Convert an array of float to int16_t

conv.c

#include <stddef.h>
#include <assert.h>
#include <immintrin.h>

static void v_f32_to_s1x15i(int16_t *dst, const float *src, size_t len)
{
	const __m128 minval = _mm_set1_ps(-1.0F);
	const __m128 maxval = _mm_set1_ps(+0x7fffff8p-27F);
	const __m128 scale  = _mm_set1_ps(32768.0F);

	__m128  *srcp = (__m128  *) __builtin_assume_aligned(src, 16);
	__m128i *dstp = (__m128i *) __builtin_assume_aligned(dst, 16);

	assert(len % 8 == 0 && "len must be a multiple of 8");

	for (size_t i = 0; i < len; i+=8) {

		/* Load samples */
		__m128 x1 = _mm_load_ps((float *) srcp++);
		/* Clamp values */
		__m128 xc1 = _mm_max_ps(_mm_min_ps(x1, maxval), minval);
		/* Multiply by scaling factor */
		__m128 y1 = _mm_mul_ps(xc1, scale);
		/* Convert to packed 32 bit integer */
		__m128i yc1 = _mm_cvtps_epi32(y1);

		/* Load samples */
		__m128 x2 = _mm_load_ps((float *) srcp++);
		/* Clamp values */
		__m128 xc2 = _mm_max_ps(_mm_min_ps(x2, maxval), minval);
		/* Multiply by scaling factor */
		__m128 y2 = _mm_mul_ps(xc2, scale);
		/* Convert to packed 32 bit integer */
		__m128i yc2= _mm_cvtps_epi32(y2);

		/* Pack them down to 16 bits */
		__m128i yp = _mm_packs_epi32(yc1, yc2);

		/* Store output */
		_mm_store_si128(dstp++, yp);
	}
}