starlight-apk/tiny_computer
1
0
Fork 0
mirror of https://github.com/Cateners/tiny_computer.git synced 2026-05-21 00:45:49 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update code to v1.0.14 (10)

Browse Source
This commit is contained in:
Caten
2024-02-29 19:35:00 +08:00
parent c2ee3b694c
commit a956d26f6d
3188 changed files with 2317293 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

148
android/extern/libjpeg-turbo/simd/arm/aarch32/jccolext-neon.c vendored Normal file
View File

@@ -0,0 +1,148 @@
/*
* jccolext-neon.c - colorspace conversion (32-bit Arm Neon)
*
* Copyright (C) 2020, Arm Limited. All Rights Reserved.
* Copyright (C) 2020, D. R. Commander. 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 jccolor-neon.c */
/* RGB -> YCbCr conversion is defined by the following equations:
* Y = 0.29900 * R + 0.58700 * G + 0.11400 * B
* Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + 128
* Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + 128
*
* Avoid floating point arithmetic by using shifted integer constants:
* 0.29899597 = 19595 * 2^-16
* 0.58700561 = 38470 * 2^-16
* 0.11399841 = 7471 * 2^-16
* 0.16874695 = 11059 * 2^-16
* 0.33125305 = 21709 * 2^-16
* 0.50000000 = 32768 * 2^-16
* 0.41868592 = 27439 * 2^-16
* 0.08131409 = 5329 * 2^-16
* These constants are defined in jccolor-neon.c
*
* We add the fixed-point equivalent of 0.5 to Cb and Cr, which effectively
* rounds up or down the result via integer truncation.
*/
void jsimd_rgb_ycc_convert_neon(JDIMENSION image_width, JSAMPARRAY input_buf,
JSAMPIMAGE output_buf, JDIMENSION output_row,
int num_rows)
{
/* Pointer to RGB(X/A) input data */
JSAMPROW inptr;
/* Pointers to Y, Cb, and Cr output data */
JSAMPROW outptr0, outptr1, outptr2;
/* Allocate temporary buffer for final (image_width % 8) pixels in row. */
ALIGN(16) uint8_t tmp_buf[8 * RGB_PIXELSIZE];
/* Set up conversion constants. */
#ifdef HAVE_VLD1_U16_X2
const uint16x4x2_t consts = vld1_u16_x2(jsimd_rgb_ycc_neon_consts);
#else
/* GCC does not currently support the intrinsic vld1_<type>_x2(). */
const uint16x4_t consts1 = vld1_u16(jsimd_rgb_ycc_neon_consts);
const uint16x4_t consts2 = vld1_u16(jsimd_rgb_ycc_neon_consts + 4);
const uint16x4x2_t consts = { { consts1, consts2 } };
#endif
const uint32x4_t scaled_128_5 = vdupq_n_u32((128 << 16) + 32767);
while (--num_rows >= 0) {
inptr = *input_buf++;
outptr0 = output_buf[0][output_row];
outptr1 = output_buf[1][output_row];
outptr2 = output_buf[2][output_row];
output_row++;
int cols_remaining = image_width;
for (; cols_remaining > 0; cols_remaining -= 8) {
/* To prevent buffer overread by the vector load instructions, the last
* (image_width % 8) columns of data are first memcopied to a temporary
* buffer large enough to accommodate the vector load.
*/
if (cols_remaining < 8) {
memcpy(tmp_buf, inptr, cols_remaining * RGB_PIXELSIZE);
inptr = tmp_buf;
}
#if RGB_PIXELSIZE == 4
uint8x8x4_t input_pixels = vld4_u8(inptr);
#else
uint8x8x3_t input_pixels = vld3_u8(inptr);
#endif
uint16x8_t r = vmovl_u8(input_pixels.val[RGB_RED]);
uint16x8_t g = vmovl_u8(input_pixels.val[RGB_GREEN]);
uint16x8_t b = vmovl_u8(input_pixels.val[RGB_BLUE]);
/* Compute Y = 0.29900 * R + 0.58700 * G + 0.11400 * B */
uint32x4_t y_low = vmull_lane_u16(vget_low_u16(r), consts.val[0], 0);
y_low = vmlal_lane_u16(y_low, vget_low_u16(g), consts.val[0], 1);
y_low = vmlal_lane_u16(y_low, vget_low_u16(b), consts.val[0], 2);
uint32x4_t y_high = vmull_lane_u16(vget_high_u16(r), consts.val[0], 0);
