X-Git-Url: https://git.dogcows.com/gitweb?a=blobdiff_plain;f=render%2Fgradient.c;h=8b6850f0e5a41adbea92d01f8f34007946069eed;hb=c564b6245fcfaf0efd523cfa17bc1a73fe640d68;hp=d1d90e418c2a0d2ff3d753922435ff9ec8198535;hpb=3611c8210cc632c2a21c67ccbf40857342c10371;p=chaz%2Fopenbox diff --git a/render/gradient.c b/render/gradient.c index d1d90e41..8b6850f0 100644 --- a/render/gradient.c +++ b/render/gradient.c @@ -41,7 +41,7 @@ void RrRender(RrAppearance *a, gint w, gint h) RrPixel32 *data = a->surface.pixel_data; RrPixel32 current; guint r,g,b; - gint off, x; + register gint off, x; switch (a->surface.grad) { case RR_SURFACE_PARENTREL: @@ -136,7 +136,7 @@ void RrRender(RrAppearance *a, gint w, gint h) static void highlight(RrSurface *s, RrPixel32 *x, RrPixel32 *y, gboolean raised) { - gint r, g, b; + register gint r, g, b; RrPixel32 *up, *down; if (raised) { @@ -171,7 +171,7 @@ static void highlight(RrSurface *s, RrPixel32 *x, RrPixel32 *y, gboolean raised) static void create_bevel_colors(RrAppearance *l) { - gint r, g, b; + register gint r, g, b; /* light color */ r = l->surface.primary->r; @@ -205,7 +205,7 @@ static void create_bevel_colors(RrAppearance *l) */ static inline void repeat_pixel(RrPixel32 *start, gint w) { - gint x; + register gint x; RrPixel32 *dest; dest = start + 1; @@ -218,8 +218,8 @@ static inline void repeat_pixel(RrPixel32 *start, gint w) /* for >= 8, then use O(log n) memcpy's... */ else { - gint len = 4; - gint lenbytes = 4 * sizeof(RrPixel32); + gchar *cdest; + gint lenbytes; /* copy the first 3 * 32 bits (3 words) ourselves - then we have 3 + the original 1 = 4 words to make copies of at a time @@ -229,16 +229,38 @@ static inline void repeat_pixel(RrPixel32 *start, gint w) for (x = 3; x > 0; --x) *(dest++) = *start; - for (x = w - 4; x > 0;) { - memcpy(dest, start, lenbytes); - x -= len; - dest += len; - len <<= 1; + /* cdest is a pointer to the pixel data that is typed char* so that + adding 1 to its position moves it only one byte + + lenbytes is the amount of bytes that we will be copying each + iteration. this doubles each time through the loop. + + x is the number of bytes left to copy into. lenbytes will alwaysa + be bounded by x + + this loop will run O(log n) times (n is the number of bytes we + need to copy into), since the size of the copy is doubled each + iteration. it seems that gcc does some nice optimizations to make + this memcpy very fast on hardware with support for vector operations + such as mmx or see. here is an idea of the kind of speed up we are + getting by doing this (splitvertical3 switches from doing + "*(data++) = color" n times to doing this memcpy thing log n times: + + % cumulative self self total + time seconds seconds calls ms/call ms/call name + 49.44 0.88 0.88 1063 0.83 0.83 splitvertical1 + 47.19 1.72 0.84 1063 0.79 0.79 splitvertical2 + 2.81 1.77 0.05 1063 0.05 0.05 splitvertical3 + */ + cdest = (gchar*)dest; + lenbytes = 4 * sizeof(RrPixel32); + for (x = (w - 4) * sizeof(RrPixel32); x > 0;) { + memcpy(cdest, start, lenbytes); + x -= lenbytes; + cdest += lenbytes; lenbytes <<= 1; - if (len > x) { - len = x; - lenbytes = x * sizeof(RrPixel32); - } + if (lenbytes > x) + lenbytes = x; } } } @@ -246,7 +268,8 @@ static inline void repeat_pixel(RrPixel32 *start, gint w) static void gradient_parentrelative(RrAppearance *a, gint w, gint h) { RrPixel32 *source, *dest; - gint sw, sh, partial_w, partial_h, i; + gint sw, sh, partial_w, partial_h; + register gint i; g_assert (a->surface.parent); g_assert (a->surface.parent->w); @@ -296,7 +319,7 @@ static void gradient_parentrelative(RrAppearance *a, gint w, gint h) static void gradient_solid(RrAppearance *l, gint w, gint h) { - gint i; + register gint i; RrPixel32 pix; RrPixel32 *data = l->surface.pixel_data; RrSurface *sp = &l->surface; @@ -392,9 +415,10 @@ static void gradient_solid(RrAppearance *l, gint w, gint h) /* * * * * * * * * * * * * * GRADIENT MAGIC WOOT * * * * * * * * * * * * * * */ -#define VARS(x) \ - guint color##x[3]; \ - gint len##x, cdelta##x[3], error##x[3] = { 0, 0, 0 }, inc##x[3]; \ +#define VARS(x) \ + register gint len##x; \ + guint color##x[3]; \ + gint cdelta##x[3], error##x[3] = { 0, 0, 0 }, inc##x[3]; \ gboolean bigslope##x[3] /* color slope > 1 */ #define SETUP(x, from, to, w) \ @@ -442,7 +466,7 @@ static void gradient_solid(RrAppearance *l, gint w, gint h) #define NEXT(x) \ { \ - gint i; \ + register gint i; \ for (i = 2; i >= 0; --i) { \ if (!cdelta##x[i]) continue; \ \ @@ -469,10 +493,10 @@ static void gradient_solid(RrAppearance *l, gint w, gint h) static void gradient_splitvertical(RrAppearance *a, gint w, gint h) { - gint y1, y2, y3; + register gint y1, y2, y3; RrSurface *sf = &a->surface; RrPixel32 *data; - gint y1sz, y2sz, y3sz; + register gint y1sz, y2sz, y3sz; VARS(y1); VARS(y2); @@ -534,12 +558,14 @@ static void gradient_splitvertical(RrAppearance *a, gint w, gint h) static void gradient_horizontal(RrSurface *sf, gint w, gint h) { - gint x, y; + register gint x, y, cpbytes; RrPixel32 *data = sf->pixel_data, *datav; + gchar *datac; VARS(x); SETUP(x, sf->primary, sf->secondary, w); + /* set the color values for the first row */ datav = data; for (x = w - 1; x > 0; --x) { /* 0 -> w - 1 */ *datav = COLOR(x); @@ -549,22 +575,32 @@ static void gradient_horizontal(RrSurface *sf, gint w, gint h) *datav = COLOR(x); ++datav; - for (y = h - 1; y > 0; --y) { /* 1 -> h */ - memcpy(datav, data, w * sizeof(RrPixel32)); - datav += w; + /* copy the first row to the rest in O(logn) copies */ + datac = (gchar*)datav; + cpbytes = 1 * w * sizeof(RrPixel32); + for (y = (h - 1) * w * sizeof(RrPixel32); y > 0;) { + memcpy(datac, data, cpbytes); + y -= cpbytes; + datac += cpbytes; + cpbytes <<= 1; + if (cpbytes > y) + cpbytes = y; } } static void gradient_mirrorhorizontal(RrSurface *sf, gint w, gint h) { - gint x, y, half1, half2; + register gint x, y, half1, half2, cpbytes; RrPixel32 *data = sf->pixel_data, *datav; + gchar *datac; VARS(x); half1 = (w + 1) / 2; half2 = w / 2; + /* set the color values for the first row */ + SETUP(x, sf->primary, sf->secondary, half1); datav = data; for (x = half1 - 1; x > 0; --x) { /* 0 -> half1 - 1 */ @@ -586,15 +622,22 @@ static void gradient_mirrorhorizontal(RrSurface *sf, gint w, gint h) ++datav; } - for (y = h - 1; y > 0; --y) { /* 1 -> h */ - memcpy(datav, data, w * sizeof(RrPixel32)); - datav += w; + /* copy the first row to the rest in O(logn) copies */ + datac = (gchar*)datav; + cpbytes = 1 * w * sizeof(RrPixel32); + for (y = (h - 1) * w * sizeof(RrPixel32); y > 0;) { + memcpy(datac, data, cpbytes); + y -= cpbytes; + datac += cpbytes; + cpbytes <<= 1; + if (cpbytes > y) + cpbytes = y; } } static void gradient_vertical(RrSurface *sf, gint w, gint h) { - gint y; + register gint y; RrPixel32 *data; VARS(y); @@ -618,10 +661,9 @@ static void gradient_vertical(RrSurface *sf, gint w, gint h) } } - static void gradient_diagonal(RrSurface *sf, gint w, gint h) { - gint x, y; + register gint x, y; RrPixel32 *data = sf->pixel_data; RrColor