diff --git a/spa/plugins/audioconvert/fmt-ops-avx2.c b/spa/plugins/audioconvert/fmt-ops-avx2.c
new file mode 100644
index 000000000..014687144
--- /dev/null
+++ b/spa/plugins/audioconvert/fmt-ops-avx2.c
@@ -0,0 +1,772 @@
+/* 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 "fmt-ops.h"
+
+#include <immintrin.h>
+
+static void
+conv_s16_to_f32d_1s_avx2(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 *d0 = dst[0];
+	uint32_t n, unrolled;
+	__m128i in;
+	__m128 out, factor = _mm_set1_ps(1.0f / S16_SCALE);
+
+	if (SPA_LIKELY(SPA_IS_ALIGNED(d0, 16)))
+		unrolled = n_samples & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in = _mm_insert_epi16(in, s[0*n_channels], 1);
+		in = _mm_insert_epi16(in, s[1*n_channels], 3);
+		in = _mm_insert_epi16(in, s[2*n_channels], 5);
+		in = _mm_insert_epi16(in, s[3*n_channels], 7);
+		in = _mm_srai_epi32(in, 16);
+		out = _mm_cvtepi32_ps(in);
+		out = _mm_mul_ps(out, factor);
+		_mm_store_ps(&d0[n], out);
+		s += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out = _mm_cvtsi32_ss(out, s[0]);
+		out = _mm_mul_ss(out, factor);
+		_mm_store_ss(&d0[n], out);
+		s += n_channels;
+	}
+}
+
+void
+conv_s16_to_f32d_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	const int16_t *s = src[0];
+	uint32_t i = 0, n_channels = conv->n_channels;
+
+	for(; i < n_channels; i++)
+		conv_s16_to_f32d_1s_avx2(conv, &dst[i], &s[i], n_channels, n_samples);
+}
+
+void
+conv_s16_to_f32d_2_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	const int16_t *s = src[0];
+	float *d0 = dst[0], *d1 = dst[1];
+	uint32_t n, unrolled;
+	__m128i in[2], t[4];
+	__m128 out[4], factor = _mm_set1_ps(1.0f / S16_SCALE);
+
+	if (SPA_IS_ALIGNED(s, 16) &&
+	    SPA_IS_ALIGNED(d0, 16) &&
+	    SPA_IS_ALIGNED(d1, 16))
+		unrolled = n_samples & ~7;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 8) {
+		in[0] = _mm_load_si128((__m128i*)(s + 0));
+		in[1] = _mm_load_si128((__m128i*)(s + 8));
+
+		t[0] = _mm_slli_epi32(in[0], 16);
+		t[0] = _mm_srai_epi32(t[0], 16);
+		out[0] = _mm_cvtepi32_ps(t[0]);
+		out[0] = _mm_mul_ps(out[0], factor);
+
+		t[1] = _mm_srai_epi32(in[0], 16);
+		out[1] = _mm_cvtepi32_ps(t[1]);
+		out[1] = _mm_mul_ps(out[1], factor);
+
+		t[2] = _mm_slli_epi32(in[1], 16);
+		t[2] = _mm_srai_epi32(t[2], 16);
+		out[2] = _mm_cvtepi32_ps(t[2]);
+		out[2] = _mm_mul_ps(out[2], factor);
+
+		t[3] = _mm_srai_epi32(in[1], 16);
+		out[3] = _mm_cvtepi32_ps(t[3]);
+		out[3] = _mm_mul_ps(out[3], factor);
+
+		_mm_store_ps(&d0[n + 0], out[0]);
+		_mm_store_ps(&d1[n + 0], out[1]);
+		_mm_store_ps(&d0[n + 4], out[2]);
+		_mm_store_ps(&d1[n + 4], out[3]);
+
+		s += 16;
+	}
+	for(; n < n_samples; n++) {
+		out[0] = _mm_cvtsi32_ss(out[0], s[0]);
+		out[0] = _mm_mul_ss(out[0], factor);
+		out[1] = _mm_cvtsi32_ss(out[1], s[1]);
+		out[1] = _mm_mul_ss(out[1], factor);
+		_mm_store_ss(&d0[n], out[0]);
+		_mm_store_ss(&d1[n], out[1]);
+		s += 2;
+	}
+}
+
+void
+conv_s24_to_f32d_1s_avx2(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 *d0 = dst[0];
+	uint32_t n, unrolled;
+	__m128i in;
+	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
+
+	if (SPA_IS_ALIGNED(d0, 16) && n_samples > 0) {
+		unrolled = n_samples & ~3;
+		if ((n_samples & 3) == 0)