y_high = vmlal_lane_u16(y_high, vget_high_u16(g), consts.val[0], 1);
y_high = vmlal_lane_u16(y_high, vget_high_u16(b), consts.val[0], 2);
/* Compute Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + 128 */
uint32x4_t cb_low = scaled_128_5;
cb_low = vmlsl_lane_u16(cb_low, vget_low_u16(r), consts.val[0], 3);
cb_low = vmlsl_lane_u16(cb_low, vget_low_u16(g), consts.val[1], 0);
cb_low = vmlal_lane_u16(cb_low, vget_low_u16(b), consts.val[1], 1);
uint32x4_t cb_high = scaled_128_5;
cb_high = vmlsl_lane_u16(cb_high, vget_high_u16(r), consts.val[0], 3);
cb_high = vmlsl_lane_u16(cb_high, vget_high_u16(g), consts.val[1], 0);
cb_high = vmlal_lane_u16(cb_high, vget_high_u16(b), consts.val[1], 1);
/* Compute Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + 128 */
uint32x4_t cr_low = scaled_128_5;
cr_low = vmlal_lane_u16(cr_low, vget_low_u16(r), consts.val[1], 1);
cr_low = vmlsl_lane_u16(cr_low, vget_low_u16(g), consts.val[1], 2);
cr_low = vmlsl_lane_u16(cr_low, vget_low_u16(b), consts.val[1], 3);
uint32x4_t cr_high = scaled_128_5;
cr_high = vmlal_lane_u16(cr_high, vget_high_u16(r), consts.val[1], 1);
cr_high = vmlsl_lane_u16(cr_high, vget_high_u16(g), consts.val[1], 2);
cr_high = vmlsl_lane_u16(cr_high, vget_high_u16(b), consts.val[1], 3);
/* Descale Y values (rounding right shift) and narrow to 16-bit. */
uint16x8_t y_u16 = vcombine_u16(vrshrn_n_u32(y_low, 16),
vrshrn_n_u32(y_high, 16));
/* Descale Cb values (right shift) and narrow to 16-bit. */
uint16x8_t cb_u16 = vcombine_u16(vshrn_n_u32(cb_low, 16),
vshrn_n_u32(cb_high, 16));
/* Descale Cr values (right shift) and narrow to 16-bit. */
uint16x8_t cr_u16 = vcombine_u16(vshrn_n_u32(cr_low, 16),
vshrn_n_u32(cr_high, 16));
/* Narrow Y, Cb, and Cr values to 8-bit and store to memory. Buffer
* overwrite is permitted up to the next multiple of ALIGN_SIZE bytes.
*/
vst1_u8(outptr0, vmovn_u16(y_u16));
vst1_u8(outptr1, vmovn_u16(cb_u16));
vst1_u8(outptr2, vmovn_u16(cr_u16));
/* Increment pointers. */
inptr += (8 * RGB_PIXELSIZE);
outptr0 += 8;
outptr1 += 8;
outptr2 += 8;
}
}
}

334
android/extern/libjpeg-turbo/simd/arm/aarch32/jchuff-neon.c vendored Normal file
View File

@@ -0,0 +1,334 @@
/*
* jchuff-neon.c - Huffman entropy encoding (32-bit 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.
*
* NOTE: All referenced figures are from
* Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994.
*/
#define JPEG_INTERNALS
#include "../../../jinclude.h"
#include "../../../jpeglib.h"
#include "../../../jsimd.h"
#include "../../../jdct.h"
#include "../../../jsimddct.h"
#include "../../jsimd.h"
#include "../jchuff.h"
#include "neon-compat.h"
#include <limits.h>
#include <arm_neon.h>
JOCTET *jsimd_huff_encode_one_block_neon(void *state, JOCTET *buffer,
JCOEFPTR block, int last_dc_val,
c_derived_tbl *dctbl,
c_derived_tbl *actbl)
{
uint8_t block_nbits[DCTSIZE2];
uint16_t block_diff[DCTSIZE2];
/* Load rows of coefficients from DCT block in zig-zag order. */
/* Compute DC coefficient difference value. (F.1.1.5.1) */
int16x8_t row0 = vdupq_n_s16(block[0] - last_dc_val);
row0 = vld1q_lane_s16(block + 1, row0, 1);
row0 = vld1q_lane_s16(block + 8, row0, 2);
row0 = vld1q_lane_s16(block + 16, row0, 3);
row0 = vld1q_lane_s16(block + 9, row0, 4);
row0 = vld1q_lane_s16(block + 2, row0, 5);
row0 = vld1q_lane_s16(block + 3, row0, 6);
row0 = vld1q_lane_s16(block + 10, row0, 7);
int16x8_t row1 = vld1q_dup_s16(block + 17);
row1 = vld1q_lane_s16(block + 24, row1, 1);
row1 = vld1q_lane_s16(block + 32, row1, 2);
