shithub: libvpx

ref: 7d0e4f9ad1ae4a1957a2ea96b2f13971cc73da52
dir: /third_party/libyuv/source/scale_neon.cc/

View raw version
/*
 *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS. All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include "libyuv/row.h"

#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif

// This module is for GCC Neon.
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__)

// NEON downscalers with interpolation.
// Provided by Fritz Koenig

// Read 32x1 throw away even pixels, and write 16x1.
void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                        uint8* dst, int dst_width) {
  asm volatile (
    ".p2align   2                              \n"
  "1:                                          \n"
    // load even pixels into q0, odd into q1
    MEMACCESS(0)
    "vld2.8     {q0, q1}, [%0]!                \n"
    "subs       %2, %2, #16                    \n"  // 16 processed per loop
    MEMACCESS(1)
    "vst1.8     {q1}, [%1]!                    \n"  // store odd pixels
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst),              // %1
    "+r"(dst_width)         // %2
  :
  : "q0", "q1"              // Clobber List
  );
}

// Read 32x2 average down and write 16x1.
void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                           uint8* dst, int dst_width) {
  asm volatile (
    // change the stride to row 2 pointer
    "add        %1, %0                         \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld1.8     {q0, q1}, [%0]!                \n"  // load row 1 and post inc
    MEMACCESS(1)
    "vld1.8     {q2, q3}, [%1]!                \n"  // load row 2 and post inc
    "subs       %3, %3, #16                    \n"  // 16 processed per loop
    "vpaddl.u8  q0, q0                         \n"  // row 1 add adjacent
    "vpaddl.u8  q1, q1                         \n"
    "vpadal.u8  q0, q2                         \n"  // row 2 add adjacent + row1
    "vpadal.u8  q1, q3                         \n"
    "vrshrn.u16 d0, q0, #2                     \n"  // downshift, round and pack
    "vrshrn.u16 d1, q1, #2                     \n"
    MEMACCESS(2)
    "vst1.8     {q0}, [%2]!                    \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(src_stride),       // %1
    "+r"(dst),              // %2
    "+r"(dst_width)         // %3
  :
  : "q0", "q1", "q2", "q3"     // Clobber List
  );
}

void ScaleRowDown4_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                        uint8* dst_ptr, int dst_width) {
  asm volatile (
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld4.8     {d0, d1, d2, d3}, [%0]!        \n" // src line 0
    "subs       %2, %2, #8                     \n" // 8 processed per loop
    MEMACCESS(1)
    "vst1.8     {d2}, [%1]!                    \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width)         // %2
  :
  : "q0", "q1", "memory", "cc"
  );
}

void ScaleRowDown4Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                           uint8* dst_ptr, int dst_width) {
  const uint8* src_ptr1 = src_ptr + src_stride;
  const uint8* src_ptr2 = src_ptr + src_stride * 2;
  const uint8* src_ptr3 = src_ptr + src_stride * 3;
asm volatile (
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld1.8     {q0}, [%0]!                    \n"   // load up 16x4
    MEMACCESS(3)
    "vld1.8     {q1}, [%3]!                    \n"
    MEMACCESS(4)
    "vld1.8     {q2}, [%4]!                    \n"
    MEMACCESS(5)
    "vld1.8     {q3}, [%5]!                    \n"
    "subs       %2, %2, #4                     \n"
    "vpaddl.u8  q0, q0                         \n"
    "vpadal.u8  q0, q1                         \n"
    "vpadal.u8  q0, q2                         \n"
    "vpadal.u8  q0, q3                         \n"
    "vpaddl.u16 q0, q0                         \n"
    "vrshrn.u32 d0, q0, #4                     \n"   // divide by 16 w/rounding
    "vmovn.u16  d0, q0                         \n"
    MEMACCESS(1)
    "vst1.32    {d0[0]}, [%1]!                 \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),   // %0
    "+r"(dst_ptr),   // %1
    "+r"(dst_width), // %2
    "+r"(src_ptr1),  // %3
    "+r"(src_ptr2),  // %4
    "+r"(src_ptr3)   // %5
  :
  : "q0", "q1", "q2", "q3", "memory", "cc"
  );
}

