python-pillow/Pillow
1
1
Fork 0
mirror of https://github.com/python-pillow/Pillow.git synced 2025-03-03 19:45:56 +03:00
Code Issues Packages Projects Releases Wiki Activity

speedup. divisions replaced by shifts.

Browse Source
This commit is contained in:
homm 2013-03-23 16:51:55 +04:00
parent e88490b4f2
commit 0663e14444
1 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
libImaging/AlphaComposite.c
View File

@ -64,14 +64,25 @@ ImagingAlphaComposite(Imaging imDst, Imaging imSrc)
// Each variable has extra meaningful bits. // Each variable has extra meaningful bits.
// Divisions are rounded. // Divisions are rounded.
UINT16 blend = dst->a * (255 - src->a); // 16 bit max // This code uses trick from Paste.c:
UINT16 outa = (src->a << 4) + ((blend << 4) + 127) / 255; // 12 // (a + (2 << (n-1)) - 1) / ((2 << n)-1)
UINT16 coef1 = ((src->a * 255) << 8) / outa; // 12 // almost equivalent to:
// tmp = a + (2 << (n-1)), ((tmp >> n) + tmp) >> n
// 0xff * 0xff = 16 meaningful bits.
UINT16 blend = dst->a * (255 - src->a);
// Shift 4 bits up, to don't loose blend precision
// on very transparent pixels.
UINT16 outa = (src->a << 4) + (((blend << 4) + (blend >> 4) + 0x80) >> 8);
UINT16 coef1 = (((src->a << 8) - src->a) << 8) / outa; // 12
UINT16 coef2 = (blend << 8) / outa; // 12 UINT16 coef2 = (blend << 8) / outa; // 12
out->r = ((src->r * coef1 + dst->r * coef2 + 0x7ff) / 255) >> 4; UINT32 tmpr = src->r * coef1 + dst->r * coef2 + 0x800;
out->g = ((src->g * coef1 + dst->g * coef2 + 0x7ff) / 255) >> 4; out->r = ((tmpr >> 8) + tmpr) >> 12;
out->b = ((src->b * coef1 + dst->b * coef2 + 0x7ff) / 255) >> 4; UINT32 tmpg = src->g * coef1 + dst->g * coef2 + 0x800;
out->g = ((tmpg >> 8) + tmpg) >> 12;
UINT32 tmpb = src->b * coef1 + dst->b * coef2 + 0x800;
out->b = ((tmpb >> 8) + tmpb) >> 12;
out->a = (outa + 0x7) >> 4; out->a = (outa + 0x7) >> 4;
} }