Remove M420 and refactor NV12ToI420

M420 is a row biplanar variation of NV12 supported on Microsoft webcams.
The code was hardcoded to bt.601 and should be jpeg, but the format is
very old and rare.  Is a variation on NV12, so if someone needs it, it
can be re-implemented easily.

Bug: libyuv:858
Change-Id: I246167dba3c190cc76af741b8e91e58e68fde28f
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/2212608
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>

14 files changed, 37 insertions, 241 deletions

diff --git a/README.chromium b/README.chromium
index abf7e6f2..32f5c883 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1756
+Version: 1757
 License: BSD
 License File: LICENSE
 
diff --git a/docs/formats.md b/docs/formats.md
index 771fb460..a29ed5c3 100644
--- a/docs/formats.md
+++ b/docs/formats.md
@@ -50,7 +50,7 @@ The following is extracted from video_common.h as a complete list of formats sup
                                                  // 10 bit lsb
 
       // 1 Secondary YUV format: row biplanar.
-      FOURCC_M420 = FOURCC('M', '4', '2', '0'),
+      FOURCC_M420 = FOURCC('M', '4', '2', '0'),  // deprecated.
 
       // 11 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp, 1 10 bpc
       FOURCC_ARGB = FOURCC('A', 'R', 'G', 'B'),
diff --git a/include/libyuv/convert.h b/include/libyuv/convert.h
index b1bb5424..91d81ac0 100644
--- a/include/libyuv/convert.h
+++ b/include/libyuv/convert.h
@@ -263,19 +263,6 @@ int AYUVToNV21(const uint8_t* src_ayuv,
                int width,
                int height);
 
-// Convert M420 to I420.
-LIBYUV_API
-int M420ToI420(const uint8_t* src_m420,
-               int src_stride_m420,
-               uint8_t* dst_y,
-               int dst_stride_y,
-               uint8_t* dst_u,
-               int dst_stride_u,
-               uint8_t* dst_v,
-               int dst_stride_v,
-               int width,
-               int height);
-
 // Convert Android420 to I420.
 LIBYUV_API
 int Android420ToI420(const uint8_t* src_y,
diff --git a/include/libyuv/convert_argb.h b/include/libyuv/convert_argb.h
index fd5f42dc..6f55e11b 100644
--- a/include/libyuv/convert_argb.h
+++ b/include/libyuv/convert_argb.h
@@ -664,15 +664,6 @@ int NV21ToRAW(const uint8_t* src_y,
               int width,
               int height);
 
-// Convert M420 to ARGB.
-LIBYUV_API
-int M420ToARGB(const uint8_t* src_m420,
-               int src_stride_m420,
-               uint8_t* dst_argb,
-               int dst_stride_argb,
-               int width,
-               int height);
-
 // Convert YUY2 to ARGB.
 LIBYUV_API
 int YUY2ToARGB(const uint8_t* src_yuy2,
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 7af8c605..dcd864f1 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1756
+#define LIBYUV_VERSION 1757
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_
diff --git a/include/libyuv/video_common.h b/include/libyuv/video_common.h
index cbfe9573..b9823d71 100644
--- a/include/libyuv/video_common.h
+++ b/include/libyuv/video_common.h
@@ -62,7 +62,7 @@ enum FourCC {
   FOURCC_I010 = FOURCC('I', '0', '1', '0'),  // bt.601 10 bit 420
   FOURCC_I210 = FOURCC('I', '0', '1', '0'),  // bt.601 10 bit 422
 
-  // 1 Secondary YUV format: row biplanar.
+  // 1 Secondary YUV format: row biplanar.  deprecated.
   FOURCC_M420 = FOURCC('M', '4', '2', '0'),
 
