Neon: Intrinsics implementation of RGB->Grayscale
There was no previous GAS implementation.
diff --git a/simd/CMakeLists.txt b/simd/CMakeLists.txt
index 2fd98d5..8f670c4 100644
--- a/simd/CMakeLists.txt
+++ b/simd/CMakeLists.txt
@@ -265,11 +265,12 @@
file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/gastest.S)
+set(SIMD_SOURCES arm/jcgray-neon.c)
if(NEON_INTRINSICS)
- set(SIMD_SOURCES arm/jccolor-neon.c)
- if(BITS EQUAL 32)
- set_source_files_properties(${SIMD_SOURCES} COMPILE_FLAGS -mfpu=neon)
- endif()
+ set(SIMD_SOURCES ${SIMD_SOURCES} arm/jccolor-neon.c)
+endif()
+if(BITS EQUAL 32)
+ set_source_files_properties(${SIMD_SOURCES} COMPILE_FLAGS -mfpu=neon)
endif()
add_library(simd OBJECT ${SIMD_SOURCES} arm/aarch${BITS}/jsimd_neon.S
diff --git a/simd/arm/aarch32/jsimd.c b/simd/arm/aarch32/jsimd.c
index 7c73267..6edea39 100644
--- a/simd/arm/aarch32/jsimd.c
+++ b/simd/arm/aarch32/jsimd.c
@@ -165,6 +165,19 @@
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;
}
@@ -247,6 +260,37 @@
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/aarch64/jsimd.c b/simd/arm/aarch64/jsimd.c
index 976bf27..d042757 100644
--- a/simd/arm/aarch64/jsimd.c
+++ b/simd/arm/aarch64/jsimd.c
@@ -191,6 +191,19 @@
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;
}
@@ -294,6 +307,37 @@
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/jcgray-neon.c b/simd/arm/jcgray-neon.c
new file mode 100644
index 0000000..71c7b2d
--- /dev/null
+++ b/simd/arm/jcgray-neon.c
@@ -0,0 +1,120 @@
+/*
+ * jcgray-neon.c - grayscale colorspace conversion (Arm Neon)
+ *
+ * Copyright (C) 2020, Arm Limited. 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 "align.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/jcgryext-neon.c b/simd/arm/jcgryext-neon.c
new file mode 100644
index 0000000..b1f00e6
--- /dev/null
+++ b/simd/arm/jcgryext-neon.c
@@ -0,0 +1,105 @@
+/*
+ * jcgryext-neon.c - grayscale colorspace conversion (Arm Neon)
+ *
+ * Copyright (C) 2020, Arm Limited. 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
+ *
+ * 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) {
+ ALIGN(16) uint8_t 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 eb3e283..cfe9625 100644
--- a/simd/jsimd.h
+++ b/simd/jsimd.h
@@ -7,6 +7,7 @@
* Copyright (C) 2014, Linaro Limited.
* Copyright (C) 2015-2016, 2018, Matthieu Darbois.
* Copyright (C) 2016-2018, Loongson Technology Corporation Limited, BeiJing.
+ * Copyright (C) 2020, Arm Limited.
*
* Based on the x86 SIMD extension for IJG JPEG library,
* Copyright (C) 1999-2006, MIYASAKA Masaru.
@@ -267,6 +268,28 @@
(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);