row1 = vld1q_lane_s16(block + 25, row1, 3);
row1 = vld1q_lane_s16(block + 18, row1, 4);
row1 = vld1q_lane_s16(block + 11, row1, 5);
row1 = vld1q_lane_s16(block + 4, row1, 6);
row1 = vld1q_lane_s16(block + 5, row1, 7);
int16x8_t row2 = vld1q_dup_s16(block + 12);
row2 = vld1q_lane_s16(block + 19, row2, 1);
row2 = vld1q_lane_s16(block + 26, row2, 2);
row2 = vld1q_lane_s16(block + 33, row2, 3);
row2 = vld1q_lane_s16(block + 40, row2, 4);
row2 = vld1q_lane_s16(block + 48, row2, 5);
row2 = vld1q_lane_s16(block + 41, row2, 6);
row2 = vld1q_lane_s16(block + 34, row2, 7);
int16x8_t row3 = vld1q_dup_s16(block + 27);
row3 = vld1q_lane_s16(block + 20, row3, 1);
row3 = vld1q_lane_s16(block + 13, row3, 2);
row3 = vld1q_lane_s16(block + 6, row3, 3);
row3 = vld1q_lane_s16(block + 7, row3, 4);
row3 = vld1q_lane_s16(block + 14, row3, 5);
row3 = vld1q_lane_s16(block + 21, row3, 6);
row3 = vld1q_lane_s16(block + 28, row3, 7);
int16x8_t abs_row0 = vabsq_s16(row0);
int16x8_t abs_row1 = vabsq_s16(row1);
int16x8_t abs_row2 = vabsq_s16(row2);
int16x8_t abs_row3 = vabsq_s16(row3);
int16x8_t row0_lz = vclzq_s16(abs_row0);
int16x8_t row1_lz = vclzq_s16(abs_row1);
int16x8_t row2_lz = vclzq_s16(abs_row2);
int16x8_t row3_lz = vclzq_s16(abs_row3);
/* Compute number of bits required to represent each coefficient. */
uint8x8_t row0_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row0_lz)));
uint8x8_t row1_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row1_lz)));
uint8x8_t row2_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row2_lz)));
uint8x8_t row3_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row3_lz)));
vst1_u8(block_nbits + 0 * DCTSIZE, row0_nbits);
vst1_u8(block_nbits + 1 * DCTSIZE, row1_nbits);
vst1_u8(block_nbits + 2 * DCTSIZE, row2_nbits);
vst1_u8(block_nbits + 3 * DCTSIZE, row3_nbits);
uint16x8_t row0_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row0, 15)),
vnegq_s16(row0_lz));
uint16x8_t row1_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row1, 15)),
vnegq_s16(row1_lz));
uint16x8_t row2_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row2, 15)),
vnegq_s16(row2_lz));
uint16x8_t row3_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row3, 15)),
vnegq_s16(row3_lz));
uint16x8_t row0_diff = veorq_u16(vreinterpretq_u16_s16(abs_row0), row0_mask);
uint16x8_t row1_diff = veorq_u16(vreinterpretq_u16_s16(abs_row1), row1_mask);
uint16x8_t row2_diff = veorq_u16(vreinterpretq_u16_s16(abs_row2), row2_mask);
uint16x8_t row3_diff = veorq_u16(vreinterpretq_u16_s16(abs_row3), row3_mask);
/* Store diff values for rows 0, 1, 2, and 3. */
vst1q_u16(block_diff + 0 * DCTSIZE, row0_diff);
vst1q_u16(block_diff + 1 * DCTSIZE, row1_diff);
vst1q_u16(block_diff + 2 * DCTSIZE, row2_diff);
vst1q_u16(block_diff + 3 * DCTSIZE, row3_diff);
/* Load last four rows of coefficients from DCT block in zig-zag order. */
int16x8_t row4 = vld1q_dup_s16(block + 35);
row4 = vld1q_lane_s16(block + 42, row4, 1);
row4 = vld1q_lane_s16(block + 49, row4, 2);
row4 = vld1q_lane_s16(block + 56, row4, 3);
row4 = vld1q_lane_s16(block + 57, row4, 4);
row4 = vld1q_lane_s16(block + 50, row4, 5);
row4 = vld1q_lane_s16(block + 43, row4, 6);
row4 = vld1q_lane_s16(block + 36, row4, 7);
int16x8_t row5 = vld1q_dup_s16(block + 29);
row5 = vld1q_lane_s16(block + 22, row5, 1);
row5 = vld1q_lane_s16(block + 15, row5, 2);
row5 = vld1q_lane_s16(block + 23, row5, 3);
row5 = vld1q_lane_s16(block + 30, row5, 4);
row5 = vld1q_lane_s16(block + 37, row5, 5);
row5 = vld1q_lane_s16(block + 44, row5, 6);
