diff options
Diffstat (limited to 'src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp')
-rw-r--r-- | src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp index e81d50fe5f..09f99748bf 100644 --- a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp @@ -149,7 +149,7 @@ inline bool run_optim_small_tensor(const ITensor *t) // Optimized convolver for 1x1 kernels used only where input width and height are both <= 8 // For big Z as in Input=7x7x832, this implementation is faster than the general code becuase it doesn't need to -// store intermidiate results in memory. Temporary results are stored in Neon registers directly and then written to the output buffer. +// store intermidiate results in memory. Temporary results are stored in SIMD registers directly and then written to the output buffer. template <unsigned int stridex> class convolver_w1x1_i8x8_f32 { |