ref: cf9e21febdb858847befa4a14d70270811c25019
dir: /test/encoder/EncUT_ExpGolomb.cpp/
#include <gtest/gtest.h> #include <math.h> #include "svc_enc_golomb.h" #include "macros.h" using namespace WelsSVCEnc; const double g_kdLog2Factor = 1.0 / log(2.0); TEST(UeExpGolombTest, TestBsSizeUeLt256) { uint32_t uiInVal = 0; for (; uiInVal < 256; ++ uiInVal) { const uint32_t uiActVal = BsSizeUE( uiInVal ); const int32_t m = static_cast<int32_t>(log( (uiInVal+1) * 1.0 ) * g_kdLog2Factor + 1e-6); const uint32_t uiExpVal = (m<<1)+1; EXPECT_EQ(uiActVal, uiExpVal); } } TEST(UeExpGolombTest, TestBsSizeUeRangeFrom256To65534) { uint32_t uiInVal = 0x100; for (; uiInVal < 0xFFFF; ++ uiInVal) { const uint32_t uiActVal = BsSizeUE( uiInVal ); const int32_t m = static_cast<int32_t>(log( (uiInVal+1) * 1.0 ) * g_kdLog2Factor + 1e-6); const uint32_t uiExpVal = (m<<1)+1; EXPECT_EQ(uiActVal, uiExpVal); } } TEST(UeExpGolombTest, TestBsSizeUeRangeFrom65535ToPlus256) { uint32_t uiInVal = 0xFFFF; const uint32_t uiCountBase = 256; const uint32_t uiInValEnd = uiInVal + uiCountBase; for (; uiInVal < uiInValEnd; ++ uiInVal) { const uint32_t uiActVal = BsSizeUE( uiInVal ); // float precision issue in case use math::log const int32_t m = WELS_LOG2(1+uiInVal); const uint32_t uiExpVal = (m<<1)+1; EXPECT_EQ(uiActVal, uiExpVal); } }