diff --git a/libfreerdp/primitives/prim_YUV.c b/libfreerdp/primitives/prim_YUV.c
index e3dfb2e46..27e3bab5b 100644
--- a/libfreerdp/primitives/prim_YUV.c
+++ b/libfreerdp/primitives/prim_YUV.c
@@ -280,47 +280,6 @@ static pstatus_t general_YUV444SplitToYUV420(
 	return PRIMITIVES_SUCCESS;
 }
 
-/**
- * | R |   ( | 256     0    403 | |    Y    | )
- * | G | = ( | 256   -48   -120 | | U - 128 | ) >> 8
- * | B |   ( | 256   475      0 | | V - 128 | )
- */
-static INLINE INT32 C(INT32 Y)
-{
-	return (Y) -   0L;
-}
-
-static INLINE INT32 D(INT32 U)
-{
-	return (U) - 128L;
-}
-
-static INLINE INT32 E(INT32 V)
-{
-	return (V) - 128L;
-}
-
-static INLINE BYTE YUV2R(INT32 Y, INT32 U, INT32 V)
-{
-	const INT32 r = (256L * C(Y) +   0L * D(U) + 403L * E(V));
-	const INT32 r8 = r >> 8L;
-	return CLIP(r8);
-}
-
-static INLINE BYTE YUV2G(INT32 Y, INT32 U, INT32 V)
-{
-	const INT32 g = (256L * C(Y) -  48L * D(U) - 120L * E(V));
-	const INT32 g8 = g >> 8L;
-	return CLIP(g8);
-}
-
-static INLINE BYTE YUV2B(INT32 Y, INT32 U, INT32 V)
-{
-	const INT32 b = (256L * C(Y) + 475L * D(U) +   0L * E(V));
-	const INT32 b8 = b >> 8L;
-	return CLIP(b8);
-}
-
 static pstatus_t general_YUV444ToRGB_8u_P3AC4R_general(
     const BYTE* pSrc[3], const UINT32 srcStep[3],
     BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
@@ -343,8 +302,8 @@ static pstatus_t general_YUV444ToRGB_8u_P3AC4R_general(
 		for (x = 0; x < nWidth; x++)
 		{
 			const BYTE Y = pY[x];
-			const INT32 U = pU[x];
-			const INT32 V = pV[x];
+			const BYTE U = pU[x];
+			const BYTE V = pV[x];
 			const BYTE r = YUV2R(Y, U, V);
 			const BYTE g = YUV2G(Y, U, V);
 			const BYTE b = YUV2B(Y, U, V);
@@ -376,8 +335,8 @@ static pstatus_t general_YUV444ToRGB_8u_P3AC4R_BGRX(
 		for (x = 0; x < nWidth; x++)
 		{
 			const BYTE Y = pY[x];
-			const INT32 U = pU[x];
-			const INT32 V = pV[x];
+			const BYTE U = pU[x];
+			const BYTE V = pV[x];
 			const BYTE r = YUV2R(Y, U, V);
 			const BYTE g = YUV2G(Y, U, V);
 			const BYTE b = YUV2B(Y, U, V);
@@ -589,7 +548,7 @@ static INLINE pstatus_t general_RGBToYUV420_BGRX(
     const BYTE* pSrc, UINT32 srcStep,
     BYTE* pDst[3], UINT32 dstStep[3], const prim_size_t* roi)
 {
-	UINT32 x, y, i, j;
+	UINT32 x, y, i;
 	size_t x1 = 0, x2 = 4, x3 = srcStep, x4 = srcStep + 4;
 	size_t y1 = 0, y2 = 1, y3 = dstStep[0], y4 = dstStep[0] + 1;
 	UINT32 max_x = roi->width - 1;
@@ -606,7 +565,6 @@ static INLINE pstatus_t general_RGBToYUV420_BGRX(
 		{
 			BYTE R, G, B;
 			INT32 Ra, Ga, Ba;
-			UINT32 color;
 			/* row 1, pixel 1 */
 			Ba = B = *(src + x1 + 0);
 			Ga = G = *(src + x1 + 1);
@@ -658,7 +616,7 @@ static INLINE pstatus_t general_RGBToYUV420_ANY(
     BYTE* pDst[3], UINT32 dstStep[3], const prim_size_t* roi)
 {
 	const UINT32 bpp = GetBytesPerPixel(srcFormat);
-	UINT32 x, y, i, j;
+	UINT32 x, y, i;
 	size_t x1 = 0, x2 = bpp, x3 = srcStep, x4 = srcStep + bpp;
 	size_t y1 = 0, y2 = 1, y3 = dstStep[0], y4 = dstStep[0] + 1;
 	UINT32 max_x = roi->width - 1;
diff --git a/libfreerdp/primitives/prim_YUV_opt.c b/libfreerdp/primitives/prim_YUV_opt.c
index f42d07ee4..6b1be6d54 100644
--- a/libfreerdp/primitives/prim_YUV_opt.c
+++ b/libfreerdp/primitives/prim_YUV_opt.c
@@ -31,22 +31,25 @@
 
 #include "prim_internal.h"
 
-static primitives_t* generic = NULL;
-
 #ifdef WITH_SSE2
 
 #include <emmintrin.h>
 #include <tmmintrin.h>
+#elif defined(WITH_NEON)
+#include <arm_neon.h>
+#endif /* WITH_SSE2 else WITH_NEON */
 
+static primitives_t* generic = NULL;
 
+#ifdef WITH_SSE2
 /****************************************************************************/
 /* SSSE3 YUV420 -> RGB conversion                                           */
 /****************************************************************************/
 
 static pstatus_t ssse3_YUV420ToRGB_BGRX(
-	const BYTE** pSrc, const UINT32* srcStep,
-	BYTE* pDst, UINT32 dstStep, UINT32 dstFormat,
-	const prim_size_t* roi)
+    const BYTE** pSrc, const UINT32* srcStep,
+    BYTE* pDst, UINT32 dstStep, UINT32 dstFormat,
+    const prim_size_t* roi)
 {
 	UINT32 lastRow, lastCol;
 	BYTE* UData, *VData, *YData;
@@ -341,9 +344,9 @@ static pstatus_t ssse3_YUV420ToRGB_BGRX(
 }
 
 static pstatus_t ssse3_YUV420ToRGB(
-	const BYTE** pSrc, const UINT32* srcStep,
-	BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
-	const prim_size_t* roi)
+    const BYTE** pSrc, const UINT32* srcStep,
+    BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
+    const prim_size_t* roi)
 {
 	switch (DstFormat)
 	{
@@ -356,8 +359,115 @@ static pstatus_t ssse3_YUV420ToRGB(
 	}
 }
 
+static pstatus_t ssse3_YUV444ToRGB_8u_P3AC4R_BGRX(
+    const BYTE** pSrc, const UINT32* srcStep,
+    BYTE* pDst, UINT32 dstStep,
+    const prim_size_t* roi)
+{
+	const UINT32 nWidth = roi->width;
+	const UINT32 nHeight = roi->height;
+	const __m128i c128 = _mm_set1_epi16(128);
+	const __m128i mapY = _mm_set_epi32(0x80800380, 0x80800280, 0x80800180, 0x80800080);
+	const __m128i map = _mm_set_epi32(0x80038002, 0x80018000, 0x80808080, 0x80808080);
+	UINT32 y;
 
+	for (y = 0; y < nHeight; y++)
+	{
+		UINT32 x;
+		__m128i* dst = (__m128i*)(pDst + dstStep * y);
+		const BYTE* YData = pSrc[0] + y * srcStep[0];
+		const BYTE* UData = pSrc[1] + y * srcStep[1];
+		const BYTE* VData = pSrc[2] + y * srcStep[2];
 