+			unrolled -= 4;
+	}
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in = _mm_setr_epi32(
+			*((uint32_t*)&s[0 * n_channels]),
+			*((uint32_t*)&s[3 * n_channels]),
+			*((uint32_t*)&s[6 * n_channels]),
+			*((uint32_t*)&s[9 * n_channels]));
+		in = _mm_slli_epi32(in, 8);
+		in = _mm_srai_epi32(in, 8);
+		out = _mm_cvtepi32_ps(in);
+		out = _mm_mul_ps(out, factor);
+		_mm_store_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_2s_avx2(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 *d0 = dst[0], *d1 = dst[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) &&
+	    SPA_IS_ALIGNED(d1, 16) &&
+	    n_samples > 0) {
+		unrolled = n_samples & ~3;
+		if ((n_samples & 3) == 0)
+			unrolled -= 4;
+	}
+	else
+		unrolled = 0;
+
+	for(n = 0; n < 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_4s_avx2(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 *d0 = dst[0], *d1 = dst[1], *d2 = dst[2], *d3 = dst[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) &&
+	    SPA_IS_ALIGNED(d1, 16) &&
+	    SPA_IS_ALIGNED(d2, 16) &&
+	    SPA_IS_ALIGNED(d3, 16) &&
+	    n_samples > 0) {
+		unrolled = n_samples & ~3;
+		if ((n_samples & 3) == 0)
+			unrolled -= 4;
+	}
+	else
+		unrolled = 0;
+
+	for(n = 0; n < 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;
+	}
+}
+
+void
+conv_s24_to_f32d_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	const int8_t *s = src[0];
+	uint32_t i = 0, n_channels = conv->n_channels;
+
+	for(; i + 3 < n_channels; i += 4)
+		conv_s24_to_f32d_4s_avx2(conv, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_s24_to_f32d_2s_avx2(conv, &dst[i], &s[3*i], n_channels, n_samples);
+	for(; i < n_channels; i++)
+		conv_s24_to_f32d_1s_avx2(conv, &dst[i], &s[3*i], n_channels, n_samples);
+}
+
+
+void
+conv_s32_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src,
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const int32_t *s = src;
+	float *d0 = dst[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 & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in = _mm_setr_epi32(s[0*n_channels],
+				    s[1*n_channels],
+				    s[2*n_channels],
+				    s[3*n_channels]);
+		in = _mm_srai_epi32(in, 8);
+		out = _mm_cvtepi32_ps(in);
+		out = _mm_mul_ps(out, factor);
+		_mm_store_ps(&d0[n], out);
+		s += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		out = _mm_cvtsi32_ss(out, s[0]>>8);
+		out = _mm_mul_ss(out, factor);
+		_mm_store_ss(&d0[n], out);
+		s += n_channels;
+	}
+}
+
+void
+conv_s32_to_f32d_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	const int32_t *s = src[0];
+	uint32_t i = 0, n_channels = conv->n_channels;
+
+	for(; i < n_channels; i++)
+		conv_s32_to_f32d_1s_avx2(conv, &dst[i], &s[i], n_channels, n_samples);
+}
+
+static void
+conv_f32d_to_s32_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0];
+	int32_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[1];
+	__m128i out[4];
+	__m128 scale = _mm_set1_ps(S32_SCALE);
+	__m128 int_min = _mm_set1_ps(S32_MIN);
+
+	if (SPA_IS_ALIGNED(s0, 16))
+		unrolled = n_samples & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale);
+		in[0] = _mm_min_ps(in[0], int_min);
+		out[0] = _mm_cvtps_epi32(in[0]);
