Add Arm NEON implementation of RGB->Grayscale

Adds an Arm NEON implementation of RGB -> Grayscale conversion. This
is new code written using compiler intrinsics; there was no previous
NEON assembly code implementation for either AArch32 or AArch64.

Bug: 922430
Change-Id: I1ce566c1741bcd15dca97a75571380d4c09c0572

7 files changed, 337 insertions, 0 deletions

diff --git a/BUILD.gn b/BUILD.gn
index b3b6470..901956a 100644
--- a/BUILD.gn
+++ b/BUILD.gn
@@ -156,6 +156,7 @@ static_library("simd") {
       "simd/arm/arm/jsimd.c",
       "simd/arm/arm/jsimd_neon.S",
       "simd/arm/common/jccolor-neon.c",
+      "simd/arm/common/jcgray-neon.c",
       "simd/arm/common/jcsample-neon.c",
       "simd/arm/common/jdcolor-neon.c",
       "simd/arm/common/jdmerge-neon.c",
@@ -171,6 +172,7 @@ static_library("simd") {
       "simd/arm/arm64/jsimd.c",
       "simd/arm/arm64/jsimd_neon.S",
       "simd/arm/common/jccolor-neon.c",
+      "simd/arm/common/jcgray-neon.c",
       "simd/arm/common/jcsample-neon.c",
       "simd/arm/common/jdcolor-neon.c",
       "simd/arm/common/jdmerge-neon.c",
diff --git a/README.chromium b/README.chromium
index fa4d87b..5e1b166 100644
--- a/README.chromium
+++ b/README.chromium
@@ -74,6 +74,7 @@ following changes which are not merged to upstream:
   - Add Arm NEON implementation of h2v1_downsample
   - Add Arm NEON implementation of h2v2_downsample
   - Implement RGB->YCbCr using Arm NEON intrinsics
+  - Add Arm NEON implementation of RGB->Grayscale
 * Patches to enable running the upstream unit tests through gtest.
   The upstream unit tests are defined here under the section 'TESTS':
   https://github.com/libjpeg-turbo/libjpeg-turbo/blob/master/CMakeLists.txt
diff --git a/simd/arm/arm/jsimd.c b/simd/arm/arm/jsimd.c
index 0ae1b57..98907a5 100644
--- a/simd/arm/arm/jsimd.c
+++ b/simd/arm/arm/jsimd.c
@@ -164,6 +164,19 @@ jsimd_can_rgb_ycc(void)
 GLOBAL(int)
 jsimd_can_rgb_gray(void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (simd_support & JSIMD_NEON)
+    return 1;
+
   return 0;
 }
 
@@ -246,6 +259,37 @@ jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
                        JSAMPIMAGE output_buf, JDIMENSION output_row,
                        int num_rows)
 {
+  void (*neonfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch (cinfo->in_color_space) {
+  case JCS_EXT_RGB:
+    neonfct = jsimd_extrgb_gray_convert_neon;
+    break;
+  case JCS_EXT_RGBX:
+  case JCS_EXT_RGBA:
+    neonfct = jsimd_extrgbx_gray_convert_neon;
+    break;
+  case JCS_EXT_BGR:
+    neonfct = jsimd_extbgr_gray_convert_neon;
+    break;
+  case JCS_EXT_BGRX:
+  case JCS_EXT_BGRA:
+    neonfct = jsimd_extbgrx_gray_convert_neon;
+    break;
+  case JCS_EXT_XBGR:
+  case JCS_EXT_ABGR:
+    neonfct = jsimd_extxbgr_gray_convert_neon;
+    break;
+  case JCS_EXT_XRGB:
+  case JCS_EXT_ARGB:
+    neonfct = jsimd_extxrgb_gray_convert_neon;
+    break;
+  default:
+    neonfct = jsimd_extrgb_gray_convert_neon;
+    break;
+  }
+
+  neonfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 
 GLOBAL(void)
diff --git a/simd/arm/arm64/jsimd.c b/simd/arm/arm64/jsimd.c
index 8cc8f69..ca29cd6 100644
--- a/simd/arm/arm64/jsimd.c
+++ b/simd/arm/arm64/jsimd.c
@@ -176,6 +176,19 @@ jsimd_can_rgb_ycc(void)
 GLOBAL(int)
 jsimd_can_rgb_gray(void)
 {
+  init_simd();
+
+  /* The code is optimised for these values only */
+  if (BITS_IN_JSAMPLE != 8)
+    return 0;
+  if (sizeof(JDIMENSION) != 4)
+    return 0;
+  if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4))
+    return 0;
+
+  if (simd_support & JSIMD_NEON)
+    return 1;
+
   return 0;
 }
 