// Down scale from 4 to 3 pixels. Use the neon multilane read/write
// to load up the every 4th pixel into a 4 different registers.
// Point samples 32 pixels to 24 pixels.
void ScaleRowDown34_NEON(const uint8* src_ptr,
                         ptrdiff_t src_stride,
                         uint8* dst_ptr, int dst_width) {
  asm volatile (
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld4.8     {d0, d1, d2, d3}, [%0]!      \n" // src line 0
    "subs       %2, %2, #24                  \n"
    "vmov       d2, d3                       \n" // order d0, d1, d2
    MEMACCESS(1)
    "vst3.8     {d0, d1, d2}, [%1]!          \n"
    "bgt        1b                           \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width)         // %2
  :
  : "d0", "d1", "d2", "d3", "memory", "cc"
  );
}

void ScaleRowDown34_0_Box_NEON(const uint8* src_ptr,
                               ptrdiff_t src_stride,
                               uint8* dst_ptr, int dst_width) {
  asm volatile (
    "vmov.u8    d24, #3                        \n"
    "add        %3, %0                         \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld4.8       {d0, d1, d2, d3}, [%0]!      \n" // src line 0
    MEMACCESS(3)
    "vld4.8       {d4, d5, d6, d7}, [%3]!      \n" // src line 1
    "subs         %2, %2, #24                  \n"

    // filter src line 0 with src line 1
    // expand chars to shorts to allow for room
    // when adding lines together
    "vmovl.u8     q8, d4                       \n"
    "vmovl.u8     q9, d5                       \n"
    "vmovl.u8     q10, d6                      \n"
    "vmovl.u8     q11, d7                      \n"

    // 3 * line_0 + line_1
    "vmlal.u8     q8, d0, d24                  \n"
    "vmlal.u8     q9, d1, d24                  \n"
    "vmlal.u8     q10, d2, d24                 \n"
    "vmlal.u8     q11, d3, d24                 \n"

    // (3 * line_0 + line_1) >> 2
    "vqrshrn.u16  d0, q8, #2                   \n"
    "vqrshrn.u16  d1, q9, #2                   \n"
    "vqrshrn.u16  d2, q10, #2                  \n"
    "vqrshrn.u16  d3, q11, #2                  \n"

    // a0 = (src[0] * 3 + s[1] * 1) >> 2
    "vmovl.u8     q8, d1                       \n"
    "vmlal.u8     q8, d0, d24                  \n"
    "vqrshrn.u16  d0, q8, #2                   \n"

    // a1 = (src[1] * 1 + s[2] * 1) >> 1
    "vrhadd.u8    d1, d1, d2                   \n"

    // a2 = (src[2] * 1 + s[3] * 3) >> 2
    "vmovl.u8     q8, d2                       \n"
    "vmlal.u8     q8, d3, d24                  \n"
    "vqrshrn.u16  d2, q8, #2                   \n"

    MEMACCESS(1)
    "vst3.8       {d0, d1, d2}, [%1]!          \n"

    "bgt          1b                           \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width),        // %2
    "+r"(src_stride)        // %3
  :
  : "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11", "d24", "memory", "cc"
  );
}

void ScaleRowDown34_1_Box_NEON(const uint8* src_ptr,
                               ptrdiff_t src_stride,
                               uint8* dst_ptr, int dst_width) {
  asm volatile (
    "vmov.u8    d24, #3                        \n"
    "add        %3, %0                         \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld4.8       {d0, d1, d2, d3}, [%0]!      \n" // src line 0
    MEMACCESS(3)
    "vld4.8       {d4, d5, d6, d7}, [%3]!      \n" // src line 1
    "subs         %2, %2, #24                  \n"
    // average src line 0 with src line 1
    "vrhadd.u8    q0, q0, q2                   \n"
    "vrhadd.u8    q1, q1, q3                   \n"

    // a0 = (src[0] * 3 + s[1] * 1) >> 2
    "vmovl.u8     q3, d1                       \n"
    "vmlal.u8     q3, d0, d24                  \n"
    "vqrshrn.u16  d0, q3, #2                   \n"

    // a1 = (src[1] * 1 + s[2] * 1) >> 1
    "vrhadd.u8    d1, d1, d2                   \n"

    // a2 = (src[2] * 1 + s[3] * 3) >> 2
    "vmovl.u8     q3, d2                       \n"
    "vmlal.u8     q3, d3, d24                  \n"
    "vqrshrn.u16  d2, q3, #2                   \n"

