michael@0: /*
michael@0:  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
michael@0:  *
michael@0:  *  Use of this source code is governed by a BSD-style license
michael@0:  *  that can be found in the LICENSE file in the root of the source
michael@0:  *  tree. An additional intellectual property rights grant can be found
michael@0:  *  in the file PATENTS. All contributing project authors may
michael@0:  *  be found in the AUTHORS file in the root of the source tree.
michael@0:  */
michael@0: 
michael@0: #include "libyuv/convert.h"
michael@0: 
michael@0: #include "libyuv/basic_types.h"
michael@0: #include "libyuv/cpu_id.h"
michael@0: #include "libyuv/planar_functions.h"
michael@0: #include "libyuv/rotate.h"
michael@0: #include "libyuv/scale.h"  // For ScalePlane()
michael@0: #include "libyuv/row.h"
michael@0: 
michael@0: #ifdef __cplusplus
michael@0: namespace libyuv {
michael@0: extern "C" {
michael@0: #endif
michael@0: 
michael@0: #define SUBSAMPLE(v, a, s) (v < 0) ? (-((-v + a) >> s)) : ((v + a) >> s)
michael@0: static __inline int Abs(int v) {
michael@0:   return v >= 0 ? v : -v;
michael@0: }
michael@0: 
michael@0: // Any I4xx To I420 format with mirroring.
michael@0: static int I4xxToI420(const uint8* src_y, int src_stride_y,
michael@0:                       const uint8* src_u, int src_stride_u,
michael@0:                       const uint8* src_v, int src_stride_v,
michael@0:                       uint8* dst_y, int dst_stride_y,
michael@0:                       uint8* dst_u, int dst_stride_u,
michael@0:                       uint8* dst_v, int dst_stride_v,
michael@0:                       int src_y_width, int src_y_height,
michael@0:                       int src_uv_width, int src_uv_height) {
michael@0:   if (src_y_width == 0 || src_y_height == 0 ||
michael@0:       src_uv_width == 0 || src_uv_height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   const int dst_y_width = Abs(src_y_width);
michael@0:   const int dst_y_height = Abs(src_y_height);
michael@0:   const int dst_uv_width = SUBSAMPLE(dst_y_width, 1, 1);
michael@0:   const int dst_uv_height = SUBSAMPLE(dst_y_height, 1, 1);
michael@0:   ScalePlane(src_y, src_stride_y, src_y_width, src_y_height,
michael@0:              dst_y, dst_stride_y, dst_y_width, dst_y_height,
michael@0:              kFilterBilinear);
michael@0:   ScalePlane(src_u, src_stride_u, src_uv_width, src_uv_height,
michael@0:              dst_u, dst_stride_u, dst_uv_width, dst_uv_height,
michael@0:              kFilterBilinear);
michael@0:   ScalePlane(src_v, src_stride_v, src_uv_width, src_uv_height,
michael@0:              dst_v, dst_stride_v, dst_uv_width, dst_uv_height,
michael@0:              kFilterBilinear);
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Copy I420 with optional flipping
michael@0: // TODO(fbarchard): Use Scale plane which supports mirroring, but ensure
michael@0: // is does row coalescing.
michael@0: LIBYUV_API
michael@0: int I420Copy(const uint8* src_y, int src_stride_y,
michael@0:              const uint8* src_u, int src_stride_u,
michael@0:              const uint8* src_v, int src_stride_v,
michael@0:              uint8* dst_y, int dst_stride_y,
michael@0:              uint8* dst_u, int dst_stride_u,
michael@0:              uint8* dst_v, int dst_stride_v,
michael@0:              int width, int height) {
michael@0:   if (!src_y || !src_u || !src_v ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     const int halfheight = (height + 1) >> 1;
michael@0:     src_y = src_y + (height - 1) * src_stride_y;
michael@0:     src_u = src_u + (halfheight - 1) * src_stride_u;
michael@0:     src_v = src_v + (halfheight - 1) * src_stride_v;
michael@0:     src_stride_y = -src_stride_y;
michael@0:     src_stride_u = -src_stride_u;
michael@0:     src_stride_v = -src_stride_v;
michael@0:   }
michael@0: 
michael@0:   if (dst_y) {
michael@0:     CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
michael@0:   }
michael@0:   // Copy UV planes.
michael@0:   const int halfwidth = (width + 1) >> 1;
michael@0:   const int halfheight = (height + 1) >> 1;
michael@0:   CopyPlane(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth, halfheight);
michael@0:   CopyPlane(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth, halfheight);
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // 422 chroma is 1/2 width, 1x height
michael@0: // 420 chroma is 1/2 width, 1/2 height
michael@0: LIBYUV_API
michael@0: int I422ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_u, int src_stride_u,
michael@0:                const uint8* src_v, int src_stride_v,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   const int src_uv_width = SUBSAMPLE(width, 1, 1);
michael@0:   return I4xxToI420(src_y, src_stride_y,
michael@0:                     src_u, src_stride_u,
michael@0:                     src_v, src_stride_v,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     width, height,
michael@0:                     src_uv_width, height);
michael@0: }
michael@0: 
michael@0: // 444 chroma is 1x width, 1x height
michael@0: // 420 chroma is 1/2 width, 1/2 height
michael@0: LIBYUV_API
michael@0: int I444ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_u, int src_stride_u,
michael@0:                const uint8* src_v, int src_stride_v,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   return I4xxToI420(src_y, src_stride_y,
michael@0:                     src_u, src_stride_u,
michael@0:                     src_v, src_stride_v,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     width, height,
michael@0:                     width, height);
michael@0: }
michael@0: 
michael@0: // 411 chroma is 1/4 width, 1x height
michael@0: // 420 chroma is 1/2 width, 1/2 height
michael@0: LIBYUV_API
michael@0: int I411ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_u, int src_stride_u,
michael@0:                const uint8* src_v, int src_stride_v,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   const int src_uv_width = SUBSAMPLE(width, 3, 2);
michael@0:   return I4xxToI420(src_y, src_stride_y,
michael@0:                     src_u, src_stride_u,
michael@0:                     src_v, src_stride_v,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     width, height,
michael@0:                     src_uv_width, height);
michael@0: }
michael@0: 
michael@0: // I400 is greyscale typically used in MJPG
michael@0: LIBYUV_API
michael@0: int I400ToI420(const uint8* src_y, int src_stride_y,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_y || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_y = src_y + (height - 1) * src_stride_y;
