diff --git a/spa/include/spa/utils/defs.h b/spa/include/spa/utils/defs.h
index 4654d4a10..9226f48de 100644
--- a/spa/include/spa/utils/defs.h
+++ b/spa/include/spa/utils/defs.h
@@ -160,6 +160,14 @@ struct spa_param_info {
 #define SPA_EXPORT
 #endif
 
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#define SPA_RESTRICT restrict
+#elif defined(__GNUC__) && __GNUC__ >= 4
+#define SPA_RESTRICT __restrict__
+#else
+#define SPA_RESTRICT
+#endif
+
 #define SPA_ROUND_DOWN_N(num,align)	((num) & ~((align) - 1))
 #define SPA_ROUND_UP_N(num,align)	SPA_ROUND_DOWN_N((num) + ((align) - 1),align)
 
diff --git a/spa/plugins/audioconvert/channelmix-ops-sse.c b/spa/plugins/audioconvert/channelmix-ops-sse.c
index 6cb6881c6..fdd2e49ff 100644
--- a/spa/plugins/audioconvert/channelmix-ops-sse.c
+++ b/spa/plugins/audioconvert/channelmix-ops-sse.c
@@ -224,7 +224,7 @@ channelmix_f32_5p1_3p1_sse(void *data, int n_dst, void *dst[n_dst],
 	float **s = (float **) src;
         __m128 mix = _mm_set1_ps(v * 0.5f);
         __m128 vol = _mm_set1_ps(v);
-	__m128 avg;
+	__m128 avg[2];
 	float *sFL = s[0], *sFR = s[1], *sFC = s[2], *sLFE = s[3], *sSL = s[4], *sSR = s[5];
 	float *dFL = d[0], *dFR = d[1], *dFC = d[2], *dLFE = d[3];
 
@@ -238,7 +238,7 @@ channelmix_f32_5p1_3p1_sse(void *data, int n_dst, void *dst[n_dst],
 	    SPA_IS_ALIGNED(dFR, 16) &&
 	    SPA_IS_ALIGNED(dFC, 16) &&
 	    SPA_IS_ALIGNED(dLFE, 16))
-		unrolled = n_samples / 4;
+		unrolled = n_samples / 8;
 	else
 		unrolled = 0;
 