+		out[1] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(0, 3, 2, 1));
+		out[2] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(1, 0, 3, 2));
+		out[3] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(2, 1, 0, 3));
+
+		d[0*n_channels] = _mm_cvtsi128_si32(out[0]);
+		d[1*n_channels] = _mm_cvtsi128_si32(out[1]);
+		d[2*n_channels] = _mm_cvtsi128_si32(out[2]);
+		d[3*n_channels] = _mm_cvtsi128_si32(out[3]);
+		d += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		in[0] = _mm_load_ss(&s0[n]);
+		in[0] = _mm_mul_ss(in[0], scale);
+		in[0] = _mm_min_ss(in[0], int_min);
+		*d = _mm_cvtss_si32(in[0]);
+		d += n_channels;
+	}
+}
+
+static void
+conv_f32d_to_s32_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0], *s1 = src[1];
+	int32_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[2];
+	__m128i out[2], t[2];
+	__m128 scale = _mm_set1_ps(S32_SCALE);
+	__m128 int_min = _mm_set1_ps(S32_MIN);
+
+	if (SPA_IS_ALIGNED(s0, 16) &&
+	    SPA_IS_ALIGNED(s1, 16))
+		unrolled = n_samples & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale);
+		in[1] = _mm_mul_ps(_mm_load_ps(&s1[n]), scale);
+
+		in[0] = _mm_min_ps(in[0], int_min);
+		in[1] = _mm_min_ps(in[1], int_min);
+
+		out[0] = _mm_cvtps_epi32(in[0]);
+		out[1] = _mm_cvtps_epi32(in[1]);
+
+		t[0] = _mm_unpacklo_epi32(out[0], out[1]);
+		t[1] = _mm_unpackhi_epi32(out[0], out[1]);
+
+		_mm_storel_pd((double*)(d + 0*n_channels), (__m128d)t[0]);
+		_mm_storeh_pd((double*)(d + 1*n_channels), (__m128d)t[0]);
+		_mm_storel_pd((double*)(d + 2*n_channels), (__m128d)t[1]);
+		_mm_storeh_pd((double*)(d + 3*n_channels), (__m128d)t[1]);
+		d += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		in[0] = _mm_load_ss(&s0[n]);
+		in[1] = _mm_load_ss(&s1[n]);
+
+		in[0] = _mm_unpacklo_ps(in[0], in[1]);
+
+		in[0] = _mm_mul_ps(in[0], scale);
+		in[0] = _mm_min_ps(in[0], int_min);
+		out[0] = _mm_cvtps_epi32(in[0]);
+		_mm_storel_epi64((__m128i*)d, out[0]);
+		d += n_channels;
+	}
+}
+
+static void
+conv_f32d_to_s32_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0], *s1 = src[1], *s2 = src[2], *s3 = src[3];
+	int32_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[4];
+	__m128i out[4];
+	__m128 scale = _mm_set1_ps(S32_SCALE);
+	__m128 int_min = _mm_set1_ps(S32_MIN);
+
+	if (SPA_IS_ALIGNED(s0, 16) &&
+	    SPA_IS_ALIGNED(s1, 16) &&
+	    SPA_IS_ALIGNED(s2, 16) &&
+	    SPA_IS_ALIGNED(s3, 16))
+		unrolled = n_samples & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 4) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), scale);
+		in[1] = _mm_mul_ps(_mm_load_ps(&s1[n]), scale);
+		in[2] = _mm_mul_ps(_mm_load_ps(&s2[n]), scale);
+		in[3] = _mm_mul_ps(_mm_load_ps(&s3[n]), scale);
+
+		in[0] = _mm_min_ps(in[0], int_min);
+		in[1] = _mm_min_ps(in[1], int_min);
+		in[2] = _mm_min_ps(in[2], int_min);
+		in[3] = _mm_min_ps(in[3], int_min);
+
+		_MM_TRANSPOSE4_PS(in[0], in[1], in[2], in[3]);
+
+		out[0] = _mm_cvtps_epi32(in[0]);
+		out[1] = _mm_cvtps_epi32(in[1]);
+		out[2] = _mm_cvtps_epi32(in[2]);
+		out[3] = _mm_cvtps_epi32(in[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]);
+		d += 4*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		in[0] = _mm_load_ss(&s0[n]);
+		in[1] = _mm_load_ss(&s1[n]);
+		in[2] = _mm_load_ss(&s2[n]);