   // 11 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp, 1 10 bpc
@@ -148,7 +148,7 @@ enum FourCCBpp {
   FOURCC_BPP_NV12 = 12,
   FOURCC_BPP_YUY2 = 16,
   FOURCC_BPP_UYVY = 16,
-  FOURCC_BPP_M420 = 12,
+  FOURCC_BPP_M420 = 12,  // deprecated
   FOURCC_BPP_Q420 = 12,
   FOURCC_BPP_ARGB = 32,
   FOURCC_BPP_BGRA = 32,
diff --git a/source/convert.cc b/source/convert.cc
index e3e282eb..3886f3f0 100644
--- a/source/convert.cc
+++ b/source/convert.cc
@@ -540,70 +540,21 @@ int I400ToNV21(const uint8_t* src_y,
   return 0;
 }
 
-static void CopyPlane2(const uint8_t* src,
-                       int src_stride_0,
-                       int src_stride_1,
-                       uint8_t* dst,
-                       int dst_stride,
-                       int width,
-                       int height) {
-  int y;
-  void (*CopyRow)(const uint8_t* src, uint8_t* dst, int width) = CopyRow_C;
-#if defined(HAS_COPYROW_SSE2)
-  if (TestCpuFlag(kCpuHasSSE2)) {
-    CopyRow = IS_ALIGNED(width, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
-  }
-#endif
-#if defined(HAS_COPYROW_AVX)
-  if (TestCpuFlag(kCpuHasAVX)) {
-    CopyRow = IS_ALIGNED(width, 64) ? CopyRow_AVX : CopyRow_Any_AVX;
-  }
-#endif
-#if defined(HAS_COPYROW_ERMS)
-  if (TestCpuFlag(kCpuHasERMS)) {
-    CopyRow = CopyRow_ERMS;
-  }
-#endif
-#if defined(HAS_COPYROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON)) {
-    CopyRow = IS_ALIGNED(width, 32) ? CopyRow_NEON : CopyRow_Any_NEON;
-  }
-#endif
-
-  // Copy plane
-  for (y = 0; y < height - 1; y += 2) {
-    CopyRow(src, dst, width);
-    CopyRow(src + src_stride_0, dst + dst_stride, width);
-    src += src_stride_0 + src_stride_1;
-    dst += dst_stride * 2;
-  }
-  if (height & 1) {
-    CopyRow(src, dst, width);
-  }
-}
-
-// Support converting from FOURCC_M420
-// Useful for bandwidth constrained transports like USB 1.0 and 2.0 and for
-// easy conversion to I420.
-// M420 format description:
-// M420 is row biplanar 420: 2 rows of Y and 1 row of UV.
-// Chroma is half width / half height. (420)
-// src_stride_m420 is row planar. Normally this will be the width in pixels.
-//   The UV plane is half width, but 2 values, so src_stride_m420 applies to
-//   this as well as the two Y planes.
-static int X420ToI420(const uint8_t* src_y,
-                      int src_stride_y0,
-                      int src_stride_y1,
-                      const uint8_t* src_uv,
-                      int src_stride_uv,
-                      uint8_t* dst_y,
-                      int dst_stride_y,
-                      uint8_t* dst_u,
-                      int dst_stride_u,
-                      uint8_t* dst_v,
-                      int dst_stride_v,
-                      int width,
-                      int height) {
+// Convert NV12 to I420.
+// TODO(fbarchard): Consider inverting destination. Faster on ARM with prfm.
+LIBYUV_API
+int NV12ToI420(const uint8_t* src_y,
+               int src_stride_y,
+               const uint8_t* src_uv,
+               int src_stride_uv,
+               uint8_t* dst_y,
+               int dst_stride_y,
+               uint8_t* dst_u,
+               int dst_stride_u,
+               uint8_t* dst_v,
+               int dst_stride_v,
+               int width,
+               int height) {
   int halfwidth = (width + 1) >> 1;
   int halfheight = (height + 1) >> 1;
   if (!src_uv || !dst_u || !dst_v || width <= 0 || height == 0) {
@@ -613,21 +564,16 @@ static int X420ToI420(const uint8_t* src_y,
   if (height < 0) {
     height = -height;
     halfheight = (height + 1) >> 1;
-    if (dst_y) {
-      dst_y = dst_y + (height - 1) * dst_stride_y;
-    }
-    dst_u = dst_u + (halfheight - 1) * dst_stride_u;
-    dst_v = dst_v + (halfheight - 1) * dst_stride_v;
-    dst_stride_y = -dst_stride_y;
-    dst_stride_u = -dst_stride_u;
-    dst_stride_v = -dst_stride_v;
+    src_y = src_y + (height - 1) * src_stride_y;
+    src_uv = src_uv + (halfheight - 1) * src_stride_uv;
+    src_stride_y = -src_stride_y;
+    src_stride_uv = -src_stride_uv;
   }
   // Coalesce rows.
-  if (src_stride_y0 == width && src_stride_y1 == width &&
-      dst_stride_y == width) {
+  if (src_stride_y == width && dst_stride_y == width) {
     width *= height;
     height = 1;
-    src_stride_y0 = src_stride_y1 = dst_stride_y = 0;
+    src_stride_y = dst_stride_y = 0;
   }
   // Coalesce rows.
   if (src_stride_uv == halfwidth * 2 && dst_stride_u == halfwidth &&
@@ -638,12 +584,7 @@ static int X420ToI420(const uint8_t* src_y,
   }
 
