diff options
Diffstat (limited to 'src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp | 24 |
1 files changed, 12 insertions, 12 deletions
diff --git a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp index 2766d698d9..8642a19f39 100644 --- a/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDirectConvolutionLayerKernel.cpp @@ -68,7 +68,7 @@ inline qint16x8_t internal_vdupq_n(qint16_t v) return vdupq_n_qs16(v); } -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 template <unsigned int stridex> float16x8_t internal_vld1q(const float16_t *in); @@ -113,7 +113,7 @@ inline float16x8_t internal_vmlal(const float16x8_t &x, const float16x8_t &y, co ARM_COMPUTE_UNUSED(fixed_point_position); return vaddq_f16(x, vmulq_f16(y, z)); } -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ template <unsigned int stridex> float32x4_t internal_vld1q(const float *in); @@ -427,7 +427,7 @@ public: } }; -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 inline float16x8x3_t load_matrix_row(const float16_t *ptr) { /* ptr is a pointer to a row in a 3x3 matrix, the function returns 3 vectors holding exactly the same value in all lanes: @@ -567,7 +567,7 @@ void accumulate_results<3>(float16_t *buffer, const float16x8x2_t &values) vst1_f16(buffer, vadd_f16(vld1_f16(buffer), vget_low_f16(values.val[0]))); } -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ inline float32x4x3_t load_matrix_row(const float *ptr) { @@ -1433,9 +1433,9 @@ void NEDirectConvolutionLayerKernel::configure(const ITensor *input, const ITens { switch(input->info()->data_type()) { -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 case DataType::F16: -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ case DataType::QS8: case DataType::QS16: _num_elems_written_per_iteration = 8; @@ -1468,9 +1468,9 @@ void NEDirectConvolutionLayerKernel::configure(const ITensor *input, const ITens _num_elems_read_per_iteration = 12; _num_elems_written_per_iteration = 16 >> conv_stride_x; break; -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 case DataType::F16: -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ case DataType::QS8: case DataType::QS16: _num_weight_elems_read_per_row = 8 + _kernel_size - 1; @@ -1532,11 +1532,11 @@ void NEDirectConvolutionLayerKernel::run(const Window &window, const ThreadInfo case DataType::F32: convolve_1x1<float, float>(window, _num_elems_read_per_iteration, _num_elems_written_per_iteration, _input, _weights, _output, _conv_info); break; -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 case DataType::F16: convolve_1x1<float16_t, float16_t>(window, _num_elems_read_per_iteration, _num_elems_written_per_iteration, _input, _weights, _output, _conv_info); break; -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ default: ARM_COMPUTE_ERROR("Data type not supported"); break; @@ -1553,11 +1553,11 @@ void NEDirectConvolutionLayerKernel::run(const Window &window, const ThreadInfo case DataType::F32: convolve_3x3<float, float>(window, _num_elems_read_per_iteration, _num_elems_written_per_iteration, _input, _weights, _output, _conv_info); break; -#ifdef ARM_COMPUTE_ENABLE_FP16 +#ifdef ARM_COMPUTE_AARCH64_V8_2 case DataType::F16: convolve_3x3<float16_t, float16_t>(window, _num_elems_read_per_iteration, _num_elems_written_per_iteration, _input, _weights, _output, _conv_info); break; -#endif /* ARM_COMPUTE_ENABLE_FP16 */ +#endif /* ARM_COMPUTE_AARCH64_V8_2 */ default: ARM_COMPUTE_ERROR("Data type not supported"); break; |