44   volatile uint32 sse;  in SumSquareError_NEON()  local
69       : "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)  in SumSquareError_NEON()
72   return sse;  in SumSquareError_NEON()
79   volatile uint32 sse;  in SumSquareError_NEON()  local
103       : "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)  in SumSquareError_NEON()
106   return sse;  in SumSquareError_NEON()
152   uint32 sse;  in SumSquareError_SSE2()  local
185         "=g"(sse)     // %3  in SumSquareError_SSE2()
193   return sse;  in SumSquareError_SSE2()
234   uint32 sse = 0u;  in SumSquareError_C()  local
237     sse += static_cast<uint32>(diff * diff);  in SumSquareError_C()
239   return sse;  in SumSquareError_C()
256   uint64 sse = 0;  in ComputeSumSquareError()  local
258 #pragma omp parallel for reduction(+ : sse)  in ComputeSumSquareError()
261     sse += SumSquareError(src_a + i, src_b + i, kBlockSize);  in ComputeSumSquareError()
267     sse += SumSquareError(src_a, src_b, remainder);  in ComputeSumSquareError()
273     sse += SumSquareError_C(src_a, src_b, remainder);  in ComputeSumSquareError()
275   return static_cast<double>(sse);  in ComputeSumSquareError()
281 double ComputePSNR(double sse, double size) {  in ComputePSNR()  argument
283   if (sse <= kMINSSE)  in ComputePSNR()
284     sse = kMINSSE;  // Produces max PSNR of 128  in ComputePSNR()
285   return 10.0 * log10(255.0 * 255.0 * size / sse);  in ComputePSNR()