		if (f & features[i].flag)
 		{
+			fputc(' ', stdout);
 			fputs(features[i].name, stdout);
 			n ++;
 		}
 	}
 	if (!n)
-		fputs("none", stdout);
+		fputs(" none", stdout);
 	fputc('\n', stdout);
 }
 
-/* called by runtime before main()...
- * TODO:
- * CPUACCEL=0 ./x11pdf disables detection
- * CPUACCEL='mmx sse' enables only mmx and sse
- * not set enables everything
- */
 void fz_cpudetect(void)
 {
 	static int hasrun = 0;
@@ -140,11 +152,14 @@ void fz_cpudetect(void)
 	int i;
 	void (*oldhandler)(int) = NULL;
 	void (*tmphandler)(int);
+	char *env;
 
 	if (hasrun)
 		return;
 	hasrun = 1;
 
+	env = getenv("CPUACCEL");
+
 	for (i = 0; i < sizeof(features) / sizeof(featuretest); i++)
 	{
 		canjump = 0;
@@ -165,7 +180,10 @@ void fz_cpudetect(void)
 		features[i].test();
 
 		/* if we got here the test succeeded */
-		flags |= features[i].flag;
+		if (enabled(env, features[i].name))
+			flags |= features[i].flag;
+		else
+			flags &= ~features[i].flag;
 	}
 
 	/* restore previous signal handler */
@@ -176,7 +194,7 @@ void fz_cpudetect(void)
 	dumpflags();
 }
 
-static __attribute__((constructor)) void fzcpudetect(void)
+static __attribute__((constructor, used)) void fzcpudetect(void)
 {
 	fz_cpudetect();
 }
diff --git a/mupdf/fontfile.c b/mupdf/fontfile.c
index ff46c8f1..cdce1a86 100644
--- a/mupdf/fontfile.c
+++ b/mupdf/fontfile.c
@@ -42,11 +42,11 @@ static char *basenames[15] =
 
 static struct { char *collection; char *serif; char *gothic; } cidfonts[5] =
 {
-	{ "Adobe-CNS1", "MOESung-Regular", "MOEKai-Regular" },
-	{ "Adobe-GB1", "gkai00mp", "gbsn00lp" },
-	{ "Adobe-Japan1", "WadaMin-Regular", "WadaMaruGo-Regular" },
-	{ "Adobe-Japan2", "WadaMin-RegularH", "WadaMaruGo-RegularH" },
-	{ "Adobe-Korea1", "Munhwa-Regular", "MunhwaGothic-Regular" },
+	{ "Adobe-CNS1", "bkai00mp.ttf", "bsmi00lp.ttf" },
+	{ "Adobe-GB1", "gkai00mp.ttf", "gbsn00lp.ttf" },
+	{ "Adobe-Japan1", "kochi-mincho.ttf", "kochi-gothic.ttf" },
+	{ "Adobe-Japan2", "kochi-mincho.ttf", "kochi-gothic.ttf" },
+	{ "Adobe-Korea1", "batang.ttf", "dotum.ttf" },
 };
 
 static void loadfontdata(int i, unsigned char **d, unsigned int *l)
@@ -130,12 +130,14 @@ printf("  load system cid font '%s'\n", filename);
 
 	fontdir = getenv("FONTDIR");
 	if (!fontdir)
-		return fz_throw("ioerror: FONTDIR environment not set");
+	{
+		fontdir = "/usr/local/share/font";
+		fz_warn("FONTDIR environment not set");
+	}
 
 	strlcpy(path, fontdir, sizeof path);
 	strlcat(path, "/", sizeof path);
 	strlcat(path, filename, sizeof path);
-	strlcat(path, ".cid.cff", sizeof path);
 
 	if (access(path, R_OK))
 		return fz_throw("ioerror: could not access file '%s'", path);
diff --git a/render/rastport.c b/render/rastport.c
index 9cae06d5..3eb89e43 100644
--- a/render/rastport.c
+++ b/render/rastport.c
@@ -264,15 +264,18 @@ static void msk_1o1(byte *src, byte *dst, int w)
 