michael@0:     src_stride_y = -src_stride_y;
michael@0:   }
michael@0:   int halfwidth = (width + 1) >> 1;
michael@0:   int halfheight = (height + 1) >> 1;
michael@0:   CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
michael@0:   SetPlane(dst_u, dst_stride_u, halfwidth, halfheight, 128);
michael@0:   SetPlane(dst_v, dst_stride_v, halfwidth, halfheight, 128);
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: static void CopyPlane2(const uint8* src, int src_stride_0, int src_stride_1,
michael@0:                        uint8* dst, int dst_stride,
michael@0:                        int width, int height) {
michael@0:   void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
michael@0: #if defined(HAS_COPYROW_X86)
michael@0:   if (TestCpuFlag(kCpuHasX86) && IS_ALIGNED(width, 4)) {
michael@0:     CopyRow = CopyRow_X86;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) &&
michael@0:       IS_ALIGNED(src, 16) &&
michael@0:       IS_ALIGNED(src_stride_0, 16) && IS_ALIGNED(src_stride_1, 16) &&
michael@0:       IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride, 16)) {
michael@0:     CopyRow = CopyRow_SSE2;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_ERMS)
michael@0:   if (TestCpuFlag(kCpuHasERMS)) {
michael@0:     CopyRow = CopyRow_ERMS;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 32)) {
michael@0:     CopyRow = CopyRow_NEON;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_MIPS)
michael@0:   if (TestCpuFlag(kCpuHasMIPS)) {
michael@0:     CopyRow = CopyRow_MIPS;
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   // Copy plane
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     CopyRow(src, dst, width);
michael@0:     CopyRow(src + src_stride_0, dst + dst_stride, width);
michael@0:     src += src_stride_0 + src_stride_1;
michael@0:     dst += dst_stride * 2;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     CopyRow(src, dst, width);
michael@0:   }
michael@0: }
michael@0: 
michael@0: // Support converting from FOURCC_M420
michael@0: // Useful for bandwidth constrained transports like USB 1.0 and 2.0 and for
michael@0: // easy conversion to I420.
michael@0: // M420 format description:
michael@0: // M420 is row biplanar 420: 2 rows of Y and 1 row of UV.
michael@0: // Chroma is half width / half height. (420)
michael@0: // src_stride_m420 is row planar. Normally this will be the width in pixels.
michael@0: //   The UV plane is half width, but 2 values, so src_stride_m420 applies to
michael@0: //   this as well as the two Y planes.
michael@0: static int X420ToI420(const uint8* src_y,
michael@0:                       int src_stride_y0, int src_stride_y1,
michael@0:                       const uint8* src_uv, int src_stride_uv,
michael@0:                       uint8* dst_y, int dst_stride_y,
michael@0:                       uint8* dst_u, int dst_stride_u,
michael@0:                       uint8* dst_v, int dst_stride_v,
michael@0:                       int width, int height) {
michael@0:   if (!src_y || !src_uv ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     int halfheight = (height + 1) >> 1;
michael@0:     dst_y = dst_y + (height - 1) * dst_stride_y;
michael@0:     dst_u = dst_u + (halfheight - 1) * dst_stride_u;
michael@0:     dst_v = dst_v + (halfheight - 1) * dst_stride_v;
michael@0:     dst_stride_y = -dst_stride_y;
michael@0:     dst_stride_u = -dst_stride_u;
michael@0:     dst_stride_v = -dst_stride_v;
michael@0:   }
michael@0:   // Coalesce rows.
michael@0:   int halfwidth = (width + 1) >> 1;
michael@0:   int halfheight = (height + 1) >> 1;
michael@0:   if (src_stride_y0 == width &&
michael@0:       src_stride_y1 == width &&
michael@0:       dst_stride_y == width) {
michael@0:     width *= height;
michael@0:     height = 1;
michael@0:     src_stride_y0 = src_stride_y1 = dst_stride_y = 0;
michael@0:   }
michael@0:   // Coalesce rows.
michael@0:   if (src_stride_uv == halfwidth * 2 &&
michael@0:       dst_stride_u == halfwidth &&
michael@0:       dst_stride_v == halfwidth) {
michael@0:     halfwidth *= halfheight;
michael@0:     halfheight = 1;
michael@0:     src_stride_uv = dst_stride_u = dst_stride_v = 0;
michael@0:   }
michael@0:   void (*SplitUVRow)(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix) =
michael@0:       SplitUVRow_C;
michael@0: #if defined(HAS_SPLITUVROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && halfwidth >= 16) {
michael@0:     SplitUVRow = SplitUVRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(halfwidth, 16)) {
michael@0:       SplitUVRow = SplitUVRow_Unaligned_SSE2;
michael@0:       if (IS_ALIGNED(src_uv, 16) && IS_ALIGNED(src_stride_uv, 16) &&
michael@0:           IS_ALIGNED(dst_u, 16) && IS_ALIGNED(dst_stride_u, 16) &&
michael@0:           IS_ALIGNED(dst_v, 16) && IS_ALIGNED(dst_stride_v, 16)) {
michael@0:         SplitUVRow = SplitUVRow_SSE2;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_SPLITUVROW_AVX2)
michael@0:   if (TestCpuFlag(kCpuHasAVX2) && halfwidth >= 32) {
michael@0:     SplitUVRow = SplitUVRow_Any_AVX2;
michael@0:     if (IS_ALIGNED(halfwidth, 32)) {
michael@0:       SplitUVRow = SplitUVRow_AVX2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_SPLITUVROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && halfwidth >= 16) {
michael@0:     SplitUVRow = SplitUVRow_Any_NEON;
michael@0:     if (IS_ALIGNED(halfwidth, 16)) {
michael@0:       SplitUVRow = SplitUVRow_NEON;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_SPLITUVROW_MIPS_DSPR2)
michael@0:   if (TestCpuFlag(kCpuHasMIPS_DSPR2) && halfwidth >= 16) {
michael@0:     SplitUVRow = SplitUVRow_Any_MIPS_DSPR2;
michael@0:     if (IS_ALIGNED(halfwidth, 16)) {
michael@0:       SplitUVRow = SplitUVRow_Unaligned_MIPS_DSPR2;
michael@0:       if (IS_ALIGNED(src_uv, 4) && IS_ALIGNED(src_stride_uv, 4) &&
michael@0:           IS_ALIGNED(dst_u, 4) && IS_ALIGNED(dst_stride_u, 4) &&
michael@0:           IS_ALIGNED(dst_v, 4) && IS_ALIGNED(dst_stride_v, 4)) {
michael@0:         SplitUVRow = SplitUVRow_MIPS_DSPR2;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   if (dst_y) {
michael@0:     if (src_stride_y0 == src_stride_y1) {
michael@0:       CopyPlane(src_y, src_stride_y0, dst_y, dst_stride_y, width, height);
michael@0:     } else {
michael@0:       CopyPlane2(src_y, src_stride_y0, src_stride_y1, dst_y, dst_stride_y,
michael@0:                  width, height);
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   for (int y = 0; y < halfheight; ++y) {
michael@0:     // Copy a row of UV.