@@ -258,6 +271,37 @@ jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf,
                        JSAMPIMAGE output_buf, JDIMENSION output_row,
                        int num_rows)
 {
+  void (*neonfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int);
+
+  switch (cinfo->in_color_space) {
+  case JCS_EXT_RGB:
+    neonfct = jsimd_extrgb_gray_convert_neon;
+    break;
+  case JCS_EXT_RGBX:
+  case JCS_EXT_RGBA:
+    neonfct = jsimd_extrgbx_gray_convert_neon;
+    break;
+  case JCS_EXT_BGR:
+    neonfct = jsimd_extbgr_gray_convert_neon;
+    break;
+  case JCS_EXT_BGRX:
+  case JCS_EXT_BGRA:
+    neonfct = jsimd_extbgrx_gray_convert_neon;
+    break;
+  case JCS_EXT_XBGR:
+  case JCS_EXT_ABGR:
+    neonfct = jsimd_extxbgr_gray_convert_neon;
+    break;
+  case JCS_EXT_XRGB:
+  case JCS_EXT_ARGB:
+    neonfct = jsimd_extxrgb_gray_convert_neon;
+    break;
+  default:
+    neonfct = jsimd_extrgb_gray_convert_neon;
+    break;
+  }
+
+  neonfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
 }
 
