From 49a4b9287346fa29cf1534c894c4ddc87d80c814 Mon Sep 17 00:00:00 2001 From: KubaPro010 Date: Thu, 27 Mar 2025 21:40:04 +0100 Subject: [PATCH] really? --- src/fm95.c | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) diff --git a/src/fm95.c b/src/fm95.c index d396e6b..c0d502f 100644 --- a/src/fm95.c +++ b/src/fm95.c @@ -49,17 +49,20 @@ static volatile sig_atomic_t to_run = 1; void uninterleave(const float *input, float *left, float *right, size_t num_samples) { #if USE_NEON size_t i = 0; - size_t vec_size = num_samples / 2; - for (; i + 4 <= vec_size; i += 4) { - float32x4x2_t interleaved = vld2q_f32(&input[i * 2]); - vst1q_f32(&left[i], interleaved.val[0]); - vst1q_f32(&right[i], interleaved.val[1]); + for (; i + 7 < num_samples; i += 8) { + float32x4x2_t interleaved1 = vld2q_f32(&input[i * 2]); + float32x4x2_t interleaved2 = vld2q_f32(&input[i * 2 + 8]); + + vst1q_f32(&left[i], interleaved1.val[0]); + vst1q_f32(&right[i], interleaved1.val[1]); + vst1q_f32(&left[i + 4], interleaved2.val[0]); + vst1q_f32(&right[i + 4], interleaved2.val[1]); } - for (; i < vec_size; i++) { - left[i] = input[i * 2]; - right[i] = input[i * 2 + 1]; + for (; i < num_samples; i += 2) { + left[i / 2] = input[i]; + right[i / 2] = input[i + 1]; } #else for (size_t i = 0; i < num_samples/2; i++) {