michael@0:     SplitUVRow(src_uv, dst_u, dst_v, halfwidth);
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:     src_uv += src_stride_uv;
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert NV12 to I420.
michael@0: LIBYUV_API
michael@0: int NV12ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_uv, int src_stride_uv,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   return X420ToI420(src_y, src_stride_y, src_stride_y,
michael@0:                     src_uv, src_stride_uv,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     width, height);
michael@0: }
michael@0: 
michael@0: // Convert NV21 to I420.  Same as NV12 but u and v pointers swapped.
michael@0: LIBYUV_API
michael@0: int NV21ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_vu, int src_stride_vu,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   return X420ToI420(src_y, src_stride_y, src_stride_y,
michael@0:                     src_vu, src_stride_vu,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     width, height);
michael@0: }
michael@0: 
michael@0: // Convert M420 to I420.
michael@0: LIBYUV_API
michael@0: int M420ToI420(const uint8* src_m420, int src_stride_m420,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   return X420ToI420(src_m420, src_stride_m420, src_stride_m420 * 2,
michael@0:                     src_m420 + src_stride_m420 * 2, src_stride_m420 * 3,
michael@0:                     dst_y, dst_stride_y,
michael@0:                     dst_u, dst_stride_u,
michael@0:                     dst_v, dst_stride_v,
michael@0:                     width, height);
michael@0: }
michael@0: 
michael@0: // Convert Q420 to I420.
michael@0: // Format is rows of YY/YUYV
michael@0: LIBYUV_API
michael@0: int Q420ToI420(const uint8* src_y, int src_stride_y,
michael@0:                const uint8* src_yuy2, int src_stride_yuy2,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_y || !src_yuy2 ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     int halfheight = (height + 1) >> 1;
michael@0:     dst_y = dst_y + (height - 1) * dst_stride_y;
michael@0:     dst_u = dst_u + (halfheight - 1) * dst_stride_u;
michael@0:     dst_v = dst_v + (halfheight - 1) * dst_stride_v;
michael@0:     dst_stride_y = -dst_stride_y;
michael@0:     dst_stride_u = -dst_stride_u;
michael@0:     dst_stride_v = -dst_stride_v;
michael@0:   }
michael@0:   // CopyRow for rows of just Y in Q420 copied to Y plane of I420.
michael@0:   void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
michael@0: #if defined(HAS_COPYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 32)) {
michael@0:     CopyRow = CopyRow_NEON;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_X86)
michael@0:   if (IS_ALIGNED(width, 4)) {
michael@0:     CopyRow = CopyRow_X86;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 32) &&
michael@0:       IS_ALIGNED(src_y, 16) && IS_ALIGNED(src_stride_y, 16) &&
michael@0:       IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:     CopyRow = CopyRow_SSE2;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_ERMS)
michael@0:   if (TestCpuFlag(kCpuHasERMS)) {
michael@0:     CopyRow = CopyRow_ERMS;
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_COPYROW_MIPS)
michael@0:   if (TestCpuFlag(kCpuHasMIPS)) {
michael@0:     CopyRow = CopyRow_MIPS;
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   void (*YUY2ToUV422Row)(const uint8* src_yuy2, uint8* dst_u, uint8* dst_v,
michael@0:       int pix) = YUY2ToUV422Row_C;
michael@0:   void (*YUY2ToYRow)(const uint8* src_yuy2, uint8* dst_y, int pix) =
michael@0:       YUY2ToYRow_C;
michael@0: #if defined(HAS_YUY2TOYROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 16) {
michael@0:     YUY2ToUV422Row = YUY2ToUV422Row_Any_SSE2;
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       YUY2ToUV422Row = YUY2ToUV422Row_Unaligned_SSE2;
michael@0:       YUY2ToYRow = YUY2ToYRow_Unaligned_SSE2;
michael@0:       if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16)) {
michael@0:         YUY2ToUV422Row = YUY2ToUV422Row_SSE2;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           YUY2ToYRow = YUY2ToYRow_SSE2;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_YUY2TOYROW_AVX2)
michael@0:   if (TestCpuFlag(kCpuHasAVX2) && width >= 32) {
michael@0:     YUY2ToUV422Row = YUY2ToUV422Row_Any_AVX2;
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_AVX2;
michael@0:     if (IS_ALIGNED(width, 32)) {
michael@0:       YUY2ToUV422Row = YUY2ToUV422Row_AVX2;
michael@0:       YUY2ToYRow = YUY2ToYRow_AVX2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_YUY2TOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_NEON;
michael@0:     if (width >= 16) {
michael@0:       YUY2ToUV422Row = YUY2ToUV422Row_Any_NEON;
michael@0:     }
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       YUY2ToYRow = YUY2ToYRow_NEON;
michael@0:       YUY2ToUV422Row = YUY2ToUV422Row_NEON;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     CopyRow(src_y, dst_y, width);
michael@0:     src_y += src_stride_y;
michael@0:     dst_y += dst_stride_y;
michael@0: 
michael@0:     YUY2ToUV422Row(src_yuy2, dst_u, dst_v, width);
michael@0:     YUY2ToYRow(src_yuy2, dst_y, width);
michael@0:     src_yuy2 += src_stride_yuy2;
michael@0:     dst_y += dst_stride_y;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     CopyRow(src_y, dst_y, width);
michael@0:     YUY2ToUV422Row(src_yuy2, dst_u, dst_v, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert YUY2 to I420.
