diff options
Diffstat (limited to 'src/core/NEON/kernels/assembly/arm_gemm.hpp')
| -rw-r--r-- | src/core/NEON/kernels/assembly/arm_gemm.hpp | 41 |
1 files changed, 22 insertions, 19 deletions
diff --git a/src/core/NEON/kernels/assembly/arm_gemm.hpp b/src/core/NEON/kernels/assembly/arm_gemm.hpp index 58db511547..f6421c12ab 100644 --- a/src/core/NEON/kernels/assembly/arm_gemm.hpp +++ b/src/core/NEON/kernels/assembly/arm_gemm.hpp @@ -122,38 +122,41 @@ public: struct Requantize32 { public: - const int32_t *bias = nullptr; - size_t bias_multi_stride = 0; - int32_t a_offset = 0; - int32_t b_offset = 0; - int32_t c_offset = 0; - bool per_channel_requant = false; - int32_t per_layer_shift = 0; - int32_t per_layer_mul = 0; - const int32_t *per_channel_shifts = nullptr; - const int32_t *per_channel_muls = nullptr; - int32_t minval = 0; - int32_t maxval = 0; + const int32_t *bias = nullptr; + size_t bias_multi_stride = 0; + int32_t a_offset = 0; + int32_t b_offset = 0; + int32_t c_offset = 0; + bool per_channel_requant = false; + int32_t per_layer_left_shift = 0; + int32_t per_layer_right_shift = 0; + int32_t per_layer_mul = 0; + const int32_t *per_channel_left_shifts = nullptr; + const int32_t *per_channel_right_shifts = nullptr; + const int32_t *per_channel_muls = nullptr; + int32_t minval = 0; + int32_t maxval = 0; Requantize32() = default; // Constructor for per-tensor quantization Requantize32(const int32_t *bias, size_t bias_multi_stride, int32_t a_offset, int32_t b_offset, int32_t c_offset, - int32_t requant_shift, int32_t requant_mul, - int32_t minv, int32_t maxv) - : bias(bias), bias_multi_stride(bias_multi_stride), a_offset(a_offset), b_offset(b_offset), c_offset(c_offset), per_channel_requant(false), per_layer_shift(requant_shift), per_layer_mul(requant_mul), - minval(minv), maxval(maxv) + int32_t requant_shift, int32_t requant_mul, int32_t minv, int32_t maxv) + : bias(bias), bias_multi_stride(bias_multi_stride), a_offset(a_offset), b_offset(b_offset), c_offset(c_offset), per_channel_requant(false), per_layer_left_shift(std::max(requant_shift, int32_t(0))), + per_layer_right_shift(std::min(requant_shift, int32_t(0))), per_layer_mul(requant_mul), minval(minv), maxval(maxv) { } // Constructor for per-channel quantization Requantize32(const int32_t *bias, size_t bias_multi_stride, int32_t a_offset, int32_t b_offset, int32_t c_offset, - const int32_t *requant_shifts, const int32_t *requant_muls, + const int32_t *requant_left_shifts, + const int32_t *requant_right_shifts, + const int32_t *requant_muls, int32_t minv, int32_t maxv) - : bias(bias), bias_multi_stride(bias_multi_stride), a_offset(a_offset), b_offset(b_offset), c_offset(c_offset), per_channel_requant(true), per_channel_shifts(requant_shifts), - per_channel_muls(requant_muls), minval(minv), maxval(maxv) + : bias(bias), bias_multi_stride(bias_multi_stride), a_offset(a_offset), b_offset(b_offset), c_offset(c_offset), per_channel_requant(true), per_channel_left_shifts(requant_left_shifts), + per_channel_right_shifts(requant_right_shifts), per_channel_muls(requant_muls), minval(minv), maxval(maxv) { } }; |