row5 = vld1q_lane_s16(block + 51, row5, 7);
int16x8_t row6 = vld1q_dup_s16(block + 58);
row6 = vld1q_lane_s16(block + 59, row6, 1);
row6 = vld1q_lane_s16(block + 52, row6, 2);
row6 = vld1q_lane_s16(block + 45, row6, 3);
row6 = vld1q_lane_s16(block + 38, row6, 4);
row6 = vld1q_lane_s16(block + 31, row6, 5);
row6 = vld1q_lane_s16(block + 39, row6, 6);
row6 = vld1q_lane_s16(block + 46, row6, 7);
int16x8_t row7 = vld1q_dup_s16(block + 53);
row7 = vld1q_lane_s16(block + 60, row7, 1);
row7 = vld1q_lane_s16(block + 61, row7, 2);
row7 = vld1q_lane_s16(block + 54, row7, 3);
row7 = vld1q_lane_s16(block + 47, row7, 4);
row7 = vld1q_lane_s16(block + 55, row7, 5);
row7 = vld1q_lane_s16(block + 62, row7, 6);
row7 = vld1q_lane_s16(block + 63, row7, 7);
int16x8_t abs_row4 = vabsq_s16(row4);
int16x8_t abs_row5 = vabsq_s16(row5);
int16x8_t abs_row6 = vabsq_s16(row6);
int16x8_t abs_row7 = vabsq_s16(row7);
int16x8_t row4_lz = vclzq_s16(abs_row4);
int16x8_t row5_lz = vclzq_s16(abs_row5);
int16x8_t row6_lz = vclzq_s16(abs_row6);
int16x8_t row7_lz = vclzq_s16(abs_row7);
/* Compute number of bits required to represent each coefficient. */
uint8x8_t row4_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row4_lz)));
uint8x8_t row5_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row5_lz)));
uint8x8_t row6_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row6_lz)));
uint8x8_t row7_nbits = vsub_u8(vdup_n_u8(16),
vmovn_u16(vreinterpretq_u16_s16(row7_lz)));
vst1_u8(block_nbits + 4 * DCTSIZE, row4_nbits);
vst1_u8(block_nbits + 5 * DCTSIZE, row5_nbits);
vst1_u8(block_nbits + 6 * DCTSIZE, row6_nbits);
vst1_u8(block_nbits + 7 * DCTSIZE, row7_nbits);
uint16x8_t row4_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row4, 15)),
vnegq_s16(row4_lz));
uint16x8_t row5_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row5, 15)),
vnegq_s16(row5_lz));
uint16x8_t row6_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row6, 15)),
vnegq_s16(row6_lz));
uint16x8_t row7_mask =
vshlq_u16(vreinterpretq_u16_s16(vshrq_n_s16(row7, 15)),
vnegq_s16(row7_lz));
uint16x8_t row4_diff = veorq_u16(vreinterpretq_u16_s16(abs_row4), row4_mask);
uint16x8_t row5_diff = veorq_u16(vreinterpretq_u16_s16(abs_row5), row5_mask);
uint16x8_t row6_diff = veorq_u16(vreinterpretq_u16_s16(abs_row6), row6_mask);
uint16x8_t row7_diff = veorq_u16(vreinterpretq_u16_s16(abs_row7), row7_mask);
/* Store diff values for rows 4, 5, 6, and 7. */
vst1q_u16(block_diff + 4 * DCTSIZE, row4_diff);
vst1q_u16(block_diff + 5 * DCTSIZE, row5_diff);
vst1q_u16(block_diff + 6 * DCTSIZE, row6_diff);
vst1q_u16(block_diff + 7 * DCTSIZE, row7_diff);
/* Construct bitmap to accelerate encoding of AC coefficients. A set bit
* means that the corresponding coefficient != 0.
*/
uint8x8_t row0_nbits_gt0 = vcgt_u8(row0_nbits, vdup_n_u8(0));
uint8x8_t row1_nbits_gt0 = vcgt_u8(row1_nbits, vdup_n_u8(0));
uint8x8_t row2_nbits_gt0 = vcgt_u8(row2_nbits, vdup_n_u8(0));
uint8x8_t row3_nbits_gt0 = vcgt_u8(row3_nbits, vdup_n_u8(0));
uint8x8_t row4_nbits_gt0 = vcgt_u8(row4_nbits, vdup_n_u8(0));
uint8x8_t row5_nbits_gt0 = vcgt_u8(row5_nbits, vdup_n_u8(0));
uint8x8_t row6_nbits_gt0 = vcgt_u8(row6_nbits, vdup_n_u8(0));
uint8x8_t row7_nbits_gt0 = vcgt_u8(row7_nbits, vdup_n_u8(0));
/* { 0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01 } */
const uint8x8_t bitmap_mask =
vreinterpret_u8_u64(vmov_n_u64(0x0102040810204080));
row0_nbits_gt0 = vand_u8(row0_nbits_gt0, bitmap_mask);
row1_nbits_gt0 = vand_u8(row1_nbits_gt0, bitmap_mask);