michael@0: LIBYUV_API
michael@0: int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_yuy2 = src_yuy2 + (height - 1) * src_stride_yuy2;
michael@0:     src_stride_yuy2 = -src_stride_yuy2;
michael@0:   }
michael@0:   void (*YUY2ToUVRow)(const uint8* src_yuy2, int src_stride_yuy2,
michael@0:                       uint8* dst_u, uint8* dst_v, int pix);
michael@0:   void (*YUY2ToYRow)(const uint8* src_yuy2,
michael@0:                      uint8* dst_y, int pix);
michael@0:   YUY2ToYRow = YUY2ToYRow_C;
michael@0:   YUY2ToUVRow = YUY2ToUVRow_C;
michael@0: #if defined(HAS_YUY2TOYROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 16) {
michael@0:     YUY2ToUVRow = YUY2ToUVRow_Any_SSE2;
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       YUY2ToUVRow = YUY2ToUVRow_Unaligned_SSE2;
michael@0:       YUY2ToYRow = YUY2ToYRow_Unaligned_SSE2;
michael@0:       if (IS_ALIGNED(src_yuy2, 16) && IS_ALIGNED(src_stride_yuy2, 16)) {
michael@0:         YUY2ToUVRow = YUY2ToUVRow_SSE2;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           YUY2ToYRow = YUY2ToYRow_SSE2;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_YUY2TOYROW_AVX2)
michael@0:   if (TestCpuFlag(kCpuHasAVX2) && width >= 32) {
michael@0:     YUY2ToUVRow = YUY2ToUVRow_Any_AVX2;
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_AVX2;
michael@0:     if (IS_ALIGNED(width, 32)) {
michael@0:       YUY2ToUVRow = YUY2ToUVRow_AVX2;
michael@0:       YUY2ToYRow = YUY2ToYRow_AVX2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_YUY2TOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     YUY2ToYRow = YUY2ToYRow_Any_NEON;
michael@0:     if (width >= 16) {
michael@0:       YUY2ToUVRow = YUY2ToUVRow_Any_NEON;
michael@0:     }
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       YUY2ToYRow = YUY2ToYRow_NEON;
michael@0:       YUY2ToUVRow = YUY2ToUVRow_NEON;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     YUY2ToUVRow(src_yuy2, src_stride_yuy2, dst_u, dst_v, width);
michael@0:     YUY2ToYRow(src_yuy2, dst_y, width);
michael@0:     YUY2ToYRow(src_yuy2 + src_stride_yuy2, dst_y + dst_stride_y, width);
michael@0:     src_yuy2 += src_stride_yuy2 * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     YUY2ToUVRow(src_yuy2, 0, dst_u, dst_v, width);
michael@0:     YUY2ToYRow(src_yuy2, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert UYVY to I420.
michael@0: LIBYUV_API
michael@0: int UYVYToI420(const uint8* src_uyvy, int src_stride_uyvy,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_uyvy = src_uyvy + (height - 1) * src_stride_uyvy;
michael@0:     src_stride_uyvy = -src_stride_uyvy;
michael@0:   }
michael@0:   void (*UYVYToUVRow)(const uint8* src_uyvy, int src_stride_uyvy,
michael@0:                       uint8* dst_u, uint8* dst_v, int pix);
michael@0:   void (*UYVYToYRow)(const uint8* src_uyvy,
michael@0:                      uint8* dst_y, int pix);
michael@0:   UYVYToYRow = UYVYToYRow_C;
michael@0:   UYVYToUVRow = UYVYToUVRow_C;
michael@0: #if defined(HAS_UYVYTOYROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 16) {
michael@0:     UYVYToUVRow = UYVYToUVRow_Any_SSE2;
michael@0:     UYVYToYRow = UYVYToYRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       UYVYToUVRow = UYVYToUVRow_Unaligned_SSE2;
michael@0:       UYVYToYRow = UYVYToYRow_Unaligned_SSE2;
michael@0:       if (IS_ALIGNED(src_uyvy, 16) && IS_ALIGNED(src_stride_uyvy, 16)) {
michael@0:         UYVYToUVRow = UYVYToUVRow_SSE2;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           UYVYToYRow = UYVYToYRow_SSE2;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_UYVYTOYROW_AVX2)
michael@0:   if (TestCpuFlag(kCpuHasAVX2) && width >= 32) {
michael@0:     UYVYToUVRow = UYVYToUVRow_Any_AVX2;
michael@0:     UYVYToYRow = UYVYToYRow_Any_AVX2;
michael@0:     if (IS_ALIGNED(width, 32)) {
michael@0:       UYVYToUVRow = UYVYToUVRow_AVX2;
michael@0:       UYVYToYRow = UYVYToYRow_AVX2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_UYVYTOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     UYVYToYRow = UYVYToYRow_Any_NEON;
michael@0:     if (width >= 16) {
michael@0:       UYVYToUVRow = UYVYToUVRow_Any_NEON;
michael@0:     }
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       UYVYToYRow = UYVYToYRow_NEON;
michael@0:       UYVYToUVRow = UYVYToUVRow_NEON;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     UYVYToUVRow(src_uyvy, src_stride_uyvy, dst_u, dst_v, width);
michael@0:     UYVYToYRow(src_uyvy, dst_y, width);
michael@0:     UYVYToYRow(src_uyvy + src_stride_uyvy, dst_y + dst_stride_y, width);
michael@0:     src_uyvy += src_stride_uyvy * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     UYVYToUVRow(src_uyvy, 0, dst_u, dst_v, width);
michael@0:     UYVYToYRow(src_uyvy, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert ARGB to I420.
michael@0: LIBYUV_API
michael@0: int ARGBToI420(const uint8* src_argb, int src_stride_argb,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_argb ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_argb = src_argb + (height - 1) * src_stride_argb;
michael@0:     src_stride_argb = -src_stride_argb;
michael@0:   }
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOYROW_SSSE3) && defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_Unaligned_SSSE3;
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) {
michael@0:         ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_ARGBTOYROW_AVX2) && defined(HAS_ARGBTOUVROW_AVX2)
michael@0:   if (TestCpuFlag(kCpuHasAVX2) && width >= 32) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_AVX2;
michael@0:     ARGBToYRow = ARGBToYRow_Any_AVX2;
michael@0:     if (IS_ALIGNED(width, 32)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_AVX2;
michael@0:       ARGBToYRow = ARGBToYRow_AVX2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: #if defined(HAS_ARGBTOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ARGBToYRow = ARGBToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       ARGBToUVRow = ARGBToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         ARGBToUVRow = ARGBToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     ARGBToUVRow(src_argb, src_stride_argb, dst_u, dst_v, width);
michael@0:     ARGBToYRow(src_argb, dst_y, width);
michael@0:     ARGBToYRow(src_argb + src_stride_argb, dst_y + dst_stride_y, width);
michael@0:     src_argb += src_stride_argb * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     ARGBToUVRow(src_argb, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(src_argb, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert BGRA to I420.
