diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/interleaves')
3 files changed, 3 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_u8q_3x3_dot.cpp b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_u8q_3x3_dot.cpp index 314f09a0c5..19264c9fce 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_u8q_3x3_dot.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_u8q_3x3_dot.cpp @@ -42,7 +42,7 @@ size_t interleave_a64_u8q_3x3_dot::get_packed_size(const DepthwiseArgs &args) { // We store 7 vectors for every <vector_of_ints> of channels. const unsigned int n = arm_gemm::roundup( - arm_gemm::iceildiv((long unsigned int) args.input_channels, + arm_gemm::iceildiv((long unsigned int) args.input_channels * args.channel_multiplier, get_vector_length<int32_t>(arm_gemm::VLType::None)), 4lu ); return n * 7 * get_vector_length<uint8_t>(arm_gemm::VLType::None); diff --git a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_s8q_3x3_dot.cpp b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_s8q_3x3_dot.cpp index 3a4999296a..5d7b54f235 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_s8q_3x3_dot.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_s8q_3x3_dot.cpp @@ -42,7 +42,7 @@ size_t interleave_sve_s8q_3x3_dot::get_packed_size(const DepthwiseArgs &args) { // We store 7 vectors for every <vector_of_ints> of channels. const unsigned int n = arm_gemm::roundup( - arm_gemm::iceildiv((long unsigned int) args.input_channels, + arm_gemm::iceildiv((long unsigned int) args.input_channels * args.channel_multiplier, get_vector_length<int32_t>(arm_gemm::VLType::SVE)), 4lu ); return n * 7 * get_vector_length<int8_t>(arm_gemm::VLType::SVE); diff --git a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_u8q_3x3_dot.cpp b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_u8q_3x3_dot.cpp index 7c5d3c4904..c3da81448b 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_u8q_3x3_dot.cpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_u8q_3x3_dot.cpp @@ -42,7 +42,7 @@ size_t interleave_sve_u8q_3x3_dot::get_packed_size(const DepthwiseArgs &args) { // We store 7 vectors for every <vector_of_ints> of channels. const unsigned int n = arm_gemm::roundup( - arm_gemm::iceildiv((long unsigned int) args.input_channels, + arm_gemm::iceildiv((long unsigned int) args.input_channels * args.channel_multiplier, get_vector_length<int32_t>(arm_gemm::VLType::SVE)), 4lu ); return n * 7 * get_vector_length<uint8_t>(arm_gemm::VLType::SVE); |