mgeppert/Tasmota
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity
4f2c1f3499
Tasmota/lib/libesp32/JPEGDEC/src/jpeg.inl
Christian Baars 7cb8a3f968
Berry: add cam module, img class (#21743) 
* cam module, img class
2024-07-07 19:50:33 +02:00

4968 lines
223 KiB
C++
Raw Blame History

//
// JPEG Decoder
//
// written by Larry Bank
// bitbank@pobox.com
// Arduino port started 8/2/2020
// Original JPEG code written 26+ years ago :)
// The goal of this code is to decode baseline JPEG images
// using no more than 18K of RAM (if sent directly to an LCD display)
//
// Copyright 2020 BitBank Software, Inc. All Rights Reserved.
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//===========================================================================
//
#include "JPEGDEC.h"
#ifdef TEENSYDUINO
#include "my_cm4_simd.h"
//#define HAS_SIMD
#endif
#if !defined(NO_SIMD) && (defined(ARM_MATH_CM4) || defined(ARM_MATH_CM7))
//#define HAS_SIMD
#endif
#if defined (ARDUINO_ARCH_ESP32) && !defined(NO_SIMD)
#include "dsps_fft2r_platform.h"
#if (dsps_fft2r_sc16_aes3_enabled == 1)
#define ESP32S3_SIMD
extern "C" {
void s3_ycbcr_convert_444(uint8_t *pY, uint8_t *pCB, uint8_t *pCR, uint16_t *pOut, int16_t *pConsts, uint8_t ucPixelType);
void s3_ycbcr_convert_420(uint8_t *pY, uint8_t *pCB, uint8_t *pCR, uint16_t *pOut, int16_t *pConsts, uint8_t ucPixelType);
void s3_dequant(int16_t *pMCU, int16_t *pQuant);
}
int16_t i16_Consts[8] = {0x80, 113, 90, 22, 46, 1,32,2048};
#endif // S3 SIMD
#endif // ESP32
#if defined( __x86_64__ ) && !defined(NO_SIMD)
#define HAS_SSE
#include <emmintrin.h>
#include <tmmintrin.h>
#include <smmintrin.h>
//#include <immintrin.h> // AVX2
#endif
#if !defined(HAS_SIMD) && !defined(NO_SIMD) && (defined(__arm64__) || defined(__aarch64__))
#include <arm_neon.h>
#define HAS_NEON
#endif
// forward references
static int JPEGInit(JPEGIMAGE *pJPEG);
static int JPEGParseInfo(JPEGIMAGE *pPage, int bExtractThumb);
static void JPEGGetMoreData(JPEGIMAGE *pPage);
static int DecodeJPEG(JPEGIMAGE *pImage);
static int32_t readRAM(JPEGFILE *pFile, uint8_t *pBuf, int32_t iLen);
static int32_t seekMem(JPEGFILE *pFile, int32_t iPosition);
#if defined (__MACH__) || defined( __LINUX__ ) || defined( __MCUXPRESSO )
static int32_t readFile(JPEGFILE *pFile, uint8_t *pBuf, int32_t iLen);
static int32_t seekFile(JPEGFILE *pFile, int32_t iPosition);
static void closeFile(void *handle);
#endif
static void JPEGDither(JPEGIMAGE *pJPEG, int iWidth, int iHeight);
/* JPEG tables */
// zigzag ordering of DCT coefficients
static const unsigned char cZigZag[64] = {0,1,5,6,14,15,27,28,
2,4,7,13,16,26,29,42,
3,8,12,17,25,30,41,43,
9,11,18,24,31,40,44,53,
10,19,23,32,39,45,52,54,
20,22,33,38,46,51,55,60,
21,34,37,47,50,56,59,61,
35,36,48,49,57,58,62,63};
// un-zigzag ordering
