AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

use integer math instead of float per-pixel. about a 2x speedup. not tested well, so this is just experimental.

Browse Source 
BUG=

Review URL: https://codereview.chromium.org/15715015

git-svn-id: http://skia.googlecode.com/svn/trunk@9317 2bbb7eff-a529-9590-31e7-b0007b416f81
This commit is contained in:
reed@google.com 2013-05-29 16:31:41 +00:00
parent a2be2cc798
commit b74706f033
1 changed files with 50 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

80
src/core/SkBitmapProcBicubic.cpp
View File

@ -15,25 +15,35 @@
#define DS(x) SkDoubleToScalar(x) #define DS(x) SkDoubleToScalar(x)
inline SkPMColor cubicBlend(const SkScalar c[16], SkScalar t, SkPMColor c0, SkPMColor c1, SkPMColor c2, SkPMColor c3) { #define MUL(a, b) ((a) * (b))
SkScalar t2 = t * t, t3 = t2 * t;
SkScalar cc[4]; static inline SkPMColor cubicBlend(const SkFixed cc[4], SkPMColor c0, SkPMColor c1, SkPMColor c2, SkPMColor c3) {
cc[0] = c[0] + SkScalarMul(c[1], t) + SkScalarMul(c[2], t2) + SkScalarMul(c[3], t3); SkFixed fa = MUL(cc[0], SkGetPackedA32(c0)) + MUL(cc[1], SkGetPackedA32(c1)) + MUL(cc[2], SkGetPackedA32(c2)) + MUL(cc[3], SkGetPackedA32(c3));
cc[1] = c[4] + SkScalarMul(c[5], t) + SkScalarMul(c[6], t2) + SkScalarMul(c[7], t3); SkFixed fr = MUL(cc[0], SkGetPackedR32(c0)) + MUL(cc[1], SkGetPackedR32(c1)) + MUL(cc[2], SkGetPackedR32(c2)) + MUL(cc[3], SkGetPackedR32(c3));
cc[2] = c[8] + SkScalarMul(c[9], t) + SkScalarMul(c[10], t2) + SkScalarMul(c[11], t3); SkFixed fg = MUL(cc[0], SkGetPackedG32(c0)) + MUL(cc[1], SkGetPackedG32(c1)) + MUL(cc[2], SkGetPackedG32(c2)) + MUL(cc[3], SkGetPackedG32(c3));
cc[3] = c[12] + SkScalarMul(c[13], t) + SkScalarMul(c[14], t2) + SkScalarMul(c[15], t3); SkFixed fb = MUL(cc[0], SkGetPackedB32(c0)) + MUL(cc[1], SkGetPackedB32(c1)) + MUL(cc[2], SkGetPackedB32(c2)) + MUL(cc[3], SkGetPackedB32(c3));
SkScalar a = SkScalarClampMax(SkScalarMul(cc[0], SkGetPackedA32(c0)) + SkScalarMul(cc[1], SkGetPackedA32(c1)) + SkScalarMul(cc[2], SkGetPackedA32(c2)) + SkScalarMul(cc[3], SkGetPackedA32(c3)), 255);
SkScalar r = SkScalarMul(cc[0], SkGetPackedR32(c0)) + SkScalarMul(cc[1], SkGetPackedR32(c1)) + SkScalarMul(cc[2], SkGetPackedR32(c2)) + SkScalarMul(cc[3], SkGetPackedR32(c3)); int a = SkClampMax(SkFixedRoundToInt(fa), 255);
SkScalar g = SkScalarMul(cc[0], SkGetPackedG32(c0)) + SkScalarMul(cc[1], SkGetPackedG32(c1)) + SkScalarMul(cc[2], SkGetPackedG32(c2)) + SkScalarMul(cc[3], SkGetPackedG32(c3)); int r = SkClampMax(SkFixedRoundToInt(fr), a);