+		for (x = 0; x < nWidth; x += 4)
+		{
+			__m128i BGRX = _mm_set_epi32(0xFF000000, 0xFF000000, 0xFF000000, 0xFF000000);
+			{
+				__m128i C, D, E;
+				/* Load Y values and expand to 32 bit */
+				{
+					const __m128i Yraw = _mm_loadu_si128((__m128i*)YData);
+					C = _mm_shuffle_epi8(Yraw, mapY); /* Reorder and multiply by 256 */
+				}
+				/* Load U values and expand to 32 bit */
+				{
+					const __m128i Uraw = _mm_loadu_si128((__m128i*)UData);
+					const __m128i U = _mm_shuffle_epi8(Uraw, map); /* Reorder dcba */
+					D = _mm_sub_epi16(U, c128); /* D = U - 128 */
+				}
+				/* Load V values and expand to 32 bit */
+				{
+					const __m128i Vraw = _mm_loadu_si128((__m128i*)VData);
+					const __m128i V = _mm_shuffle_epi8(Vraw, map); /* Reorder dcba */
+					E = _mm_sub_epi16(V, c128); /* E = V - 128 */
+				}
+				/* Get the R value */
+				{
+					const __m128i c403 = _mm_set1_epi16(403);
+					const __m128i e403 = _mm_unpackhi_epi16(_mm_mullo_epi16(E, c403), _mm_mulhi_epi16(E, c403));
+					const __m128i Rs = _mm_add_epi32(C, e403);
+					const __m128i R32 = _mm_srai_epi32(Rs, 8);
+					const __m128i R16 = _mm_packs_epi32(R32, _mm_setzero_si128());
+					const __m128i R = _mm_packus_epi16(R16, _mm_setzero_si128());
+					const __m128i mask = _mm_set_epi32(0x80038080, 0x80028080, 0x80018080, 0x80008080);
+					const __m128i packed = _mm_shuffle_epi8(R, mask);
+					BGRX = _mm_or_si128(BGRX, packed);
+				}
+				/* Get the G value */
+				{
+					const __m128i c48 = _mm_set1_epi16(48);
+					const __m128i d48 = _mm_unpackhi_epi16(_mm_mullo_epi16(D, c48), _mm_mulhi_epi16(D, c48));
+					const __m128i c120 = _mm_set1_epi16(120);
+					const __m128i e120 = _mm_unpackhi_epi16(_mm_mullo_epi16(E, c120), _mm_mulhi_epi16(E, c120));
+					const __m128i de = _mm_add_epi32(d48, e120);
+					const __m128i Gs = _mm_sub_epi32(C, de);
+					const __m128i G32 = _mm_srai_epi32(Gs, 8);
+					const __m128i G16 = _mm_packs_epi32(G32, _mm_setzero_si128());
+					const __m128i G = _mm_packus_epi16(G16, _mm_setzero_si128());
+					const __m128i mask = _mm_set_epi32(0x80800380, 0x80800280, 0x80800180, 0x80800080);
+					const __m128i packed = _mm_shuffle_epi8(G, mask);
+					BGRX = _mm_or_si128(BGRX, packed);
+				}
+				/* Get the B value */
+				{
+					const __m128i c475 = _mm_set1_epi16(475);
+					const __m128i d475 = _mm_unpackhi_epi16(_mm_mullo_epi16(D, c475), _mm_mulhi_epi16(D, c475));
+					const __m128i Bs = _mm_add_epi32(C, d475);
+					const __m128i B32 = _mm_srai_epi32(Bs, 8);
+					const __m128i B16 = _mm_packs_epi32(B32, _mm_setzero_si128());
+					const __m128i B = _mm_packus_epi16(B16, _mm_setzero_si128());
+					const __m128i mask = _mm_set_epi32(0x80808003, 0x80808002, 0x80808001, 0x80808000);
+					const __m128i packed = _mm_shuffle_epi8(B, mask);
+					BGRX = _mm_or_si128(BGRX, packed);
+				}
+			}
+			_mm_storeu_si128(dst++, BGRX);
+			YData += 4;
+			UData += 4;
+			VData += 4;
+		}
+	}
+
+	return PRIMITIVES_SUCCESS;
+}
+
+static pstatus_t ssse3_YUV444ToRGB_8u_P3AC4R(const BYTE** pSrc, const UINT32* srcStep,
+        BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
+        const prim_size_t* roi)
+{
+	if (roi->width % 4 != 0)
+		return generic->YUV444ToRGB_8u_P3AC4R(pSrc, srcStep, pDst, dstStep, DstFormat, roi);
+
+	switch (DstFormat)
+	{
+		case PIXEL_FORMAT_BGRX32:
+		case PIXEL_FORMAT_BGRA32:
+			return ssse3_YUV444ToRGB_8u_P3AC4R_BGRX(pSrc, srcStep, pDst, dstStep, roi);
+
+		default:
+			return generic->YUV444ToRGB_8u_P3AC4R(pSrc, srcStep, pDst, dstStep, DstFormat, roi);
+	}
+}
 
 /****************************************************************************/
 /* SSSE3 RGB -> YUV420 conversion                                          **/
@@ -383,18 +493,22 @@ static pstatus_t ssse3_YUV420ToRGB(
  *
  */
 
-PRIM_ALIGN_128 static const BYTE bgrx_y_factors[] = {
-	   9,  92,  27,   0,   9,  92,  27,   0,   9,  92,  27,   0,   9,  92,  27,   0
+PRIM_ALIGN_128 static const BYTE bgrx_y_factors[] =
+{
+	9,  92,  27,   0,   9,  92,  27,   0,   9,  92,  27,   0,   9,  92,  27,   0
 };
-PRIM_ALIGN_128 static const BYTE bgrx_u_factors[] = {
-	  64, -49, -15,   0,  64, -49, -15,   0,  64, -49, -15,   0,  64, -49, -15,   0
+PRIM_ALIGN_128 static const BYTE bgrx_u_factors[] =
+{
+	64, -49, -15,   0,  64, -49, -15,   0,  64, -49, -15,   0,  64, -49, -15,   0
 };
-PRIM_ALIGN_128 static const BYTE bgrx_v_factors[] = {
-	  -6, -58,  64,   0,  -6, -58,  64,   0,  -6, -58,  64,   0,  -6, -58,  64,   0
+PRIM_ALIGN_128 static const BYTE bgrx_v_factors[] =
+{
+	-6, -58,  64,   0,  -6, -58,  64,   0,  -6, -58,  64,   0,  -6, -58,  64,   0
 };
 
-PRIM_ALIGN_128 static const BYTE const_buf_128b[] = {
-	 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128
+PRIM_ALIGN_128 static const BYTE const_buf_128b[] =
+{
+	128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128
 };
 