static const unsigned char cZigZag2[64] = {0,1,8,16,9,2,3,10,
17,24,32,25,18,11,4,5,
12,19,26,33,40,48,41,34,
27,20,13,6,7,14,21,28,
35,42,49,56,57,50,43,36,
29,22,15,23,30,37,44,51,
58,59,52,45,38,31,39,46,
53,60,61,54,47,55,62,63};
#ifdef HAS_NEON
// 16-bit constants for NEON ycc->rgb conversion
static const int16_t __attribute__((aligned(16))) sYCCRGBConstants[4] = {5742/2, -2925/2, -1409/2, 7258/2};
// 16-bit constants for IDCT calculation
static const int16_t __attribute__((aligned(16))) s0414[8] = {1697*2,1697*2,1697*2,1697*2,1697*2,1697*2,1697*2,1697*2}; // 1.414213562 - 1.0
static const int16_t __attribute__((aligned(16))) s1414[8] = {5793*2,5793*2,5793*2,5793*2,5793*2,5793*2,5793*2,5793*2}; // 1.414213562
static const int16_t __attribute__((aligned(16))) s1847[8] = {7568*2,7568*2,7568*2,7568*2,7568*2,7568*2,7568*2,7568*2}; // 1.8477
static const int16_t __attribute__((aligned(16))) s2613[8] = {-10703,-10703,-10703,-10703,-10703,-10703,-10703,-10703}; // -2.6131259
static const int16_t __attribute__((aligned(16))) sp2613[8] = {10703,10703,10703,10703,10703,10703,10703,10703}; // 2.6131259
static const int16_t __attribute__((aligned(16))) s1082[8] = {4433*2,4433*2,4433*2,4433*2,4433*2,4433*2,4433*2,4433*2}; // 1.08239
#endif // HAS_NEON
#ifdef HAS_SSE
#if defined ( __GNUC__ ) || defined( _GCC_ANDROID ) || defined( __APPLE__)
signed short s1402[8] __attribute__((aligned(16))) = { 5742, 5742, 5742, 5742, 5742, 5742, 5742, 5742 };
signed short s0714[8] __attribute__((aligned(16))) = { -2925, -2925, -2925, -2925, -2925, -2925, -2925, -2925 };
signed short s0344[8] __attribute__((aligned(16))) = { -1409, -1409, -1409, -1409, -1409, -1409, -1409, -1409 };
signed short s1772[8] __attribute__((aligned(16))) = { 7258, 7258, 7258, 7258, 7258, 7258, 7258, 7258 };
// 16-bit constants for IDCT calculation
signed short s0414[8] __attribute__((aligned(16))) = { 1697 * 4, 1697 * 4, 1697 * 4, 1697 * 4, 1697 * 4, 1697 * 4, 1697 * 4, 1697 * 4 }; // 1.414213562 - 1.0
signed short s1414[8] __attribute__((aligned(16))) = { 5793 * 4, 5793 * 4, 5793 * 4, 5793 * 4, 5793 * 4, 5793 * 4, 5793 * 4, 5793 * 4 }; // 1.414213562
signed short s1847[8] __attribute__((aligned(16))) = { 7568 * 4, 7568 * 4, 7568 * 4, 7568 * 4, 7568 * 4, 7568 * 4, 7568 * 4, 7568 * 4 }; // 1.8477
signed short s2613[8] __attribute__((aligned(16))) = { -10703 * 2, -10703 * 2, -10703 * 2, -10703 * 2, -10703 * 2, -10703 * 2, -10703 * 2, -10703 * 2 }; // -2.6131259
signed short sp2613[8] __attribute__((aligned(16))) = { 10703 * 2, 10703 * 2, 10703 * 2, 10703 * 2, 10703 * 2, 10703 * 2, 10703 * 2, 10703 * 2 }; // 2.6131259
signed short s1082[8] __attribute__((aligned(16))) = { 4433 * 4, 4433 * 4, 4433 * 4, 4433 * 4, 4433 * 4, 4433 * 4, 4433 * 4, 4433 * 4 }; // 1.08239
