experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Slightly optimize quantized add.

Browse Source 
PiperOrigin-RevId: 312028385
Change-Id: Ie1fbb3071e4e258c24db78440e1275168694fda9
This commit is contained in:
Renjie Liu 2020-05-18 00:28:56 -07:00 committed by TensorFlower Gardener
parent ea4ef0e6fa
commit 344f898250
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
tensorflow/lite/kernels/internal/optimized/integer_ops/add.h
View File

@ -47,6 +47,9 @@ inline void AddElementwise(int size, const ArithmeticParams& params,
const int32x4_t input1_left_dup = vdupq_n_s32(input1_left_shift);
const int32x4_t input2_left_dup = vdupq_n_s32(input2_left_shift);
const int16x8_t input1_offset_dup = vdupq_n_s16(params.input1_offset);
const int16x8_t input2_offset_dup = vdupq_n_s16(params.input2_offset);
for (; i <= size - 16; i += 16) {
const int8x16_t input1_val_original = vld1q_s8(input1_data + i);
const int8x16_t input2_val_original = vld1q_s8(input2_data + i);
@ -61,13 +64,13 @@ inline void AddElementwise(int size, const ArithmeticParams& params,
const int16x8_t input2_val_s16_low =
vmovl_s8(vget_low_s8(input2_val_original));
const int16x8_t input1_val_high =
vaddq_s16(input1_val_s16_high, vdupq_n_s16(params.input1_offset));
vaddq_s16(input1_val_s16_high, input1_offset_dup);
const int16x8_t input2_val_high =
vaddq_s16(input2_val_s16_high, vdupq_n_s16(params.input2_offset));
vaddq_s16(input2_val_s16_high, input2_offset_dup);
const int16x8_t input1_val_low =
vaddq_s16(input1_val_s16_low, vdupq_n_s16(params.input1_offset));
vaddq_s16(input1_val_s16_low, input1_offset_dup);
const int16x8_t input2_val_low =
vaddq_s16(input2_val_s16_low, vdupq_n_s16(params.input2_offset));
vaddq_s16(input2_val_s16_low, input2_offset_dup);
const int16x4_t input1_val_high_high = vget_high_s16(input1_val_high);
const int16x4_t input1_val_high_low = vget_low_s16(input1_val_high);
const int16x4_t input1_val_low_high = vget_high_s16(input1_val_low);