diff options
| author | Sang-Hoon Park <sang-hoon.park@arm.com> | 2020-03-13 14:56:05 +0000 |
|---|---|---|
| committer | Sang-Hoon Park <sang-hoon.park@arm.com> | 2020-04-07 09:00:09 +0000 |
| commit | 0d008f77b0085619c446d0ab5dc1228a80776706 (patch) | |
| tree | e1f6e91bf8da63e8ef98e11ab8eb6a6972a284f2 /arm_compute/core/utils/quantization/AsymmHelpers.h | |
| parent | 4df2cf3177129d10500d30056bf8404418f703d6 (diff) | |
| download | ComputeLibrary-0d008f77b0085619c446d0ab5dc1228a80776706.tar.gz | |
COMPMID-3281: Implement QSYMM16 Layer Normalization for NEON QLSTM
- Reference kernel is modified to use the same algorithm as NEON kernel.
- NEON kernel is implemented.
- Tests for validation and run are added.
Change-Id: I3533bc2bd12c6e9cc75d837ecf193f74ceddf796
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2948
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'arm_compute/core/utils/quantization/AsymmHelpers.h')
| -rw-r--r-- | arm_compute/core/utils/quantization/AsymmHelpers.h | 15 |
1 files changed, 13 insertions, 2 deletions
diff --git a/arm_compute/core/utils/quantization/AsymmHelpers.h b/arm_compute/core/utils/quantization/AsymmHelpers.h index 0f0ec72b60..a7bbf9b137 100644 --- a/arm_compute/core/utils/quantization/AsymmHelpers.h +++ b/arm_compute/core/utils/quantization/AsymmHelpers.h @@ -128,7 +128,7 @@ int32_t saturating_rounding_doubling_highmul(int32_t a, int32_t b); * * @return The multiplied value */ -int32_t multiply_by_quantized_multipler(int32_t input, int32_t qmul, int32_t shift); +int32_t multiply_by_quantized_multiplier(int32_t input, int32_t qmul, int32_t shift); /** Compute the value multiplied the power-of-two * @@ -137,7 +137,18 @@ int32_t multiply_by_quantized_multipler(int32_t input, int32_t qmul, int32_t shi * * @return The multiplied value */ -int32_t saturating_rounding_multiply_by_pow2(int exponent, int32_t v); +int32_t saturating_rounding_multiply_by_pow2(int32_t exponent, int32_t v); + +/** Compute quantized multiplier and shift for the inverse square root of input. + * Using 3-bit fixed point and 5 iteration of Newton-Raphson method. + * + * @param[in] input Input to use + * @param[in] reverse_shift -1 to reverse the shift direction + * @param[out] output_inv_sqrt Quantized multiplier for inverse square root + * @param[out] output_shift Shift for inverse square root + * + */ +void get_invsqrt_quantized_multiplier_exp(int32_t input, int32_t reverse_shift, int32_t &output_inv_sqrt, int32_t &output_shift); } // namespace quantization } // namespace arm_compute |