@@ -247,37 +247,49 @@ channelmix_f32_5p1_3p1_sse(void *data, int n_dst, void *dst[n_dst],
 			memset(d[i], 0, n_samples * sizeof(float));
 	}
 	else if (v == VOLUME_NORM) {
-		for(n = 0; unrolled--; n += 4) {
-			avg = _mm_add_ps(_mm_load_ps(&sFL[n]), _mm_load_ps(&sSL[n]));
-			_mm_store_ps(&dFL[n], _mm_mul_ps(avg, mix));
-			avg = _mm_add_ps(_mm_load_ps(&sFR[n]), _mm_load_ps(&sSR[n]));
-			_mm_store_ps(&dFR[n], _mm_mul_ps(avg, mix));
+		for(n = 0; unrolled--; n += 8) {
+			avg[0] = _mm_add_ps(_mm_load_ps(&sFL[n]), _mm_load_ps(&sSL[n]));
+			avg[1] = _mm_add_ps(_mm_load_ps(&sFL[n+4]), _mm_load_ps(&sSL[n+4]));
+			_mm_store_ps(&dFL[n], _mm_mul_ps(avg[0], mix));
+			_mm_store_ps(&dFL[n+4], _mm_mul_ps(avg[1], mix));
+			avg[0] = _mm_add_ps(_mm_load_ps(&sFR[n]), _mm_load_ps(&sSR[n]));
+			avg[1] = _mm_add_ps(_mm_load_ps(&sFR[n+4]), _mm_load_ps(&sSR[n+4]));
+			_mm_store_ps(&dFR[n], _mm_mul_ps(avg[0], mix));
+			_mm_store_ps(&dFR[n+4], _mm_mul_ps(avg[1], mix));
 			_mm_store_ps(&dFC[n], _mm_load_ps(&sFC[n]));
+			_mm_store_ps(&dFC[n+4], _mm_load_ps(&sFC[n+4]));
 			_mm_store_ps(&dLFE[n], _mm_load_ps(&sLFE[n]));
+			_mm_store_ps(&dLFE[n+4], _mm_load_ps(&sLFE[n+4]));
 		}
 		for(; n < n_samples; n++) {
-			avg = _mm_add_ss(_mm_load_ss(&sFL[n]), _mm_load_ss(&sSL[n]));
-			_mm_store_ss(&dFL[n], _mm_mul_ss(avg, mix));
-			avg = _mm_add_ss(_mm_load_ss(&sFR[n]), _mm_load_ss(&sSR[n]));
-			_mm_store_ss(&dFR[n], _mm_mul_ss(avg, mix));
+			avg[0] = _mm_add_ss(_mm_load_ss(&sFL[n]), _mm_load_ss(&sSL[n]));
+			_mm_store_ss(&dFL[n], _mm_mul_ss(avg[0], mix));
+			avg[0] = _mm_add_ss(_mm_load_ss(&sFR[n]), _mm_load_ss(&sSR[n]));
+			_mm_store_ss(&dFR[n], _mm_mul_ss(avg[0], mix));
 			_mm_store_ss(&dFC[n], _mm_load_ss(&sFC[n]));
 			_mm_store_ss(&dLFE[n], _mm_load_ss(&sLFE[n]));
 		}
 	}
 	else {
-		for(n = 0; unrolled--; n += 4) {
-			avg = _mm_add_ps(_mm_load_ps(&sFL[n]), _mm_load_ps(&sSL[n]));
-			_mm_store_ps(&dFL[n], _mm_mul_ps(avg, mix));
-			avg = _mm_add_ps(_mm_load_ps(&sFR[n]), _mm_load_ps(&sSR[n]));
-			_mm_store_ps(&dFR[n], _mm_mul_ps(avg, mix));
+		for(n = 0; unrolled--; n += 8) {
+			avg[0] = _mm_add_ps(_mm_load_ps(&sFL[n]), _mm_load_ps(&sSL[n]));
+			avg[1] = _mm_add_ps(_mm_load_ps(&sFL[n+4]), _mm_load_ps(&sSL[n+4]));
+			_mm_store_ps(&dFL[n], _mm_mul_ps(avg[0], mix));
+			_mm_store_ps(&dFL[n+4], _mm_mul_ps(avg[1], mix));
+			avg[0] = _mm_add_ps(_mm_load_ps(&sFR[n]), _mm_load_ps(&sSR[n]));
+			avg[1] = _mm_add_ps(_mm_load_ps(&sFR[n+4]), _mm_load_ps(&sSR[n+4]));
+			_mm_store_ps(&dFR[n], _mm_mul_ps(avg[0], mix));
+			_mm_store_ps(&dFR[n+4], _mm_mul_ps(avg[1], mix));
 			_mm_store_ps(&dFC[n], _mm_mul_ps(_mm_load_ps(&sFC[n]), vol));
+			_mm_store_ps(&dFC[n+4], _mm_mul_ps(_mm_load_ps(&sFC[n+4]), vol));
 			_mm_store_ps(&dLFE[n], _mm_mul_ps(_mm_load_ps(&sLFE[n]), vol));
+			_mm_store_ps(&dLFE[n+4], _mm_mul_ps(_mm_load_ps(&sLFE[n+4]), vol));
 		}
 		for(; n < n_samples; n++) {
-			avg = _mm_add_ss(_mm_load_ss(&sFL[n]), _mm_load_ss(&sSL[n]));
-			_mm_store_ss(&dFL[n], _mm_mul_ss(avg, mix));
-			avg = _mm_add_ss(_mm_load_ss(&sFR[n]), _mm_load_ss(&sSR[n]));
-			_mm_store_ss(&dFR[n], _mm_mul_ss(avg, mix));
+			avg[0] = _mm_add_ss(_mm_load_ss(&sFL[n]), _mm_load_ss(&sSL[n]));
+			_mm_store_ss(&dFL[n], _mm_mul_ss(avg[0], mix));
+			avg[0] = _mm_add_ss(_mm_load_ss(&sFR[n]), _mm_load_ss(&sSR[n]));
+			_mm_store_ss(&dFR[n], _mm_mul_ss(avg[0], mix));
 			_mm_store_ss(&dFC[n], _mm_mul_ss(_mm_load_ss(&sFC[n]), vol));
 			_mm_store_ss(&dLFE[n], _mm_mul_ss(_mm_load_ss(&sLFE[n]), vol));
 		}
diff --git a/spa/plugins/audioconvert/fmt-ops-sse2.c b/spa/plugins/audioconvert/fmt-ops-sse2.c
index e0cef1fc8..9528c410d 100644
--- a/spa/plugins/audioconvert/fmt-ops-sse2.c
+++ b/spa/plugins/audioconvert/fmt-ops-sse2.c
@@ -30,12 +30,12 @@
 #include <emmintrin.h>
 
 static void
-conv_s16_to_f32d_1_sse2(void *data, void *dst[], const void *src, int n_channels, int n_samples)
+conv_s16_to_f32d_1_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t *s = src;
 	float **d = (float **) dst;
 	float *d0 = d[0];
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S16_SCALE);
 
@@ -64,12 +64,12 @@ conv_s16_to_f32d_1_sse2(void *data, void *dst[], const void *src, int n_channels
 }
 
 static void
-conv_s16_to_f32d_2_sse2(void *data, void *dst[], const void *src, int n_channels, int n_samples)
+conv_s16_to_f32d_2_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t *s = src;
 	float **d = (float **) dst;
 	float *d0 = d[0], *d1 = d[1];
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128i in, t[2];
 	__m128 out[2], factor = _mm_set1_ps(1.0f / S16_SCALE);
 
@@ -110,10 +110,10 @@ conv_s16_to_f32d_2_sse2(void *data, void *dst[], const void *src, int n_channels
 }
 
 static void
-conv_s16_to_f32d_sse2(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s16_to_f32d_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t *s = src[0];
-	int i = 0;
+	uint32_t i = 0;
 
 	for(; i + 1 < n_channels; i += 2)
 		conv_s16_to_f32d_2_sse2(data, &dst[i], &s[i], n_channels, n_samples);
@@ -122,16 +122,16 @@ conv_s16_to_f32d_sse2(void *data, void *dst[], const void *src[], int n_channels
 }
 
 static void
-conv_s24_to_f32d_1_sse2(void *data, void *dst[], const void *src, int n_channels, int n_samples)
+conv_s24_to_f32d_1_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src;
 	float **d = (float **) dst;
 	float *d0 = d[0];
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
 
