/* * Copyright 2011 Google Inc. * * Use of this source code is governed by a BSD-style license that can be * found in the LICENSE file. */ #include "include/private/SkColorData.h" #include "tests/Test.h" // our std SkAlpha255To256 static int test_srcover0(unsigned dst, unsigned alpha) { return alpha + SkAlphaMul(dst, SkAlpha255To256(255 - alpha)); } // faster hack +1 static int test_srcover1(unsigned dst, unsigned alpha) { return alpha + SkAlphaMul(dst, 256 - alpha); } // slower "correct" static int test_srcover2(unsigned dst, unsigned alpha) { return alpha + SkMulDiv255Round(dst, 255 - alpha); } DEF_TEST(SrcOver, reporter) { /* Here's the idea. Can we ensure that when we blend on top of an opaque dst, that the result always stay's opaque (i.e. exactly 255)? */ unsigned i; int opaqueCounter0 = 0; int opaqueCounter1 = 0; int opaqueCounter2 = 0; for (i = 0; i <= 255; i++) { unsigned result0 = test_srcover0(0xFF, i); unsigned result1 = test_srcover1(0xFF, i); unsigned result2 = test_srcover2(0xFF, i); opaqueCounter0 += (result0 == 0xFF); opaqueCounter1 += (result1 == 0xFF); opaqueCounter2 += (result2 == 0xFF); } #if 0 INFOF(reporter, "---- opaque test: [%d %d %d]\n", opaqueCounter0, opaqueCounter1, opaqueCounter2); #endif // we acknowledge that technique0 does not always return opaque REPORTER_ASSERT(reporter, opaqueCounter0 == 256); REPORTER_ASSERT(reporter, opaqueCounter1 == 256); REPORTER_ASSERT(reporter, opaqueCounter2 == 256); // Now ensure that we never over/underflow a byte for (i = 0; i <= 255; i++) { for (unsigned dst = 0; dst <= 255; dst++) { unsigned r0 = test_srcover0(dst, i); unsigned r1 = test_srcover1(dst, i); unsigned r2 = test_srcover2(dst, i); unsigned max = std::max(dst, i); // ignore the known failure if (dst != 255) { REPORTER_ASSERT(reporter, r0 <= 255 && r0 >= max); } REPORTER_ASSERT(reporter, r1 <= 255 && r1 >= max); REPORTER_ASSERT(reporter, r2 <= 255 && r2 >= max); #if 0 // this shows where r1 (faster) differs from r2 (more exact) if (r1 != r2) { INFOF(reporter, "--- dst=%d i=%d r1=%d r2=%d exact=%g\n", dst, i, r1, r2, i + dst - dst*i/255.0f); } #endif } } }