+		in[3] = _mm_load_ss(&s3[n]);
+
+		in[0] = _mm_unpacklo_ps(in[0], in[2]);
+		in[1] = _mm_unpacklo_ps(in[1], in[3]);
+		in[0] = _mm_unpacklo_ps(in[0], in[1]);
+
+		in[0] = _mm_mul_ps(in[0], scale);
+		in[0] = _mm_min_ps(in[0], int_min);
+		out[0] = _mm_cvtps_epi32(in[0]);
+		_mm_storeu_si128((__m128i*)d, out[0]);
+		d += n_channels;
+	}
+}
+
+void
+conv_f32d_to_s32_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	int32_t *d = dst[0];
+	uint32_t i = 0, n_channels = conv->n_channels;
+
+	for(; i + 3 < n_channels; i += 4)
+		conv_f32d_to_s32_4s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_f32d_to_s32_2s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+	for(; i < n_channels; i++)
+		conv_f32d_to_s32_1s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+}
+
+static void
+conv_f32d_to_s16_1s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0];
+	int16_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[2];
+	__m128i out[2];
+	__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))
+		unrolled = n_samples & ~7;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 8) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n]), int_max);
+		in[1] = _mm_mul_ps(_mm_load_ps(&s0[n+4]), int_max);
+		out[0] = _mm_cvtps_epi32(in[0]);
+		out[1] = _mm_cvtps_epi32(in[1]);
+		out[0] = _mm_packs_epi32(out[0], out[1]);
+
+		d[0*n_channels] = _mm_extract_epi16(out[0], 0);
+		d[1*n_channels] = _mm_extract_epi16(out[0], 1);
+		d[2*n_channels] = _mm_extract_epi16(out[0], 2);
+		d[3*n_channels] = _mm_extract_epi16(out[0], 3);
+		d[4*n_channels] = _mm_extract_epi16(out[0], 4);
+		d[5*n_channels] = _mm_extract_epi16(out[0], 5);
+		d[6*n_channels] = _mm_extract_epi16(out[0], 6);
+		d[7*n_channels] = _mm_extract_epi16(out[0], 7);
+		d += 8*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		in[0] = _mm_mul_ss(_mm_load_ss(&s0[n]), int_max);
+		in[0] = _mm_min_ss(int_max, _mm_max_ss(in[0], int_min));
+		*d = _mm_cvtss_si32(in[0]);
+		d += n_channels;
+	}
+}
+
+static void
+conv_f32d_to_s16_2s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0], *s1 = src[1];
+	int16_t *d = dst;
+	uint32_t n, unrolled;
+	__m128 in[2];
+	__m128i out[4], t[2];
+	__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))
+		unrolled = n_samples & ~3;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < 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);
+
+		t[0] = _mm_cvtps_epi32(in[0]);
+		t[1] = _mm_cvtps_epi32(in[1]);
+
+		t[0] = _mm_packs_epi32(t[0], t[0]);
+		t[1] = _mm_packs_epi32(t[1], t[1]);
+
+		out[0] = _mm_unpacklo_epi16(t[0], t[1]);
+		out[1] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(0, 3, 2, 1));
+		out[2] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(1, 0, 3, 2));
+		out[3] = _mm_shuffle_epi32(out[0], _MM_SHUFFLE(2, 1, 0, 3));
+
+		*((int32_t*)(d + 0*n_channels)) = _mm_cvtsi128_si32(out[0]);
+		*((int32_t*)(d + 1*n_channels)) = _mm_cvtsi128_si32(out[1]);
+		*((int32_t*)(d + 2*n_channels)) = _mm_cvtsi128_si32(out[2]);
+		*((int32_t*)(d + 3*n_channels)) = _mm_cvtsi128_si32(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[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));
+		d[0] = _mm_cvtss_si32(in[0]);