-	if (SPA_IS_ALIGNED(d0, 16) && n_samples > 4) {
+	if (SPA_IS_ALIGNED(d0, 16)) {
 		unrolled = n_samples / 4;
 		if ((n_samples & 3) == 0)
 			unrolled--;
@@ -161,22 +161,167 @@ conv_s24_to_f32d_1_sse2(void *data, void *dst[], const void *src, int n_channels
 }
 
 static void
-conv_s24_to_f32d_sse2(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_to_f32d_2_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
+{
+	const uint8_t *s = src;
+	float **d = (float **) dst;
+	float *d0 = d[0], *d1 = d[1];
+	uint32_t n, unrolled;
+	__m128i in[2];
+	__m128 out[2], factor = _mm_set1_ps(1.0f / S24_SCALE);
+
+	if (SPA_IS_ALIGNED(d0, 16)) {
+		unrolled = n_samples / 4;
+		if ((n_samples & 3) == 0)
+			unrolled--;
+	}
+	else
+		unrolled = 0;
+
+	for(n = 0; unrolled--; n += 4) {
+		in[0] = _mm_setr_epi32(
+			*((uint32_t*)&s[0 + 0*n_channels]),
+			*((uint32_t*)&s[0 + 3*n_channels]),
+			*((uint32_t*)&s[0 + 6*n_channels]),
+			*((uint32_t*)&s[0 + 9*n_channels]));
+		in[1] = _mm_setr_epi32(
+			*((uint32_t*)&s[3 + 0*n_channels]),
+			*((uint32_t*)&s[3 + 3*n_channels]),
+			*((uint32_t*)&s[3 + 6*n_channels]),
+			*((uint32_t*)&s[3 + 9*n_channels]));
+
+		in[0] = _mm_slli_epi32(in[0], 8);
+		in[1] = _mm_slli_epi32(in[1], 8);
+
+		in[0] = _mm_srai_epi32(in[0], 8);
+		in[1] = _mm_srai_epi32(in[1], 8);
+
+		out[0] = _mm_cvtepi32_ps(in[0]);
+		out[1] = _mm_cvtepi32_ps(in[1]);
+
+		out[0] = _mm_mul_ps(out[0], factor);
+		out[1] = _mm_mul_ps(out[1], factor);
+
+		_mm_store_ps(&d0[n], out[0]);
+		_mm_store_ps(&d1[n], out[1]);
+
+		s += 12 * n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
+		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
+		out[0] = _mm_mul_ss(out[0], factor);
+		out[1] = _mm_mul_ss(out[1], factor);
+		_mm_store_ss(&d0[n], out[0]);
+		_mm_store_ss(&d1[n], out[1]);
+		s += 3 * n_channels;
+	}
+}
+static void
+conv_s24_to_f32d_4_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
+{
+	const uint8_t *s = src;
+	float **d = (float **) dst;
+	float *d0 = d[0], *d1 = d[1], *d2 = d[2], *d3 = d[3];
+	uint32_t n, unrolled;
+	__m128i in[4];
+	__m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE);
+
+	if (SPA_IS_ALIGNED(d0, 16)) {
+		unrolled = n_samples / 4;
+		if ((n_samples & 3) == 0)
+			unrolled--;
+	}
+	else
+		unrolled = 0;
+
+	for(n = 0; unrolled--; n += 4) {
+		in[0] = _mm_setr_epi32(
+			*((uint32_t*)&s[0 + 0*n_channels]),
+			*((uint32_t*)&s[0 + 3*n_channels]),
+			*((uint32_t*)&s[0 + 6*n_channels]),
+			*((uint32_t*)&s[0 + 9*n_channels]));
+		in[1] = _mm_setr_epi32(
+			*((uint32_t*)&s[3 + 0*n_channels]),
+			*((uint32_t*)&s[3 + 3*n_channels]),
+			*((uint32_t*)&s[3 + 6*n_channels]),
+			*((uint32_t*)&s[3 + 9*n_channels]));
+		in[2] = _mm_setr_epi32(
+			*((uint32_t*)&s[6 + 0*n_channels]),
+			*((uint32_t*)&s[6 + 3*n_channels]),
+			*((uint32_t*)&s[6 + 6*n_channels]),
+			*((uint32_t*)&s[6 + 9*n_channels]));
+		in[3] = _mm_setr_epi32(
+			*((uint32_t*)&s[9 + 0*n_channels]),
+			*((uint32_t*)&s[9 + 3*n_channels]),
+			*((uint32_t*)&s[9 + 6*n_channels]),
+			*((uint32_t*)&s[9 + 9*n_channels]));
+
+		in[0] = _mm_slli_epi32(in[0], 8);
+		in[1] = _mm_slli_epi32(in[1], 8);
+		in[2] = _mm_slli_epi32(in[2], 8);
+		in[3] = _mm_slli_epi32(in[3], 8);
+
+		in[0] = _mm_srai_epi32(in[0], 8);
+		in[1] = _mm_srai_epi32(in[1], 8);
+		in[2] = _mm_srai_epi32(in[2], 8);
+		in[3] = _mm_srai_epi32(in[3], 8);
+
+		out[0] = _mm_cvtepi32_ps(in[0]);
+		out[1] = _mm_cvtepi32_ps(in[1]);
+		out[2] = _mm_cvtepi32_ps(in[2]);
+		out[3] = _mm_cvtepi32_ps(in[3]);
+
+		out[0] = _mm_mul_ps(out[0], factor);
+		out[1] = _mm_mul_ps(out[1], factor);
+		out[2] = _mm_mul_ps(out[2], factor);
+		out[3] = _mm_mul_ps(out[3], factor);
+
+		_mm_store_ps(&d0[n], out[0]);
+		_mm_store_ps(&d1[n], out[1]);
+		_mm_store_ps(&d2[n], out[2]);
+		_mm_store_ps(&d3[n], out[3]);
+
+		s += 12 * n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
+		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
+		out[2] = _mm_cvtsi32_ss(out[2], read_s24(s+6));
+		out[3] = _mm_cvtsi32_ss(out[3], read_s24(s+9));
+		out[0] = _mm_mul_ss(out[0], factor);
+		out[1] = _mm_mul_ss(out[1], factor);
+		out[2] = _mm_mul_ss(out[2], factor);
+		out[3] = _mm_mul_ss(out[3], factor);
+		_mm_store_ss(&d0[n], out[0]);
+		_mm_store_ss(&d1[n], out[1]);
+		_mm_store_ss(&d2[n], out[2]);
+		_mm_store_ss(&d3[n], out[3]);
+		s += 3 * n_channels;
+	}
+}
+
+static void
+conv_s24_to_f32d_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int8_t *s = src[0];
-	int i = 0;
+	uint32_t i = 0;
 
