memops: Use right-aligned S24LE to float conversion

ALSA expects right-aligned samples (0x00******) as mentioned in the
source code for SND_PCM_FORMAT_S24_LE:
Signed 24 bit Little Endian using low three bytes in 32-bit word
See http://git.alsa-project.org/?p=alsa-
lib.git;a=blob;f=include/pcm.h;h=5b0782315585de1d5ab82c9f2036b62c168f5a48;hb=HEAD#l140

Signed-off-by: Timo Wischer <twischer@de.adit-jv.com>

common/memops.c

@@ -73,8 +73,8 @@
    So, for now (October 2008) we use 2^(N-1)-1 as the scaling factor.
 */
 
-#define SAMPLE_24BIT_SCALING  8388607.0f
-#define SAMPLE_16BIT_SCALING  32767.0f
+#define SAMPLE_24BIT_SCALING  8388607
+#define SAMPLE_16BIT_SCALING  32767
 
 /* these are just values to use if the floating point value was out of range
    
@@ -414,8 +414,10 @@ void sample_move_d32u24_sS (char *dst, jack_default_audio_sample_t *src, unsigne
 
 void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
 {
+	const jack_default_audio_sample_t scaling = 1.0 / (SAMPLE_24BIT_SCALING << 8);
+
 #if defined (__ARM_NEON__) || defined (__ARM_NEON)
-	float32x4_t factor = vdupq_n_f32(1.0 / SAMPLE_24BIT_SCALING);
+	float32x4_t factor = vdupq_n_f32(scaling);
 	unsigned long unrolled = nsamples / 4;
 	while (unrolled--) {
 		int32x4_t src128;
@@ -435,7 +437,8 @@ void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsign
 				break;
 		}
 		src128 = vreinterpretq_s32_u8(vrev32q_u8(vreinterpretq_u8_s32(src128)));
-		int32x4_t shifted = vshrq_n_s32(src128, 8);
+		/* sign extension - left shift will be reverted by scaling */
+		int32x4_t shifted = vshlq_n_s32(src128, 8);
 		float32x4_t as_float = vcvtq_f32_s32(shifted);
 		float32x4_t divided = vmulq_f32(as_float, factor);
 		vst1q_f32(dst, divided);
@@ -448,8 +451,6 @@ void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsign
 
 	/* ALERT: signed sign-extension portability !!! */
 
-	const jack_default_audio_sample_t scaling = 1.0/SAMPLE_24BIT_SCALING;
-
 	while (nsamples--) {
 		int x;
 #if __BYTE_ORDER == __LITTLE_ENDIAN
@@ -469,7 +470,8 @@ void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsign
 		x <<= 8;
 		x |= (unsigned char)(src[0]);
 #endif
-		*dst = (x >> 8) * scaling;
+		/* sign extension - left shift will be reverted by scaling */
+		*dst = (x << 8) * scaling;
 		dst++;
 		src += src_skip;
 	}
@@ -477,10 +479,11 @@ void sample_move_dS_s32u24s (jack_default_audio_sample_t *dst, char *src, unsign
 
 void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigned long nsamples, unsigned long src_skip)
 {
+	const jack_default_audio_sample_t scaling = 1.0 / (SAMPLE_24BIT_SCALING << 8);
+
 #if defined (__SSE2__) && !defined (__sun__)
 	unsigned long unrolled = nsamples / 4;
-	static float inv_sample_max_24bit = 1.0 / SAMPLE_24BIT_SCALING;
-	__m128 factor = _mm_set1_ps(inv_sample_max_24bit);
+	__m128 factor = _mm_set1_ps(scaling);
 	while (unrolled--)
 	{
 		int i1 = *((int *) src);
@@ -493,7 +496,8 @@ void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigne
 		src+= src_skip;
 
 		__m128i src = _mm_set_epi32(i4, i3, i2, i1);
-		__m128i shifted = _mm_srai_epi32(src, 8);
+		/* sign extension - left shift will be reverted by scaling */
+		__m128i shifted = _mm_slli_epi32(src, 8);
 
 		__m128 as_float = _mm_cvtepi32_ps(shifted);
 		__m128 divided = _mm_mul_ps(as_float, factor);
@@ -505,7 +509,7 @@ void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigne
 	nsamples = nsamples & 3;
 #elif defined (__ARM_NEON__) || defined (__ARM_NEON)
 	unsigned long unrolled = nsamples / 4;
-	float32x4_t factor = vdupq_n_f32(1.0 / SAMPLE_24BIT_SCALING);
+	float32x4_t factor = vdupq_n_f32(scaling);
 	while (unrolled--) {
 		int32x4_t src128;
 		switch(src_skip) {
@@ -522,7 +526,8 @@ void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigne
 				src128 = vld1q_lane_s32((int32_t*)(src+3*src_skip), src128, 3);
 				break;
 		}
-		int32x4_t shifted = vshrq_n_s32(src128, 8);
+		/* sign extension  - left shift will be reverted by scaling */
+		int32x4_t shifted = vshlq_n_s32(src128, 8);
 		float32x4_t as_float = vcvtq_f32_s32(shifted);
 		float32x4_t divided = vmulq_f32(as_float, factor);
 		vst1q_f32(dst, divided);
@@ -535,9 +540,9 @@ void sample_move_dS_s32u24 (jack_default_audio_sample_t *dst, char *src, unsigne
 
 	/* ALERT: signed sign-extension portability !!! */
 
-	const jack_default_audio_sample_t scaling = 1.0/SAMPLE_24BIT_SCALING;
 	while (nsamples--) {
-		*dst = (*((int *) src) >> 8) * scaling;
+		/* sign extension  - left shift will be reverted by scaling */
+		*dst = (*((int *) src) << 8) * scaling;
 		dst++;
 		src += src_skip;
 	}