signed short sfastDCT[8] __attribute__((aligned(16))) = { 4096, 4096, 4096, 4096, -815, 2320, 3472, 4096 };
#else
// 16-bit Constants for SSE ycc->rgb conversion
__declspec(align(16)) signed short s1402[8] = {5742,5742,5742,5742,5742,5742,5742,5742};
__declspec(align(16)) signed short s0714[8] = {-2925,-2925,-2925,-2925,-2925,-2925,-2925,-2925};
__declspec(align(16)) signed short s0344[8] = {-1409,-1409,-1409,-1409,-1409,-1409,-1409,-1409};
__declspec(align(16)) signed short s1772[8] = {7258,7258,7258,7258,7258,7258,7258,7258};
// 16-bit constants for IDCT calculation
__declspec(align(16)) signed short s0414[8] = {1697*4,1697*4,1697*4,1697*4,1697*4,1697*4,1697*4,1697*4}; // 1.414213562 - 1.0
__declspec(align(16)) signed short s1414[8] = {5793*4,5793*4,5793*4,5793*4,5793*4,5793*4,5793*4,5793*4}; // 1.414213562
__declspec(align(16)) signed short s1847[8] = {7568*4,7568*4,7568*4,7568*4,7568*4,7568*4,7568*4,7568*4}; // 1.8477
__declspec(align(16)) signed short s2613[8] = {-10703*2,-10703*2,-10703*2,-10703*2,-10703*2,-10703*2,-10703*2,-10703*2}; // -2.6131259
__declspec(align(16)) signed short sp2613[8] = {10703*2,10703*2,10703*2,10703*2,10703*2,10703*2,10703*2,10703*2}; // 2.6131259
__declspec(align(16)) signed short s1082[8] = {4433*4,4433*4,4433*4,4433*4,4433*4,4433*4,4433*4,4433*4}; // 1.08239
__declspec(align(16)) signed short sfastDCT[8] = {4096,4096,4096,4096,-815,2320,3472,4096};
#endif // GCC
#endif // HAS_SSE
// For AA&N IDCT method, multipliers are equal to quantization
// coefficients scaled by scalefactor[row]*scalefactor[col], where
// scalefactor[0] = 1
// scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
// For integer operation, the multiplier table is to be scaled by
// IFAST_SCALE_BITS.
static const int iScaleBits[64] = {16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247};
//
// Range clip and shift for RGB565 output
// input value is 0 to 255, then another 256 for overflow to FF, then 512 more for negative values wrapping around
// Trims a few instructions off the final output stage
//
static const uint8_t ucRangeTable[] = {0x80,0x81,0x82,0x83,0x84,0x85,0x86,0x87,0x88,0x89,0x8a,0x8b,0x8c,0x8d,0x8e,0x8f,
0x90,0x91,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9a,0x9b,0x9c,0x9d,0x9e,0x9f,
0xa0,0xa1,0xa2,0xa3,0xa4,0xa5,0xa6,0xa7,0xa8,0xa9,0xaa,0xab,0xac,0xad,0xae,0xaf,
0xb0,0xb1,0xb2,0xb3,0xb4,0xb5,0xb6,0xb7,0xb8,0xb9,0xba,0xbb,0xbc,0xbd,0xbe,0xbf,
0xc0,0xc1,0xc2,0xc3,0xc4,0xc5,0xc6,0xc7,0xc8,0xc9,0xca,0xcb,0xcc,0xcd,0xce,0xcf,
0xd0,0xd1,0xd2,0xd3,0xd4,0xd5,0xd6,0xd7,0xd8,0xd9,0xda,0xdb,0xdc,0xdd,0xde,0xdf,
0xe0,0xe1,0xe2,0xe3,0xe4,0xe5,0xe6,0xe7,0xe8,0xe9,0xea,0xeb,0xec,0xed,0xee,0xef,
0xf0,0xf1,0xf2,0xf3,0xf4,0xf5,0xf6,0xf7,0xf8,0xf9,0xfa,0xfb,0xfc,0xfd,0xfe,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,0xff,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,
