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

BUG=

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

git-svn-id: http://skia.googlecode.com/svn/trunk@9317 2bbb7eff-a529-9590-31e7-b0007b416f81

1 files changed, 50 insertions, 30 deletions

diff --git a/src/core/SkBitmapProcBicubic.cpp b/src/core/SkBitmapProcBicubic.cpp
index de2b0ff9a4..c00fd0afa2 100644
--- a/src/core/SkBitmapProcBicubic.cpp
+++ b/src/core/SkBitmapProcBicubic.cpp
@@ -15,25 +15,35 @@
 
 #define DS(x) SkDoubleToScalar(x)
 
-inline SkPMColor cubicBlend(const SkScalar c[16], SkScalar t, SkPMColor c0, SkPMColor c1, SkPMColor c2, SkPMColor c3) {
-    SkScalar t2 = t * t, t3 = t2 * t;
-    SkScalar cc[4];
-    cc[0] = c[0]  + SkScalarMul(c[1], t) + SkScalarMul(c[2], t2) + SkScalarMul(c[3], t3);
-    cc[1] = c[4]  + SkScalarMul(c[5], t) + SkScalarMul(c[6], t2) + SkScalarMul(c[7], t3);
-    cc[2] = c[8]  + SkScalarMul(c[9], t) + SkScalarMul(c[10], t2) + SkScalarMul(c[11], t3);
-    cc[3] = c[12] + SkScalarMul(c[13], t) + SkScalarMul(c[14], t2) + SkScalarMul(c[15], t3);
-    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));
-    SkScalar g = SkScalarMul(cc[0], SkGetPackedG32(c0)) + SkScalarMul(cc[1], SkGetPackedG32(c1)) + SkScalarMul(cc[2], SkGetPackedG32(c2)) + SkScalarMul(cc[3], SkGetPackedG32(c3));
-    SkScalar b = SkScalarMul(cc[0], SkGetPackedB32(c0)) + SkScalarMul(cc[1], SkGetPackedB32(c1)) + SkScalarMul(cc[2], SkGetPackedB32(c2)) + SkScalarMul(cc[3], SkGetPackedB32(c3));
-    return SkPackARGB32(SkScalarRoundToInt(a),
-                        SkScalarRoundToInt(SkScalarClampMax(r, a)),
-                        SkScalarRoundToInt(SkScalarClampMax(g, a)),
-                        SkScalarRoundToInt(SkScalarClampMax(b, a)));
+#define MUL(a, b)   ((a) * (b))
+
+static inline SkPMColor cubicBlend(const SkFixed cc[4], SkPMColor c0, SkPMColor c1, SkPMColor c2, SkPMColor c3) {
+    SkFixed fa = MUL(cc[0], SkGetPackedA32(c0)) + MUL(cc[1], SkGetPackedA32(c1)) + MUL(cc[2], SkGetPackedA32(c2)) + MUL(cc[3], SkGetPackedA32(c3));
+    SkFixed fr = MUL(cc[0], SkGetPackedR32(c0)) + MUL(cc[1], SkGetPackedR32(c1)) + MUL(cc[2], SkGetPackedR32(c2)) + MUL(cc[3], SkGetPackedR32(c3));
+    SkFixed fg = MUL(cc[0], SkGetPackedG32(c0)) + MUL(cc[1], SkGetPackedG32(c1)) + MUL(cc[2], SkGetPackedG32(c2)) + MUL(cc[3], SkGetPackedG32(c3));
+    SkFixed fb = MUL(cc[0], SkGetPackedB32(c0)) + MUL(cc[1], SkGetPackedB32(c1)) + MUL(cc[2], SkGetPackedB32(c2)) + MUL(cc[3], SkGetPackedB32(c3));
+
+    int a = SkClampMax(SkFixedRoundToInt(fa), 255);
+    int r = SkClampMax(SkFixedRoundToInt(fr), a);
+    int g = SkClampMax(SkFixedRoundToInt(fg), a);
+    int b = SkClampMax(SkFixedRoundToInt(fb), a);
+    
+    return SkPackARGB32(a, r, g, b);
+}
+
+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,
-                          SkPMColor colors[], int count) {
+                          SkPMColor* SK_RESTRICT colors, int count) {
 
     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),
@@ -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),
     };
 
-    SkPMColor *dptr = &(colors[0]);
+    const int maxX = s.fBitmap->width() - 1;
+    const int maxY = s.fBitmap->height() - 1;
+    
     while (count-- > 0) {
         SkPoint srcPt;
         s.fInvProc(*s.fInvMatrix, SkIntToScalar(x),
@@ -51,37 +63,45 @@ static void bicubicFilter(const SkBitmapProcState& s, int x, int y,
         srcPt.fY -= SK_ScalarHalf;
         SkScalar fractx = srcPt.fX - SkScalarFloorToScalar(srcPt.fX);
         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 sy = SkScalarFloorToInt(srcPt.fY);
-        int x0 = SkClampMax(sx - 1, s.fBitmap->width() - 1);
-        int x1 = SkClampMax(sx    , s.fBitmap->width() - 1);
-        int x2 = SkClampMax(sx + 1, s.fBitmap->width() - 1);
-        int x3 = SkClampMax(sx + 2, s.fBitmap->width() - 1);
-        int y0 = SkClampMax(sy - 1, s.fBitmap->height() - 1);
-        int y1 = SkClampMax(sy    , s.fBitmap->height() - 1);
-        int y2 = SkClampMax(sy + 1, s.fBitmap->height() - 1);
-        int y3 = SkClampMax(sy + 2, s.fBitmap->height() - 1);
+        
+        // Here is where we can support other tile modes (e.g. repeat or mirror)
+        int x0 = SkClampMax(sx - 1, maxX);
+        int x1 = SkClampMax(sx    , maxX);
+        int x2 = SkClampMax(sx + 1, maxX);
+        int x3 = SkClampMax(sx + 2, maxX);
+        int y0 = SkClampMax(sy - 1, maxY);
+        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 s10 = *s.fBitmap->getAddr32(x1, y0);
         SkPMColor s20 = *s.fBitmap->getAddr32(x2, 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 s11 = *s.fBitmap->getAddr32(x1, y1);
         SkPMColor s21 = *s.fBitmap->getAddr32(x2, 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 s12 = *s.fBitmap->getAddr32(x1, y2);
         SkPMColor s22 = *s.fBitmap->getAddr32(x2, 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 s13 = *s.fBitmap->getAddr32(x1, y3);
         SkPMColor s23 = *s.fBitmap->getAddr32(x2, y3);
         SkPMColor s33 = *s.fBitmap->getAddr32(x3, y3);
-        SkPMColor s3 = cubicBlend(coefficients, fractx, s03, s13, s23, s33);
-        *dptr++ = cubicBlend(coefficients, fracty, s0, s1, s2, s3);
+        SkPMColor s3 = cubicBlend(coeffX, s03, s13, s23, s33);
+        *colors++ = cubicBlend(coeffY, s0, s1, s2, s3);
         x++;
     }
 }