cherry-pick from 7241: speed-up matrix44::setconcat when both inputs are
scale+translate by inlining the zero-assignments, rather than calling sk_bzero.
bench: matrix44_setconcat 2x faster
git-svn-id: http://skia.googlecode.com/svn/trunk@7251 2bbb7eff-a529-9590-31e7-b0007b416f81
diff --git a/src/utils/SkMatrix44.cpp b/src/utils/SkMatrix44.cpp
index 1906593..92c8715 100644
--- a/src/utils/SkMatrix44.cpp
+++ b/src/utils/SkMatrix44.cpp
@@ -360,11 +360,14 @@
SkMScalar storage[16];
SkMScalar* result = useStorage ? storage : &fMat[0][0];
+ // Both matrices are at most scale+translate
if (bits_isonly(a_mask | b_mask, kScale_Mask | kTranslate_Mask)) {
- sk_bzero(result, sizeof(storage));
result[0] = a.fMat[0][0] * b.fMat[0][0];
+ result[1] = result[2] = result[3] = result[4] = 0;
result[5] = a.fMat[1][1] * b.fMat[1][1];
+ result[6] = result[7] = result[8] = result[9] = 0;
result[10] = a.fMat[2][2] * b.fMat[2][2];
+ result[11] = 0;
result[12] = a.fMat[0][0] * b.fMat[3][0] + a.fMat[3][0];
result[13] = a.fMat[1][1] * b.fMat[3][1] + a.fMat[3][1];
result[14] = a.fMat[2][2] * b.fMat[3][2] + a.fMat[3][2];