0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0a,0x0b,0x0c,0x0d,0x0e,0x0f,
0x10,0x11,0x12,0x13,0x14,0x15,0x16,0x17,0x18,0x19,0x1a,0x1b,0x1c,0x1d,0x1e,0x1f,
0x20,0x21,0x22,0x23,0x24,0x25,0x26,0x27,0x28,0x29,0x2a,0x2b,0x2c,0x2d,0x2e,0x2f,
0x30,0x31,0x32,0x33,0x34,0x35,0x36,0x37,0x38,0x39,0x3a,0x3b,0x3c,0x3d,0x3e,0x3f,
0x40,0x41,0x42,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4a,0x4b,0x4c,0x4d,0x4e,0x4f,
0x50,0x51,0x52,0x53,0x54,0x55,0x56,0x57,0x58,0x59,0x5a,0x5b,0x5c,0x5d,0x5e,0x5f,
0x60,0x61,0x62,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6a,0x6b,0x6c,0x6d,0x6e,0x6f,
0x70,0x71,0x72,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7a,0x7b,0x7c,0x7d,0x7e,0x7f};
//
// Convert 8-bit grayscale into RGB565
//
static const uint16_t usGrayTo565[] = {0x0000,0x0000,0x0000,0x0000,0x0020,0x0020,0x0020,0x0020, // 0
0x0841,0x0841,0x0841,0x0841,0x0861,0x0861,0x0861,0x0861,
0x1082,0x1082,0x1082,0x1082,0x10a2,0x10a2,0x10a2,0x10a2,
0x18c3,0x18c3,0x18c3,0x18c3,0x18e3,0x18e3,0x18e3,0x18e3,
0x2104,0x2104,0x2104,0x2104,0x2124,0x2124,0x2124,0x2124,
0x2945,0x2945,0x2945,0x2945,0x2965,0x2965,0x2965,0x2965,
0x3186,0x3186,0x3186,0x3186,0x31a6,0x31a6,0x31a6,0x31a6,
0x39c7,0x39c7,0x39c7,0x39c7,0x39e7,0x39e7,0x39e7,0x39e7,
0x4208,0x4208,0x4208,0x4208,0x4228,0x4228,0x4228,0x4228,
0x4a49,0x4a49,0x4a49,0x4a49,0x4a69,0x4a69,0x4a69,0x4a69,
0x528a,0x528a,0x528a,0x528a,0x52aa,0x52aa,0x52aa,0x52aa,
0x5acb,0x5acb,0x5acb,0x5acb,0x5aeb,0x5aeb,0x5aeb,0x5aeb,
0x630c,0x630c,0x630c,0x630c,0x632c,0x632c,0x632c,0x632c,
0x6b4d,0x6b4d,0x6b4d,0x6b4d,0x6b6d,0x6b6d,0x6b6d,0x6b6d,
0x738e,0x738e,0x738e,0x738e,0x73ae,0x73ae,0x73ae,0x73ae,
0x7bcf,0x7bcf,0x7bcf,0x7bcf,0x7bef,0x7bef,0x7bef,0x7bef,
0x8410,0x8410,0x8410,0x8410,0x8430,0x8430,0x8430,0x8430,
0x8c51,0x8c51,0x8c51,0x8c51,0x8c71,0x8c71,0x8c71,0x8c71,
0x9492,0x9492,0x9492,0x9492,0x94b2,0x94b2,0x94b2,0x94b2,
0x9cd3,0x9cd3,0x9cd3,0x9cd3,0x9cf3,0x9cf3,0x9cf3,0x9cf3,
0xa514,0xa514,0xa514,0xa514,0xa534,0xa534,0xa534,0xa534,
0xad55,0xad55,0xad55,0xad55,0xad75,0xad75,0xad75,0xad75,
0xb596,0xb596,0xb596,0xb596,0xb5b6,0xb5b6,0xb5b6,0xb5b6,
0xbdd7,0xbdd7,0xbdd7,0xbdd7,0xbdf7,0xbdf7,0xbdf7,0xbdf7,
0xc618,0xc618,0xc618,0xc618,0xc638,0xc638,0xc638,0xc638,
0xce59,0xce59,0xce59,0xce59,0xce79,0xce79,0xce79,0xce79,
0xd69a,0xd69a,0xd69a,0xd69a,0xd6ba,0xd6ba,0xd6ba,0xd6ba,
0xdedb,0xdedb,0xdedb,0xdedb,0xdefb,0xdefb,0xdefb,0xdefb,
0xe71c,0xe71c,0xe71c,0xe71c,0xe73c,0xe73c,0xe73c,0xe73c,
0xef5d,0xef5d,0xef5d,0xef5d,0xef7d,0xef7d,0xef7d,0xef7d,
0xf79e,0xf79e,0xf79e,0xf79e,0xf7be,0xf7be,0xf7be,0xf7be,
0xffdf,0xffdf,0xffdf,0xffdf,0xffff,0xffff,0xffff,0xffff};
//
// Clip and convert red value into 5-bits for RGB565
//
static const uint16_t usRangeTableR[] = {0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, // 0
