1 /*
2 * jcgryext-neon.c - grayscale colorspace conversion (Arm Neon)
3 *
4 * Copyright (C) 2020, Arm Limited. All Rights Reserved.
5 *
6 * This software is provided 'as-is', without any express or implied
7 * warranty. In no event will the authors be held liable for any damages
8 * arising from the use of this software.
9 *
10 * Permission is granted to anyone to use this software for any purpose,
11 * including commercial applications, and to alter it and redistribute it
12 * freely, subject to the following restrictions:
13 *
14 * 1. The origin of this software must not be misrepresented; you must not
15 * claim that you wrote the original software. If you use this software
16 * in a product, an acknowledgment in the product documentation would be
17 * appreciated but is not required.
18 * 2. Altered source versions must be plainly marked as such, and must not be
19 * misrepresented as being the original software.
20 * 3. This notice may not be removed or altered from any source distribution.
21 */
22
23 /* This file is included by jcgray-neon.c */
24
25
26 /* RGB -> Grayscale conversion is defined by the following equation:
27 * Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
28 *
29 * Avoid floating point arithmetic by using shifted integer constants:
30 * 0.29899597 = 19595 * 2^-16
31 * 0.58700561 = 38470 * 2^-16
32 * 0.11399841 = 7471 * 2^-16
33 * These constants are defined in jcgray-neon.c
34 *
35 * This is the same computation as the RGB -> Y portion of RGB -> YCbCr.
36 */
37
jsimd_rgb_gray_convert_neon(JDIMENSION image_width,JSAMPARRAY input_buf,JSAMPIMAGE output_buf,JDIMENSION output_row,int num_rows)38 void jsimd_rgb_gray_convert_neon(JDIMENSION image_width, JSAMPARRAY input_buf,
39 JSAMPIMAGE output_buf, JDIMENSION output_row,
40 int num_rows)
41 {
42 JSAMPROW inptr;
43 JSAMPROW outptr;
44 /* Allocate temporary buffer for final (image_width % 16) pixels in row. */
45 ALIGN(16) uint8_t tmp_buf[16 * RGB_PIXELSIZE];
46
47 while (--num_rows >= 0) {
48 inptr = *input_buf++;
49 outptr = output_buf[0][output_row];
50 output_row++;
51
52 int cols_remaining = image_width;
53 for (; cols_remaining > 0; cols_remaining -= 16) {
54
55 /* To prevent buffer overread by the vector load instructions, the last
56 * (image_width % 16) columns of data are first memcopied to a temporary
57 * buffer large enough to accommodate the vector load.
58 */
59 if (cols_remaining < 16) {
60 memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
61 inptr = tmp_buf;
62 }
63
64 #if RGB_PIXELSIZE == 4
65 uint8x16x4_t input_pixels = vld4q_u8(inptr);
66 #else
67 uint8x16x3_t input_pixels = vld3q_u8(inptr);
68 #endif
69 uint16x8_t r_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_RED]));
70 uint16x8_t r_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_RED]));
71 uint16x8_t g_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_GREEN]));
72 uint16x8_t g_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_GREEN]));
73 uint16x8_t b_l = vmovl_u8(vget_low_u8(input_pixels.val[RGB_BLUE]));
74 uint16x8_t b_h = vmovl_u8(vget_high_u8(input_pixels.val[RGB_BLUE]));
75
76 /* Compute Y = 0.29900 * R + 0.58700 * G + 0.11400 * B */
77 uint32x4_t y_ll = vmull_n_u16(vget_low_u16(r_l), F_0_298);
78 uint32x4_t y_lh = vmull_n_u16(vget_high_u16(r_l), F_0_298);
79 uint32x4_t y_hl = vmull_n_u16(vget_low_u16(r_h), F_0_298);
80 uint32x4_t y_hh = vmull_n_u16(vget_high_u16(r_h), F_0_298);
81 y_ll = vmlal_n_u16(y_ll, vget_low_u16(g_l), F_0_587);
82 y_lh = vmlal_n_u16(y_lh, vget_high_u16(g_l), F_0_587);
83 y_hl = vmlal_n_u16(y_hl, vget_low_u16(g_h), F_0_587);
84 y_hh = vmlal_n_u16(y_hh, vget_high_u16(g_h), F_0_587);
85 y_ll = vmlal_n_u16(y_ll, vget_low_u16(b_l), F_0_113);
86 y_lh = vmlal_n_u16(y_lh, vget_high_u16(b_l), F_0_113);
87 y_hl = vmlal_n_u16(y_hl, vget_low_u16(b_h), F_0_113);
88 y_hh = vmlal_n_u16(y_hh, vget_high_u16(b_h), F_0_113);
89
90 /* Descale Y values (rounding right shift) and narrow to 16-bit. */
91 uint16x8_t y_l = vcombine_u16(vrshrn_n_u32(y_ll, 16),
92 vrshrn_n_u32(y_lh, 16));
93 uint16x8_t y_h = vcombine_u16(vrshrn_n_u32(y_hl, 16),
94 vrshrn_n_u32(y_hh, 16));
95
96 /* Narrow Y values to 8-bit and store to memory. Buffer overwrite is
97 * permitted up to the next multiple of ALIGN_SIZE bytes.
98 */
99 vst1q_u8(outptr, vcombine_u8(vmovn_u16(y_l), vmovn_u16(y_h)));
100
101 /* Increment pointers. */
102 inptr += (16 * RGB_PIXELSIZE);
103 outptr += 16;
104 }
105 }
106 }
107