Searched refs:rhs_scale (Results 1 – 3 of 3) sorted by relevance
| /external/gemmlowp/doc/ |
| D | quantization.md | 149 `rhs_scale`, `lhs_zero_point`, `rhs_zero_point`, so that their matrix entries
154 rhs_real_value[i] = rhs_scale * (rhs_quantized_value[i] - rhs_zero_point)
164 rhs_scale * (rhs_quantized_value[i] - rhs_zero_point)
166 = lhs_scale * rhs_scale * Sum_over_i(
195 (lhs_scale * rhs_scale / result_scale) *
247 (lhs_scale * rhs_scale / result_scale) * int32_accumulator (7)
250 The difficulty here is of course that `(lhs_scale * rhs_scale / result_scale)`
271 (lhs_scale * rhs_scale / result_scale) * int32_accumulator
|
| /external/tensorflow/tensorflow/compiler/mlir/tosa/transforms/ |
| D | legalize_common.cc | 647 double lhs_scale, rhs_scale, result_scale; in convertConcatV2Op() local
659 rhs_scale = static_cast<double>(rhs_quant_type.getScale()); in convertConcatV2Op()
675 if (rhs_scale != result_scale) { in convertConcatV2Op()
680 rhs_scale / result_scale, rhs_zeropoint, in convertConcatV2Op()
707 rhs_scale = static_cast<float>(rhs_quant_type.getScale()); in convertConcatV2Op()
710 if (rhs_scale != result_scale) { in convertConcatV2Op()
715 rhs_scale / result_scale, rhs_zeropoint, in convertConcatV2Op()
|
| /external/tensorflow/tensorflow/compiler/mlir/tosa/g3doc/ |
| D | legalization.md | 3181 double rhs_scale = double(1 << input_shift) * %rhs.scale / max_scale_2x
3185 compute_scale_32(rhs_scale, rhs_multiplier, rhs_shift)
|