    MEMACCESS(1)
    "vst3.8       {d0, d1, d2}, [%1]!          \n"
    "bgt          1b                           \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width),        // %2
    "+r"(src_stride)        // %3
  :
  : "r4", "q0", "q1", "q2", "q3", "d24", "memory", "cc"
  );
}

#define HAS_SCALEROWDOWN38_NEON
static uvec8 kShuf38 =
  { 0, 3, 6, 8, 11, 14, 16, 19, 22, 24, 27, 30, 0, 0, 0, 0 };
static uvec8 kShuf38_2 =
  { 0, 8, 16, 2, 10, 17, 4, 12, 18, 6, 14, 19, 0, 0, 0, 0 };
static vec16 kMult38_Div6 =
  { 65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12,
    65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12 };
static vec16 kMult38_Div9 =
  { 65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18,
    65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18 };

// 32 -> 12
void ScaleRowDown38_NEON(const uint8* src_ptr,
                         ptrdiff_t src_stride,
                         uint8* dst_ptr, int dst_width) {
  asm volatile (
    MEMACCESS(3)
    "vld1.8     {q3}, [%3]                     \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld1.8     {d0, d1, d2, d3}, [%0]!        \n"
    "subs       %2, %2, #12                    \n"
    "vtbl.u8    d4, {d0, d1, d2, d3}, d6       \n"
    "vtbl.u8    d5, {d0, d1, d2, d3}, d7       \n"
    MEMACCESS(1)
    "vst1.8     {d4}, [%1]!                    \n"
    MEMACCESS(1)
    "vst1.32    {d5[0]}, [%1]!                 \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width)         // %2
  : "r"(&kShuf38)           // %3
  : "d0", "d1", "d2", "d3", "d4", "d5", "memory", "cc"
  );
}

// 32x3 -> 12x1
void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
                                      ptrdiff_t src_stride,
                                      uint8* dst_ptr, int dst_width) {
  const uint8* src_ptr1 = src_ptr + src_stride * 2;

  asm volatile (
    MEMACCESS(5)
    "vld1.16    {q13}, [%5]                    \n"
    MEMACCESS(6)
    "vld1.8     {q14}, [%6]                    \n"
    MEMACCESS(7)
    "vld1.8     {q15}, [%7]                    \n"
    "add        %3, %0                         \n"
    ".p2align   2                              \n"
  "1:                                          \n"

    // d0 = 00 40 01 41 02 42 03 43
    // d1 = 10 50 11 51 12 52 13 53
    // d2 = 20 60 21 61 22 62 23 63
    // d3 = 30 70 31 71 32 72 33 73
    MEMACCESS(0)
    "vld4.8       {d0, d1, d2, d3}, [%0]!      \n"
    MEMACCESS(3)
    "vld4.8       {d4, d5, d6, d7}, [%3]!      \n"
    MEMACCESS(4)
    "vld4.8       {d16, d17, d18, d19}, [%4]!  \n"
    "subs         %2, %2, #12                  \n"

    // Shuffle the input data around to get align the data
    //  so adjacent data can be added. 0,1 - 2,3 - 4,5 - 6,7
    // d0 = 00 10 01 11 02 12 03 13
    // d1 = 40 50 41 51 42 52 43 53
    "vtrn.u8      d0, d1                       \n"
    "vtrn.u8      d4, d5                       \n"
    "vtrn.u8      d16, d17                     \n"

    // d2 = 20 30 21 31 22 32 23 33
    // d3 = 60 70 61 71 62 72 63 73
    "vtrn.u8      d2, d3                       \n"
    "vtrn.u8      d6, d7                       \n"
    "vtrn.u8      d18, d19                     \n"

    // d0 = 00+10 01+11 02+12 03+13
    // d2 = 40+50 41+51 42+52 43+53
    "vpaddl.u8    q0, q0                       \n"
    "vpaddl.u8    q2, q2                       \n"
    "vpaddl.u8    q8, q8                       \n"

    // d3 = 60+70 61+71 62+72 63+73
    "vpaddl.u8    d3, d3                       \n"
    "vpaddl.u8    d7, d7                       \n"
    "vpaddl.u8    d19, d19                     \n"

    // combine source lines
    "vadd.u16     q0, q2                       \n"
    "vadd.u16     q0, q8                       \n"
    "vadd.u16     d4, d3, d7                   \n"
    "vadd.u16     d4, d19                      \n"

