radio95/fm95
0
1
Fork 0
mirror of https://github.com/radio95-rnt/fm95.git synced 2026-02-26 19:23:51 +01:00
Code Issues Activity

that neon used more cpu than the normal version???

Browse Source
This commit is contained in:
KubaPro010
2025-03-27 21:21:55 +01:00
parent 4121657da9
commit 7e64dcaab3
1 changed files with 11 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
lib/filters.c
View File

@@ -33,29 +33,20 @@ void init_lpf(LPFFilter *filter, float cutoff, int sample_rate) {
float process_lpf(LPFFilter *filter, float x) {
#if USE_NEON
float32_t input = x;
float32x4_t y_vec = vdupq_n_f32(x);
for (int i = 0; i < LPF_ORDER; i += 4) {
float32x4_t d1_vec = vld1q_f32(&filter->d1[i]);
float32x4_t d2_vec = vld1q_f32(&filter->d2[i]);
float32x4_t w1_vec = vld1q_f32(&filter->w1[i]);
float32x4_t w2_vec = vld1q_f32(&filter->w2[i]);
float32x4_t A_vec = vld1q_f32(&filter->A[i]);
float32x4_t y_vec = vdupq_n_f32(input);
float32x4_t w0_new_vec = vmlaq_f32(vmlaq_f32(y_vec, d1_vec, w1_vec), d2_vec, w2_vec);
y_vec = vmulq_f32(A_vec, vaddq_f32(vaddq_f32(w0_new_vec, w1_vec), vaddq_f32(w1_vec, w2_vec)));
for (int i = 0; i < LPF_ORDER; i++) {
float32x4_t d1_vec = vdupq_n_f32(filter->d1[i]);
float32x4_t d2_vec = vdupq_n_f32(filter->d2[i]);
float32x4_t w1_vec = vdupq_n_f32(filter->w1[i]);
float32x4_t w2_vec = vdupq_n_f32(filter->w2[i]);
float32x4_t A_vec = vdupq_n_f32(filter->A[i]);
float32x4_t w1_term = vmulq_f32(d1_vec, w1_vec);
float32x4_t w2_term = vmulq_f32(d2_vec, w2_vec);
float32x4_t w0_new_vec = vaddq_f32(vaddq_f32(w1_term, w2_term), y_vec);
float32x4_t two_w1 = vmulq_n_f32(w1_vec, 2.0f);
float32x4_t output_term = vaddq_f32(w0_new_vec, vaddq_f32(two_w1, w2_vec));
y_vec = vmulq_f32(A_vec, output_term);
filter->w2[i] = filter->w1[i];
filter->w1[i] = vgetq_lane_f32(w0_new_vec, 0);
vst1q_f32(&filter->w2[i], w1_vec);
vst1q_f32(&filter->w1[i], w0_new_vec);
}
return vgetq_lane_f32(y_vec, 0);
#else
float y = x;