 /*
@@ -418,13 +532,12 @@ PRIM_ALIGN_128 static const BYTE rgbx_v_factors[] = {
 /* compute the luma (Y) component from a single rgb source line */
 
 static INLINE void ssse3_RGBToYUV420_BGRX_Y(
-	const BYTE* src, BYTE* dst, UINT32 width)
+    const BYTE* src, BYTE* dst, UINT32 width)
 {
 	UINT32 x;
 	__m128i y_factors, x0, x1, x2, x3;
 	const __m128i* argb = (const __m128i*) src;
 	__m128i* ydst = (__m128i*) dst;
-
 	y_factors = _mm_load_si128((__m128i*)bgrx_y_factors);
 
 	for (x = 0; x < width; x += 16)
@@ -434,24 +547,19 @@ static INLINE void ssse3_RGBToYUV420_BGRX_Y(
 		x1 = _mm_load_si128(argb++); // 2nd 4 pixels
 		x2 = _mm_load_si128(argb++); // 3rd 4 pixels
 		x3 = _mm_load_si128(argb++); // 4th 4 pixels
-
 		/* multiplications and subtotals */
 		x0 = _mm_maddubs_epi16(x0, y_factors);
 		x1 = _mm_maddubs_epi16(x1, y_factors);
 		x2 = _mm_maddubs_epi16(x2, y_factors);
 		x3 = _mm_maddubs_epi16(x3, y_factors);
-
 		/* the total sums */
 		x0 = _mm_hadd_epi16(x0, x1);
 		x2 = _mm_hadd_epi16(x2, x3);
-
 		/* shift the results */
 		x0 = _mm_srli_epi16(x0, 7);
 		x2 = _mm_srli_epi16(x2, 7);
-
 		/* pack the 16 words into bytes */
 		x0 = _mm_packus_epi16(x0, x2);
-
 		/* save to y plane */
 		_mm_storeu_si128(ydst++, x0);
 	}
@@ -460,18 +568,15 @@ static INLINE void ssse3_RGBToYUV420_BGRX_Y(
 /* compute the chrominance (UV) components from two rgb source lines */
 
 static INLINE void ssse3_RGBToYUV420_BGRX_UV(
-	const BYTE* src1, const BYTE* src2,
-	BYTE* dst1, BYTE* dst2, UINT32 width)
+    const BYTE* src1, const BYTE* src2,
+    BYTE* dst1, BYTE* dst2, UINT32 width)
 {
 	UINT32 x;
 	__m128i vector128, u_factors, v_factors, x0, x1, x2, x3, x4, x5;
-
 	const __m128i* rgb1 = (const __m128i*)src1;
 	const __m128i* rgb2 = (const __m128i*)src2;
-
 	__m64* udst = (__m64*)dst1;
 	__m64* vdst = (__m64*)dst2;
-
 	vector128 = _mm_load_si128((__m128i*)const_buf_128b);
 	u_factors = _mm_load_si128((__m128i*)bgrx_u_factors);
 	v_factors = _mm_load_si128((__m128i*)bgrx_v_factors);
@@ -482,68 +587,53 @@ static INLINE void ssse3_RGBToYUV420_BGRX_UV(
 		x0 = _mm_load_si128(rgb1++);
 		x4 = _mm_load_si128(rgb2++);
 		x0 = _mm_avg_epu8(x0, x4);
-
 		x1 = _mm_load_si128(rgb1++);
 		x4 = _mm_load_si128(rgb2++);
 		x1 = _mm_avg_epu8(x1, x4);
-
 		x2 = _mm_load_si128(rgb1++);
 		x4 = _mm_load_si128(rgb2++);
 		x2 = _mm_avg_epu8(x2, x4);
-
 		x3 = _mm_load_si128(rgb1++);
 		x4 = _mm_load_si128(rgb2++);
 		x3 = _mm_avg_epu8(x3, x4);
-
 		// subsample these 16x1 pixels into 8x1 pixels */
-
 		/**
 		 * shuffle controls
 		 * c = a[0],a[2],b[0],b[2] == 10 00 10 00 = 0x88
 		 * c = a[1],a[3],b[1],b[3] == 11 01 11 01 = 0xdd
 		 */
-
-		x4 = _mm_castps_si128( _mm_shuffle_ps(_mm_castsi128_ps(x0), _mm_castsi128_ps(x1), 0x88) );
-		x0 = _mm_castps_si128( _mm_shuffle_ps(_mm_castsi128_ps(x0), _mm_castsi128_ps(x1), 0xdd) );
+		x4 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(x0), _mm_castsi128_ps(x1), 0x88));
+		x0 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(x0), _mm_castsi128_ps(x1), 0xdd));
 		x0 = _mm_avg_epu8(x0, x4);
-
-		x4 = _mm_castps_si128( _mm_shuffle_ps(_mm_castsi128_ps(x2), _mm_castsi128_ps(x3), 0x88) );
-		x1 = _mm_castps_si128( _mm_shuffle_ps(_mm_castsi128_ps(x2), _mm_castsi128_ps(x3), 0xdd) );
+		x4 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(x2), _mm_castsi128_ps(x3), 0x88));
+		x1 = _mm_castps_si128(_mm_shuffle_ps(_mm_castsi128_ps(x2), _mm_castsi128_ps(x3), 0xdd));
 		x1 = _mm_avg_epu8(x1, x4);
-
 		/* multiplications and subtotals */
 		x2 = _mm_maddubs_epi16(x0, u_factors);
 		x3 = _mm_maddubs_epi16(x1, u_factors);
-
 		x4 = _mm_maddubs_epi16(x0, v_factors);
 		x5 = _mm_maddubs_epi16(x1, v_factors);
-
 		/* the total sums */
 		x0 = _mm_hadd_epi16(x2, x3);
 		x1 = _mm_hadd_epi16(x4, x5);
-
 		/* shift the results */
 		x0 = _mm_srai_epi16(x0, 7);
 		x1 = _mm_srai_epi16(x1, 7);
-
 		/* pack the 16 words into bytes */
 		x0 = _mm_packs_epi16(x0, x1);
-
 		/* add 128 */
 		x0 = _mm_add_epi8(x0, vector128);
-
 		/* the lower 8 bytes go to the u plane */
 		_mm_storel_pi(udst++, _mm_castsi128_ps(x0));
-
 		/* the upper 8 bytes go to the v plane */
 		_mm_storeh_pi(vdst++, _mm_castsi128_ps(x0));
 	}
 }
 
 static pstatus_t ssse3_RGBToYUV420_BGRX(
-	const BYTE* pSrc, UINT32 srcFormat, UINT32 srcStep,
-	BYTE* pDst[3], UINT32 dstStep[3],
-	const prim_size_t* roi)
+    const BYTE* pSrc, UINT32 srcFormat, UINT32 srcStep,
+    BYTE* pDst[3], UINT32 dstStep[3],
+    const prim_size_t* roi)
 {
 	UINT32 y;
 	const BYTE* argb = pSrc;
@@ -561,15 +651,13 @@ static pstatus_t ssse3_RGBToYUV420_BGRX(
 		return generic->RGBToYUV420_8u_P3AC4R(pSrc, srcFormat, srcStep, pDst, dstStep, roi);
 	}
 
-	for (y = 0; y < roi->height-1; y+=2)
+	for (y = 0; y < roi->height - 1; y += 2)
 	{
 		const BYTE* line1 = argb;
 		const BYTE* line2 = argb + srcStep;
-
 		ssse3_RGBToYUV420_BGRX_UV(line1, line2, udst, vdst, roi->width);
 		ssse3_RGBToYUV420_BGRX_Y(line1, ydst, roi->width);
 		ssse3_RGBToYUV420_BGRX_Y(line2, ydst + dstStep[0], roi->width);
-
 		argb += 2 * srcStep;
 		ydst += 2 * dstStep[0];
 		udst += 1 * dstStep[1];
@@ -587,21 +675,569 @@ static pstatus_t ssse3_RGBToYUV420_BGRX(
 }
 
 static pstatus_t ssse3_RGBToYUV420(
-	const BYTE* pSrc, UINT32 srcFormat, UINT32 srcStep,
-	BYTE* pDst[3], UINT32 dstStep[3],
-	const prim_size_t* roi)
+    const BYTE* pSrc, UINT32 srcFormat, UINT32 srcStep,
+    BYTE* pDst[3], UINT32 dstStep[3],
+    const prim_size_t* roi)
 {
 	switch (srcFormat)
 	{
 		case PIXEL_FORMAT_BGRX32:
 		case PIXEL_FORMAT_BGRA32:
 			return ssse3_RGBToYUV420_BGRX(pSrc, srcFormat, srcStep, pDst, dstStep, roi);
+
 		default:
 			return generic->RGBToYUV420_8u_P3AC4R(pSrc, srcFormat, srcStep, pDst, dstStep, roi);
 	}
 }
 