    // dst_ptr[3] = (s[6 + st * 0] + s[7 + st * 0]
    //             + s[6 + st * 1] + s[7 + st * 1]
    //             + s[6 + st * 2] + s[7 + st * 2]) / 6
    "vqrdmulh.s16 q2, q2, q13                  \n"
    "vmovn.u16    d4, q2                       \n"

    // Shuffle 2,3 reg around so that 2 can be added to the
    //  0,1 reg and 3 can be added to the 4,5 reg. This
    //  requires expanding from u8 to u16 as the 0,1 and 4,5
    //  registers are already expanded. Then do transposes
    //  to get aligned.
    // q2 = xx 20 xx 30 xx 21 xx 31 xx 22 xx 32 xx 23 xx 33
    "vmovl.u8     q1, d2                       \n"
    "vmovl.u8     q3, d6                       \n"
    "vmovl.u8     q9, d18                      \n"

    // combine source lines
    "vadd.u16     q1, q3                       \n"
    "vadd.u16     q1, q9                       \n"

    // d4 = xx 20 xx 30 xx 22 xx 32
    // d5 = xx 21 xx 31 xx 23 xx 33
    "vtrn.u32     d2, d3                       \n"

    // d4 = xx 20 xx 21 xx 22 xx 23
    // d5 = xx 30 xx 31 xx 32 xx 33
    "vtrn.u16     d2, d3                       \n"

    // 0+1+2, 3+4+5
    "vadd.u16     q0, q1                       \n"

    // Need to divide, but can't downshift as the the value
    //  isn't a power of 2. So multiply by 65536 / n
    //  and take the upper 16 bits.
    "vqrdmulh.s16 q0, q0, q15                  \n"

    // Align for table lookup, vtbl requires registers to
    //  be adjacent
    "vmov.u8      d2, d4                       \n"

    "vtbl.u8      d3, {d0, d1, d2}, d28        \n"
    "vtbl.u8      d4, {d0, d1, d2}, d29        \n"

    MEMACCESS(1)
    "vst1.8       {d3}, [%1]!                  \n"
    MEMACCESS(1)
    "vst1.32      {d4[0]}, [%1]!               \n"
    "bgt          1b                           \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst_ptr),          // %1
    "+r"(dst_width),        // %2
    "+r"(src_stride),       // %3
    "+r"(src_ptr1)          // %4
  : "r"(&kMult38_Div6),     // %5
    "r"(&kShuf38_2),        // %6
    "r"(&kMult38_Div9)      // %7
  : "q0", "q1", "q2", "q3", "q8", "q9", "q13", "q14", "q15", "memory", "cc"
  );
}

// 32x2 -> 12x1
void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
                               ptrdiff_t src_stride,
                               uint8* dst_ptr, int dst_width) {
  asm volatile (
    MEMACCESS(4)
    "vld1.16    {q13}, [%4]                    \n"
    MEMACCESS(5)
    "vld1.8     {q14}, [%5]                    \n"
    "add        %3, %0                         \n"
    ".p2align   2                              \n"
  "1:                                          \n"

    // d0 = 00 40 01 41 02 42 03 43
    // d1 = 10 50 11 51 12 52 13 53
    // d2 = 20 60 21 61 22 62 23 63
    // d3 = 30 70 31 71 32 72 33 73
    MEMACCESS(0)
    "vld4.8       {d0, d1, d2, d3}, [%0]!      \n"
    MEMACCESS(3)
    "vld4.8       {d4, d5, d6, d7}, [%3]!      \n"
    "subs         %2, %2, #12                  \n"

    // Shuffle the input data around to get align the data
    //  so adjacent data can be added. 0,1 - 2,3 - 4,5 - 6,7
    // d0 = 00 10 01 11 02 12 03 13
    // d1 = 40 50 41 51 42 52 43 53
    "vtrn.u8      d0, d1                       \n"
    "vtrn.u8      d4, d5                       \n"

    // d2 = 20 30 21 31 22 32 23 33
    // d3 = 60 70 61 71 62 72 63 73
    "vtrn.u8      d2, d3                       \n"
    "vtrn.u8      d6, d7                       \n"

    // d0 = 00+10 01+11 02+12 03+13
    // d2 = 40+50 41+51 42+52 43+53
    "vpaddl.u8    q0, q0                       \n"
    "vpaddl.u8    q2, q2                       \n"

    // d3 = 60+70 61+71 62+72 63+73
    "vpaddl.u8    d3, d3                       \n"
    "vpaddl.u8    d7, d7                       \n"