row2_nbits_gt0 = vand_u8(row2_nbits_gt0, bitmap_mask);
row3_nbits_gt0 = vand_u8(row3_nbits_gt0, bitmap_mask);
row4_nbits_gt0 = vand_u8(row4_nbits_gt0, bitmap_mask);
row5_nbits_gt0 = vand_u8(row5_nbits_gt0, bitmap_mask);
row6_nbits_gt0 = vand_u8(row6_nbits_gt0, bitmap_mask);
row7_nbits_gt0 = vand_u8(row7_nbits_gt0, bitmap_mask);
uint8x8_t bitmap_rows_10 = vpadd_u8(row1_nbits_gt0, row0_nbits_gt0);
uint8x8_t bitmap_rows_32 = vpadd_u8(row3_nbits_gt0, row2_nbits_gt0);
uint8x8_t bitmap_rows_54 = vpadd_u8(row5_nbits_gt0, row4_nbits_gt0);
uint8x8_t bitmap_rows_76 = vpadd_u8(row7_nbits_gt0, row6_nbits_gt0);
uint8x8_t bitmap_rows_3210 = vpadd_u8(bitmap_rows_32, bitmap_rows_10);
uint8x8_t bitmap_rows_7654 = vpadd_u8(bitmap_rows_76, bitmap_rows_54);
uint8x8_t bitmap = vpadd_u8(bitmap_rows_7654, bitmap_rows_3210);
/* Shift left to remove DC bit. */
bitmap = vreinterpret_u8_u64(vshl_n_u64(vreinterpret_u64_u8(bitmap), 1));
/* Move bitmap to 32-bit scalar registers. */
uint32_t bitmap_1_32 = vget_lane_u32(vreinterpret_u32_u8(bitmap), 1);
uint32_t bitmap_33_63 = vget_lane_u32(vreinterpret_u32_u8(bitmap), 0);
/* Set up state and bit buffer for output bitstream. */
working_state *state_ptr = (working_state *)state;
int free_bits = state_ptr->cur.free_bits;
size_t put_buffer = state_ptr->cur.put_buffer;
/* Encode DC coefficient. */
unsigned int nbits = block_nbits[0];
/* Emit Huffman-coded symbol and additional diff bits. */
unsigned int diff = block_diff[0];
PUT_CODE(dctbl->ehufco[nbits], dctbl->ehufsi[nbits], diff)
/* Encode AC coefficients. */
unsigned int r = 0; /* r = run length of zeros */
unsigned int i = 1; /* i = number of coefficients encoded */
/* Code and size information for a run length of 16 zero coefficients */
const unsigned int code_0xf0 = actbl->ehufco[0xf0];
const unsigned int size_0xf0 = actbl->ehufsi[0xf0];
while (bitmap_1_32 != 0) {
r = BUILTIN_CLZ(bitmap_1_32);
i += r;
bitmap_1_32 <<= r;
nbits = block_nbits[i];
diff = block_diff[i];
while (r > 15) {
/* If run length > 15, emit special run-length-16 codes. */
PUT_BITS(code_0xf0, size_0xf0)
r -= 16;
}
/* Emit Huffman symbol for run length / number of bits. (F.1.2.2.1) */
unsigned int rs = (r << 4) + nbits;
PUT_CODE(actbl->ehufco[rs], actbl->ehufsi[rs], diff)
i++;
bitmap_1_32 <<= 1;
}
r = 33 - i;
i = 33;
while (bitmap_33_63 != 0) {
unsigned int leading_zeros = BUILTIN_CLZ(bitmap_33_63);
r += leading_zeros;
i += leading_zeros;
bitmap_33_63 <<= leading_zeros;
nbits = block_nbits[i];
diff = block_diff[i];
while (r > 15) {
/* If run length > 15, emit special run-length-16 codes. */
PUT_BITS(code_0xf0, size_0xf0)
r -= 16;
}
/* Emit Huffman symbol for run length / number of bits. (F.1.2.2.1) */
unsigned int rs = (r << 4) + nbits;
PUT_CODE(actbl->ehufco[rs], actbl->ehufsi[rs], diff)
r = 0;
i++;
bitmap_33_63 <<= 1;
}
/* If the last coefficient(s) were zero, emit an end-of-block (EOB) code.
* The value of RS for the EOB code is 0.
*/
if (i != 64) {
PUT_BITS(actbl->ehufco[0], actbl->ehufsi[0])
}
state_ptr->cur.put_buffer = put_buffer;
state_ptr->cur.free_bits = free_bits;
return buffer;
}

976
android/extern/libjpeg-turbo/simd/arm/aarch32/jsimd.c vendored Normal file
View File

@@ -0,0 +1,976 @@
/*
* jsimd_arm.c
*
* Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
* Copyright (C) 2011, Nokia Corporation and/or its subsidiary(-ies).
* Copyright (C) 2009-2011, 2013-2014, 2016, 2018, 2022, D. R. Commander.
* Copyright (C) 2015-2016, 2018, 2022, Matthieu Darbois.
* Copyright (C) 2019, Google LLC.
