Improved font caching I made a patch that improves the performance of font caching mechanism. This is based on a funny behaviour of FontConfig: it was handling FcCharSet in a somewhat unexpected way. So, we are currently adding "a character" to a new FcCharSet, and then add it to a FcPattern. However, if we toss the FcPattern to FontConfig, it loads the entire language(charset) that contains the character we gave. That is, we don't always have to load a new font for each unknown character. Instead, we can reused cached fonts, and this significantly reduces the number of calls to extremely slow FontConfig matching functions. One more thing. I found that, in libXft, there's a function called XftCharExists. XftCharIndex internally calls this function, and does more stuffs if the character does exist. Since the returned index is never used in st, we should call XftCharExists instead of XftCharIndex. Please note that I already made this change in the patch.
Eon S. Jeon esjeon@lavabit.com
1 files changed,
22 insertions(+),
41 deletions(-)
jump to
M
st.c
→
st.c
@@ -462,17 +462,12 @@ };
typedef struct { XftFont *font; - long c; int flags; } Fontcache; -/* - * Fontcache is a ring buffer, with frccur as current position and frclen as - * the current length of used elements. - */ - -static Fontcache frc[1024]; -static int frccur = -1, frclen = 0; +/* Fontcache is an array now. A new font will be appended to the array. */ +static Fontcache frc[16]; +static int frclen = 0; ssize_t xwrite(int fd, char *s, size_t len) {@@ -2781,18 +2776,12 @@ }
void xunloadfonts(void) { - int i, ip; + int i; - /* - * Free the loaded fonts in the font cache. This is done backwards - * from the frccur. - */ - for(i = 0, ip = frccur; i < frclen; i++, ip--) { - if(ip < 0) - ip = LEN(frc) - 1; - XftFontClose(xw.dpy, frc[ip].font); + /* Free the loaded fonts in the font cache. */ + for(i = 0; i < frclen; i++) { + XftFontClose(xw.dpy, frc[i].font); } - frccur = -1; frclen = 0; xunloadfont(&dc.font);@@ -2918,7 +2907,7 @@ void
xdraws(char *s, Glyph base, int x, int y, int charlen, int bytelen) { int winx = borderpx + x * xw.cw, winy = borderpx + y * xw.ch, width = charlen * xw.cw, xp, i; - int frp, frcflags; + int frcflags; int u8fl, u8fblen, u8cblen, doesexist; char *u8c, *u8fs; long u8char;@@ -3044,7 +3033,7 @@ u8cblen = utf8decode(s, &u8char);
s += u8cblen; bytelen -= u8cblen; - doesexist = XftCharIndex(xw.dpy, font->match, u8char); + doesexist = XftCharExists(xw.dpy, font->match, u8char); if(!doesexist || bytelen <= 0) { if(bytelen <= 0) { if(doesexist) {@@ -3071,14 +3060,10 @@ }
if(doesexist) break; - frp = frccur; /* Search the font cache. */ - for(i = 0; i < frclen; i++, frp--) { - if(frp <= 0) - frp = LEN(frc) - 1; - - if(frc[frp].c == u8char - && frc[frp].flags == frcflags) { + for(i = 0; i < frclen; i++) { + if(XftCharExists(xw.dpy, frc[i].font, u8char) + && frc[i].flags == frcflags) { break; } }@@ -3113,28 +3098,24 @@
/* * Overwrite or create the new cache entry. */ - frccur++; - frclen++; - if(frccur >= LEN(frc)) - frccur = 0; - if(frclen > LEN(frc)) { - frclen = LEN(frc); - XftFontClose(xw.dpy, frc[frccur].font); + if(frclen >= LEN(frc)) { + frclen = LEN(frc) - 1; + XftFontClose(xw.dpy, frc[frclen].font); } - frc[frccur].font = XftFontOpenPattern(xw.dpy, + frc[frclen].font = XftFontOpenPattern(xw.dpy, fontpattern); - frc[frccur].c = u8char; - frc[frccur].flags = frcflags; + frc[frclen].flags = frcflags; + + i = frclen; + frclen++; FcPatternDestroy(fcpattern); FcCharSetDestroy(fccharset); - - frp = frccur; } - XftDrawStringUtf8(xw.draw, fg, frc[frp].font, - xp, winy + frc[frp].font->ascent, + XftDrawStringUtf8(xw.draw, fg, frc[i].font, + xp, winy + frc[i].font->ascent, (FcChar8 *)u8c, u8cblen); xp += font->width;