    // combine source lines
    "vadd.u16     q0, q2                       \n"
    "vadd.u16     d4, d3, d7                   \n"

    // dst_ptr[3] = (s[6] + s[7] + s[6+st] + s[7+st]) / 4
    "vqrshrn.u16  d4, q2, #2                   \n"

    // Shuffle 2,3 reg around so that 2 can be added to the
    //  0,1 reg and 3 can be added to the 4,5 reg. This
    //  requires expanding from u8 to u16 as the 0,1 and 4,5
    //  registers are already expanded. Then do transposes
    //  to get aligned.
    // q2 = xx 20 xx 30 xx 21 xx 31 xx 22 xx 32 xx 23 xx 33
    "vmovl.u8     q1, d2                       \n"
    "vmovl.u8     q3, d6                       \n"

    // combine source lines
    "vadd.u16     q1, q3                       \n"

    // d4 = xx 20 xx 30 xx 22 xx 32
    // d5 = xx 21 xx 31 xx 23 xx 33
    "vtrn.u32     d2, d3                       \n"

    // d4 = xx 20 xx 21 xx 22 xx 23
    // d5 = xx 30 xx 31 xx 32 xx 33
    "vtrn.u16     d2, d3                       \n"

    // 0+1+2, 3+4+5
    "vadd.u16     q0, q1                       \n"

    // Need to divide, but can't downshift as the the value
    //  isn't a power of 2. So multiply by 65536 / n
    //  and take the upper 16 bits.
    "vqrdmulh.s16 q0, q0, q13                  \n"

    // Align for table lookup, vtbl requires registers to
    //  be adjacent
    "vmov.u8      d2, d4                       \n"

    "vtbl.u8      d3, {d0, d1, d2}, d28        \n"
    "vtbl.u8      d4, {d0, d1, d2}, d29        \n"

    MEMACCESS(1)
    "vst1.8       {d3}, [%1]!                  \n"
    MEMACCESS(1)
    "vst1.32      {d4[0]}, [%1]!               \n"
    "bgt          1b                           \n"
  : "+r"(src_ptr),       // %0
    "+r"(dst_ptr),       // %1
    "+r"(dst_width),     // %2
    "+r"(src_stride)     // %3
  : "r"(&kMult38_Div6),  // %4
    "r"(&kShuf38_2)      // %5
  : "q0", "q1", "q2", "q3", "q13", "q14", "memory", "cc"
  );
}

// 16x2 -> 16x1
void ScaleFilterRows_NEON(uint8* dst_ptr,
                          const uint8* src_ptr, ptrdiff_t src_stride,
                          int dst_width, int source_y_fraction) {
  asm volatile (
    "cmp          %4, #0                       \n"
    "beq          100f                         \n"
    "add          %2, %1                       \n"
    "cmp          %4, #64                      \n"
    "beq          75f                          \n"
    "cmp          %4, #128                     \n"
    "beq          50f                          \n"
    "cmp          %4, #192                     \n"
    "beq          25f                          \n"

    "vdup.8       d5, %4                       \n"
    "rsb          %4, #256                     \n"
    "vdup.8       d4, %4                       \n"
    // General purpose row blend.
  "1:                                          \n"
    MEMACCESS(1)
    "vld1.8       {q0}, [%1]!                  \n"
    MEMACCESS(2)
    "vld1.8       {q1}, [%2]!                  \n"
    "subs         %3, %3, #16                  \n"
    "vmull.u8     q13, d0, d4                  \n"
    "vmull.u8     q14, d1, d4                  \n"
    "vmlal.u8     q13, d2, d5                  \n"
    "vmlal.u8     q14, d3, d5                  \n"
    "vrshrn.u16   d0, q13, #8                  \n"
    "vrshrn.u16   d1, q14, #8                  \n"
    MEMACCESS(0)
    "vst1.8       {q0}, [%0]!                  \n"
    "bgt          1b                           \n"
    "b            99f                          \n"

    // Blend 25 / 75.
  "25:                                         \n"
    MEMACCESS(1)
    "vld1.8       {q0}, [%1]!                  \n"
    MEMACCESS(2)
    "vld1.8       {q1}, [%2]!                  \n"
    "subs         %3, %3, #16                  \n"
    "vrhadd.u8    q0, q1                       \n"
    "vrhadd.u8    q0, q1                       \n"
    MEMACCESS(0)
    "vst1.8       {q0}, [%0]!                  \n"
    "bgt          25b                          \n"
    "b            99f                          \n"