 static void msk_w3i1o4(byte *rgb, byte *src, byte *dst, int n)
 {
+	byte rgb0 = rgb[0];
+	byte rgb1 = rgb[1];
+	byte rgb2 = rgb[2];
 	byte sa, ssa;
 	while (n--)
 	{
 		sa = src[0];
 		ssa = 255 - sa;
 		dst[0] = sa + fz_mul255(dst[0], ssa);
-		dst[1] = rgb[0] + fz_mul255((short)dst[1] - rgb[0], ssa);
-		dst[2] = rgb[1] + fz_mul255((short)dst[2] - rgb[1], ssa);
-		dst[3] = rgb[2] + fz_mul255((short)dst[3] - rgb[2], ssa);
+		dst[1] = rgb0 + fz_mul255((short)dst[1] - rgb0, ssa);
+		dst[2] = rgb1 + fz_mul255((short)dst[2] - rgb1, ssa);
+		dst[3] = rgb2 + fz_mul255((short)dst[3] - rgb2, ssa);
 		src ++;
 		dst += 4;
 	}
@@ -496,6 +499,9 @@ static void img_4o4(FZ_PSRC, FZ_PDST, FZ_PCTM)
 
 static void img_w3i1o4(byte *rgb, FZ_PSRC, FZ_PDST, FZ_PCTM)
 {
+	byte rgb0 = rgb[0];
+	byte rgb1 = rgb[1];
+	byte rgb2 = rgb[2];
 	byte sa, ssa;
 	while (h--)
 	{
@@ -508,9 +514,9 @@ static void img_w3i1o4(byte *rgb, FZ_PSRC, FZ_PDST, FZ_PCTM)
 			sa = samplemask(src, srcw, srch, u, v);
 			ssa = 255 - sa;
 			dstp[0] = sa + fz_mul255(dstp[0], ssa);
-			dstp[1] = rgb[0] + fz_mul255((short)dstp[1] - rgb[0], ssa);
-			dstp[2] = rgb[1] + fz_mul255((short)dstp[2] - rgb[1], ssa);
-			dstp[3] = rgb[2] + fz_mul255((short)dstp[3] - rgb[2], ssa);
+			dstp[1] = rgb0 + fz_mul255((short)dstp[1] - rgb0, ssa);
+			dstp[2] = rgb1 + fz_mul255((short)dstp[2] - rgb1, ssa);
+			dstp[3] = rgb2 + fz_mul255((short)dstp[3] - rgb2, ssa);
 			dstp += 4;
 			u += fa;
 			v += fb;
diff --git a/render/rastppc.c b/render/rastppc.c
index f26e5b66..276ee2d7 100644
--- a/render/rastppc.c
+++ b/render/rastppc.c
@@ -1,6 +1,7 @@
 /*
-PowerPC specific render optims live here
-*/
+ * PowerPC specific render optims live here
+ */
+
 #include <fitz.h>
 
 #ifdef HAVE_ALTIVEC
@@ -14,6 +15,7 @@ fz_accelrastfuncs(fz_rastfuncs *tab)
 #  ifdef HAVE_ALTIVEC
 	if (fz_cpuflags & HAVE_ALTIVEC)
 	{
+		puts("installed altivec rastfuncs");
 	}
 #  endif
 }
diff --git a/render/rastx86.c b/render/rastx86.c
index 79020fb1..9360b5e2 100644
--- a/render/rastx86.c
+++ b/render/rastx86.c
@@ -3,6 +3,8 @@ x86 specific render optims live here
 */
 #include <fitz.h>
 
+typedef unsigned char byte;
+
 /* always surround cpu specific code with HAVE_XXX */
 #ifdef HAVE_MMX
 
@@ -10,15 +12,77 @@ x86 specific render optims live here
    shouldn't require anything */
 #include <mmintrin.h>
 