* Copyright (C) 2020, Arm Limited.
*
* Based on the x86 SIMD extension for IJG JPEG library,
* Copyright (C) 1999-2006, MIYASAKA Masaru.
* For conditions of distribution and use, see copyright notice in jsimdext.inc
*
* This file contains the interface between the "normal" portions
* of the library and the SIMD implementations when running on a
* 32-bit Arm architecture.
*/
#define JPEG_INTERNALS
#include "../../../jinclude.h"
#include "../../../jpeglib.h"
#include "../../../jsimd.h"
#include "../../../jdct.h"
#include "../../../jsimddct.h"
#include "../../jsimd.h"
#include <ctype.h>
static THREAD_LOCAL unsigned int simd_support = ~0;
static THREAD_LOCAL unsigned int simd_huffman = 1;
#if !defined(__ARM_NEON__) && (defined(__linux__) || defined(ANDROID) || defined(__ANDROID__))
#define SOMEWHAT_SANE_PROC_CPUINFO_SIZE_LIMIT (1024 * 1024)
LOCAL(int)
check_feature(char *buffer, char *feature)
{
char *p;
if (*feature == 0)
return 0;
if (strncmp(buffer, "Features", 8) != 0)
return 0;
buffer += 8;
while (isspace(*buffer))
buffer++;
/* Check if 'feature' is present in the buffer as a separate word */
while ((p = strstr(buffer, feature))) {
if (p > buffer && !isspace(*(p - 1))) {
buffer++;
continue;
}
p += strlen(feature);
if (*p != 0 && !isspace(*p)) {
buffer++;
continue;
}
return 1;
}
return 0;
}
LOCAL(int)
parse_proc_cpuinfo(int bufsize)
{
char *buffer = (char *)malloc(bufsize);
FILE *fd;
simd_support = 0;
if (!buffer)
return 0;
fd = fopen("/proc/cpuinfo", "r");
if (fd) {
while (fgets(buffer, bufsize, fd)) {
if (!strchr(buffer, '\n') && !feof(fd)) {
/* "impossible" happened - insufficient size of the buffer! */
fclose(fd);
free(buffer);
return 0;
}
if (check_feature(buffer, "neon"))
simd_support |= JSIMD_NEON;
}
fclose(fd);
}
free(buffer);
return 1;
}
#endif
/*
* Check what SIMD accelerations are supported.
*/
LOCAL(void)
init_simd(void)
{
#ifndef NO_GETENV
char env[2] = { 0 };
#endif
#if !defined(__ARM_NEON__) && (defined(__linux__) || defined(ANDROID) || defined(__ANDROID__))
int bufsize = 1024; /* an initial guess for the line buffer size limit */
#endif
if (simd_support != ~0U)
return;
simd_support = 0;
#if defined(__ARM_NEON__)
simd_support |= JSIMD_NEON;
#elif defined(__linux__) || defined(ANDROID) || defined(__ANDROID__)
/* We still have a chance to use Neon regardless of globally used
* -mcpu/-mfpu options passed to gcc by performing runtime detection via
* /proc/cpuinfo parsing on linux/android */
while (!parse_proc_cpuinfo(bufsize)) {
bufsize *= 2;
if (bufsize > SOMEWHAT_SANE_PROC_CPUINFO_SIZE_LIMIT)
break;
}
#endif
#ifndef NO_GETENV
/* Force different settings through environment variables */
if (!GETENV_S(env, 2, "JSIMD_FORCENEON") && !strcmp(env, "1"))
simd_support = JSIMD_NEON;
if (!GETENV_S(env, 2, "JSIMD_FORCENONE") && !strcmp(env, "1"))
simd_support = 0;
if (!GETENV_S(env, 2, "JSIMD_NOHUFFENC") && !strcmp(env, "1"))
simd_huffman = 0;
#endif
}
GLOBAL(int)
jsimd_can_rgb_ycc(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;
}
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;
}
GLOBAL(int)
jsimd_can_ycc_rgb(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;
}
GLOBAL(int)
jsimd_can_ycc_rgb565(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_rgb_ycc_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_ycc_convert_neon;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
neonfct = jsimd_extrgbx_ycc_convert_neon;
break;
case JCS_EXT_BGR:
neonfct = jsimd_extbgr_ycc_convert_neon;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
neonfct = jsimd_extbgrx_ycc_convert_neon;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
neonfct = jsimd_extxbgr_ycc_convert_neon;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
neonfct = jsimd_extxrgb_ycc_convert_neon;
break;
default:
neonfct = jsimd_extrgb_ycc_convert_neon;
break;
}
neonfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows);
}
GLOBAL(void)
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)
jsimd_ycc_rgb_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
JDIMENSION input_row, JSAMPARRAY output_buf,
int num_rows)
{
void (*neonfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int);
switch (cinfo->out_color_space) {
case JCS_EXT_RGB:
neonfct = jsimd_ycc_extrgb_convert_neon;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
neonfct = jsimd_ycc_extrgbx_convert_neon;
break;
case JCS_EXT_BGR:
neonfct = jsimd_ycc_extbgr_convert_neon;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
neonfct = jsimd_ycc_extbgrx_convert_neon;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