left, right; RrColor extracorner; @@ -668,7 +710,7 @@ static void gradient_diagonal(RrSurface *sf, gint w, gint h) static void gradient_crossdiagonal(RrSurface *sf, gint w, gint h) { - gint x, y; + register gint x, y; RrPixel32 *data = sf->pixel_data; RrColor left, right; RrColor extracorner; @@ -713,14 +755,13 @@ static void gradient_crossdiagonal(RrSurface *sf, gint w, gint h) *data = COLOR(x); } -static void gradient_pyramid(RrSurface *sf, gint inw, gint inh) +static void gradient_pyramid(RrSurface *sf, gint w, gint h) { - gint x, y, w = (inw >> 1) + 1, h = (inh >> 1) + 1; - RrPixel32 *data = sf->pixel_data; - RrPixel32 *end = data + inw*inh - 1; - RrPixel32 current; + RrPixel32 *ldata, *rdata; + RrPixel32 *cp; RrColor left, right; RrColor extracorner; + register gint x, y, halfw, halfh, midx, midy; VARS(lefty); VARS(righty); @@ -730,54 +771,64 @@ static void gradient_pyramid(RrSurface *sf, gint inw, gint inh) extracorner.g = (sf->primary->g + sf->secondary->g) / 2; extracorner.b = (sf->primary->b + sf->secondary->b) / 2; - SETUP(lefty, (&extracorner), sf->secondary, h); - SETUP(righty, sf->primary, (&extracorner), h); + halfw = w >> 1; + halfh = h >> 1; + midx = w - halfw - halfw; /* 0 or 1, depending if w is even or odd */ + midy = h - halfh - halfh; /* 0 or 1, depending if h is even or odd */ + + SETUP(lefty, sf->primary, (&extracorner), halfh + midy); + SETUP(righty, (&extracorner), sf->secondary, halfh + midy); + + /* draw the top half + + it is faster to draw both top quarters together than to draw one and + then copy it over to the other side. + */ + + ldata = sf->pixel_data; + rdata = ldata + w - 1; + for (y = halfh + midy; y > 0; --y) { /* 0 -> (h+1)/2 */ + RrPixel32 c; - for (y = h - 1; y > 0; --y) { /* 0 -> h-1 */ COLOR_RR(lefty, (&left)); COLOR_RR(righty, (&right)); - SETUP(x, (&left), (&right), w); + SETUP(x, (&left), (&right), halfw + midx); - for (x = w - 1; x > 0; --x) { /* 0 -> w-1 */ - current = COLOR(x); - *(data+x) = current; - *(data+inw-x) = current; - *(end-x) = current; - *(end-(inw-x)) = current; + for (x = halfw + midx - 1; x > 0; --x) { /* 0 -> (w+1)/2 */ + c = COLOR(x); + *(ldata++) = *(rdata--) = c; NEXT(x); } - current = COLOR(x); - *(data+x) = current; - *(data+inw-x) = current; - *(end-x) = current; - *(end-(inw-x)) = current; - - data+=inw; - end-=inw; + c = COLOR(x); + *ldata = *rdata = c; + ldata += halfw + 1; + rdata += halfw - 1 + midx + w; NEXT(lefty); NEXT(righty); } - COLOR_RR(lefty, (&left)); - COLOR_RR(righty, (&right)); - SETUP(x, (&left), (&right), w); + /* copy the top half into the bottom half, mirroring it, so we can only + copy one row at a time - for (x = w - 1; x > 0; --x) { /* 0 -> w-1 */ - current = COLOR(x); - *(data+x) = current; - *(data+inw-x) = current; - *(end-x) = current; - *(end-(inw-x)) = current; + it is faster, to move the writing pointer forward, and the reading + pointer backward - NEXT(x); + this is the current code, moving the write pointer forward and read + pointer backward + 41.78 4.26 1.78 504 3.53 3.53 gradient_pyramid2 + this is the opposite, moving the read pointer forward and the write + pointer backward + 42.27 4.40 1.86 504 3.69 3.69 gradient_pyramid2 + + */ + ldata = sf->pixel_data + (halfh - 1) * w; + cp = ldata + (midy + 1) * w; + for (y = halfh; y > 0; --y) { + memcpy(cp, ldata, w * sizeof(RrPixel32)); + ldata -= w; + cp += w; } - current = COLOR(x); - *(data+x) = current; - *(data+inw-x) = current; - *(end-x) = current; - *(end-(inw-x)) = current; } -