| 1 | #pragma once |
|---|---|
| 2 | #ifndef __CVTT_AGGREGATEDERROR_H__ |
| 3 | #define __CVTT_AGGREGATEDERROR_H__ |
| 4 | |
| 5 | #include "ConvectionKernels_ParallelMath.h" |
| 6 | |
| 7 | namespace cvtt |
| 8 | { |
| 9 | namespace Internal |
| 10 | { |
| 11 | template<int TVectorSize> |
| 12 | class AggregatedError |
| 13 | { |
| 14 | public: |
| 15 | typedef ParallelMath::UInt16 MUInt16; |
| 16 | typedef ParallelMath::UInt31 MUInt31; |
| 17 | typedef ParallelMath::Float MFloat; |
| 18 | |
| 19 | AggregatedError() |
| 20 | { |
| 21 | for (int ch = 0; ch < TVectorSize; ch++) |
| 22 | m_errorUnweighted[ch] = ParallelMath::MakeUInt31(0); |
| 23 | } |
| 24 | |
| 25 | void Add(const MUInt16 &channelErrorUnweighted, int ch) |
| 26 | { |
| 27 | m_errorUnweighted[ch] = m_errorUnweighted[ch] + ParallelMath::ToUInt31(channelErrorUnweighted); |
| 28 | } |
| 29 | |
| 30 | MFloat Finalize(uint32_t flags, const float channelWeightsSq[TVectorSize]) const |
| 31 | { |
| 32 | if (flags & cvtt::Flags::Uniform) |
| 33 | { |
| 34 | MUInt31 total = m_errorUnweighted[0]; |
| 35 | for (int ch = 1; ch < TVectorSize; ch++) |
| 36 | total = total + m_errorUnweighted[ch]; |
| 37 | return ParallelMath::ToFloat(total); |
| 38 | } |
| 39 | else |
| 40 | { |
| 41 | MFloat total = ParallelMath::ToFloat(m_errorUnweighted[0]) * channelWeightsSq[0]; |
| 42 | for (int ch = 1; ch < TVectorSize; ch++) |
| 43 | total = total + ParallelMath::ToFloat(m_errorUnweighted[ch]) * channelWeightsSq[ch]; |
| 44 | return total; |
| 45 | } |
| 46 | } |
| 47 | |
| 48 | private: |
| 49 | MUInt31 m_errorUnweighted[TVectorSize]; |
| 50 | }; |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | #endif |
| 55 | |
| 56 |