+#elif defined(WITH_NEON)
 
+static INLINE uint8x8_t neon_YUV2R(int32x4_t Ch, int32x4_t Cl,
+                                   int16x4_t Dh, int16x4_t Dl,
+                                   int16x4_t Eh, int16x4_t El)
+{
+	/* R = (256 * Y + 403 * (V - 128)) >> 8 */
+	const int16x4_t c403 = vdup_n_s16(403);
+	const int32x4_t CEh = vmlal_s16(Ch, Eh, c403);
+	const int32x4_t CEl = vmlal_s16(Cl, El, c403);
+	const int32x4_t Rh = vrshrq_n_s32(CEh, 8);
+	const int32x4_t Rl = vrshrq_n_s32(CEl, 8);
+	const int16x8_t R = vcombine_s16(vqmovn_s32(Rl), vqmovn_s32(Rh));
+	return vqmovun_s16(R);
+}
+
+static INLINE uint8x8_t neon_YUV2G(int32x4_t Ch, int32x4_t Cl,
+                                   int16x4_t Dh, int16x4_t Dl,
+                                   int16x4_t Eh, int16x4_t El)
+{
+	/* G = (256L * Y -  48 * (U - 128) - 120 * (V - 128)) >> 8 */
+	const int16x4_t c48 = vdup_n_s16(48);
+	const int16x4_t c120 = vdup_n_s16(120);
+	const int32x4_t CDh = vmlsl_s16(Ch, Dh, c48);
+	const int32x4_t CDl = vmlsl_s16(Cl, Dl, c48);
+	const int32x4_t CDEh = vmlsl_s16(CDh, Eh, c120);
+	const int32x4_t CDEl = vmlsl_s16(CDl, El, c120);
+	const int32x4_t Gh = vrshrq_n_s32(CDEh, 8);
+	const int32x4_t Gl = vrshrq_n_s32(CDEl, 8);
+	const int16x8_t G = vcombine_s16(vqmovn_s32(Gl), vqmovn_s32(Gh));
+	return vqmovun_s16(G);
+}
+
+static INLINE uint8x8_t neon_YUV2B(int32x4_t Ch, int32x4_t Cl,
+                                   int16x4_t Dh, int16x4_t Dl,
+                                   int16x4_t Eh, int16x4_t El)
+{
+	/* B = (256L * Y + 475 * (U - 128)) >> 8*/
+	const int16x4_t c475 = vdup_n_s16(475);
+	const int32x4_t CDh = vmlal_s16(Ch, Dh, c475);
+	const int32x4_t CDl = vmlal_s16(Ch, Dl, c475);
+	const int32x4_t Bh = vrshrq_n_s32(CDh, 8);
+	const int32x4_t Bl = vrshrq_n_s32(CDl, 8);
+	const int16x8_t B = vcombine_s16(vqmovn_s32(Bl), vqmovn_s32(Bh));
+	return vqmovun_s16(B);
+}
+
+static INLINE pstatus_t neon_YUV420ToX(
+    const BYTE* pSrc[3], const UINT32 srcStep[3],
+    BYTE* pDst, UINT32 dstStep,
+    const prim_size_t* roi, const uint8_t rPos,  const uint8_t gPos,
+    const uint8_t bPos, const uint8_t aPos)
+{
+	UINT32 y;
+	const UINT32 nWidth = roi->width;
+	const UINT32 nHeight = roi->height;
+	const int16x8_t c128 = vdupq_n_s16(128);
+
+	for (y = 0; y < nHeight; y += 2)
+	{
+		const uint8_t* pY1 = pSrc[0] + y * srcStep[0];
+		const uint8_t* pY2 = pY1 + srcStep[0];
+		const uint8_t* pU = pSrc[1] + (y / 2) * srcStep[1];
+		const uint8_t* pV = pSrc[2] + (y / 2) * srcStep[2];
+		uint8_t* pRGB1 = pDst + y * dstStep;
+		uint8_t* pRGB2 = pRGB1 + dstStep;
+		UINT32 x;
+		const BOOL lastY = y >= nHeight - 1;
+
+		for (x = 0; x < nWidth;)
+		{
+			const BOOL lastX = (nWidth - x) < 16;
+			const uint8x8_t Uraw = vld1_u8(pU);
+			const uint8x8x2_t Uu = vzip_u8(Uraw, Uraw);
+			const int16x8_t U1 = vreinterpretq_s16_u16(vmovl_u8(Uu.val[0]));
+			const int16x8_t U2 = vreinterpretq_s16_u16(vmovl_u8(Uu.val[1]));
+			const uint8x8_t Vraw = vld1_u8(pV);
+			const uint8x8x2_t Vu = vzip_u8(Vraw, Vraw);
+			const int16x8_t V1 = vreinterpretq_s16_u16(vmovl_u8(Vu.val[0]));
+			const int16x8_t V2 = vreinterpretq_s16_u16(vmovl_u8(Vu.val[1]));
+			const int16x8_t D1 = vsubq_s16(U1, c128);
+			const int16x4_t D1h = vget_high_s16(D1);
+			const int16x4_t D1l = vget_low_s16(D1);
+			const int16x8_t E1 = vsubq_s16(V1, c128);
+			const int16x4_t E1h = vget_high_s16(E1);
+			const int16x4_t E1l = vget_low_s16(E1);
+			const int16x8_t D2 = vsubq_s16(U2, c128);
+			const int16x4_t D2h = vget_high_s16(D2);
+			const int16x4_t D2l = vget_low_s16(D2);
+			const int16x8_t E2 = vsubq_s16(V2, c128);
+			const int16x4_t E2h = vget_high_s16(E2);
+			const int16x4_t E2l = vget_low_s16(E2);
+			uint8x8x4_t bgrx;
+			bgrx.val[aPos] = vdup_n_u8(0xFF);
+			{
+				const uint8x8_t Y1u = vld1_u8(pY1);
+				const int16x8_t Y1 = vreinterpretq_s16_u16(vmovl_u8(Y1u));
+				const int32x4_t Ch = vmulq_n_s32(vmovl_s16(vget_high_s16(Y1)), 256);  /* Y * 256 */
+				const int32x4_t Cl = vmulq_n_s32(vmovl_s16(vget_low_s16(Y1)), 256);  /* Y * 256 */
+				bgrx.val[rPos] = neon_YUV2R(Ch, Cl, D1h, D1l, E1h, E1l);
+				bgrx.val[gPos] = neon_YUV2G(Ch, Cl, D1h, D1l, E1h, E1l);
+				bgrx.val[bPos] = neon_YUV2B(Ch, Cl, D1h, D1l, E1h, E1l);
+				vst4_u8(pRGB1, bgrx);
+				pRGB1 += 32;
+				pY1 += 8;
+				x += 8;
+			}
+
+			if (!lastX)
+			{
+				const uint8x8_t Y1u = vld1_u8(pY1);
+				const int16x8_t Y1 = vreinterpretq_s16_u16(vmovl_u8(Y1u));
+				const int32x4_t Ch = vmulq_n_s32(vmovl_s16(vget_high_s16(Y1)), 256);  /* Y * 256 */
+				const int32x4_t Cl = vmulq_n_s32(vmovl_s16(vget_low_s16(Y1)), 256);  /* Y * 256 */
+				bgrx.val[rPos] = neon_YUV2R(Ch, Cl, D2h, D2l, E2h, E2l);
+				bgrx.val[gPos] = neon_YUV2G(Ch, Cl, D2h, D2l, E2h, E2l);
+				bgrx.val[bPos] = neon_YUV2B(Ch, Cl, D2h, D2l, E2h, E2l);
+				vst4_u8(pRGB1, bgrx);
+				pRGB1 += 32;
+				pY1 += 8;
+				x += 8;
+			}
+
+			if (!lastY)
+			{
+				const uint8x8_t Y2u = vld1_u8(pY2);
+				const int16x8_t Y2 = vreinterpretq_s16_u16(vmovl_u8(Y2u));
+				const int32x4_t Ch = vmulq_n_s32(vmovl_s16(vget_high_s16(Y2)), 256);  /* Y * 256 */
+				const int32x4_t Cl = vmulq_n_s32(vmovl_s16(vget_low_s16(Y2)), 256);  /* Y * 256 */
+				bgrx.val[rPos] = neon_YUV2R(Ch, Cl, D1h, D1l, E1h, E1l);
+				bgrx.val[gPos] = neon_YUV2G(Ch, Cl, D1h, D1l, E1h, E1l);
+				bgrx.val[bPos] = neon_YUV2B(Ch, Cl, D1h, D1l, E1h, E1l);
+				vst4_u8(pRGB2, bgrx);
+				pRGB2 += 32;
+				pY2 += 8;
+
+				if (!lastX)
+				{
+					const uint8x8_t Y2u = vld1_u8(pY2);
+					const int16x8_t Y2 = vreinterpretq_s16_u16(vmovl_u8(Y2u));
+					const int32x4_t Ch = vmulq_n_s32(vmovl_s16(vget_high_s16(Y2)), 256);  /* Y * 256 */
+					const int32x4_t Cl = vmulq_n_s32(vmovl_s16(vget_low_s16(Y2)), 256);  /* Y * 256 */
