dengzhihao
/
CMSIS-DSP
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

CMSIS-DSP: Correct loading of FP16 data.

Browse Source
pull/19/head
David Green 5 years ago committed by Christophe Favergeon
parent 1ca28fa995
commit 67edaf27e6
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File

8
ComputeLibrary/Include/NEMath.h
Unescape Escape View File

@ -365,10 +365,10 @@ inline float16x8_t vtanhq_f16(float16x8_t val)
inline float16x8_t vtaylor_polyq_f16(float16x8_t x, const float16_t *coeffs) inline float16x8_t vtaylor_polyq_f16(float16x8_t x, const float16_t *coeffs)
{ {
const float16x8_t A = vaddq_f16(&coeffs[8*0], vmulq_f16(&coeffs[8*4], x)); const float16x8_t A = vaddq_f16(vld1q_f16(&coeffs[8*0]), vmulq_f16(vld1q_f16(&coeffs[8*4]), x));
const float16x8_t B = vaddq_f16(&coeffs[8*2], vmulq_f16(&coeffs[8*6], x)); const float16x8_t B = vaddq_f16(vld1q_f16(&coeffs[8*2]), vmulq_f16(vld1q_f16(&coeffs[8*6]), x));
const float16x8_t C = vaddq_f16(&coeffs[8*1], vmulq_f16(&coeffs[8*5], x)); const float16x8_t C = vaddq_f16(vld1q_f16(&coeffs[8*1]), vmulq_f16(vld1q_f16(&coeffs[8*5]), x));
const float16x8_t D = vaddq_f16(&coeffs[8*3], vmulq_f16(&coeffs[8*7], x)); const float16x8_t D = vaddq_f16(vld1q_f16(&coeffs[8*3]), vmulq_f16(vld1q_f16(&coeffs[8*7]), x));
const float16x8_t x2 = vmulq_f16(x, x); const float16x8_t x2 = vmulq_f16(x, x);
const float16x8_t x4 = vmulq_f16(x2, x2); const float16x8_t x4 = vmulq_f16(x2, x2);
const float16x8_t res = vaddq_f16(vaddq_f16(A, vmulq_f16(B, x2)), vmulq_f16(vaddq_f16(C, vmulq_f16(D, x2)), x4)); const float16x8_t res = vaddq_f16(vaddq_f16(A, vmulq_f16(B, x2)), vmulq_f16(vaddq_f16(C, vmulq_f16(D, x2)), x4));

Write Preview
Loading…
Cancel
Save