SkScalar b = SkScalarMul(cc[0], SkGetPackedB32(c0)) + SkScalarMul(cc[1], SkGetPackedB32(c1)) + SkScalarMul(cc[2], SkGetPackedB32(c2)) + SkScalarMul(cc[3], SkGetPackedB32(c3)); int g = SkClampMax(SkFixedRoundToInt(fg), a);
return SkPackARGB32(SkScalarRoundToInt(a), int b = SkClampMax(SkFixedRoundToInt(fb), a);
SkScalarRoundToInt(SkScalarClampMax(r, a)),
SkScalarRoundToInt(SkScalarClampMax(g, a)), return SkPackARGB32(a, r, g, b);
SkScalarRoundToInt(SkScalarClampMax(b, a))); }
static float poly_eval(const float cc[4], float t) {
return cc[0] + t * (cc[1] + t * (cc[2] + t * cc[3]));
}
static void build_coeff4(SkFixed dst[4], const float src[16], float t) {
dst[0] = SkFloatToFixed(poly_eval(&src[ 0], t));
dst[1] = SkFloatToFixed(poly_eval(&src[ 4], t));
dst[2] = SkFloatToFixed(poly_eval(&src[ 8], t));
dst[3] = SkFloatToFixed(poly_eval(&src[12], t));
} }
static void bicubicFilter(const SkBitmapProcState& s, int x, int y, static void bicubicFilter(const SkBitmapProcState& s, int x, int y,
SkPMColor colors[], int count) { SkPMColor* SK_RESTRICT colors, int count) {
static const SkScalar coefficients[16] = { static const SkScalar coefficients[16] = {
DS( 1.0 / 18.0), DS(-9.0 / 18.0), DS( 15.0 / 18.0), DS( -7.0 / 18.0), DS( 1.0 / 18.0), DS(-9.0 / 18.0), DS( 15.0 / 18.0), DS( -7.0 / 18.0),
@ -42,7 +52,9 @@ static void bicubicFilter(const SkBitmapProcState& s, int x, int y,
DS( 0.0 / 18.0), DS( 0.0 / 18.0), DS( -6.0 / 18.0), DS( 7.0 / 18.0), DS( 0.0 / 18.0), DS( 0.0 / 18.0), DS( -6.0 / 18.0), DS( 7.0 / 18.0),
}; };
SkPMColor *dptr = &(colors[0]); const int maxX = s.fBitmap->width() - 1;
const int maxY = s.fBitmap->height() - 1;
while (count-- > 0) { while (count-- > 0) {
SkPoint srcPt; SkPoint srcPt;
s.fInvProc(*s.fInvMatrix, SkIntToScalar(x), s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
@ -51,37 +63,45 @@ static void bicubicFilter(const SkBitmapProcState& s, int x, int y,
srcPt.fY -= SK_ScalarHalf; srcPt.fY -= SK_ScalarHalf;
SkScalar fractx = srcPt.fX - SkScalarFloorToScalar(srcPt.fX); SkScalar fractx = srcPt.fX - SkScalarFloorToScalar(srcPt.fX);
SkScalar fracty = srcPt.fY - SkScalarFloorToScalar(srcPt.fY); SkScalar fracty = srcPt.fY - SkScalarFloorToScalar(srcPt.fY);
SkFixed coeffX[4], coeffY[4];
build_coeff4(coeffX, coefficients, fractx);
build_coeff4(coeffY, coefficients, fracty);
int sx = SkScalarFloorToInt(srcPt.fX); int sx = SkScalarFloorToInt(srcPt.fX);
int sy = SkScalarFloorToInt(srcPt.fY); int sy = SkScalarFloorToInt(srcPt.fY);
int x0 = SkClampMax(sx - 1, s.fBitmap->width() - 1);
int x1 = SkClampMax(sx , s.fBitmap->width() - 1); // Here is where we can support other tile modes (e.g. repeat or mirror)
int x2 = SkClampMax(sx + 1, s.fBitmap->width() - 1); int x0 = SkClampMax(sx - 1, maxX);