+static void duff_4i1o4mmx(byte *sp0, int sw, byte *mp0, int mw, byte *dp0, int dw, int w0, int h)
+{
+	/*
+	  rendering all pages of
+	  x11pdf ~/doc/OpenGL/Presentations/CEDEC2003_Venus_and_Vulcan.pdf
+	    %     cumulative  self	     self     total
+	  time     seconds   seconds    calls ms/call ms/call  name
+	  30.50     20.04    20.04      261    76.76    76.76  duff_4i1o4
+	  21.67     22.02    10.95      221    49.55    49.55  duff_4i1o4mmx
+	*/
+	__m64 mzero = _mm_setzero_si64();
+	while (h--)
+	{
+		byte *sp = sp0;
+		byte *mp = mp0;
+		byte *dp = dp0;
+
+		unsigned *s = (unsigned *)sp;
+		unsigned *d = (unsigned *)dp;
+
+		int w = w0;
+
+		/* TODO: unroll and process two pixels/iteration */
+		while (w--)
+		{
+			int ts = *s++;
+			int ma = *mp++ + 1;
+			int sa = ((ts & 0xff) * ma) >> 8;
+			int ssa = 254 - sa;
+
+			__m64 d0 = _mm_cvtsi32_si64(*d);
+			__m64 s0 = _mm_cvtsi32_si64(ts);
+
+			/* 4 x 9 bit alpha value */
+			__m64 mma = _mm_set1_pi16(ma);
+			__m64 mssa = _mm_set1_pi16(ssa);
+
+			/* unpack 0000argb => a0r0g0b0 */
+			__m64 d1 = _mm_unpacklo_pi8(d0, mzero);
+			__m64 s1 = _mm_unpacklo_pi8(s0, mzero);
+
+			/* s1 * ma => a0r0g0b0 */
+			__m64 msma = _mm_mullo_pi16(s1, mma);
+			/* d1 * mssa */
+			__m64 mdssa = _mm_mullo_pi16(d1, mssa);
+
+			__m64 res0 = _mm_add_pi16(msma, mdssa);
+			/* TODO: is it possible to get rid of the shift? */
+			__m64 res1 = _mm_srli_pi16(res0, 8);
+
+			/* pack */
+			__m64 res2 = _mm_packs_pu16(res1, mzero);
+
+			*d++ = _mm_cvtsi64_si32(res2);
+		}
+
+		sp0 += sw;
+		mp0 += mw;
+		dp0 += dw;
+	}
+
+	_mm_empty();
+}
+
 static inline unsigned
 getargb(unsigned *s, int w, int h, int u, int v)
 {
-	if (u < 0 || u >= w) return 0;
-	if (v < 0 || v >= h) return 0;
+	if (u < 0 | u >= w | v < 0 | v >= h) return 0;
 	return s[w * v + u];
 }
 
-/* this code has not been tested since refactoring */
 static void img_4o4mmx(FZ_PSRC, FZ_PDST, FZ_PCTM)
 {
 	/* since mmx does not have an unsigned multiply instruction we use
@@ -33,12 +97,16 @@ static void img_4o4mmx(FZ_PSRC, FZ_PDST, FZ_PCTM)
 		unsigned *d = (unsigned *)dst0;
 		int u = u0;
 		int v = v0;
-				int w = w0;
+		int w = w0;
+
+		__m64 mzero = _mm_setzero_si64();
+		__m64 m256 = _mm_set1_pi16(256);
+		__m64 malphamask = _mm_cvtsi32_si64(0xff);
 
 		while (w--)
 		{
 			int iu = u >> 17;
-			int iv = u >> 17;
+			int iv = v >> 17;
 
 			int fu = u & 0x7fff;
 			int fv = v & 0x7fff;
@@ -75,7 +143,6 @@ static void img_4o4mmx(FZ_PSRC, FZ_PDST, FZ_PCTM)
 			}
 
 			/* unpack src into 4x16bit vectors */
-			__m64 mzero = _mm_setzero_si64();
 			__m64 ms0 = _mm_unpackhi_pi8(ms0s1, mzero);
 			__m64 ms1 = _mm_unpacklo_pi8(ms0s1, mzero);
 			__m64 ms2 = _mm_unpackhi_pi8(ms2s3, mzero);
@@ -110,14 +177,12 @@ static void img_4o4mmx(FZ_PSRC, FZ_PDST, FZ_PCTM)
 			__m64 d1 = _mm_unpacklo_pi8(d0, mzero);
 
 			/* get src alpha */
-			__m64 m256 = _mm_set1_pi16(256);
-			__m64 malphamask = _mm_cvtsi32_si64(0xff);
 