+					bgrx.val[rPos] = neon_YUV2R(Ch, Cl, D2h, D2l, E2h, E2l);
+					bgrx.val[gPos] = neon_YUV2G(Ch, Cl, D2h, D2l, E2h, E2l);
+					bgrx.val[bPos] = neon_YUV2B(Ch, Cl, D2h, D2l, E2h, E2l);
+					vst4_u8(pRGB2, bgrx);
+					pRGB2 += 32;
+					pY2 += 8;
+				}
+			}
+
+			pU += 8;
+			pV += 8;
+		}
+	}
+
+	return PRIMITIVES_SUCCESS;
+}
+
+static pstatus_t neon_YUV420ToRGB_8u_P3AC4R(
+    const BYTE* pSrc[3], const UINT32 srcStep[3],
+    BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
+    const prim_size_t* roi)
+{
+	switch (DstFormat)
+	{
+		case PIXEL_FORMAT_BGRA32:
+		case PIXEL_FORMAT_BGRX32:
+			return neon_YUV420ToX(pSrc, srcStep, pDst, dstStep, roi, 2, 1, 0, 3);
+
+		case PIXEL_FORMAT_RGBA32:
+		case PIXEL_FORMAT_RGBX32:
+			return neon_YUV420ToX(pSrc, srcStep, pDst, dstStep, roi, 0, 1, 2, 3);
+
+		case PIXEL_FORMAT_ARGB32:
+		case PIXEL_FORMAT_XRGB32:
+			return neon_YUV420ToX(pSrc, srcStep, pDst, dstStep, roi, 1, 2, 3, 0);
+
+		case PIXEL_FORMAT_ABGR32:
+		case PIXEL_FORMAT_XBGR32:
+			return neon_YUV420ToX(pSrc, srcStep, pDst, dstStep, roi, 3, 2, 1, 0);
+
+		default:
+			return generic->YUV420ToRGB_8u_P3AC4R(pSrc, srcStep, pDst, dstStep, DstFormat, roi);
+	}
+}
+
+static INLINE pstatus_t neon_YUV444ToX(
+    const BYTE* pSrc[3], const UINT32 srcStep[3],
+    BYTE* pDst, UINT32 dstStep,
+    const prim_size_t* roi, const uint8_t rPos,  const uint8_t gPos,
+    const uint8_t bPos, const uint8_t aPos)
+{
+	UINT32 y;
+	const UINT32 nWidth = roi->width;
+	const UINT32 nHeight = roi->height;
+	const UINT32 yPad = srcStep[0] - roi->width;
+	const UINT32 uPad = srcStep[1] - roi->width;
+	const UINT32 vPad = srcStep[2] - roi->width;
+	const UINT32 dPad = dstStep - roi->width * 4;
+	const uint8_t* pY = pSrc[0];
+	const uint8_t* pU = pSrc[1];
+	const uint8_t* pV = pSrc[2];
+	uint8_t* pRGB = pDst;
+	const int16x8_t c128 = vdupq_n_s16(128);
+	const int16x4_t c48 = vdup_n_s16(48);
+	const int16x4_t c120 = vdup_n_s16(120);
+	const int16x4_t c403 = vdup_n_s16(403);
+	const int16x4_t c475 = vdup_n_s16(475);
+	const DWORD pad = nWidth % 8;
+
+	for (y = 0; y < nHeight; y++)
+	{
+		UINT32 x;
+
+		for (x = 0; x < nWidth - pad; x += 8)
+		{
+			const uint8x8_t Yu = vld1_u8(pY);
+			const int16x8_t Y = vreinterpretq_s16_u16(vmovl_u8(Yu));
+			const uint8x8_t Uu = vld1_u8(pU);
+			const int16x8_t U = vreinterpretq_s16_u16(vmovl_u8(Uu));
+			const uint8x8_t Vu = vld1_u8(pV);
+			const int16x8_t V = vreinterpretq_s16_u16(vmovl_u8(Vu));
+			/* Do the calculations on Y in 32bit width, the result of 255 * 256 does not fit
+			 * a signed 16 bit value. */
+			const int32x4_t Ch = vmulq_n_s32(vmovl_s16(vget_high_s16(Y)), 256);  /* Y * 256 */
+			const int32x4_t Cl = vmulq_n_s32(vmovl_s16(vget_low_s16(Y)), 256);  /* Y * 256 */
+			const int16x8_t D = vsubq_s16(U, c128);
+			const int16x4_t Dh = vget_high_s16(D);
+			const int16x4_t Dl = vget_low_s16(D);
+			const int16x8_t E = vsubq_s16(V, c128);
+			const int16x4_t Eh = vget_high_s16(E);
+			const int16x4_t El = vget_low_s16(E);
+			uint8x8x4_t bgrx;
+			{
+				/* B = (256L * Y + 475 * (U - 128)) >> 8*/
+				const int32x4_t CDh = vmlal_s16(Ch, Dh, c475);
+				const int32x4_t CDl = vmlal_s16(Cl, Dl, c475);
+				const int32x4_t Bh = vrshrq_n_s32(CDh, 8);
+				const int32x4_t Bl = vrshrq_n_s32(CDl, 8);
+				const int16x8_t B = vcombine_s16(vqmovn_s32(Bl), vqmovn_s32(Bh));
+				bgrx.val[bPos] = vqmovun_s16(B);
+			}
+			{
+				/* G = (256L * Y -  48 * (U - 128) - 120 * (V - 128)) >> 8 */
+				const int32x4_t CDh = vmlsl_s16(Ch, Dh, c48);
+				const int32x4_t CDl = vmlsl_s16(Cl, Dl, c48);
+				const int32x4_t CDEh = vmlsl_s16(CDh, Eh, c120);
+				const int32x4_t CDEl = vmlsl_s16(CDl, El, c120);
+				const int32x4_t Gh = vrshrq_n_s32(CDEh, 8);
+				const int32x4_t Gl = vrshrq_n_s32(CDEl, 8);
+				const int16x8_t G = vcombine_s16(vqmovn_s32(Gl), vqmovn_s32(Gh));
+				bgrx.val[gPos] = vqmovun_s16(G);
+			}
+			{
+				/* R = (256 * Y + 403 * (V - 128)) >> 8 */
+				const int32x4_t CEh = vmlal_s16(Ch, Eh, c403);
+				const int32x4_t CEl = vmlal_s16(Cl, El, c403);
+				const int32x4_t Rh = vrshrq_n_s32(CEh, 8);
+				const int32x4_t Rl = vrshrq_n_s32(CEl, 8);
+				const int16x8_t R = vcombine_s16(vqmovn_s32(Rl), vqmovn_s32(Rh));
+				bgrx.val[rPos] = vqmovun_s16(R);
+			}
+			{
+				/* A */
+				bgrx.val[aPos] = vdup_n_u8(0xFF);
+			}
+			vst4_u8(pRGB, bgrx);
+			pRGB += 32;
+			pY += 8;
+			pU += 8;
+			pV += 8;
+		}
+
+		for (x = 0; x < pad; x++)
+		{
+			const BYTE Y = *pY++;
+			const BYTE U = *pU++;
+			const BYTE V = *pV++;
+			const BYTE r = YUV2R(Y, U, V);
+			const BYTE g = YUV2G(Y, U, V);
+			const BYTE b = YUV2B(Y, U, V);
+			pRGB[aPos] = 0xFF;
+			pRGB[rPos] = r;
+			pRGB[gPos] = g;
+			pRGB[bPos] = b;
+			pRGB += 4;
+		}
+
+		pRGB += dPad;
+		pY += yPad;
+		pU += uPad;
+		pV += vPad;
+	}
+
+	return PRIMITIVES_SUCCESS;
+}
+
+static pstatus_t neon_YUV444ToRGB_8u_P3AC4R(
+    const BYTE* pSrc[3], const UINT32 srcStep[3],
+    BYTE* pDst, UINT32 dstStep, UINT32 DstFormat,
+    const prim_size_t* roi)
+{
+	switch (DstFormat)
+	{
+		case PIXEL_FORMAT_BGRA32:
+		case PIXEL_FORMAT_BGRX32:
+			return neon_YUV444ToX(pSrc, srcStep, pDst, dstStep, roi, 2, 1, 0, 3);
+
+		case PIXEL_FORMAT_RGBA32:
+		case PIXEL_FORMAT_RGBX32:
+			return neon_YUV444ToX(pSrc, srcStep, pDst, dstStep, roi, 0, 1, 2, 3);
+
+		case PIXEL_FORMAT_ARGB32:
+		case PIXEL_FORMAT_XRGB32:
+			return neon_YUV444ToX(pSrc, srcStep, pDst, dstStep, roi, 1, 2, 3, 0);
+
+		case PIXEL_FORMAT_ABGR32:
+		case PIXEL_FORMAT_XBGR32:
+			return neon_YUV444ToX(pSrc, srcStep, pDst, dstStep, roi, 3, 2, 1, 0);
+
+		default:
+			return generic->YUV444ToRGB_8u_P3AC4R(pSrc, srcStep, pDst, dstStep, DstFormat, roi);
+	}