neonfct = jsimd_ycc_extxbgr_convert_neon;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
neonfct = jsimd_ycc_extxrgb_convert_neon;
break;
default:
neonfct = jsimd_ycc_extrgb_convert_neon;
break;
}
neonfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows);
}
GLOBAL(void)
jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
JDIMENSION input_row, JSAMPARRAY output_buf,
int num_rows)
{
jsimd_ycc_rgb565_convert_neon(cinfo->output_width, input_buf, input_row,
output_buf, num_rows);
}
GLOBAL(int)
jsimd_can_h2v2_downsample(void)
{
init_simd();
/* The code is optimised for these values only */
if (BITS_IN_JSAMPLE != 8)
return 0;
if (DCTSIZE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_h2v1_downsample(void)
{
init_simd();
/* The code is optimised for these values only */
if (BITS_IN_JSAMPLE != 8)
return 0;
if (DCTSIZE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
jsimd_h2v2_downsample_neon(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor, compptr->width_in_blocks,
input_data, output_data);
}
GLOBAL(void)
jsimd_h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY output_data)
{
jsimd_h2v1_downsample_neon(cinfo->image_width, cinfo->max_v_samp_factor,
compptr->v_samp_factor, compptr->width_in_blocks,
input_data, output_data);
}
GLOBAL(int)
jsimd_can_h2v2_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_h2v1_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
{
jsimd_h2v2_upsample_neon(cinfo->max_v_samp_factor, cinfo->output_width,
input_data, output_data_ptr);
}
GLOBAL(void)
jsimd_h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
{
jsimd_h2v1_upsample_neon(cinfo->max_v_samp_factor, cinfo->output_width,
input_data, output_data_ptr);
}
GLOBAL(int)
jsimd_can_h2v2_fancy_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_h2v1_fancy_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_h1v2_fancy_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
{
jsimd_h2v2_fancy_upsample_neon(cinfo->max_v_samp_factor,
compptr->downsampled_width, input_data,
output_data_ptr);
}
GLOBAL(void)
jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
{
jsimd_h2v1_fancy_upsample_neon(cinfo->max_v_samp_factor,
compptr->downsampled_width, input_data,
output_data_ptr);
}
GLOBAL(void)
jsimd_h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr)
{
jsimd_h1v2_fancy_upsample_neon(cinfo->max_v_samp_factor,
compptr->downsampled_width, input_data,
output_data_ptr);
}
GLOBAL(int)
jsimd_can_h2v2_merged_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_h2v1_merged_upsample(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 (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
{
void (*neonfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch (cinfo->out_color_space) {
case JCS_EXT_RGB:
neonfct = jsimd_h2v2_extrgb_merged_upsample_neon;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
neonfct = jsimd_h2v2_extrgbx_merged_upsample_neon;
break;
case JCS_EXT_BGR:
neonfct = jsimd_h2v2_extbgr_merged_upsample_neon;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
neonfct = jsimd_h2v2_extbgrx_merged_upsample_neon;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
neonfct = jsimd_h2v2_extxbgr_merged_upsample_neon;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
neonfct = jsimd_h2v2_extxrgb_merged_upsample_neon;
break;
default:
neonfct = jsimd_h2v2_extrgb_merged_upsample_neon;
break;
}
neonfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
}
GLOBAL(void)
jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf,
JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf)
{
void (*neonfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY);
switch (cinfo->out_color_space) {
case JCS_EXT_RGB:
neonfct = jsimd_h2v1_extrgb_merged_upsample_neon;
break;
case JCS_EXT_RGBX:
case JCS_EXT_RGBA:
neonfct = jsimd_h2v1_extrgbx_merged_upsample_neon;
break;
case JCS_EXT_BGR:
neonfct = jsimd_h2v1_extbgr_merged_upsample_neon;
break;
case JCS_EXT_BGRX:
case JCS_EXT_BGRA:
neonfct = jsimd_h2v1_extbgrx_merged_upsample_neon;
break;
case JCS_EXT_XBGR:
case JCS_EXT_ABGR:
neonfct = jsimd_h2v1_extxbgr_merged_upsample_neon;
break;
case JCS_EXT_XRGB:
case JCS_EXT_ARGB:
neonfct = jsimd_h2v1_extxrgb_merged_upsample_neon;
break;
default:
neonfct = jsimd_h2v1_extrgb_merged_upsample_neon;
break;
}
neonfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf);
}
GLOBAL(int)
jsimd_can_convsamp(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (BITS_IN_JSAMPLE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (sizeof(DCTELEM) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_convsamp_float(void)