-			/* splat alpha TODO: better way? */
+			/* splat alpha */
 			__m64 a0001 = _mm_and_si64(malphamask, t8);
-			__m64 a0010 = _mm_slli_si64(a0001, 16);
-			__m64 a0011 = _mm_or_si64(a0001, a0010);
+			__m64 a0011 = _mm_unpacklo_pi16(a0001, a0001);
 			__m64 a1111 = _mm_unpacklo_pi16(a0011, a0011);
+
 			/* 255+1 - sa */
 			__m64 sna = _mm_sub_pi16(m256, a1111);
 
@@ -152,9 +217,9 @@ fz_accelrastfuncs(fz_rastfuncs *tab)
 #  ifdef HAVE_MMX
 	if (fz_cpuflags & HAVE_MMX)
 	{
+		tab->duff_4i1o4 = duff_4i1o4mmx;
 		tab->img_4o4 = img_4o4mmx;
 	}
 #  endif
 }
 #endif
-
diff --git a/test/ximage.c b/test/ximage.c
index 1c16bfda..3b4a9b8b 100644
--- a/test/ximage.c
+++ b/test/ximage.c
@@ -246,7 +246,7 @@ select_mode(void)
 			info.mode = byteorder == MSBFirst ? RGBA8888 : ABGR8888;
 	}
 
-	printf("argb:8888 -> %s\n", modename[info.mode]);
+	printf("convert ARGB8888 to %s\n", modename[info.mode]);
 
 	/* select conversion function */
 	info.convert_func = ximage_convert_funcs[info.mode];
@@ -414,7 +414,6 @@ ximage_blit(Drawable d, GC gc,
 /*
  *
  */
-
 #ifndef _C99
 #ifdef __GNUC__
 #define restrict __restrict__
@@ -456,20 +455,16 @@ ximage_convert_bgra8888(PARAMS)
 	int x, y;
 	unsigned *s = (unsigned *)src;
 	unsigned *d = (unsigned *)dst;
+	unsigned val;
 	for (y = 0; y < h; y++) {
 		for (x = 0; x < w; x++) {
-			unsigned val = s[x];
-			unsigned a0g0 = val & 0xff00ff00;
-			unsigned gb00 = val << 16;
-			unsigned zzar = val >> 16;
-			unsigned gbar = gb00 | zzar;
-			d[x] = (gbar & 0x00ff00ff) | a0g0;
-/*
+			val = s[x];
 			d[x] =
 				(val >> 24) |
-				((val >> 8) & 0xff) |
-				((val << 8) & 0xff0000) |
-				(val << 24);
+				((val >> 8) & 0xff00) |
+				(val << 24) |
+				((val << 8) & 0xff0000);
+/*
 			d[x] =
 				(((val >> 24) & 0xff) <<  0) |
 				(((val >> 16) & 0xff) <<  8) |
@@ -487,7 +482,6 @@ ximage_convert_bgra8888(PARAMS)
 static void
 ximage_convert_abgr8888(PARAMS)
 {
-#if 1
 	int x, y;
 	unsigned *s = (unsigned *)src;
 	unsigned *d = (unsigned *)dst;
@@ -496,26 +490,16 @@ ximage_convert_abgr8888(PARAMS)
 	for (y = 0; y < h; y++) {
 		for (x = 0; x < w; x++) {
 			val = s[x];
-			/* bigendian... */
+#if 1 /* FZ_MSB */
 			d[x] = (val & 0xff00ff00) |
 				(((val << 16) | (val >> 16)) & 0x00ff00ff);
+#else /* FZ_LSB */
+			d[x] = (val << 24) | ((val >> 8) & 0xff);
+#endif
 		}
 		d += dststride>>2;
 		s += srcstride>>2;
 	}
-#else
-	int x, y;
-	for (y = 0; y < h; y++) {
-		for (x = 0; x < w; x++) {
-			dst[x * 4 + 0] = src[x * 4 + 0];
-			dst[x * 4 + 1] = src[x * 4 + 3];
-			dst[x * 4 + 2] = src[x * 4 + 2];
-			dst[x * 4 + 3] = src[x * 4 + 1];
-		}
-		dst += dststride;
-		src += srcstride;
-	}
-#endif
 }
 
 static void