+}
+
+static pstatus_t neon_YUV420CombineToYUV444(
+    const BYTE* pMainSrc[3], const UINT32 srcMainStep[3],
+    const BYTE* pAuxSrc[3], const UINT32 srcAuxStep[3],
+    BYTE* pDst[3], const UINT32 dstStep[3],
+    const prim_size_t* roi)
+{
+	UINT32 x, y;
+	const UINT32 nWidth = roi->width;
+	const UINT32 nHeight = roi->height;
+	const UINT32 halfWidth = nWidth / 2, halfHeight = nHeight / 2;
+	const UINT32 oddY = 1;
+	const UINT32 evenY = 0;
+	/* The auxilary frame is aligned to multiples of 16x16.
+	 * We need the padded height for B4 and B5 conversion. */
+	const UINT32 padHeigth = roi->height + 16 - roi->height % 16;
+
+	if (pMainSrc && pMainSrc[0] && pMainSrc[1] && pMainSrc[2])
+	{
+		/* Y data is already here... */
+		/* B1 */
+		for (y = 0; y < nHeight; y++)
+		{
+			const BYTE* Ym = pMainSrc[0] + srcMainStep[0] * y;
+			BYTE* pY = pDst[0] + dstStep[0] * y;
+			memcpy(pY, Ym, nWidth);
+		}
+
+		/* The first half of U, V are already here part of this frame. */
+		/* B2 and B3 */
+		for (y = 0; y < halfHeight; y++)
+		{
+			const UINT32 val2y = (2 * y + evenY);
+			const BYTE* Um = pMainSrc[1] + srcMainStep[1] * y;
+			const BYTE* Vm = pMainSrc[2] + srcMainStep[2] * y;
+			BYTE* pU = pDst[1] + dstStep[1] * val2y;
+			BYTE* pV = pDst[2] + dstStep[2] * val2y;
+			BYTE* pU1 = pU + dstStep[1];
+			BYTE* pV1 = pV + dstStep[2];
+
+			for (x = 0; x + 16 < halfWidth; x += 16)
+			{
+				{
+					const uint8x16_t u = vld1q_u8(Um);
+					uint8x16x2_t u2x;
+					u2x.val[0] = u;
+					u2x.val[1] = u;
+					vst2q_u8(pU, u2x);
+					vst2q_u8(pU1, u2x);
+					Um += 16;
+					pU += 32;
+					pU1 += 32;
+				}
+				{
+					const uint8x16_t v = vld1q_u8(Vm);
+					uint8x16x2_t v2x;
+					v2x.val[0] = v;
+					v2x.val[1] = v;
+					vst2q_u8(pV, v2x);
+					vst2q_u8(pV1, v2x);
+					Vm += 16;
+					pV += 32;
+					pV1 += 32;
+				}
+			}
+
+			for (; x < halfWidth; x++)
+			{
+				const BYTE u = *Um++;
+				const BYTE v = *Vm++;
+				*pU++ = u;
+				*pU++ = u;
+				*pU1++ = u;
+				*pU1++ = u;
+				*pV++ = v;
+				*pV++ = v;
+				*pV1++ = v;
+				*pV1++ = v;
+			}
+		}
+	}
+
+	if (!pAuxSrc || !pAuxSrc[0] || !pAuxSrc[1] || !pAuxSrc[2])
+		return PRIMITIVES_SUCCESS;
+
+	/* The second half of U and V is a bit more tricky... */
+	/* B4 and B5 */
+	for (y = 0; y < padHeigth; y += 16)
+	{
+		const BYTE* Ya = pAuxSrc[0] + srcAuxStep[0] * y;
+		UINT32 x;
+		BYTE* pU = pDst[1] + dstStep[1] * (y + 1);
+		BYTE* pV = pDst[2] + dstStep[2] * (y + 1);
+
+		for (x = 0; x < 8; x++)
+		{
+			if (y + x >= nHeight)
+				continue;
+
+			memcpy(pU, Ya, nWidth);
+			pU += dstStep[1] * 2;
+			Ya += srcAuxStep[0];
+		}
+
+		for (x = 0; x < 8; x++)
+		{
+			if (y + x >= nHeight)
+				continue;
+
+			memcpy(pV, Ya, nWidth);
+			pV += dstStep[1] * 2;
+			Ya += srcAuxStep[0];
+		}
+	}
+
+	/* B6 and B7 */
+	for (y = 0; y < halfHeight; y++)
+	{
+		const UINT32 val2y = (y * 2 + evenY);
+		const BYTE* Ua = pAuxSrc[1] + srcAuxStep[1] * y;
+		const BYTE* Va = pAuxSrc[2] + srcAuxStep[2] * y;
+		BYTE* pU = pDst[1] + dstStep[1] * val2y;
+		BYTE* pV = pDst[2] + dstStep[2] * val2y;
+
+		for (x = 0; x + 16 < halfWidth; x += 16)
+		{
+			{
+				const uint8x16_t u = vld1q_u8(Ua);
+				uint8x16x2_t uu = vld2q_u8(pU);
+				uu.val[1] = u;
+				vst2q_u8(pU, uu);
+				Ua += 16;
+				pU += 32;
+			}
+			{
+				const uint8x16_t v = vld1q_u8(Va);
+				uint8x16x2_t vv = vld2q_u8(pV);
+				vv.val[1] = v;
+				vst2q_u8(pV, vv);
+				Va += 16;
+				pV += 32;
+			}
+		}
+
+		for (; x < halfWidth; x++)
+		{
+			pU++;
+			*pU++ = *Ua++;
+			pV++;
+			*pV++ = *Va++;
+		}
+	}
+
+	/* Filter */
+	for (y = 0; y < halfHeight; y++)
+	{
+		const UINT32 val2y = (y * 2 + evenY);
+		const UINT32 val2y1 = val2y + oddY;
+		BYTE* pU = pDst[1] + dstStep[1] * val2y;
+		BYTE* pV = pDst[2] + dstStep[2] * val2y;
+		BYTE* pU1 = pU + dstStep[1];
+		BYTE* pV1 = pV + dstStep[2];
+
+		if (val2y1 > nHeight)
+			continue;
+
+		for (x = 0; x + 16 < halfWidth; x += 16)
+		{
+			{
+				/* U = (U2x,2y << 2) - U2x1,2y - U2x,2y1 - U2x1,2y1 */
+				uint8x8x2_t u = vld2_u8(pU);
+				const int16x8_t up = vreinterpretq_s16_u16(vshll_n_u8(u.val[0], 2)); /* Ux2,2y << 2 */
+				const uint8x8x2_t u1 = vld2_u8(pU1);
+				const uint16x8_t usub = vaddl_u8(u1.val[1], u1.val[0]); /* U2x,2y1 + U2x1,2y1 */
+				const int16x8_t us = vreinterpretq_s16_u16(vaddw_u8(usub,
+				                     u.val[1])); /* U2x1,2y + U2x,2y1 + U2x1,2y1 */
+				const int16x8_t un = vsubq_s16(up, us);
+				const uint8x8_t u8 = vqmovun_s16(un); /* CLIP(un) */
+				u.val[0] = u8;
+				vst2_u8(pU, u);
+				pU += 16;
+				pU1 += 16;
+			}
+			{
+				/* V = (V2x,2y << 2) - V2x1,2y - V2x,2y1 - V2x1,2y1 */
+				uint8x8x2_t v = vld2_u8(pV);
+				const int16x8_t vp = vreinterpretq_s16_u16(vshll_n_u8(v.val[0], 2)); /* Vx2,2y << 2 */
+				const uint8x8x2_t v1 = vld2_u8(pV1);
+				const uint16x8_t vsub = vaddl_u8(v1.val[1], v1.val[0]); /* V2x,2y1 + V2x1,2y1 */
+				const int16x8_t vs = vreinterpretq_s16_u16(vaddw_u8(vsub,
+				                     v.val[1])); /* V2x1,2y + V2x,2y1 + V2x1,2y1 */
+				const int16x8_t vn = vsubq_s16(vp, vs);
+				const uint8x8_t v8 = vqmovun_s16(vn); /* CLIP(vn) */
+				v.val[0] = v8;
+				vst2_u8(pV, v);
+				pV += 16;
+				pV1 += 16;
+			}
+		}
+
+		for (; x < halfWidth; x++)
+		{
+			const UINT32 val2x = (x * 2);
+			const UINT32 val2x1 = val2x + 1;
+			const INT32 up = pU[val2x] * 4;
+			const INT32 vp = pV[val2x] * 4;
+			INT32 u2020;
+			INT32 v2020;
+
+			if (val2x1 > nWidth)
+				continue;
+
+			u2020 = up - pU[val2x1] - pU1[val2x] - pU1[val2x1];
+			v2020 = vp - pV[val2x1] - pV1[val2x] - pV1[val2x1];
+			*pU = CLIP(u2020);
+			*pV = CLIP(v2020);
+			pU += 2;
+			pV += 2;
+		}
+	}
+
+	return PRIMITIVES_SUCCESS;
+}
 #endif
 