   if (dst_y) {
-    if (src_stride_y0 == src_stride_y1) {
-      CopyPlane(src_y, src_stride_y0, dst_y, dst_stride_y, width, height);
-    } else {
-      CopyPlane2(src_y, src_stride_y0, src_stride_y1, dst_y, dst_stride_y,
-                 width, height);
-    }
+    CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
   }
 
   // Split UV plane - NV12 / NV21
@@ -653,25 +594,6 @@ static int X420ToI420(const uint8_t* src_y,
   return 0;
 }
 
-// Convert NV12 to I420.
-LIBYUV_API
-int NV12ToI420(const uint8_t* src_y,
-               int src_stride_y,
-               const uint8_t* src_uv,
-               int src_stride_uv,
-               uint8_t* dst_y,
-               int dst_stride_y,
-               uint8_t* dst_u,
-               int dst_stride_u,
-               uint8_t* dst_v,
-               int dst_stride_v,
-               int width,
-               int height) {
-  return X420ToI420(src_y, src_stride_y, src_stride_y, src_uv, src_stride_uv,
-                    dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
-                    dst_stride_v, width, height);
-}
-
 // Convert NV21 to I420.  Same as NV12 but u and v pointers swapped.
 LIBYUV_API
 int NV21ToI420(const uint8_t* src_y,
@@ -686,27 +608,11 @@ int NV21ToI420(const uint8_t* src_y,
                int dst_stride_v,
                int width,
                int height) {
-  return X420ToI420(src_y, src_stride_y, src_stride_y, src_vu, src_stride_vu,
-                    dst_y, dst_stride_y, dst_v, dst_stride_v, dst_u,
-                    dst_stride_u, width, height);
-}
-
-// Convert M420 to I420.
-LIBYUV_API
-int M420ToI420(const uint8_t* src_m420,
-               int src_stride_m420,
-               uint8_t* dst_y,
-               int dst_stride_y,
-               uint8_t* dst_u,
-               int dst_stride_u,
-               uint8_t* dst_v,
-               int dst_stride_v,
-               int width,
-               int height) {
-  return X420ToI420(src_m420, src_stride_m420, src_stride_m420 * 2,
-                    src_m420 + src_stride_m420 * 2, src_stride_m420 * 3, dst_y,
-                    dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v,
-                    width, height);
+  return NV12ToI420(src_y, src_stride_y,
+                    src_vu, src_stride_vu,
+                    dst_y, dst_stride_y,
+                    dst_v, dst_stride_v,
+                    dst_u, dst_stride_u, width, height);
 }
 
 // Convert YUY2 to I420.
diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index 14fae1db..5e7225fa 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -2753,83 +2753,6 @@ int NV21ToYUV24(const uint8_t* src_y,
   return 0;
 }
 