michael@0: LIBYUV_API
michael@0: int BGRAToI420(const uint8* src_bgra, int src_stride_bgra,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_bgra ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_bgra = src_bgra + (height - 1) * src_stride_bgra;
michael@0:     src_stride_bgra = -src_stride_bgra;
michael@0:   }
michael@0:   void (*BGRAToUVRow)(const uint8* src_bgra0, int src_stride_bgra,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = BGRAToUVRow_C;
michael@0:   void (*BGRAToYRow)(const uint8* src_bgra, uint8* dst_y, int pix) =
michael@0:       BGRAToYRow_C;
michael@0: #if defined(HAS_BGRATOYROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     BGRAToUVRow = BGRAToUVRow_Any_SSSE3;
michael@0:     BGRAToYRow = BGRAToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       BGRAToUVRow = BGRAToUVRow_Unaligned_SSSE3;
michael@0:       BGRAToYRow = BGRAToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(src_bgra, 16) && IS_ALIGNED(src_stride_bgra, 16)) {
michael@0:         BGRAToUVRow = BGRAToUVRow_SSSE3;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           BGRAToYRow = BGRAToYRow_SSSE3;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #elif defined(HAS_BGRATOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     BGRAToYRow = BGRAToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       BGRAToYRow = BGRAToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       BGRAToUVRow = BGRAToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         BGRAToUVRow = BGRAToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     BGRAToUVRow(src_bgra, src_stride_bgra, dst_u, dst_v, width);
michael@0:     BGRAToYRow(src_bgra, dst_y, width);
michael@0:     BGRAToYRow(src_bgra + src_stride_bgra, dst_y + dst_stride_y, width);
michael@0:     src_bgra += src_stride_bgra * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     BGRAToUVRow(src_bgra, 0, dst_u, dst_v, width);
michael@0:     BGRAToYRow(src_bgra, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert ABGR to I420.
michael@0: LIBYUV_API
michael@0: int ABGRToI420(const uint8* src_abgr, int src_stride_abgr,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_abgr ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_abgr = src_abgr + (height - 1) * src_stride_abgr;
michael@0:     src_stride_abgr = -src_stride_abgr;
michael@0:   }
michael@0:   void (*ABGRToUVRow)(const uint8* src_abgr0, int src_stride_abgr,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ABGRToUVRow_C;
michael@0:   void (*ABGRToYRow)(const uint8* src_abgr, uint8* dst_y, int pix) =
michael@0:       ABGRToYRow_C;
michael@0: #if defined(HAS_ABGRTOYROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ABGRToUVRow = ABGRToUVRow_Any_SSSE3;
michael@0:     ABGRToYRow = ABGRToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ABGRToUVRow = ABGRToUVRow_Unaligned_SSSE3;
michael@0:       ABGRToYRow = ABGRToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(src_abgr, 16) && IS_ALIGNED(src_stride_abgr, 16)) {
michael@0:         ABGRToUVRow = ABGRToUVRow_SSSE3;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           ABGRToYRow = ABGRToYRow_SSSE3;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #elif defined(HAS_ABGRTOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     ABGRToYRow = ABGRToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ABGRToYRow = ABGRToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       ABGRToUVRow = ABGRToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         ABGRToUVRow = ABGRToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     ABGRToUVRow(src_abgr, src_stride_abgr, dst_u, dst_v, width);
michael@0:     ABGRToYRow(src_abgr, dst_y, width);
michael@0:     ABGRToYRow(src_abgr + src_stride_abgr, dst_y + dst_stride_y, width);
michael@0:     src_abgr += src_stride_abgr * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     ABGRToUVRow(src_abgr, 0, dst_u, dst_v, width);
michael@0:     ABGRToYRow(src_abgr, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert RGBA to I420.
michael@0: LIBYUV_API
michael@0: int RGBAToI420(const uint8* src_rgba, int src_stride_rgba,
michael@0:                uint8* dst_y, int dst_stride_y,
michael@0:                uint8* dst_u, int dst_stride_u,
michael@0:                uint8* dst_v, int dst_stride_v,
michael@0:                int width, int height) {
michael@0:   if (!src_rgba ||
michael@0:       !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_rgba = src_rgba + (height - 1) * src_stride_rgba;
michael@0:     src_stride_rgba = -src_stride_rgba;
michael@0:   }
michael@0:   void (*RGBAToUVRow)(const uint8* src_rgba0, int src_stride_rgba,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = RGBAToUVRow_C;
michael@0:   void (*RGBAToYRow)(const uint8* src_rgba, uint8* dst_y, int pix) =
michael@0:       RGBAToYRow_C;
michael@0: #if defined(HAS_RGBATOYROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     RGBAToUVRow = RGBAToUVRow_Any_SSSE3;
michael@0:     RGBAToYRow = RGBAToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       RGBAToUVRow = RGBAToUVRow_Unaligned_SSSE3;
michael@0:       RGBAToYRow = RGBAToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(src_rgba, 16) && IS_ALIGNED(src_stride_rgba, 16)) {
michael@0:         RGBAToUVRow = RGBAToUVRow_SSSE3;
michael@0:         if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:           RGBAToYRow = RGBAToYRow_SSSE3;
michael@0:         }
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #elif defined(HAS_RGBATOYROW_NEON)
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     RGBAToYRow = RGBAToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       RGBAToYRow = RGBAToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       RGBAToUVRow = RGBAToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         RGBAToUVRow = RGBAToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0:     RGBAToUVRow(src_rgba, src_stride_rgba, dst_u, dst_v, width);
michael@0:     RGBAToYRow(src_rgba, dst_y, width);
michael@0:     RGBAToYRow(src_rgba + src_stride_rgba, dst_y + dst_stride_y, width);
michael@0:     src_rgba += src_stride_rgba * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0:     RGBAToUVRow(src_rgba, 0, dst_u, dst_v, width);
michael@0:     RGBAToYRow(src_rgba, dst_y, width);
michael@0:   }
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert RGB24 to I420.
