From 507e551ebdfb1642ddf507ecaf6399eaf1bf9fd3 Mon Sep 17 00:00:00 2001
From: "H. Peter Anvin" <hpa@zytor.com>
Date: Fri, 15 Sep 2006 23:00:58 -0700
Subject: [PATCH] Actually do proper gamma-corrected alpha blending.  It seems
 fast enough.

---
 com32/lib/Makefile             |  6 ++++-
 com32/lib/sys/vesa/alphatbl.pl | 54 ++++++++++++++++++++++++++++++++++++++++++
 com32/lib/sys/vesa/drawtxt.c   | 16 +++++--------
 com32/lib/sys/vesa/video.h     |  3 ++-
 4 files changed, 67 insertions(+), 12 deletions(-)
 create mode 100644 com32/lib/sys/vesa/alphatbl.pl

diff --git a/com32/lib/Makefile b/com32/lib/Makefile
index d20bc3b..e3cef23 100644
--- a/com32/lib/Makefile
+++ b/com32/lib/Makefile
@@ -42,6 +42,7 @@ LIBOBJS = \
 	\
 	sys/vesacon_write.o sys/vesaserial_write.o 			\
 	sys/vesa/initvesa.o sys/vesa/drawtxt.o	sys/vesa/background.o	\
+	sys/vesa/alphatbl.o						\
 	\
 	pci/cfgtype.o 							\
 	pci/readb.o pci/readw.o pci/readl.o pci/readbios.o 		\
@@ -77,7 +78,7 @@ libcom32.a : $(LIBOBJS)
 	$(RANLIB) $@
 
 tidy:
-	rm -f *.o .*.d */*.o */.*.d lib/sys/vesa/alphatbl.c
+	rm -f *.o .*.d */*.o */.*.d sys/vesa/alphatbl.c
 
 clean: tidy
 	rm -f *.a
@@ -95,6 +96,9 @@ install: all
 sys/vesa/drawtxt.o: sys/vesa/drawtxt.c
 	$(CC) $(CFLAGS) -O3 -c -o $@ $<
 
+sys/vesa/alphatbl.c: sys/vesa/alphatbl.pl
+	$(PERL) $< > $@
+
 jpeg/jidctflt.o: jpeg/jidctflt.c
 	$(CC) $(CFLAGS) -O3 -c -o $@ $<
 
diff --git a/com32/lib/sys/vesa/alphatbl.pl b/com32/lib/sys/vesa/alphatbl.pl
new file mode 100644
index 0000000..dd71648
--- /dev/null
+++ b/com32/lib/sys/vesa/alphatbl.pl
@@ -0,0 +1,54 @@
+#!/usr/bin/perl
+#
+# Produce gamma-correction tables for alpha blending, assuming sRGB space.
+#
+
+sub srgb_to_linear($)
+{
+    my($s) = @_;
+
+    if ($s <= 10) {
+	return $s/(255*12.92);
+    } else {
+	return (($s+14.025)/269.025)**2.4;
+    }
+}
+
+sub linear_to_srgb($)
+{
+    my($l) = @_;
+    my $s;
+
+    if ($l <= 0.00304) {
+	$s = 12.92*$l;
+    } else {
+	$s = 1.055*$l**(1.0/2.4) - 0.055;
+    }
+
+    return int($s*255+0.5);
+}
+
+# Header
+print "#include <inttypes.h>\n\n";
+
+#
+# Table 1: convert 8-bit sRGB values to 16-bit linear values
+#
+
+print "const uint16_t __vesacon_srgb_to_linear[256] = {\n";
+for ($i = 0; $i <= 255; $i++) {
+    printf "\t%5d,\n", int(srgb_to_linear($i)*65535+0.5);
+}
+print "};\n\n";
+
+#
+# Table 2: convert linear values in the range [0, 65535*255],
+#          shifted right by 12 bits, to sRGB
+#
+
+print "const uint8_t __vesacon_linear_to_srgb[4080] = {\n";
+for ($i = 0; $i <= 4079; $i++) {
+    printf "\t%3d,\n", linear_to_srgb(($i+0.5)/4079.937744);
+}
+print "};\n\n";
+
diff --git a/com32/lib/sys/vesa/drawtxt.c b/com32/lib/sys/vesa/drawtxt.c
index 794ebe0..f35e1b7 100644
--- a/com32/lib/sys/vesa/drawtxt.c
+++ b/com32/lib/sys/vesa/drawtxt.c
@@ -45,19 +45,15 @@ static inline void *copy_dword(void *dst, void *src, size_t dword_count)
   return dst;			/* Updated destination pointer */
 }
 
-/*
- * Linear alpha blending.  Useless for anything that's actually
- * depends on color accuracy (because of gamma), but it's fine for
- * what we want.
- *
- * This algorithm is exactly equivalent to (alpha*fg+(255-alpha)*bg)/255
- * for all 8-bit values, but is substantially faster.
- */
 static inline __attribute__((always_inline))
   uint8_t alpha_val(uint8_t fg, uint8_t bg, uint8_t alpha)
 {
-  unsigned int tmp = 1 + alpha*fg + (255-alpha)*bg;
-  return (tmp + (tmp >> 8)) >> 8;
+  unsigned int tmp;
+
+  tmp  = __vesacon_srgb_to_linear[fg] * alpha;
+  tmp += __vesacon_srgb_to_linear[bg] * (255-alpha);
+  
+  return __vesacon_linear_to_srgb[tmp >> 12];
 }
 
 static uint32_t alpha_pixel(uint32_t fg, uint32_t bg)
diff --git a/com32/lib/sys/vesa/video.h b/com32/lib/sys/vesa/video.h
index 13449d8..5070f9e 100644
--- a/com32/lib/sys/vesa/video.h
+++ b/com32/lib/sys/vesa/video.h
@@ -69,7 +69,8 @@ extern uint8_t __vesacon_graphics_font[FONT_MAX_CHARS][FONT_MAX_HEIGHT];
 extern uint32_t __vesacon_background[VIDEO_Y_SIZE][VIDEO_X_SIZE];
 extern uint32_t __vesacon_shadowfb[VIDEO_Y_SIZE][VIDEO_X_SIZE];
 
-extern unsigned char __vesacon_alpha_tbl[256][4];
+extern const uint16_t __vesacon_srgb_to_linear[256];
+extern const uint8_t  __vesacon_linear_to_srgb[4080];
 
 int __vesacon_init_background(void);
 int vesacon_load_background(const char *);
-- 
2.7.4