 void primitives_init_YUV_opt(primitives_t* prims)
@@ -615,6 +1251,16 @@ void primitives_init_YUV_opt(primitives_t* prims)
 	{
 		prims->RGBToYUV420_8u_P3AC4R = ssse3_RGBToYUV420;
 		prims->YUV420ToRGB_8u_P3AC4R = ssse3_YUV420ToRGB;
+		prims->YUV444ToRGB_8u_P3AC4R = ssse3_YUV444ToRGB_8u_P3AC4R;
+	}
+
+#elif defined(WITH_NEON)
+
+	if (IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE))
+	{
+		prims->YUV420ToRGB_8u_P3AC4R = neon_YUV420ToRGB_8u_P3AC4R;
+		prims->YUV444ToRGB_8u_P3AC4R = neon_YUV444ToRGB_8u_P3AC4R;
+		prims->YUV420CombineToYUV444 = neon_YUV420CombineToYUV444;
 	}
 
 #endif
diff --git a/libfreerdp/primitives/prim_internal.h b/libfreerdp/primitives/prim_internal.h
index 1bf4cd342..f931212e3 100644
--- a/libfreerdp/primitives/prim_internal.h
+++ b/libfreerdp/primitives/prim_internal.h
@@ -129,6 +129,47 @@ static INLINE BYTE CLIP(INT32 X)
 	return X;
 }
 