michael@0: LIBYUV_API
michael@0: int RGB24ToI420(const uint8* src_rgb24, int src_stride_rgb24,
michael@0:                 uint8* dst_y, int dst_stride_y,
michael@0:                 uint8* dst_u, int dst_stride_u,
michael@0:                 uint8* dst_v, int dst_stride_v,
michael@0:                 int width, int height) {
michael@0:   if (!src_rgb24 || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_rgb24 = src_rgb24 + (height - 1) * src_stride_rgb24;
michael@0:     src_stride_rgb24 = -src_stride_rgb24;
michael@0:   }
michael@0: 
michael@0: #if defined(HAS_RGB24TOYROW_NEON)
michael@0:   void (*RGB24ToUVRow)(const uint8* src_rgb24, int src_stride_rgb24,
michael@0:       uint8* dst_u, uint8* dst_v, int width) = RGB24ToUVRow_C;
michael@0:   void (*RGB24ToYRow)(const uint8* src_rgb24, uint8* dst_y, int pix) =
michael@0:       RGB24ToYRow_C;
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     RGB24ToYRow = RGB24ToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       RGB24ToYRow = RGB24ToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       RGB24ToUVRow = RGB24ToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         RGB24ToUVRow = RGB24ToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #else  // HAS_RGB24TOYROW_NEON
michael@0: 
michael@0:   // Allocate 2 rows of ARGB.
michael@0:   const int kRowSize = (width * 4 + 15) & ~15;
michael@0:   align_buffer_64(row, kRowSize * 2);
michael@0: 
michael@0:   void (*RGB24ToARGBRow)(const uint8* src_rgb, uint8* dst_argb, int pix) =
michael@0:       RGB24ToARGBRow_C;
michael@0: #if defined(HAS_RGB24TOARGBROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     RGB24ToARGBRow = RGB24ToARGBRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       RGB24ToARGBRow = RGB24ToARGBRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:         ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif  // HAS_ARGBTOUVROW_SSSE3
michael@0: #endif  // HAS_RGB24TOYROW_NEON
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0: #if defined(HAS_RGB24TOYROW_NEON)
michael@0:     RGB24ToUVRow(src_rgb24, src_stride_rgb24, dst_u, dst_v, width);
michael@0:     RGB24ToYRow(src_rgb24, dst_y, width);
michael@0:     RGB24ToYRow(src_rgb24 + src_stride_rgb24, dst_y + dst_stride_y, width);
michael@0: #else
michael@0:     RGB24ToARGBRow(src_rgb24, row, width);
michael@0:     RGB24ToARGBRow(src_rgb24 + src_stride_rgb24, row + kRowSize, width);
michael@0:     ARGBToUVRow(row, kRowSize, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0:     ARGBToYRow(row + kRowSize, dst_y + dst_stride_y, width);
michael@0: #endif
michael@0:     src_rgb24 += src_stride_rgb24 * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0: #if defined(HAS_RGB24TOYROW_NEON)
michael@0:     RGB24ToUVRow(src_rgb24, 0, dst_u, dst_v, width);
michael@0:     RGB24ToYRow(src_rgb24, dst_y, width);
michael@0: #else
michael@0:     RGB24ToARGBRow(src_rgb24, row, width);
michael@0:     ARGBToUVRow(row, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0: #endif
michael@0:   }
michael@0: #if !defined(HAS_RGB24TOYROW_NEON)
michael@0:   free_aligned_buffer_64(row);
michael@0: #endif
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert RAW to I420.
michael@0: LIBYUV_API
michael@0: int RAWToI420(const uint8* src_raw, int src_stride_raw,
michael@0:               uint8* dst_y, int dst_stride_y,
michael@0:               uint8* dst_u, int dst_stride_u,
michael@0:               uint8* dst_v, int dst_stride_v,
michael@0:               int width, int height) {
michael@0:   if (!src_raw || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_raw = src_raw + (height - 1) * src_stride_raw;
michael@0:     src_stride_raw = -src_stride_raw;
michael@0:   }
michael@0: 
michael@0: #if defined(HAS_RAWTOYROW_NEON)
michael@0:   void (*RAWToUVRow)(const uint8* src_raw, int src_stride_raw,
michael@0:       uint8* dst_u, uint8* dst_v, int width) = RAWToUVRow_C;
michael@0:   void (*RAWToYRow)(const uint8* src_raw, uint8* dst_y, int pix) =
michael@0:       RAWToYRow_C;
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     RAWToYRow = RAWToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       RAWToYRow = RAWToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       RAWToUVRow = RAWToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         RAWToUVRow = RAWToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #else  // HAS_RAWTOYROW_NEON
michael@0: 
michael@0:   // Allocate 2 rows of ARGB.
michael@0:   const int kRowSize = (width * 4 + 15) & ~15;
michael@0:   align_buffer_64(row, kRowSize * 2);
michael@0: 
michael@0:   void (*RAWToARGBRow)(const uint8* src_rgb, uint8* dst_argb, int pix) =
michael@0:       RAWToARGBRow_C;
michael@0: #if defined(HAS_RAWTOARGBROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     RAWToARGBRow = RAWToARGBRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       RAWToARGBRow = RAWToARGBRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:         ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif  // HAS_ARGBTOUVROW_SSSE3
michael@0: #endif  // HAS_RAWTOYROW_NEON
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0: #if defined(HAS_RAWTOYROW_NEON)
michael@0:     RAWToUVRow(src_raw, src_stride_raw, dst_u, dst_v, width);
michael@0:     RAWToYRow(src_raw, dst_y, width);
michael@0:     RAWToYRow(src_raw + src_stride_raw, dst_y + dst_stride_y, width);
michael@0: #else
michael@0:     RAWToARGBRow(src_raw, row, width);
michael@0:     RAWToARGBRow(src_raw + src_stride_raw, row + kRowSize, width);
michael@0:     ARGBToUVRow(row, kRowSize, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0:     ARGBToYRow(row + kRowSize, dst_y + dst_stride_y, width);
michael@0: #endif
michael@0:     src_raw += src_stride_raw * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0: #if defined(HAS_RAWTOYROW_NEON)
michael@0:     RAWToUVRow(src_raw, 0, dst_u, dst_v, width);
michael@0:     RAWToYRow(src_raw, dst_y, width);
michael@0: #else
michael@0:     RAWToARGBRow(src_raw, row, width);
michael@0:     ARGBToUVRow(row, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0: #endif
michael@0:   }
michael@0: #if !defined(HAS_RAWTOYROW_NEON)
michael@0:   free_aligned_buffer_64(row);
michael@0: #endif
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert RGB565 to I420.