+		d[1] = _mm_cvtss_si32(in[1]);
+		d += n_channels;
+	}
+}
+
+static void
+conv_f32d_to_s16_4s_avx2(void *data, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
+		uint32_t n_channels, uint32_t n_samples)
+{
+	const float *s0 = src[0], *s1 = src[1], *s2 = src[2], *s3 = src[3];
+	int16_t *d = dst;
+	uint32_t n, unrolled;
+	__m256 in[4];
+	__m256i out[4], t[4];
+	__m256 int_max = _mm256_set1_ps(S16_MAX_F);
+
+	if (SPA_IS_ALIGNED(s0, 16) &&
+	    SPA_IS_ALIGNED(s1, 16) &&
+	    SPA_IS_ALIGNED(s2, 16) &&
+	    SPA_IS_ALIGNED(s3, 16))
+		unrolled = n_samples & ~7;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 8) {
+		in[0] = _mm256_mul_ps(_mm256_load_ps(&s0[n]), int_max);
+		in[1] = _mm256_mul_ps(_mm256_load_ps(&s1[n]), int_max);
+		in[2] = _mm256_mul_ps(_mm256_load_ps(&s2[n]), int_max);
+		in[3] = _mm256_mul_ps(_mm256_load_ps(&s3[n]), int_max);
+
+		t[0] = _mm256_cvtps_epi32(in[0]);
+		t[1] = _mm256_cvtps_epi32(in[1]);
+		t[2] = _mm256_cvtps_epi32(in[2]);
+		t[3] = _mm256_cvtps_epi32(in[3]);
+
+		t[0] = _mm256_packs_epi32(t[0], t[2]);
+		t[1] = _mm256_packs_epi32(t[1], t[3]);
+
+		out[0] = _mm256_unpacklo_epi16(t[0], t[1]);
+		out[1] = _mm256_unpackhi_epi16(t[0], t[1]);
+		out[2] = _mm256_unpacklo_epi32(out[0], out[1]);
+		out[3] = _mm256_unpackhi_epi32(out[0], out[1]);
+
+		*(int64_t*)(d + 0*n_channels) = _mm256_extract_epi64(out[2], 0);
+		*(int64_t*)(d + 1*n_channels) = _mm256_extract_epi64(out[2], 1);
+		*(int64_t*)(d + 2*n_channels) = _mm256_extract_epi64(out[3], 0);
+		*(int64_t*)(d + 3*n_channels) = _mm256_extract_epi64(out[3], 1);
+		*(int64_t*)(d + 4*n_channels) = _mm256_extract_epi64(out[2], 2);
+		*(int64_t*)(d + 5*n_channels) = _mm256_extract_epi64(out[2], 3);
+		*(int64_t*)(d + 6*n_channels) = _mm256_extract_epi64(out[3], 2);
+		*(int64_t*)(d + 7*n_channels) = _mm256_extract_epi64(out[3], 3);
+
+		d += 8*n_channels;
+	}
+	for(; n < n_samples; n++) {
+		__m128 in[4];
+		__m128 int_max = _mm_set1_ps(S16_MAX_F);
+	        __m128 int_min = _mm_sub_ps(_mm_setzero_ps(), int_max);
+
+		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;
+	}
+}
+
+void
+conv_f32d_to_s16_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	int16_t *d = dst[0];
+	uint32_t i = 0, n_channels = conv->n_channels;
+
+	for(; i + 3 < n_channels; i += 4)
+		conv_f32d_to_s16_4s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+	for(; i + 1 < n_channels; i += 2)
+		conv_f32d_to_s16_2s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+	for(; i < n_channels; i++)
+		conv_f32d_to_s16_1s_avx2(conv, &d[i], &src[i], n_channels, n_samples);
+}
+
+void
+conv_f32d_to_s16_2_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
+		uint32_t n_samples)
+{
+	const float *s0 = src[0], *s1 = src[1];
+	int16_t *d = dst[0];
+	uint32_t n, unrolled;
+	__m128 in[4];
+	__m128i out[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))
+		unrolled = n_samples & ~7;
+	else
+		unrolled = 0;
+
+	for(n = 0; n < unrolled; n += 8) {
+		in[0] = _mm_mul_ps(_mm_load_ps(&s0[n+0]), int_max);
+		in[1] = _mm_mul_ps(_mm_load_ps(&s1[n+0]), int_max);
+		in[2] = _mm_mul_ps(_mm_load_ps(&s0[n+4]), int_max);
+		in[3] = _mm_mul_ps(_mm_load_ps(&s1[n+4]), int_max);
+
+		out[0] = _mm_cvtps_epi32(in[0]);
+		out[1] = _mm_cvtps_epi32(in[1]);