+	for(; i + 3 < n_channels; i += 4)
+		conv_s24_to_f32d_4_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_s24_to_f32d_2_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
 	for(; i < n_channels; i++)
 		conv_s24_to_f32d_1_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
 }
 
 static void
-conv_f32d_to_s32_1_sse2(void *data, void *dst, const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32_1_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	const float *s0 = s[0];
 	int32_t *d = dst;
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128 in[1];
 	__m128i out[4];
 	__m128 int_max = _mm_set1_ps(S24_MAX_F);
@@ -212,12 +357,12 @@ conv_f32d_to_s32_1_sse2(void *data, void *dst, const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s32_2_sse2(void *data, void *dst, const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32_2_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	const float *s0 = s[0], *s1 = s[1];
 	int32_t *d = dst;
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128 in[2];
 	__m128i out[2], t[2];
 	__m128 int_max = _mm_set1_ps(S24_MAX_F);
@@ -265,12 +410,12 @@ conv_f32d_to_s32_2_sse2(void *data, void *dst, const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s32_4_sse2(void *data, void *dst, const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32_4_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	const float *s0 = s[0], *s1 = s[1], *s2 = s[2], *s3 = s[3];
 	int32_t *d = dst;
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128 in[4];
 	__m128i out[4], t[4];
 	__m128 int_max = _mm_set1_ps(S24_MAX_F);
@@ -279,7 +424,8 @@ conv_f32d_to_s32_4_sse2(void *data, void *dst, const void *src[], int n_channels
 	if (SPA_IS_ALIGNED(s0, 16) &&
 	    SPA_IS_ALIGNED(s1, 16) &&
 	    SPA_IS_ALIGNED(s2, 16) &&
-	    SPA_IS_ALIGNED(s3, 16))
+	    SPA_IS_ALIGNED(s3, 16) &&
+	    SPA_IS_ALIGNED(d, 16))
 		unrolled = n_samples / 4;
 	else
 		unrolled = 0;
@@ -310,10 +456,10 @@ conv_f32d_to_s32_4_sse2(void *data, void *dst, const void *src[], int n_channels
 		out[2] = _mm_unpacklo_epi64(t[2], t[3]);
 		out[3] = _mm_unpackhi_epi64(t[2], t[3]);
 
-		_mm_storeu_si128((__m128i*)(d + 0*n_channels), out[0]);
-		_mm_storeu_si128((__m128i*)(d + 1*n_channels), out[1]);
-		_mm_storeu_si128((__m128i*)(d + 2*n_channels), out[2]);
-		_mm_storeu_si128((__m128i*)(d + 3*n_channels), out[3]);
+		_mm_store_si128((__m128i*)(d + 0*n_channels), out[0]);
+		_mm_store_si128((__m128i*)(d + 1*n_channels), out[1]);
+		_mm_store_si128((__m128i*)(d + 2*n_channels), out[2]);
+		_mm_store_si128((__m128i*)(d + 3*n_channels), out[3]);
 		d += 4*n_channels;
 	}
 	for(; n < n_samples; n++) {
@@ -335,10 +481,10 @@ conv_f32d_to_s32_4_sse2(void *data, void *dst, const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s32_sse2(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	int32_t *d = dst[0];
-	int i = 0;
+	uint32_t i = 0;
 