{
return 0;
}
GLOBAL(void)
jsimd_convsamp(JSAMPARRAY sample_data, JDIMENSION start_col,
DCTELEM *workspace)
{
jsimd_convsamp_neon(sample_data, start_col, workspace);
}
GLOBAL(void)
jsimd_convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col,
FAST_FLOAT *workspace)
{
}
GLOBAL(int)
jsimd_can_fdct_islow(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(DCTELEM) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_fdct_ifast(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(DCTELEM) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_fdct_float(void)
{
return 0;
}
GLOBAL(void)
jsimd_fdct_islow(DCTELEM *data)
{
jsimd_fdct_islow_neon(data);
}
GLOBAL(void)
jsimd_fdct_ifast(DCTELEM *data)
{
jsimd_fdct_ifast_neon(data);
}
GLOBAL(void)
jsimd_fdct_float(FAST_FLOAT *data)
{
}
GLOBAL(int)
jsimd_can_quantize(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (sizeof(DCTELEM) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_quantize_float(void)
{
return 0;
}
GLOBAL(void)
jsimd_quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
{
jsimd_quantize_neon(coef_block, divisors, workspace);
}
GLOBAL(void)
jsimd_quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors,
FAST_FLOAT *workspace)
{
}
GLOBAL(int)
jsimd_can_idct_2x2(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (BITS_IN_JSAMPLE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (sizeof(ISLOW_MULT_TYPE) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_idct_4x4(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (BITS_IN_JSAMPLE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (sizeof(ISLOW_MULT_TYPE) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_idct_2x2(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col)
{
jsimd_idct_2x2_neon(compptr->dct_table, coef_block, output_buf, output_col);
}
GLOBAL(void)
jsimd_idct_4x4(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col)
{
jsimd_idct_4x4_neon(compptr->dct_table, coef_block, output_buf, output_col);
}
GLOBAL(int)
jsimd_can_idct_islow(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (BITS_IN_JSAMPLE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (sizeof(ISLOW_MULT_TYPE) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_idct_ifast(void)
{
init_simd();
/* The code is optimised for these values only */
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (BITS_IN_JSAMPLE != 8)
return 0;
if (sizeof(JDIMENSION) != 4)
return 0;
if (sizeof(IFAST_MULT_TYPE) != 2)
return 0;
if (IFAST_SCALE_BITS != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_can_idct_float(void)
{
return 0;
}
GLOBAL(void)
jsimd_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col)
{
jsimd_idct_islow_neon(compptr->dct_table, coef_block, output_buf,
output_col);
}
GLOBAL(void)
jsimd_idct_ifast(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col)
{
jsimd_idct_ifast_neon(compptr->dct_table, coef_block, output_buf,
output_col);
}
GLOBAL(void)
jsimd_idct_float(j_decompress_ptr cinfo, jpeg_component_info *compptr,
JCOEFPTR coef_block, JSAMPARRAY output_buf,
JDIMENSION output_col)
{
}
GLOBAL(int)
jsimd_can_huff_encode_one_block(void)
{
init_simd();
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (simd_support & JSIMD_NEON && simd_huffman)
return 1;
return 0;
}
GLOBAL(JOCTET *)
jsimd_huff_encode_one_block(void *state, JOCTET *buffer, JCOEFPTR block,
int last_dc_val, c_derived_tbl *dctbl,
c_derived_tbl *actbl)
{
return jsimd_huff_encode_one_block_neon(state, buffer, block, last_dc_val,
dctbl, actbl);
}
GLOBAL(int)
jsimd_can_encode_mcu_AC_first_prepare(void)
{
init_simd();
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(void)
jsimd_encode_mcu_AC_first_prepare(const JCOEF *block,
const int *jpeg_natural_order_start, int Sl,
int Al, UJCOEF *values, size_t *zerobits)
{
jsimd_encode_mcu_AC_first_prepare_neon(block, jpeg_natural_order_start,
Sl, Al, values, zerobits);
}
GLOBAL(int)
jsimd_can_encode_mcu_AC_refine_prepare(void)
{
init_simd();
if (DCTSIZE != 8)
return 0;
if (sizeof(JCOEF) != 2)
return 0;
if (simd_support & JSIMD_NEON)
return 1;
return 0;
}
GLOBAL(int)
jsimd_encode_mcu_AC_refine_prepare(const JCOEF *block,
const int *jpeg_natural_order_start, int Sl,
int Al, UJCOEF *absvalues, size_t *bits)
{
return jsimd_encode_mcu_AC_refine_prepare_neon(block,
jpeg_natural_order_start, Sl,
Al, absvalues, bits);
}

1200
android/extern/libjpeg-turbo/simd/arm/aarch32/jsimd_neon.S vendored Normal file
View File

File diff suppressed because it is too large Load Diff