michael@0: LIBYUV_API
michael@0: int RGB565ToI420(const uint8* src_rgb565, int src_stride_rgb565,
michael@0:                 uint8* dst_y, int dst_stride_y,
michael@0:                 uint8* dst_u, int dst_stride_u,
michael@0:                 uint8* dst_v, int dst_stride_v,
michael@0:                 int width, int height) {
michael@0:   if (!src_rgb565 || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_rgb565 = src_rgb565 + (height - 1) * src_stride_rgb565;
michael@0:     src_stride_rgb565 = -src_stride_rgb565;
michael@0:   }
michael@0: 
michael@0: #if defined(HAS_RGB565TOYROW_NEON)
michael@0:   void (*RGB565ToUVRow)(const uint8* src_rgb565, int src_stride_rgb565,
michael@0:       uint8* dst_u, uint8* dst_v, int width) = RGB565ToUVRow_C;
michael@0:   void (*RGB565ToYRow)(const uint8* src_rgb565, uint8* dst_y, int pix) =
michael@0:       RGB565ToYRow_C;
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     RGB565ToYRow = RGB565ToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       RGB565ToYRow = RGB565ToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       RGB565ToUVRow = RGB565ToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         RGB565ToUVRow = RGB565ToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #else  // HAS_RGB565TOYROW_NEON
michael@0: 
michael@0:   // Allocate 2 rows of ARGB.
michael@0:   const int kRowSize = (width * 4 + 15) & ~15;
michael@0:   align_buffer_64(row, kRowSize * 2);
michael@0: 
michael@0:   void (*RGB565ToARGBRow)(const uint8* src_rgb, uint8* dst_argb, int pix) =
michael@0:       RGB565ToARGBRow_C;
michael@0: #if defined(HAS_RGB565TOARGBROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 8) {
michael@0:     RGB565ToARGBRow = RGB565ToARGBRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       RGB565ToARGBRow = RGB565ToARGBRow_SSE2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:         ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif  // HAS_ARGBTOUVROW_SSSE3
michael@0: #endif  // HAS_RGB565TOYROW_NEON
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0: #if defined(HAS_RGB565TOYROW_NEON)
michael@0:     RGB565ToUVRow(src_rgb565, src_stride_rgb565, dst_u, dst_v, width);
michael@0:     RGB565ToYRow(src_rgb565, dst_y, width);
michael@0:     RGB565ToYRow(src_rgb565 + src_stride_rgb565, dst_y + dst_stride_y, width);
michael@0: #else
michael@0:     RGB565ToARGBRow(src_rgb565, row, width);
michael@0:     RGB565ToARGBRow(src_rgb565 + src_stride_rgb565, row + kRowSize, width);
michael@0:     ARGBToUVRow(row, kRowSize, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0:     ARGBToYRow(row + kRowSize, dst_y + dst_stride_y, width);
michael@0: #endif
michael@0:     src_rgb565 += src_stride_rgb565 * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0: #if defined(HAS_RGB565TOYROW_NEON)
michael@0:     RGB565ToUVRow(src_rgb565, 0, dst_u, dst_v, width);
michael@0:     RGB565ToYRow(src_rgb565, dst_y, width);
michael@0: #else
michael@0:     RGB565ToARGBRow(src_rgb565, row, width);
michael@0:     ARGBToUVRow(row, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0: #endif
michael@0:   }
michael@0: #if !defined(HAS_RGB565TOYROW_NEON)
michael@0:   free_aligned_buffer_64(row);
michael@0: #endif
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert ARGB1555 to I420.
michael@0: LIBYUV_API
michael@0: int ARGB1555ToI420(const uint8* src_argb1555, int src_stride_argb1555,
michael@0:                    uint8* dst_y, int dst_stride_y,
michael@0:                    uint8* dst_u, int dst_stride_u,
michael@0:                    uint8* dst_v, int dst_stride_v,
michael@0:                    int width, int height) {
michael@0:   if (!src_argb1555 || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_argb1555 = src_argb1555 + (height - 1) * src_stride_argb1555;
michael@0:     src_stride_argb1555 = -src_stride_argb1555;
michael@0:   }
michael@0: 
michael@0: #if defined(HAS_ARGB1555TOYROW_NEON)
michael@0:   void (*ARGB1555ToUVRow)(const uint8* src_argb1555, int src_stride_argb1555,
michael@0:       uint8* dst_u, uint8* dst_v, int width) = ARGB1555ToUVRow_C;
michael@0:   void (*ARGB1555ToYRow)(const uint8* src_argb1555, uint8* dst_y, int pix) =
michael@0:       ARGB1555ToYRow_C;
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     ARGB1555ToYRow = ARGB1555ToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ARGB1555ToYRow = ARGB1555ToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       ARGB1555ToUVRow = ARGB1555ToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         ARGB1555ToUVRow = ARGB1555ToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #else  // HAS_ARGB1555TOYROW_NEON
michael@0: 
michael@0:   // Allocate 2 rows of ARGB.
michael@0:   const int kRowSize = (width * 4 + 15) & ~15;
michael@0:   align_buffer_64(row, kRowSize * 2);
michael@0: 
michael@0:   void (*ARGB1555ToARGBRow)(const uint8* src_rgb, uint8* dst_argb, int pix) =
michael@0:       ARGB1555ToARGBRow_C;
michael@0: #if defined(HAS_ARGB1555TOARGBROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 8) {
michael@0:     ARGB1555ToARGBRow = ARGB1555ToARGBRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ARGB1555ToARGBRow = ARGB1555ToARGBRow_SSE2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:         ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif  // HAS_ARGBTOUVROW_SSSE3
michael@0: #endif  // HAS_ARGB1555TOYROW_NEON
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0: #if defined(HAS_ARGB1555TOYROW_NEON)
michael@0:     ARGB1555ToUVRow(src_argb1555, src_stride_argb1555, dst_u, dst_v, width);
michael@0:     ARGB1555ToYRow(src_argb1555, dst_y, width);
michael@0:     ARGB1555ToYRow(src_argb1555 + src_stride_argb1555, dst_y + dst_stride_y,
michael@0:                    width);
michael@0: #else
michael@0:     ARGB1555ToARGBRow(src_argb1555, row, width);
michael@0:     ARGB1555ToARGBRow(src_argb1555 + src_stride_argb1555, row + kRowSize,
michael@0:                       width);
michael@0:     ARGBToUVRow(row, kRowSize, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0:     ARGBToYRow(row + kRowSize, dst_y + dst_stride_y, width);
michael@0: #endif
michael@0:     src_argb1555 += src_stride_argb1555 * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0: #if defined(HAS_ARGB1555TOYROW_NEON)
michael@0:     ARGB1555ToUVRow(src_argb1555, 0, dst_u, dst_v, width);
michael@0:     ARGB1555ToYRow(src_argb1555, dst_y, width);
michael@0: #else
michael@0:     ARGB1555ToARGBRow(src_argb1555, row, width);
michael@0:     ARGBToUVRow(row, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0: #endif
michael@0:   }
michael@0: #if !defined(HAS_ARGB1555TOYROW_NEON)
michael@0:   free_aligned_buffer_64(row);
michael@0: #endif
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: // Convert ARGB4444 to I420.