 	for(; i + 3 < n_channels; i += 4)
 		conv_f32d_to_s32_4_sse2(data, &d[i], &src[i], n_channels, n_samples);
@@ -349,12 +495,12 @@ conv_f32d_to_s32_sse2(void *data, void *dst[], const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s16_1_sse2(void *data, void *dst, const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s16_1_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	const float *s0 = s[0];
 	int16_t *d = dst;
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128 in[2];
 	__m128i out[2];
 	__m128 int_max = _mm_set1_ps(S16_MAX_F);
@@ -391,12 +537,12 @@ conv_f32d_to_s16_1_sse2(void *data, void *dst, const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s16_2_sse2(void *data, void *dst, const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s16_2_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	const float *s0 = s[0], *s1 = s[1];
 	int16_t *d = dst;
-	int n, unrolled;
+	uint32_t n, unrolled;
 	__m128 in[2];
 	__m128i out[4], t[2];
 	__m128 int_max = _mm_set1_ps(S16_MAX_F);
@@ -441,11 +587,76 @@ conv_f32d_to_s16_2_sse2(void *data, void *dst, const void *src[], int n_channels
 }
 
 static void
-conv_f32d_to_s16_sse2(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s16_4_sse2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
+{
+	const float **s = (const float **) src;
+	const float *s0 = s[0], *s1 = s[1], *s2 = s[2], *s3 = s[3];
+	int16_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[4];
+	__m128i out[4], t[4];
+	__m128 int_max = _mm_set1_ps(S16_MAX_F);
+        __m128 int_min = _mm_sub_ps(_mm_setzero_ps(), int_max);
+
+	if (SPA_IS_ALIGNED(s0, 16) &&
+	    SPA_IS_ALIGNED(s1, 16) &&
+	    SPA_IS_ALIGNED(s2, 16) &&
+	    SPA_IS_ALIGNED(s3, 16))
+		unrolled = n_samples / 4;
+	else
+		unrolled = 0;
+
+	for(n = 0; unrolled--; n += 4) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), int_max);
+		in[1] = _mm_mul_ps(_mm_load_ps(&s1[n]), int_max);
+		in[2] = _mm_mul_ps(_mm_load_ps(&s2[n]), int_max);
+		in[3] = _mm_mul_ps(_mm_load_ps(&s3[n]), int_max);
+
+		t[0] = _mm_cvtps_epi32(in[0]);
+		t[1] = _mm_cvtps_epi32(in[1]);
+		t[2] = _mm_cvtps_epi32(in[2]);
+		t[3] = _mm_cvtps_epi32(in[3]);
+
+		t[0] = _mm_packs_epi32(t[0], t[2]);
+		t[1] = _mm_packs_epi32(t[1], t[3]);
+
+		out[0] = _mm_unpacklo_epi16(t[0], t[1]);
+		out[1] = _mm_unpackhi_epi16(t[0], t[1]);
+		out[2] = _mm_unpacklo_epi32(out[0], out[1]);
+		out[3] = _mm_unpackhi_epi32(out[0], out[1]);
+
+		_mm_storel_pi((__m64*)(d + 0*n_channels), (__m128)out[2]);
+		_mm_storeh_pi((__m64*)(d + 1*n_channels), (__m128)out[2]);
+		_mm_storel_pi((__m64*)(d + 2*n_channels), (__m128)out[3]);
+		_mm_storeh_pi((__m64*)(d + 3*n_channels), (__m128)out[3]);
+
+		d += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		in[0] = _mm_mul_ss(_mm_load_ss(&s0[n]), int_max);
+		in[1] = _mm_mul_ss(_mm_load_ss(&s1[n]), int_max);
+		in[2] = _mm_mul_ss(_mm_load_ss(&s2[n]), int_max);
+		in[3] = _mm_mul_ss(_mm_load_ss(&s3[n]), int_max);
+		in[0] = _mm_min_ss(int_max, _mm_max_ss(in[0], int_min));
+		in[1] = _mm_min_ss(int_max, _mm_max_ss(in[1], int_min));
+		in[2] = _mm_min_ss(int_max, _mm_max_ss(in[2], int_min));
+		in[3] = _mm_min_ss(int_max, _mm_max_ss(in[3], int_min));
+		d[0] = _mm_cvtss_si32(in[0]);
+		d[1] = _mm_cvtss_si32(in[1]);
+		d[2] = _mm_cvtss_si32(in[2]);
+		d[3] = _mm_cvtss_si32(in[3]);
+		d += n_channels;
+	}
+}
+
+static void
+conv_f32d_to_s16_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	int16_t *d = dst[0];
-	int i = 0;
+	uint32_t i = 0;
 