    // Blend 50 / 50.
  "50:                                         \n"
    MEMACCESS(1)
    "vld1.8       {q0}, [%1]!                  \n"
    MEMACCESS(2)
    "vld1.8       {q1}, [%2]!                  \n"
    "subs         %3, %3, #16                  \n"
    "vrhadd.u8    q0, q1                       \n"
    MEMACCESS(0)
    "vst1.8       {q0}, [%0]!                  \n"
    "bgt          50b                          \n"
    "b            99f                          \n"

    // Blend 75 / 25.
  "75:                                         \n"
    MEMACCESS(1)
    "vld1.8       {q1}, [%1]!                  \n"
    MEMACCESS(2)
    "vld1.8       {q0}, [%2]!                  \n"
    "subs         %3, %3, #16                  \n"
    "vrhadd.u8    q0, q1                       \n"
    "vrhadd.u8    q0, q1                       \n"
    MEMACCESS(0)
    "vst1.8       {q0}, [%0]!                  \n"
    "bgt          75b                          \n"
    "b            99f                          \n"

    // Blend 100 / 0 - Copy row unchanged.
  "100:                                        \n"
    MEMACCESS(1)
    "vld1.8       {q0}, [%1]!                  \n"
    "subs         %3, %3, #16                  \n"
    MEMACCESS(0)
    "vst1.8       {q0}, [%0]!                  \n"
    "bgt          100b                         \n"

  "99:                                         \n"
    MEMACCESS(0)
    "vst1.8       {d1[7]}, [%0]                \n"
  : "+r"(dst_ptr),          // %0
    "+r"(src_ptr),          // %1
    "+r"(src_stride),       // %2
    "+r"(dst_width),        // %3
    "+r"(source_y_fraction) // %4
  :
  : "q0", "q1", "d4", "d5", "q13", "q14", "memory", "cc"
  );
}

void ScaleARGBRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                            uint8* dst, int dst_width) {
  asm volatile (
    ".p2align   2                              \n"
  "1:                                          \n"
    // load even pixels into q0, odd into q1
    MEMACCESS(0)
    "vld2.32    {q0, q1}, [%0]!                \n"
    MEMACCESS(0)
    "vld2.32    {q2, q3}, [%0]!                \n"
    "subs       %2, %2, #8                     \n"  // 8 processed per loop
    MEMACCESS(1)
    "vst1.8     {q1}, [%1]!                    \n"  // store odd pixels
    MEMACCESS(1)
    "vst1.8     {q3}, [%1]!                    \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(dst),              // %1
    "+r"(dst_width)         // %2
  :
  : "memory", "cc", "q0", "q1", "q2", "q3"  // Clobber List
  );
}

void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
                               uint8* dst, int dst_width) {
  asm volatile (
    // change the stride to row 2 pointer
    "add        %1, %1, %0                     \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld4.8     {d0, d2, d4, d6}, [%0]!        \n"  // load 8 ARGB pixels.
    MEMACCESS(0)
    "vld4.8     {d1, d3, d5, d7}, [%0]!        \n"  // load next 8 ARGB pixels.
    "subs       %3, %3, #8                     \n"  // 8 processed per loop.
    "vpaddl.u8  q0, q0                         \n"  // B 16 bytes -> 8 shorts.
    "vpaddl.u8  q1, q1                         \n"  // G 16 bytes -> 8 shorts.
    "vpaddl.u8  q2, q2                         \n"  // R 16 bytes -> 8 shorts.
    "vpaddl.u8  q3, q3                         \n"  // A 16 bytes -> 8 shorts.
    MEMACCESS(1)
    "vld4.8     {d16, d18, d20, d22}, [%1]!    \n"  // load 8 more ARGB pixels.
    MEMACCESS(1)
    "vld4.8     {d17, d19, d21, d23}, [%1]!    \n"  // load last 8 ARGB pixels.
    "vpadal.u8  q0, q8                         \n"  // B 16 bytes -> 8 shorts.
    "vpadal.u8  q1, q9                         \n"  // G 16 bytes -> 8 shorts.
    "vpadal.u8  q2, q10                        \n"  // R 16 bytes -> 8 shorts.
    "vpadal.u8  q3, q11                        \n"  // A 16 bytes -> 8 shorts.
    "vrshrn.u16 d0, q0, #2                     \n"  // downshift, round and pack
    "vrshrn.u16 d1, q1, #2                     \n"
    "vrshrn.u16 d2, q2, #2                     \n"
    "vrshrn.u16 d3, q3, #2                     \n"
    MEMACCESS(2)
    "vst4.8     {d0, d1, d2, d3}, [%2]!        \n"
    "bgt        1b                             \n"
  : "+r"(src_ptr),          // %0
    "+r"(src_stride),       // %1
    "+r"(dst),              // %2
    "+r"(dst_width)         // %3
  :
  : "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
  );
}