-// Convert M420 to ARGB.
-LIBYUV_API
-int M420ToARGB(const uint8_t* src_m420,
-               int src_stride_m420,
-               uint8_t* dst_argb,
-               int dst_stride_argb,
-               int width,
-               int height) {
-  int y;
-  void (*NV12ToARGBRow)(
-      const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* rgb_buf,
-      const struct YuvConstants* yuvconstants, int width) = NV12ToARGBRow_C;
-  if (!src_m420 || !dst_argb || width <= 0 || height == 0) {
-    return -1;
-  }
-  // Negative height means invert the image.
-  if (height < 0) {
-    height = -height;
-    dst_argb = dst_argb + (height - 1) * dst_stride_argb;
-    dst_stride_argb = -dst_stride_argb;
-  }
-#if defined(HAS_NV12TOARGBROW_SSSE3)
-  if (TestCpuFlag(kCpuHasSSSE3)) {
-    NV12ToARGBRow = NV12ToARGBRow_Any_SSSE3;
-    if (IS_ALIGNED(width, 8)) {
-      NV12ToARGBRow = NV12ToARGBRow_SSSE3;
-    }
-  }
-#endif
-#if defined(HAS_NV12TOARGBROW_AVX2)
-  if (TestCpuFlag(kCpuHasAVX2)) {
-    NV12ToARGBRow = NV12ToARGBRow_Any_AVX2;
-    if (IS_ALIGNED(width, 16)) {
-      NV12ToARGBRow = NV12ToARGBRow_AVX2;
-    }
-  }
-#endif
-#if defined(HAS_NV12TOARGBROW_NEON)
-  if (TestCpuFlag(kCpuHasNEON)) {
-    NV12ToARGBRow = NV12ToARGBRow_Any_NEON;
-    if (IS_ALIGNED(width, 8)) {
-      NV12ToARGBRow = NV12ToARGBRow_NEON;
-    }
-  }
-#endif
-#if defined(HAS_NV12TOARGBROW_MMI)
-  if (TestCpuFlag(kCpuHasMMI)) {
-    NV12ToARGBRow = NV12ToARGBRow_Any_MMI;
-    if (IS_ALIGNED(width, 4)) {
-      NV12ToARGBRow = NV12ToARGBRow_MMI;
-    }
-  }
-#endif
-#if defined(HAS_NV12TOARGBROW_MSA)
-  if (TestCpuFlag(kCpuHasMSA)) {
-    NV12ToARGBRow = NV12ToARGBRow_Any_MSA;
-    if (IS_ALIGNED(width, 8)) {
-      NV12ToARGBRow = NV12ToARGBRow_MSA;
-    }
-  }
-#endif
-
-  for (y = 0; y < height - 1; y += 2) {
-    NV12ToARGBRow(src_m420, src_m420 + src_stride_m420 * 2, dst_argb,
-                  &kYuvI601Constants, width);
-    NV12ToARGBRow(src_m420 + src_stride_m420, src_m420 + src_stride_m420 * 2,
-                  dst_argb + dst_stride_argb, &kYuvI601Constants, width);
-    dst_argb += dst_stride_argb * 2;
-    src_m420 += src_stride_m420 * 3;
-  }
-  if (height & 1) {
-    NV12ToARGBRow(src_m420, src_m420 + src_stride_m420 * 2, dst_argb,
-                  &kYuvI601Constants, width);
-  }
-  return 0;
-}
-
 // Convert YUY2 to ARGB.
 LIBYUV_API
 int YUY2ToARGB(const uint8_t* src_yuy2,
diff --git a/source/convert_from.cc b/source/convert_from.cc
index 140c90d9..f2cfc1d8 100644
--- a/source/convert_from.cc
+++ b/source/convert_from.cc
@@ -645,7 +645,6 @@ int ConvertFromI420(const uint8_t* y,
                      height);
       break;
     }
-    // TODO(fbarchard): Add M420.
     // Triplanar formats
     case FOURCC_I420:
     case FOURCC_YV12: {
diff --git a/source/convert_to_argb.cc b/source/convert_to_argb.cc
index c08f6101..84df16c8 100644
--- a/source/convert_to_argb.cc
+++ b/source/convert_to_argb.cc
@@ -180,12 +180,6 @@ int ConvertToARGB(const uint8_t* sample,
       r = NV21ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
                      dst_stride_argb, crop_width, inv_crop_height);
       break;
-    case FOURCC_M420:
-      src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
-      r = M420ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
-                     inv_crop_height);
-      break;
-
     // Triplanar formats
     case FOURCC_I420:
     case FOURCC_YV12: {
diff --git a/source/convert_to_i420.cc b/source/convert_to_i420.cc
index 584be0ac..ac6eeab2 100644
--- a/source/convert_to_i420.cc
+++ b/source/convert_to_i420.cc
@@ -179,11 +179,6 @@ int ConvertToI420(const uint8_t* sample,
                            dst_stride_y, dst_v, dst_stride_v, dst_u,
                            dst_stride_u, crop_width, inv_crop_height, rotation);
       break;
-    case FOURCC_M420:
-      src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
-      r = M420ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u,
-                     dst_v, dst_stride_v, crop_width, inv_crop_height);
-      break;
     // Triplanar formats
     case FOURCC_I420:
     case FOURCC_YV12: {
diff --git a/source/row_gcc.cc b/source/row_gcc.cc
index a5de6af2..9a18d594 100644
--- a/source/row_gcc.cc
+++ b/source/row_gcc.cc
@@ -3140,8 +3140,7 @@ void I400ToARGBRow_AVX2(const uint8_t* y_buf,
                         int width) {
   asm volatile(
       "vmovdqa    192(%3),%%ymm2                 \n"  // yg = 18997 = 1.164
-      "vmovdqa    224(%3),%%ymm3                 \n"  // ygb = -1160 = 1.164 *
-                                                      // 16
+      "vmovdqa    224(%3),%%ymm3                 \n"  // ygb = -1160 = 1.164*16
       "vpcmpeqb   %%ymm4,%%ymm4,%%ymm4           \n"  // 0xff000000
       "vpslld     $0x18,%%ymm4,%%ymm4            \n"
 