int x3 = SkClampMax(sx + 2, s.fBitmap->width() - 1); int x1 = SkClampMax(sx , maxX);
int y0 = SkClampMax(sy - 1, s.fBitmap->height() - 1); int x2 = SkClampMax(sx + 1, maxX);
int y1 = SkClampMax(sy , s.fBitmap->height() - 1); int x3 = SkClampMax(sx + 2, maxX);
int y2 = SkClampMax(sy + 1, s.fBitmap->height() - 1); int y0 = SkClampMax(sy - 1, maxY);
int y3 = SkClampMax(sy + 2, s.fBitmap->height() - 1); int y1 = SkClampMax(sy , maxY);
int y2 = SkClampMax(sy + 1, maxY);
int y3 = SkClampMax(sy + 2, maxY);
SkPMColor s00 = *s.fBitmap->getAddr32(x0, y0); SkPMColor s00 = *s.fBitmap->getAddr32(x0, y0);
SkPMColor s10 = *s.fBitmap->getAddr32(x1, y0); SkPMColor s10 = *s.fBitmap->getAddr32(x1, y0);
SkPMColor s20 = *s.fBitmap->getAddr32(x2, y0); SkPMColor s20 = *s.fBitmap->getAddr32(x2, y0);
SkPMColor s30 = *s.fBitmap->getAddr32(x3, y0); SkPMColor s30 = *s.fBitmap->getAddr32(x3, y0);
SkPMColor s0 = cubicBlend(coefficients, fractx, s00, s10, s20, s30); SkPMColor s0 = cubicBlend(coeffX, s00, s10, s20, s30);
SkPMColor s01 = *s.fBitmap->getAddr32(x0, y1); SkPMColor s01 = *s.fBitmap->getAddr32(x0, y1);
SkPMColor s11 = *s.fBitmap->getAddr32(x1, y1); SkPMColor s11 = *s.fBitmap->getAddr32(x1, y1);
SkPMColor s21 = *s.fBitmap->getAddr32(x2, y1); SkPMColor s21 = *s.fBitmap->getAddr32(x2, y1);
SkPMColor s31 = *s.fBitmap->getAddr32(x3, y1); SkPMColor s31 = *s.fBitmap->getAddr32(x3, y1);
SkPMColor s1 = cubicBlend(coefficients, fractx, s01, s11, s21, s31); SkPMColor s1 = cubicBlend(coeffX, s01, s11, s21, s31);
SkPMColor s02 = *s.fBitmap->getAddr32(x0, y2); SkPMColor s02 = *s.fBitmap->getAddr32(x0, y2);
SkPMColor s12 = *s.fBitmap->getAddr32(x1, y2); SkPMColor s12 = *s.fBitmap->getAddr32(x1, y2);
SkPMColor s22 = *s.fBitmap->getAddr32(x2, y2); SkPMColor s22 = *s.fBitmap->getAddr32(x2, y2);
SkPMColor s32 = *s.fBitmap->getAddr32(x3, y2); SkPMColor s32 = *s.fBitmap->getAddr32(x3, y2);
SkPMColor s2 = cubicBlend(coefficients, fractx, s02, s12, s22, s32); SkPMColor s2 = cubicBlend(coeffX, s02, s12, s22, s32);
SkPMColor s03 = *s.fBitmap->getAddr32(x0, y3); SkPMColor s03 = *s.fBitmap->getAddr32(x0, y3);
SkPMColor s13 = *s.fBitmap->getAddr32(x1, y3); SkPMColor s13 = *s.fBitmap->getAddr32(x1, y3);
SkPMColor s23 = *s.fBitmap->getAddr32(x2, y3); SkPMColor s23 = *s.fBitmap->getAddr32(x2, y3);
SkPMColor s33 = *s.fBitmap->getAddr32(x3, y3); SkPMColor s33 = *s.fBitmap->getAddr32(x3, y3);
SkPMColor s3 = cubicBlend(coefficients, fractx, s03, s13, s23, s33); SkPMColor s3 = cubicBlend(coeffX, s03, s13, s23, s33);
*dptr++ = cubicBlend(coefficients, fracty, s0, s1, s2, s3); *colors++ = cubicBlend(coeffY, s0, s1, s2, s3);
x++; x++;
} }
} }