michael@0: LIBYUV_API
michael@0: int ARGB4444ToI420(const uint8* src_argb4444, int src_stride_argb4444,
michael@0:                    uint8* dst_y, int dst_stride_y,
michael@0:                    uint8* dst_u, int dst_stride_u,
michael@0:                    uint8* dst_v, int dst_stride_v,
michael@0:                    int width, int height) {
michael@0:   if (!src_argb4444 || !dst_y || !dst_u || !dst_v ||
michael@0:       width <= 0 || height == 0) {
michael@0:     return -1;
michael@0:   }
michael@0:   // Negative height means invert the image.
michael@0:   if (height < 0) {
michael@0:     height = -height;
michael@0:     src_argb4444 = src_argb4444 + (height - 1) * src_stride_argb4444;
michael@0:     src_stride_argb4444 = -src_stride_argb4444;
michael@0:   }
michael@0: 
michael@0: #if defined(HAS_ARGB4444TOYROW_NEON)
michael@0:   void (*ARGB4444ToUVRow)(const uint8* src_argb4444, int src_stride_argb4444,
michael@0:       uint8* dst_u, uint8* dst_v, int width) = ARGB4444ToUVRow_C;
michael@0:   void (*ARGB4444ToYRow)(const uint8* src_argb4444, uint8* dst_y, int pix) =
michael@0:       ARGB4444ToYRow_C;
michael@0:   if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
michael@0:     ARGB4444ToYRow = ARGB4444ToYRow_Any_NEON;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ARGB4444ToYRow = ARGB4444ToYRow_NEON;
michael@0:     }
michael@0:     if (width >= 16) {
michael@0:       ARGB4444ToUVRow = ARGB4444ToUVRow_Any_NEON;
michael@0:       if (IS_ALIGNED(width, 16)) {
michael@0:         ARGB4444ToUVRow = ARGB4444ToUVRow_NEON;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #else  // HAS_ARGB4444TOYROW_NEON
michael@0: 
michael@0:   // Allocate 2 rows of ARGB.
michael@0:   const int kRowSize = (width * 4 + 15) & ~15;
michael@0:   align_buffer_64(row, kRowSize * 2);
michael@0: 
michael@0:   void (*ARGB4444ToARGBRow)(const uint8* src_rgb, uint8* dst_argb, int pix) =
michael@0:       ARGB4444ToARGBRow_C;
michael@0: #if defined(HAS_ARGB4444TOARGBROW_SSE2)
michael@0:   if (TestCpuFlag(kCpuHasSSE2) && width >= 8) {
michael@0:     ARGB4444ToARGBRow = ARGB4444ToARGBRow_Any_SSE2;
michael@0:     if (IS_ALIGNED(width, 8)) {
michael@0:       ARGB4444ToARGBRow = ARGB4444ToARGBRow_SSE2;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToUVRow)(const uint8* src_argb0, int src_stride_argb,
michael@0:                       uint8* dst_u, uint8* dst_v, int width) = ARGBToUVRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToUVRow = ARGBToUVRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToUVRow = ARGBToUVRow_SSSE3;
michael@0:     }
michael@0:   }
michael@0: #endif
michael@0:   void (*ARGBToYRow)(const uint8* src_argb, uint8* dst_y, int pix) =
michael@0:       ARGBToYRow_C;
michael@0: #if defined(HAS_ARGBTOUVROW_SSSE3)
michael@0:   if (TestCpuFlag(kCpuHasSSSE3) && width >= 16) {
michael@0:     ARGBToYRow = ARGBToYRow_Any_SSSE3;
michael@0:     if (IS_ALIGNED(width, 16)) {
michael@0:       ARGBToYRow = ARGBToYRow_Unaligned_SSSE3;
michael@0:       if (IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
michael@0:         ARGBToYRow = ARGBToYRow_SSSE3;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: #endif  // HAS_ARGBTOUVROW_SSSE3
michael@0: #endif  // HAS_ARGB4444TOYROW_NEON
michael@0: 
michael@0:   for (int y = 0; y < height - 1; y += 2) {
michael@0: #if defined(HAS_ARGB4444TOYROW_NEON)
michael@0:     ARGB4444ToUVRow(src_argb4444, src_stride_argb4444, dst_u, dst_v, width);
michael@0:     ARGB4444ToYRow(src_argb4444, dst_y, width);
michael@0:     ARGB4444ToYRow(src_argb4444 + src_stride_argb4444, dst_y + dst_stride_y,
michael@0:                    width);
michael@0: #else
michael@0:     ARGB4444ToARGBRow(src_argb4444, row, width);
michael@0:     ARGB4444ToARGBRow(src_argb4444 + src_stride_argb4444, row + kRowSize,
michael@0:                       width);
michael@0:     ARGBToUVRow(row, kRowSize, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0:     ARGBToYRow(row + kRowSize, dst_y + dst_stride_y, width);
michael@0: #endif
michael@0:     src_argb4444 += src_stride_argb4444 * 2;
michael@0:     dst_y += dst_stride_y * 2;
michael@0:     dst_u += dst_stride_u;
michael@0:     dst_v += dst_stride_v;
michael@0:   }
michael@0:   if (height & 1) {
michael@0: #if defined(HAS_ARGB4444TOYROW_NEON)
michael@0:     ARGB4444ToUVRow(src_argb4444, 0, dst_u, dst_v, width);
michael@0:     ARGB4444ToYRow(src_argb4444, dst_y, width);
michael@0: #else
michael@0:     ARGB4444ToARGBRow(src_argb4444, row, width);
michael@0:     ARGBToUVRow(row, 0, dst_u, dst_v, width);
michael@0:     ARGBToYRow(row, dst_y, width);
michael@0: #endif
michael@0:   }
michael@0: #if !defined(HAS_ARGB4444TOYROW_NEON)
michael@0:   free_aligned_buffer_64(row);
michael@0: #endif
michael@0:   return 0;
michael@0: }
michael@0: 
michael@0: #ifdef __cplusplus
michael@0: }  // extern "C"
michael@0: }  // namespace libyuv
michael@0: #endif