+	for(; i + 3 < n_channels; i += 4)
+		conv_f32d_to_s16_4_sse2(data, &d[i], &src[i], n_channels, n_samples);
 	for(; i + 1 < n_channels; i += 2)
 		conv_f32d_to_s16_2_sse2(data, &d[i], &src[i], n_channels, n_samples);
 	for(; i < n_channels; i++)
diff --git a/spa/plugins/audioconvert/fmt-ops-sse41.c b/spa/plugins/audioconvert/fmt-ops-sse41.c
new file mode 100644
index 000000000..567f368bb
--- /dev/null
+++ b/spa/plugins/audioconvert/fmt-ops-sse41.c
@@ -0,0 +1,79 @@
+/* Spa
+ *
+ * Copyright © 2018 Wim Taymans
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include <string.h>
+#include <stdio.h>
+
+#include <spa/utils/defs.h>
+
+#include <smmintrin.h>
+
+static void
+conv_s24_to_f32d_1_sse41(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
+{
+	const uint8_t *s = src;
+	float **d = (float **) dst;
+	float *d0 = d[0];
+	uint32_t n, unrolled;
+	__m128i in;
+	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
+
+	if (SPA_IS_ALIGNED(d0, 16))
+		unrolled = n_samples / 4;
+	else
+		unrolled = 0;
+
+	for(n = 0; unrolled--; n += 4) {
+		in = _mm_insert_epi32(in, *((uint32_t*)&s[0 * n_channels]), 0);
+		in = _mm_insert_epi32(in, *((uint32_t*)&s[3 * n_channels]), 1);
+		in = _mm_insert_epi32(in, *((uint32_t*)&s[6 * n_channels]), 2);
+		in = _mm_insert_epi32(in, *((uint32_t*)&s[9 * n_channels]), 3);
+		in = _mm_slli_epi32(in, 8);
+		in = _mm_srai_epi32(in, 8);
+		out = _mm_cvtepi32_ps(in);
+		out = _mm_mul_ps(out, factor);
+		_mm_storeu_ps(&d0[n], out);
+		s += 12 * n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out = _mm_cvtsi32_ss(out, read_s24(s));
+		out = _mm_mul_ss(out, factor);
+		_mm_store_ss(&d0[n], out);
+		s += 3 * n_channels;
+	}
+}
+
+static void
+conv_s24_to_f32d_sse41(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
+{
+	const int8_t *s = src[0];
+	uint32_t i = 0;
+
+	for(; i + 3 < n_channels; i += 4)
+		conv_s24_to_f32d_4_ssse3(data, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_s24_to_f32d_2_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i < n_channels; i++)
+		conv_s24_to_f32d_1_sse41(data, &dst[i], &s[3*i], n_channels, n_samples);
+}
diff --git a/spa/plugins/audioconvert/fmt-ops-ssse3.c b/spa/plugins/audioconvert/fmt-ops-ssse3.c
new file mode 100644
index 000000000..93b30178e
--- /dev/null
+++ b/spa/plugins/audioconvert/fmt-ops-ssse3.c
@@ -0,0 +1,108 @@
+/* Spa
+ *
+ * Copyright © 2018 Wim Taymans
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include <string.h>
+#include <stdio.h>
+
+#include <spa/utils/defs.h>
+
+#include <smmintrin.h>
+
+static void
+conv_s24_to_f32d_4_ssse3(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src, uint32_t n_channels, uint32_t n_samples)
+{
+	const uint8_t *s = src;
+	float **d = (float **) dst;
+	float *d0 = d[0], *d1 = d[1], *d2 = d[2], *d3 = d[3];
+	uint32_t n, unrolled;
+	__m128i in[4];
+	__m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE);
+	//const __m128i mask = _mm_setr_epi8(-1, 0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11);
+	const __m128i mask = _mm_set_epi8(15, 14, 13, -1, 12, 11, 10, -1, 9, 8, 7, -1, 6, 5, 4, -1);
+
+	if (SPA_IS_ALIGNED(d0, 16))
+		unrolled = n_samples / 4;
+	else
+		unrolled = 0;
+
+	for(n = 0; unrolled--; n += 4) {
+                in[0] = _mm_loadu_si128((__m128i*)(s + 0*n_channels));
+                in[1] = _mm_loadu_si128((__m128i*)(s + 3*n_channels));
+                in[2] = _mm_loadu_si128((__m128i*)(s + 6*n_channels));
+                in[3] = _mm_loadu_si128((__m128i*)(s + 9*n_channels));
+		in[0] = _mm_shuffle_epi8(in[0], mask);
+		in[1] = _mm_shuffle_epi8(in[1], mask);
+		in[2] = _mm_shuffle_epi8(in[2], mask);
+		in[3] = _mm_shuffle_epi8(in[3], mask);
+		in[0] = _mm_srai_epi32(in[0], 8);
+		in[1] = _mm_srai_epi32(in[1], 8);
+		in[2] = _mm_srai_epi32(in[2], 8);
+		in[3] = _mm_srai_epi32(in[3], 8);
+		out[0] = _mm_cvtepi32_ps(in[0]);
+		out[1] = _mm_cvtepi32_ps(in[1]);
+		out[2] = _mm_cvtepi32_ps(in[2]);
+		out[3] = _mm_cvtepi32_ps(in[3]);
+		out[0] = _mm_mul_ps(out[0], factor);
+		out[1] = _mm_mul_ps(out[1], factor);
+		out[2] = _mm_mul_ps(out[2], factor);
+		out[3] = _mm_mul_ps(out[3], factor);
+
+		_MM_TRANSPOSE4_PS(out[0], out[1], out[2], out[3]);
+
+		_mm_store_ps(&d0[n], out[0]);
+		_mm_store_ps(&d1[n], out[1]);
+		_mm_store_ps(&d2[n], out[2]);
+		_mm_store_ps(&d3[n], out[3]);
+		s += 12 * n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
+		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
+		out[2] = _mm_cvtsi32_ss(out[2], read_s24(s+6));
+		out[3] = _mm_cvtsi32_ss(out[3], read_s24(s+9));
+		out[0] = _mm_mul_ss(out[0], factor);
+		out[1] = _mm_mul_ss(out[1], factor);
+		out[2] = _mm_mul_ss(out[2], factor);
+		out[3] = _mm_mul_ss(out[3], factor);
+		_mm_store_ss(&d0[n], out[0]);
+		_mm_store_ss(&d1[n], out[1]);
+		_mm_store_ss(&d2[n], out[2]);
+		_mm_store_ss(&d3[n], out[3]);
+		s += 3 * n_channels;
+	}
+}
+
+static void
+conv_s24_to_f32d_ssse3(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
+{
+	const int8_t *s = src[0];
+	uint32_t i = 0;
+
+	for(; i + 3 < n_channels; i += 4)
+		conv_s24_to_f32d_4_ssse3(data, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_s24_to_f32d_2_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i < n_channels; i++)
+		conv_s24_to_f32d_1_sse2(data, &dst[i], &s[3*i], n_channels, n_samples);
+}
diff --git a/spa/plugins/audioconvert/fmt-ops.c b/spa/plugins/audioconvert/fmt-ops.c
index 9017ecf96..ff0be8862 100644
--- a/spa/plugins/audioconvert/fmt-ops.c
+++ b/spa/plugins/audioconvert/fmt-ops.c
@@ -83,68 +83,74 @@ static inline void write_s24(void *dst, int32_t val)
 #if defined (__SSE2__)
 #include "fmt-ops-sse2.c"
 #endif
+#if defined (__SSSE3__)
+#include "fmt-ops-ssse3.c"
+#endif
+#if defined (__SSE4_1__)
+#include "fmt-ops-sse41.c"
+#endif
 