// Reads 4 pixels at a time.
// Alignment requirement: src_argb 4 byte aligned.
void ScaleARGBRowDownEven_NEON(const uint8* src_argb,  ptrdiff_t src_stride,
                               int src_stepx, uint8* dst_argb, int dst_width) {
  asm volatile (
    "mov        r12, %3, lsl #2                \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld1.32    {d0[0]}, [%0], r12             \n"
    MEMACCESS(0)
    "vld1.32    {d0[1]}, [%0], r12             \n"
    MEMACCESS(0)
    "vld1.32    {d1[0]}, [%0], r12             \n"
    MEMACCESS(0)
    "vld1.32    {d1[1]}, [%0], r12             \n"
    "subs       %2, %2, #4                     \n"  // 4 pixels per loop.
    MEMACCESS(1)
    "vst1.8     {q0}, [%1]!                    \n"
    "bgt        1b                             \n"
  : "+r"(src_argb),    // %0
    "+r"(dst_argb),    // %1
    "+r"(dst_width)    // %2
  : "r"(src_stepx)     // %3
  : "memory", "cc", "r12", "q0"
  );
}

// Reads 4 pixels at a time.
// Alignment requirement: src_argb 4 byte aligned.
void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb, ptrdiff_t src_stride,
                                  int src_stepx,
                                  uint8* dst_argb, int dst_width) {
  asm volatile (
    "mov        r12, %4, lsl #2                \n"
    "add        %1, %1, %0                     \n"
    ".p2align   2                              \n"
  "1:                                          \n"
    MEMACCESS(0)
    "vld1.8     {d0}, [%0], r12                \n"  // Read 4 2x2 blocks -> 2x1
    MEMACCESS(1)
    "vld1.8     {d1}, [%1], r12                \n"
    MEMACCESS(0)
    "vld1.8     {d2}, [%0], r12                \n"
    MEMACCESS(1)
    "vld1.8     {d3}, [%1], r12                \n"
    MEMACCESS(0)
    "vld1.8     {d4}, [%0], r12                \n"
    MEMACCESS(1)
    "vld1.8     {d5}, [%1], r12                \n"
    MEMACCESS(0)
    "vld1.8     {d6}, [%0], r12                \n"
    MEMACCESS(1)
    "vld1.8     {d7}, [%1], r12                \n"
    "vaddl.u8   q0, d0, d1                     \n"
    "vaddl.u8   q1, d2, d3                     \n"
    "vaddl.u8   q2, d4, d5                     \n"
    "vaddl.u8   q3, d6, d7                     \n"
    "vswp.8     d1, d2                         \n"  // ab_cd -> ac_bd
    "vswp.8     d5, d6                         \n"  // ef_gh -> eg_fh
    "vadd.u16   q0, q0, q1                     \n"  // (a+b)_(c+d)
    "vadd.u16   q2, q2, q3                     \n"  // (e+f)_(g+h)
    "vrshrn.u16 d0, q0, #2                     \n"  // first 2 pixels.
    "vrshrn.u16 d1, q2, #2                     \n"  // next 2 pixels.
    "subs       %3, %3, #4                     \n"  // 4 pixels per loop.
    MEMACCESS(2)
    "vst1.8     {q0}, [%2]!                    \n"
    "bgt        1b                             \n"
  : "+r"(src_argb),    // %0
    "+r"(src_stride),  // %1
    "+r"(dst_argb),    // %2
    "+r"(dst_width)    // %3
  : "r"(src_stepx)     // %4
  : "memory", "cc", "r12", "q0", "q1", "q2", "q3"
  );
}

#endif  // __ARM_NEON__

#ifdef __cplusplus
}  // extern "C"
}  // namespace libyuv
#endif