+		out[2] = _mm_cvtps_epi32(in[2]);
+		out[3] = _mm_cvtps_epi32(in[3]);
+
+		out[0] = _mm_packs_epi32(out[0], out[2]);
+		out[1] = _mm_packs_epi32(out[1], out[3]);
+
+		out[2] = _mm_unpacklo_epi16(out[0], out[1]);
+		out[3] = _mm_unpackhi_epi16(out[0], out[1]);
+
+		_mm_storeu_si128((__m128i*)(d+0), out[2]);
+		_mm_storeu_si128((__m128i*)(d+8), out[3]);
+
+		d += 16;
+	}
+	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[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));
+		d[0] = _mm_cvtss_si32(in[0]);
+		d[1] = _mm_cvtss_si32(in[1]);
+		d += 2;
+	}
+}
diff --git a/spa/plugins/audioconvert/fmt-ops-sse2.c b/spa/plugins/audioconvert/fmt-ops-sse2.c
index 104bf6827..93ac0fa5d 100644
--- a/spa/plugins/audioconvert/fmt-ops-sse2.c
+++ b/spa/plugins/audioconvert/fmt-ops-sse2.c
@@ -31,8 +31,7 @@ conv_s16_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		uint32_t n_channels, uint32_t n_samples)
 {
 	const int16_t *s = src;
-	float **d = (float **) dst;
-	float *d0 = d[0];
+	float *d0 = dst[0];
 	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S16_SCALE);
@@ -77,8 +76,7 @@ conv_s16_to_f32d_2_sse2(struct convert *conv, void * SPA_RESTRICT dst[], const v
 		uint32_t n_samples)
 {
 	const int16_t *s = src[0];
-	float **d = (float **) dst;
-	float *d0 = d[0], *d1 = d[1];
+	float *d0 = dst[0], *d1 = dst[1];
 	uint32_t n, unrolled;
 	__m128i in[2], t[4];
 	__m128 out[4], factor = _mm_set1_ps(1.0f / S16_SCALE);
@@ -135,8 +133,7 @@ conv_s24_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src;
-	float **d = (float **) dst;
-	float *d0 = d[0];
+	float *d0 = dst[0];
 	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
@@ -175,8 +172,7 @@ conv_s24_to_f32d_2s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src;
-	float **d = (float **) dst;
-	float *d0 = d[0], *d1 = d[1];
+	float *d0 = dst[0], *d1 = dst[1];
 	uint32_t n, unrolled;
 	__m128i in[2];
 	__m128 out[2], factor = _mm_set1_ps(1.0f / S24_SCALE);
@@ -235,8 +231,7 @@ conv_s24_to_f32d_4s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		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];
+	float *d0 = dst[0], *d1 = dst[1], *d2 = dst[2], *d3 = dst[3];
 	uint32_t n, unrolled;
 	__m128i in[4];
 	__m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE);
@@ -340,8 +335,7 @@ conv_s32_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		uint32_t n_channels, uint32_t n_samples)
 {
 	const int32_t *s = src;
-	float **d = (float **) dst;
-	float *d0 = d[0];
+	float *d0 = dst[0];
 	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
@@ -385,8 +379,7 @@ static void
 conv_f32d_to_s32_1s_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];
+	const float *s0 = src[0];
 	int32_t *d = dst;
 	uint32_t n, unrolled;
 	__m128 in[1];
@@ -426,8 +419,7 @@ static void
 conv_f32d_to_s32_2s_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];
+	const float *s0 = src[0], *s1 = src[1];
 	int32_t *d = dst;
 	uint32_t n, unrolled;
 	__m128 in[2];
@@ -478,8 +470,7 @@ static void
 conv_f32d_to_s32_4s_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];
+	const float *s0 = src[0], *s1 = src[1], *s2 = src[2], *s3 = src[3];
 	int32_t *d = dst;
 	uint32_t n, unrolled;
 	__m128 in[4];