+/**
+ * | R |   ( | 256     0    403 | |    Y    | )
+ * | G | = ( | 256   -48   -120 | | U - 128 | ) >> 8
+ * | B |   ( | 256   475      0 | | V - 128 | )
+ */
+static INLINE INT32 C(INT32 Y)
+{
+	return (Y) -   0L;
+}
+
+static INLINE INT32 D(INT32 U)
+{
+	return (U) - 128L;
+}
+
+static INLINE INT32 E(INT32 V)
+{
+	return (V) - 128L;
+}
+
+static INLINE BYTE YUV2R(INT32 Y, INT32 U, INT32 V)
+{
+	const INT32 r = (256L * C(Y) +   0L * D(U) + 403L * E(V));
+	const INT32 r8 = r >> 8L;
+	return CLIP(r8);
+}
+
+static INLINE BYTE YUV2G(INT32 Y, INT32 U, INT32 V)
+{
+	const INT32 g = (256L * C(Y) -  48L * D(U) - 120L * E(V));
+	const INT32 g8 = g >> 8L;
+	return CLIP(g8);
+}
+
+static INLINE BYTE YUV2B(INT32 Y, INT32 U, INT32 V)
+{
+	const INT32 b = (256L * C(Y) + 475L * D(U) +   0L * E(V));
+	const INT32 b8 = b >> 8L;
+	return CLIP(b8);
+}
+
 /* Function prototypes for all the init/deinit routines. */
 FREERDP_LOCAL void primitives_init_copy(primitives_t* prims);
 FREERDP_LOCAL void primitives_init_set(primitives_t* prims);
diff --git a/libfreerdp/primitives/test/TestPrimitivesYUV.c b/libfreerdp/primitives/test/TestPrimitivesYUV.c
index 310bd3e2c..67599362d 100644
--- a/libfreerdp/primitives/test/TestPrimitivesYUV.c
+++ b/libfreerdp/primitives/test/TestPrimitivesYUV.c
@@ -22,7 +22,7 @@ static BOOL similar(const BYTE* src, const BYTE* dst, size_t size)
 	{
 		int diff = src[x] - dst[x];
 
-		if (abs(diff) > 2)
+		if (abs(diff) > 4)
 		{
 			fprintf(stderr, "%"PRIuz" %02"PRIX8" : %02"PRIX8" diff=%d\n", x, src[x], dst[x], abs(diff));
 			return FALSE;
@@ -78,13 +78,14 @@ static BOOL similarRGB(const BYTE* src, const BYTE* dst, size_t size, UINT32 for
 	return TRUE;
 }
 
-static void get_size(UINT32* width, UINT32* height)
+static void get_size(BOOL large, UINT32* width, UINT32* height)
 {
+	UINT32 shift = large ? 8 : 1;
 	winpr_RAND((BYTE*)width, sizeof(*width));
 	winpr_RAND((BYTE*)height, sizeof(*height));
 	// TODO: Algorithm only works on even resolutions...
-	*width = (*width % 64 + 1) << 1;
-	*height = (*height % 64 + 1) << 1;
+	*width = (*width % 64 + 1) << shift;
+	*height = (*height % 64 + 1) << shift;
 }
 
 static BOOL check_padding(const BYTE* psrc, size_t size, size_t padding,
@@ -170,7 +171,7 @@ static void free_padding(void* src, size_t padding)
 
 /* Create 2 pseudo YUV420 frames of same size.
  * Combine them and check, if the data is at the expected position. */
-static BOOL TestPrimitiveYUVCombine(void)
+static BOOL TestPrimitiveYUVCombine(primitives_t* prims, prim_size_t roi)
 {
 	UINT32 x, y, i;
 	UINT32 awidth, aheight;
@@ -186,9 +187,6 @@ static BOOL TestPrimitiveYUVCombine(void)
 	size_t padding = 10000;
 	PROFILER_DEFINE(yuvCombine);
 	PROFILER_DEFINE(yuvSplit);
-	prim_size_t roi;
-	primitives_t* prims = primitives_get();
-	get_size(&roi.width, &roi.height);
 	awidth = roi.width + 16 - roi.width % 16;
 	aheight = roi.height + 16 - roi.height % 16;
 	fprintf(stderr, "Running YUVCombine on frame size %"PRIu32"x%"PRIu32" [%"PRIu32"x%"PRIu32"]\n",
@@ -360,18 +358,16 @@ fail:
 	return rc;
 }
 
-static BOOL TestPrimitiveYUV(BOOL use444)
+static BOOL TestPrimitiveYUV(primitives_t* prims, prim_size_t roi, BOOL use444)
 {
 	BOOL rc = FALSE;
 	UINT32 x, y;
 	UINT32 awidth, aheight;
 	BYTE* yuv[3] = {0};
 	UINT32 yuv_step[3];
-	prim_size_t roi;
 	BYTE* rgb = NULL;
 	BYTE* rgb_dst = NULL;
 	size_t size;
-	primitives_t* prims = primitives_get();
 	size_t uvsize, uvwidth;
 	size_t padding = 10000;
 	size_t stride;
@@ -390,7 +386,6 @@ static BOOL TestPrimitiveYUV(BOOL use444)
 	PROFILER_DEFINE(rgbToYUV444);
 	PROFILER_DEFINE(yuv420ToRGB);
 	PROFILER_DEFINE(yuv444ToRGB);
-	get_size(&roi.width, &roi.height);
 	/* Buffers need to be 16x16 aligned. */
 	awidth = roi.width + 16 - roi.width % 16;
 	aheight = roi.height + 16 - roi.height % 16;
@@ -457,6 +452,7 @@ static BOOL TestPrimitiveYUV(BOOL use444)
 	for (x = 0; x < sizeof(formats) / sizeof(formats[0]); x++)
 	{
 		const UINT32 DstFormat = formats[x];
+		printf("Testing destination color format %s\n", GetColorFormatName(DstFormat));
 
 		if (use444)
 		{
@@ -561,26 +557,73 @@ fail:
 
 int TestPrimitivesYUV(int argc, char* argv[])
 {
+	BOOL large = (argc > 1);
 	UINT32 x;
 	int rc = -1;
 	prim_test_setup(FALSE);
+	primitives_t* prims = primitives_get();
+	primitives_t* generic = primitives_get_generic();
 
 	for (x = 0; x < 10; x++)
 	{
-		if (!TestPrimitiveYUV(TRUE)) {
+		prim_size_t roi;
+		get_size(large, &roi.width, &roi.height);
+		printf("-------------------- GENERIC ------------------------\n");
+
+		if (!TestPrimitiveYUV(generic, roi, TRUE))
+		{
 			printf("TestPrimitiveYUV (444) failed.\n");
 			goto end;
 		}
 
-		if (!TestPrimitiveYUV(FALSE)) {
+		printf("---------------------- END --------------------------\n");
+#if 1
+		printf("------------------- OPTIMIZED -----------------------\n");
+
+		if (!TestPrimitiveYUV(prims, roi, TRUE))
+		{
+			printf("TestPrimitiveYUV (444) failed.\n");
+			goto end;
+		}
+
+		printf("---------------------- END --------------------------\n");
+#endif
+		printf("-------------------- GENERIC ------------------------\n");
+
+		if (!TestPrimitiveYUV(generic, roi, FALSE))
+		{
 			printf("TestPrimitiveYUV (420) failed.\n");
 			goto end;
 		}
 
-		if (!TestPrimitiveYUVCombine()) {
+		printf("---------------------- END --------------------------\n");
+		printf("------------------- OPTIMIZED -----------------------\n");
+
+		if (!TestPrimitiveYUV(prims, roi, FALSE))
+		{
+			printf("TestPrimitiveYUV (420) failed.\n");
+			goto end;
+		}
+
+		printf("---------------------- END --------------------------\n");
+		printf("-------------------- GENERIC ------------------------\n");
+
+		if (!TestPrimitiveYUVCombine(generic, roi))
+		{
 			printf("TestPrimitiveYUVCombine failed.\n");
 			goto end;
 		}
+
+		printf("---------------------- END --------------------------\n");
+		printf("------------------- OPTIMIZED -----------------------\n");
+
+		if (!TestPrimitiveYUVCombine(prims, roi))
+		{
+			printf("TestPrimitiveYUVCombine failed.\n");
+			goto end;
+		}
+
+		printf("---------------------- END --------------------------\n");
 	}
 
 	rc = 0;