diff --git a/source/row_neon.cc b/source/row_neon.cc
index ae2d2830..a9a377a1 100644
--- a/source/row_neon.cc
+++ b/source/row_neon.cc
@@ -351,7 +351,9 @@ void I400ToARGBRow_NEON(const uint8_t* src_y,
   asm volatile(
       YUVTORGB_SETUP
       "vmov.u8    d23, #255                      \n"
-      "1:                                        \n" READYUV400 YUVTORGB
+      "1:                                        \n"
+      READYUV400
+      YUVTORGB
       "subs       %2, %2, #8                     \n"
       "vst4.8     {d20, d21, d22, d23}, [%1]!    \n"
       "bgt        1b                             \n"
diff --git a/unit_test/video_common_test.cc b/unit_test/video_common_test.cc
index a84206a2..eb183aaa 100644
--- a/unit_test/video_common_test.cc
+++ b/unit_test/video_common_test.cc
@@ -65,7 +65,7 @@ TEST_F(LibYUVBaseTest, TestFourCC) {
   EXPECT_TRUE(TestValidFourCC(FOURCC_NV12, FOURCC_BPP_NV12));
   EXPECT_TRUE(TestValidFourCC(FOURCC_YUY2, FOURCC_BPP_YUY2));
   EXPECT_TRUE(TestValidFourCC(FOURCC_UYVY, FOURCC_BPP_UYVY));
-  EXPECT_TRUE(TestValidFourCC(FOURCC_M420, FOURCC_BPP_M420));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_M420, FOURCC_BPP_M420));  // deprecated.
   EXPECT_TRUE(TestValidFourCC(FOURCC_Q420, FOURCC_BPP_Q420));  // deprecated.
   EXPECT_TRUE(TestValidFourCC(FOURCC_ARGB, FOURCC_BPP_ARGB));
   EXPECT_TRUE(TestValidFourCC(FOURCC_BGRA, FOURCC_BPP_BGRA));