@@ -556,8 +547,7 @@ static void
 conv_f32d_to_s16_1s_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];
+	const float *s0 = src[0];
 	int16_t *d = dst;
 	uint32_t n, unrolled;
 	__m128 in[2];
@@ -599,8 +589,7 @@ static void
 conv_f32d_to_s16_2s_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];
+	const float *s0 = src[0], *s1 = src[1];
 	int16_t *d = dst;
 	uint32_t n, unrolled;
 	__m128 in[2];
@@ -724,7 +713,6 @@ conv_f32d_to_s16_sse2(struct convert *conv, void * SPA_RESTRICT dst[], const voi
 		conv_f32d_to_s16_1s_sse2(conv, &d[i], &src[i], n_channels, n_samples);
 }
 
-
 void
 conv_f32d_to_s16_2_sse2(struct convert *conv, void * SPA_RESTRICT dst[], const void * SPA_RESTRICT src[],
 		uint32_t n_samples)
diff --git a/spa/plugins/audioconvert/fmt-ops-sse41.c b/spa/plugins/audioconvert/fmt-ops-sse41.c
index eef05df21..0478555e8 100644
--- a/spa/plugins/audioconvert/fmt-ops-sse41.c
+++ b/spa/plugins/audioconvert/fmt-ops-sse41.c
@@ -31,8 +31,7 @@ conv_s24_to_f32d_1s_sse41(void *data, void * SPA_RESTRICT dst[], const void * SP
 		uint32_t n_channels, uint32_t n_samples)
 {
 	const uint8_t *s = src;
-	float **d = (float **) dst;
-	float *d0 = d[0];
+	float *d0 = dst[0];
 	uint32_t n, unrolled;
 	__m128i in;
 	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
diff --git a/spa/plugins/audioconvert/fmt-ops-ssse3.c b/spa/plugins/audioconvert/fmt-ops-ssse3.c
index 6383147fc..6a7fc2e05 100644
--- a/spa/plugins/audioconvert/fmt-ops-ssse3.c
+++ b/spa/plugins/audioconvert/fmt-ops-ssse3.c
@@ -31,8 +31,7 @@ conv_s24_to_f32d_4s_ssse3(void *data, void * SPA_RESTRICT dst[], const void * SP
 		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];
+	float *d0 = dst[0], *d1 = dst[1], *d2 = dst[2], *d3 = dst[3];
 	uint32_t n, unrolled;
 	__m128i in[4];
 	__m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE);
diff --git a/spa/plugins/audioconvert/fmt-ops.c b/spa/plugins/audioconvert/fmt-ops.c
index 0cadb8f32..6fd247e64 100644
--- a/spa/plugins/audioconvert/fmt-ops.c
+++ b/spa/plugins/audioconvert/fmt-ops.c
@@ -55,6 +55,10 @@ static struct conv_info conv_table[] =
 
 	{ SPA_AUDIO_FORMAT_S16, SPA_AUDIO_FORMAT_F32, 0, 0, conv_s16_to_f32_c },
 	{ SPA_AUDIO_FORMAT_S16P, SPA_AUDIO_FORMAT_F32P, 0, 0, conv_s16d_to_f32d_c },
+#if defined (HAVE_AVX2)
+	{ SPA_AUDIO_FORMAT_S16, SPA_AUDIO_FORMAT_F32P, 2, SPA_CPU_FLAG_SSE2, conv_s16_to_f32d_2_avx2 },
+	{ SPA_AUDIO_FORMAT_S16, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSE2, conv_s16_to_f32d_avx2 },
+#endif
 #if defined (HAVE_SSE2)
 	{ SPA_AUDIO_FORMAT_S16, SPA_AUDIO_FORMAT_F32P, 2, SPA_CPU_FLAG_SSE2, conv_s16_to_f32d_2_sse2 },
 	{ SPA_AUDIO_FORMAT_S16, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSE2, conv_s16_to_f32d_sse2 },
@@ -67,6 +71,9 @@ static struct conv_info conv_table[] =
 	{ SPA_AUDIO_FORMAT_F32, SPA_AUDIO_FORMAT_F32P, 0, 0, conv_deinterleave_32_c },
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_F32, 0, 0, conv_interleave_32_c },
 
+#if defined (HAVE_AVX2)
+	{ SPA_AUDIO_FORMAT_S32, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSE2, conv_s32_to_f32d_avx2 },
+#endif
 #if defined (HAVE_SSE2)
 	{ SPA_AUDIO_FORMAT_S32, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSE2, conv_s32_to_f32d_sse2 },
