| 14 | } |
| 15 | |
| 16 | LocalImpl::float_noncontig_batch_kern LocalImpl::dispatch_float_noncontig_batch( |
| 17 | const TensorLayout& src, const TensorLayout&, const TensorLayout&) { |
| 18 | megdnn_assert( |
| 19 | src.stride[0] > 0 && |
| 20 | static_cast<size_t>(src.stride[0]) >= src.total_nr_elems() / src.shape[0]); |
| 21 | |
| 22 | if (param().mode == Mode::CROSS_CORRELATION) { |
| 23 | if (is_supported(SIMDType::FMA)) { |
| 24 | return local_xcorr_FMA; |
| 25 | } else if (is_supported(SIMDType::AVX)) { |
| 26 | return local_xcorr_AVX; |
| 27 | } else if (is_supported(SIMDType::SSE)) { |
| 28 | return local_xcorr_SSE; |
| 29 | } else { |
| 30 | megdnn_throw("no fma/avx/sse detected"); |
| 31 | } |
| 32 | } else { |
| 33 | if (is_supported(SIMDType::FMA)) { |
| 34 | return local_conv_FMA; |
| 35 | } else if (is_supported(SIMDType::AVX)) { |
| 36 | return local_conv_AVX; |
| 37 | } else if (is_supported(SIMDType::SSE)) { |
| 38 | return local_conv_SSE; |
| 39 | } else { |
| 40 | megdnn_throw("no fma/avx/sse detected"); |
| 41 | } |
| 42 | } |
| 43 | } |
| 44 | |
| 45 | void LocalImpl::exec( |
| 46 | _megdnn_tensor_in src, _megdnn_tensor_in filter, _megdnn_tensor_out dst, |
no test coverage detected