 static void
-conv_copy8d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy8d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i;
+	uint32_t i;
 	for (i = 0; i < n_channels; i++)
 		memcpy(dst[i], src[i], n_samples);
 }
 
 static void
-conv_copy8(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy8(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	memcpy(dst[0], src[0], n_samples * n_channels);
 }
 
 
 static void
-conv_copy16d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy16d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i;
+	uint32_t i;
 	for (i = 0; i < n_channels; i++)
 		memcpy(dst[i], src[i], n_samples * sizeof(int16_t));
 }
 
 static void
-conv_copy16(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy16(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	memcpy(dst[0], src[0], n_samples * sizeof(int16_t) * n_channels);
 }
 
 static void
-conv_copy24d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy24d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i;
+	uint32_t i;
 	for (i = 0; i < n_channels; i++)
 		memcpy(dst[i], src[i], n_samples * 3);
 }
 
 static void
-conv_copy24(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy24(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	memcpy(dst[0], src[0], n_samples * 3 * n_channels);
 }
 
 static void
-conv_copy32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i;
+	uint32_t i;
 	for (i = 0; i < n_channels; i++)
 		memcpy(dst[i], src[i], n_samples * sizeof(int32_t));
 }
 
 static void
-conv_copy32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_copy32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	memcpy(dst[0], src[0], n_samples * sizeof(int32_t) * n_channels);
 }
 
 static void
-conv_u8d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_u8d_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const uint8_t *s = src[i];
@@ -156,17 +162,17 @@ conv_u8d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_u8_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_u8_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_u8d_to_f32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_u8_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_u8_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src[0];
 	float **d = (float **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -175,11 +181,11 @@ conv_u8_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_u8d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_u8d_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t **s = (const uint8_t **) src;
 	float *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -188,9 +194,9 @@ conv_u8d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s16d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s16d_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const int16_t *s = src[i];
@@ -201,17 +207,17 @@ conv_s16d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_s16_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s16_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_s16d_to_f32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_s16_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s16_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t *s = src[0];
 	float **d = (float **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -220,11 +226,11 @@ conv_s16_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s16d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s16d_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t **s = (const int16_t **) src;
 	float *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -233,9 +239,9 @@ conv_s16d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s32d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s32d_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const int32_t *s = src[i];
@@ -247,17 +253,17 @@ conv_s32d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_s32_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s32_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_s32d_to_f32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_s32_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s32_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t *s = src[0];
 	float **d = (float **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -266,11 +272,11 @@ conv_s32_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s32d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s32d_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t **s = (const int32_t **) src;
 	float *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -279,9 +285,9 @@ conv_s32d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s24d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24d_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const int8_t *s = src[i];
@@ -295,17 +301,17 @@ conv_s24d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_s24_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_s24d_to_f32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_s24_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src[0];
 	float **d = (float **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -316,11 +322,11 @@ conv_s24_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s24d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24d_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t **s = (const uint8_t **) src;
 	float *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -330,9 +336,9 @@ conv_s24d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_s24_32d_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_32d_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const int32_t *s = src[i];
@@ -344,17 +350,17 @@ conv_s24_32d_to_f32d(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-conv_s24_32_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_32_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_s24_32d_to_f32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_s24_32_to_f32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_32_to_f32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t *s = src[0];
 	float **d = (float **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -363,11 +369,11 @@ conv_s24_32_to_f32d(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-conv_s24_32d_to_f32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_s24_32d_to_f32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t **s = (const int32_t **) src;
 	float *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -376,9 +382,9 @@ conv_s24_32d_to_f32(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-conv_f32d_to_u8d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_u8d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const float *s = src[i];
@@ -390,17 +396,17 @@ conv_f32d_to_u8d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32_to_u8(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_u8(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_f32d_to_u8d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_f32_to_u8d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_u8d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float *s = src[0];
 	uint8_t **d = (uint8_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -409,11 +415,11 @@ conv_f32_to_u8d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_u8(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_u8(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	uint8_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -422,9 +428,9 @@ conv_f32d_to_u8(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_s16d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s16d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const float *s = src[i];
@@ -436,17 +442,17 @@ conv_f32d_to_s16d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_f32_to_s16(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s16(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_f32d_to_s16d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_f32_to_s16d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s16d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float *s = src[0];
 	int16_t **d = (int16_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -455,11 +461,11 @@ conv_f32_to_s16d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_s16(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s16(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	int16_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -468,9 +474,9 @@ conv_f32d_to_s16(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_s32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const float *s = src[i];
@@ -482,17 +488,17 @@ conv_f32d_to_s32d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_f32_to_s32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_f32d_to_s32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_f32_to_s32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float *s = src[0];
 	int32_t **d = (int32_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -501,11 +507,11 @@ conv_f32_to_s32d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_s32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	int32_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -516,9 +522,9 @@ conv_f32d_to_s32(void *data, void *dst[], const void *src[], int n_channels, int
 