 #endif
@@ -77,6 +84,9 @@ static struct conv_info conv_table[] =
 
 	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32, 0, 0, conv_s24_to_f32_c },
 	{ SPA_AUDIO_FORMAT_S24P, SPA_AUDIO_FORMAT_F32P, 0, 0, conv_s24d_to_f32d_c },
+#if defined (HAVE_AVX2)
+	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSE2, conv_s24_to_f32d_avx2 },
+#endif
 #if defined (HAVE_SSSE3)
 //	{ SPA_AUDIO_FORMAT_S24, SPA_AUDIO_FORMAT_F32P, 0, SPA_CPU_FLAG_SSSE3, conv_s24_to_f32d_ssse3 },
 #endif
@@ -105,6 +115,10 @@ static struct conv_info conv_table[] =
 	{ SPA_AUDIO_FORMAT_F32, SPA_AUDIO_FORMAT_S16, 0, 0, conv_f32_to_s16_c },
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S16P, 0, 0, conv_f32d_to_s16d_c },
 	{ SPA_AUDIO_FORMAT_F32, SPA_AUDIO_FORMAT_S16P, 0, 0, conv_f32_to_s16d_c },
+#if defined (HAVE_AVX2)
+	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S16, 2, SPA_CPU_FLAG_SSE2, conv_f32d_to_s16_2_avx2 },
+	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S16, 0, SPA_CPU_FLAG_SSE2, conv_f32d_to_s16_avx2 },
+#endif
 #if defined (HAVE_SSE2)
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S16, 2, SPA_CPU_FLAG_SSE2, conv_f32d_to_s16_2_sse2 },
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S16, 0, SPA_CPU_FLAG_SSE2, conv_f32d_to_s16_sse2 },
@@ -114,6 +128,9 @@ static struct conv_info conv_table[] =
 	{ SPA_AUDIO_FORMAT_F32, SPA_AUDIO_FORMAT_S32, 0, 0, conv_f32_to_s32_c },
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S32P, 0, 0, conv_f32d_to_s32d_c },
 	{ SPA_AUDIO_FORMAT_F32, SPA_AUDIO_FORMAT_S32P, 0, 0, conv_f32_to_s32d_c },
+#if defined (HAVE_AVX2)
+	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S32, 0, SPA_CPU_FLAG_SSE2, conv_f32d_to_s32_avx2 },
+#endif
 #if defined (HAVE_SSE2)
 	{ SPA_AUDIO_FORMAT_F32P, SPA_AUDIO_FORMAT_S32, 0, SPA_CPU_FLAG_SSE2, conv_f32d_to_s32_sse2 },
 #endif
diff --git a/spa/plugins/audioconvert/fmt-ops.h b/spa/plugins/audioconvert/fmt-ops.h
index d8bb509c7..e296ea508 100644
--- a/spa/plugins/audioconvert/fmt-ops.h
+++ b/spa/plugins/audioconvert/fmt-ops.h
@@ -200,5 +200,13 @@ DEFINE_FUNCTION(s24_to_f32d, ssse3);
 #endif
 #if defined(HAVE_SSE41)
 DEFINE_FUNCTION(s24_to_f32d, sse41);
-
+#endif
+#if defined(HAVE_AVX2)
+DEFINE_FUNCTION(s16_to_f32d_2, avx2);
+DEFINE_FUNCTION(s16_to_f32d, avx2);
+DEFINE_FUNCTION(s24_to_f32d, avx2);
+DEFINE_FUNCTION(s32_to_f32d, avx2);
+DEFINE_FUNCTION(f32d_to_s32, avx2);
+DEFINE_FUNCTION(f32d_to_s16_2, avx2);
+DEFINE_FUNCTION(f32d_to_s16, avx2);
 #endif
diff --git a/spa/plugins/audioconvert/meson.build b/spa/plugins/audioconvert/meson.build
index 35515bd9a..ccb06abd0 100644
--- a/spa/plugins/audioconvert/meson.build
+++ b/spa/plugins/audioconvert/meson.build
@@ -74,6 +74,16 @@ if have_avx and have_fma
 	simd_cargs += ['-DHAVE_AVX', '-DHAVE_FMA']
 	simd_dependencies += audioconvert_avx
 endif
+if have_avx2
+	audioconvert_avx2 = static_library('audioconvert_avx2',
+		['fmt-ops-avx2.c'],
+		c_args : [avx2_args, '-O3', '-DHAVE_AVX2'],
+		include_directories : [spa_inc],
+		install : false
+	)
+	simd_cargs += ['-DHAVE_AVX2']
+	simd_dependencies += audioconvert_avx2
+endif
 
 audioconvertlib = shared_library('spa-audioconvert',
                           audioconvert_sources,