 GLOBAL(void)
diff --git a/simd/arm/common/jcgray-neon.c b/simd/arm/common/jcgray-neon.c
new file mode 100644
index 0000000..4cba0ef
--- /dev/null
+++ b/simd/arm/common/jcgray-neon.c
@@ -0,0 +1,117 @@
+/*
+ * jcgray-neon.c - grayscale colorspace conversion (Arm NEON)
+ *
+ * Copyright 2020 The Chromium Authors. All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+#define JPEG_INTERNALS
+#include "../../../jinclude.h"
+#include "../../../jpeglib.h"
+#include "../../../jsimd.h"
+#include "../../../jdct.h"
+#include "../../../jsimddct.h"
+#include "../../jsimd.h"
+
+#include <arm_neon.h>
+
+/* RGB -> Grayscale conversion constants. */
+
+#define F_0_298  19595
+#define F_0_587  38470
+#define F_0_113  7471
+
+/* Include inline routines for colorspace extensions. */
+
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED  EXT_RGB_RED
+#define RGB_GREEN  EXT_RGB_GREEN
+#define RGB_BLUE  EXT_RGB_BLUE
+#define RGB_PIXELSIZE  EXT_RGB_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extrgb_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
+
+#define RGB_RED  EXT_RGBX_RED
+#define RGB_GREEN  EXT_RGBX_GREEN
+#define RGB_BLUE  EXT_RGBX_BLUE
+#define RGB_PIXELSIZE  EXT_RGBX_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extrgbx_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
+
+#define RGB_RED  EXT_BGR_RED
+#define RGB_GREEN  EXT_BGR_GREEN
+#define RGB_BLUE  EXT_BGR_BLUE
+#define RGB_PIXELSIZE  EXT_BGR_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extbgr_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
+
+#define RGB_RED  EXT_BGRX_RED
+#define RGB_GREEN  EXT_BGRX_GREEN
+#define RGB_BLUE  EXT_BGRX_BLUE
+#define RGB_PIXELSIZE  EXT_BGRX_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extbgrx_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
+
+#define RGB_RED  EXT_XBGR_RED
+#define RGB_GREEN  EXT_XBGR_GREEN
+#define RGB_BLUE  EXT_XBGR_BLUE
+#define RGB_PIXELSIZE  EXT_XBGR_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extxbgr_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
+
+#define RGB_RED  EXT_XRGB_RED
+#define RGB_GREEN  EXT_XRGB_GREEN
+#define RGB_BLUE  EXT_XRGB_BLUE
+#define RGB_PIXELSIZE  EXT_XRGB_PIXELSIZE
+#define jsimd_rgb_gray_convert_neon  jsimd_extxrgb_gray_convert_neon
+#include "jcgryext-neon.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef jsimd_rgb_gray_convert_neon
diff --git a/simd/arm/common/jcgryext-neon.c b/simd/arm/common/jcgryext-neon.c
new file mode 100644
index 0000000..35ba0ac
--- /dev/null
+++ b/simd/arm/common/jcgryext-neon.c
@@ -0,0 +1,107 @@
+/*
+ * jcgryext-neon.c - grayscale colorspace conversion (Arm NEON)
+ *
+ * Copyright 2020 The Chromium Authors. All Rights Reserved.
+ *
+ * This software is provided 'as-is', without any express or implied
+ * warranty.  In no event will the authors be held liable for any damages
+ * arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose,
+ * including commercial applications, and to alter it and redistribute it
+ * freely, subject to the following restrictions:
+ *
+ * 1. The origin of this software must not be misrepresented; you must not
+ *    claim that you wrote the original software. If you use this software
+ *    in a product, an acknowledgment in the product documentation would be
+ *    appreciated but is not required.
+ * 2. Altered source versions must be plainly marked as such, and must not be
+ *    misrepresented as being the original software.
+ * 3. This notice may not be removed or altered from any source distribution.
+ */
+
+/* This file is included by jcgray-neon.c */
+
+/*
+ * RGB -> Grayscale conversion is defined by the following equation:
+ *    Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+ *
+ * Avoid floating point arithmetic by using shifted integer constants:
+ *    0.29899597 = 19595 * 2^-16
+ *    0.58700561 = 38470 * 2^-16
+ *    0.11399841 =  7471 * 2^-16
+ * These constants are defined in jcgray-neon.c
+ *
+ * We use rounding later to get correct values.
+ *
+ * This is the same computation as the RGB -> Y portion of RGB -> YCbCr.
+ */
+
+void jsimd_rgb_gray_convert_neon(JDIMENSION image_width,
+                                 JSAMPARRAY input_buf,
+                                 JSAMPIMAGE output_buf,
+                                 JDIMENSION output_row,
+                                 int num_rows)
+{
+  JSAMPROW inptr;
+  JSAMPROW outptr;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+
+    int cols_remaining = image_width;
+    for (; cols_remaining > 0; cols_remaining -= 16) {
+
+      /* To prevent buffer overread by the vector load instructions, the */
+      /* last (image_width % 16) columns of data are first memcopied to a */
+      /* temporary buffer large enough to accommodate the vector load. */
+      if (cols_remaining < 16) {
+        uint8_t __attribute__((aligned(8))) tmp_buf[16 * RGB_PIXELSIZE];
+        memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
+        inptr = tmp_buf;
+      }
+
+#if RGB_PIXELSIZE == 4
+      uint8x16x4_t input_pixels = vld4q_u8(inptr);
+#else
+      uint8x16x3_t input_pixels = vld3q_u8(inptr);
+#endif
+      uint16x8_t r_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_RED]));
+      uint16x8_t r_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_RED]));
+      uint16x8_t g_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_GREEN]));
+      uint16x8_t g_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_GREEN]));
+      uint16x8_t b_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_BLUE]));
+      uint16x8_t b_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_BLUE]));
+
+      /* Compute Y = 0.29900 * R + 0.58700 * G + 0.11400 * B */
+      uint32x4_t y_ll = vmull_n_u16(vget_low_u16(r_l), F_0_298);
+      uint32x4_t y_lh = vmull_n_u16(vget_high_u16(r_l), F_0_298);
+      uint32x4_t y_hl = vmull_n_u16(vget_low_u16(r_h), F_0_298);
+      uint32x4_t y_hh = vmull_n_u16(vget_high_u16(r_h), F_0_298);
+      y_ll = vmlal_n_u16(y_ll, vget_low_u16(g_l), F_0_587);
+      y_lh = vmlal_n_u16(y_lh, vget_high_u16(g_l), F_0_587);
+      y_hl = vmlal_n_u16(y_hl, vget_low_u16(g_h), F_0_587);
+      y_hh = vmlal_n_u16(y_hh, vget_high_u16(g_h), F_0_587);
+      y_ll = vmlal_n_u16(y_ll, vget_low_u16(b_l), F_0_113);
+      y_lh = vmlal_n_u16(y_lh, vget_high_u16(b_l), F_0_113);
+      y_hl = vmlal_n_u16(y_hl, vget_low_u16(b_h), F_0_113);
+      y_hh = vmlal_n_u16(y_hh, vget_high_u16(b_h), F_0_113);
+
+      /* Descale Y values (rounding right shift) and narrow to 16-bit. */
+      uint16x8_t y_l = vcombine_u16(vrshrn_n_u32(y_ll, 16),
+                                    vrshrn_n_u32(y_lh, 16));
+      uint16x8_t y_h = vcombine_u16(vrshrn_n_u32(y_hl, 16),
+                                    vrshrn_n_u32(y_hh, 16));
+
+      /* Narrow Y values to 8-bit and store to memory. Buffer overwrite is */
+      /* permitted up to the next multiple of ALIGN_SIZE bytes. */
+      vst1q_u8(outptr, vcombine_u8(vmovn_u16(y_l), vmovn_u16(y_h)));
+
+      /* Increment pointers. */
+      inptr += (16 * RGB_PIXELSIZE);
+      outptr += 16;
+    }
+  }
+}
diff --git a/simd/jsimd.h b/simd/jsimd.h
index 50a3c5d..99c8801 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -256,6 +256,28 @@ EXTERN(void) jsimd_extxrgb_gray_convert_avx2
   (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
    JDIMENSION output_row, int num_rows);
 
+EXTERN(void) jsimd_rgb_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgb_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extrgbx_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgr_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extbgrx_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxbgr_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+EXTERN(void) jsimd_extxrgb_gray_convert_neon
+  (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
+   JDIMENSION output_row, int num_rows);
+
 EXTERN(void) jsimd_rgb_gray_convert_dspr2
   (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf,
    JDIMENSION output_row, int num_rows);