 
 static void
-conv_f32d_to_s24d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s24d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const float *s = src[i];
@@ -532,17 +538,17 @@ conv_f32d_to_s24d(void *data, void *dst[], const void *src[], int n_channels, in
 }
 
 static void
-conv_f32_to_s24(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s24(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_f32d_to_s24d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_f32_to_s24d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s24d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float *s = src[0];
 	uint8_t **d = (uint8_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -552,11 +558,11 @@ conv_f32_to_s24d(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-conv_f32d_to_s24(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s24(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	uint8_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -568,9 +574,9 @@ conv_f32d_to_s24(void *data, void *dst[], const void *src[], int n_channels, int
 
 
 static void
-conv_f32d_to_s24_32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s24_32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
-	int i, j;
+	uint32_t i, j;
 
 	for (i = 0; i < n_channels; i++) {
 		const float *s = src[i];
@@ -582,17 +588,17 @@ conv_f32d_to_s24_32d(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-conv_f32_to_s24_32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s24_32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	conv_f32d_to_s24_32d(data, dst, src, 1, n_samples * n_channels);
 }
 
 static void
-conv_f32_to_s24_32d(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32_to_s24_32d(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float *s = src[0];
 	int32_t **d = (int32_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -601,11 +607,11 @@ conv_f32_to_s24_32d(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-conv_f32d_to_s24_32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+conv_f32d_to_s24_32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const float **s = (const float **) src;
 	int32_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -614,11 +620,11 @@ conv_f32d_to_s24_32(void *data, void *dst[], const void *src[], int n_channels,
 }
 
 static void
-deinterleave_8(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+deinterleave_8(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src[0];
 	uint8_t **d = (uint8_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -627,11 +633,11 @@ deinterleave_8(void *data, void *dst[], const void *src[], int n_channels, int n
 }
 
 static void
-deinterleave_16(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+deinterleave_16(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint16_t *s = src[0];
 	uint16_t **d = (uint16_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -640,11 +646,11 @@ deinterleave_16(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-deinterleave_24(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+deinterleave_24(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src[0];
 	uint8_t **d = (uint8_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -655,11 +661,11 @@ deinterleave_24(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-deinterleave_32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+deinterleave_32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const uint32_t *s = src[0];
 	uint32_t **d = (uint32_t **) dst;
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -668,11 +674,11 @@ deinterleave_32(void *data, void *dst[], const void *src[], int n_channels, int
 }
 
 static void
-interleave_8(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+interleave_8(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int8_t **s = (const int8_t **) src;
 	uint8_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -681,11 +687,11 @@ interleave_8(void *data, void *dst[], const void *src[], int n_channels, int n_s
 }
 
 static void
-interleave_16(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+interleave_16(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t **s = (const int16_t **) src;
 	uint16_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -694,11 +700,11 @@ interleave_16(void *data, void *dst[], const void *src[], int n_channels, int n_
 }
 
 static void
-interleave_24(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+interleave_24(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int8_t **s = (const int8_t **) src;
 	uint8_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++) {
@@ -709,11 +715,11 @@ interleave_24(void *data, void *dst[], const void *src[], int n_channels, int n_
 }
 
 static void
-interleave_32(void *data, void *dst[], const void *src[], int n_channels, int n_samples)
+interleave_32(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[], uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t **s = (const int32_t **) src;
 	uint32_t *d = dst[0];
-	int i, j;
+	uint32_t i, j;
 
 	for (j = 0; j < n_samples; j++) {
 		for (i = 0; i < n_channels; i++)
@@ -722,8 +728,8 @@ interleave_32(void *data, void *dst[], const void *src[], int n_channels, int n_
 }
 
 
-typedef void (*convert_func_t) (void *data, void *dst[], const void *src[],
-		int n_channels, int n_samples);
+typedef void (*convert_func_t) (void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples);
 
 static const struct conv_info {
 	uint32_t src_fmt;
@@ -761,6 +767,12 @@ static const struct conv_info {
 
 	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32, 0, conv_s24_to_f32 },
 	{ SPA_AUDIO_FORMAT_S24P, SPA_AUDIO_FORMAT_F32P, 0, conv_s24d_to_f32d },
+#if defined (__SSE4_1__)
+	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32P, FEATURE_SSE2, conv_s24_to_f32d_sse41 },
+#endif
+#if defined (__SSSE3__)
+	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32P, FEATURE_SSE2, conv_s24_to_f32d_ssse3 },
+#endif
 #if defined (__SSE2__)
 	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32P, FEATURE_SSE2, conv_s24_to_f32d_sse2 },
 #endif