0x0800,0x0800,0x0800,0x0800,0x0800,0x0800,0x0800,0x0800,
0x1000,0x1000,0x1000,0x1000,0x1000,0x1000,0x1000,0x1000,
0x1800,0x1800,0x1800,0x1800,0x1800,0x1800,0x1800,0x1800,
0x2000,0x2000,0x2000,0x2000,0x2000,0x2000,0x2000,0x2000,
0x2800,0x2800,0x2800,0x2800,0x2800,0x2800,0x2800,0x2800,
0x3000,0x3000,0x3000,0x3000,0x3000,0x3000,0x3000,0x3000,
0x3800,0x3800,0x3800,0x3800,0x3800,0x3800,0x3800,0x3800,
0x4000,0x4000,0x4000,0x4000,0x4000,0x4000,0x4000,0x4000,
0x4800,0x4800,0x4800,0x4800,0x4800,0x4800,0x4800,0x4800,
0x5000,0x5000,0x5000,0x5000,0x5000,0x5000,0x5000,0x5000,
0x5800,0x5800,0x5800,0x5800,0x5800,0x5800,0x5800,0x5800,
0x6000,0x6000,0x6000,0x6000,0x6000,0x6000,0x6000,0x6000,
0x6800,0x6800,0x6800,0x6800,0x6800,0x6800,0x6800,0x6800,
0x7000,0x7000,0x7000,0x7000,0x7000,0x7000,0x7000,0x7000,
0x7800,0x7800,0x7800,0x7800,0x7800,0x7800,0x7800,0x7800,
0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
0x8800,0x8800,0x8800,0x8800,0x8800,0x8800,0x8800,0x8800,
0x9000,0x9000,0x9000,0x9000,0x9000,0x9000,0x9000,0x9000,
0x9800,0x9800,0x9800,0x9800,0x9800,0x9800,0x9800,0x9800,
0xa000,0xa000,0xa000,0xa000,0xa000,0xa000,0xa000,0xa000,
0xa800,0xa800,0xa800,0xa800,0xa800,0xa800,0xa800,0xa800,
0xb000,0xb000,0xb000,0xb000,0xb000,0xb000,0xb000,0xb000,
0xb800,0xb800,0xb800,0xb800,0xb800,0xb800,0xb800,0xb800,
0xc000,0xc000,0xc000,0xc000,0xc000,0xc000,0xc000,0xc000,
0xc800,0xc800,0xc800,0xc800,0xc800,0xc800,0xc800,0xc800,
0xd000,0xd000,0xd000,0xd000,0xd000,0xd000,0xd000,0xd000,
0xd800,0xd800,0xd800,0xd800,0xd800,0xd800,0xd800,0xd800,
0xe000,0xe000,0xe000,0xe000,0xe000,0xe000,0xe000,0xe000,
0xe800,0xe800,0xe800,0xe800,0xe800,0xe800,0xe800,0xe800,
0xf000,0xf000,0xf000,0xf000,0xf000,0xf000,0xf000,0xf000,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800, // 256
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,0xf800,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 512
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 768
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
//
// Clip and convert green value into 5-bits for RGB565
//
static const uint16_t usRangeTableG[] = {0x0000,0x0000,0x0000,0x0000,0x0020,0x0020,0x0020,0x0020, // 0
0x0040,0x0040,0x0040,0x0040,0x0060,0x0060,0x0060,0x0060,
0x0080,0x0080,0x0080,0x0080,0x00a0,0x00a0,0x00a0,0x00a0,
0x00c0,0x00c0,0x00c0,0x00c0,0x00e0,0x00e0,0x00e0,0x00e0,
0x0100,0x0100,0x0100,0x0100,0x0120,0x0120,0x0120,0x0120,
0x0140,0x0140,0x0140,0x0140,0x0160,0x0160,0x0160,0x0160,
0x0180,0x0180,0x0180,0x0180,0x01a0,0x01a0,0x01a0,0x01a0,
0x01c0,0x01c0,0x01c0,0x01c0,0x01e0,0x01e0,0x01e0,0x01e0,
0x0200,0x0200,0x0200,0x0200,0x0220,0x0220,0x0220,0x0220,
0x0240,0x0240,0x0240,0x0240,0x0260,0x0260,0x0260,0x0260,
0x0280,0x0280,0x0280,0x0280,0x02a0,0x02a0,0x02a0,0x02a0,
0x02c0,0x02c0,0x02c0,0x02c0,0x02e0,0x02e0,0x02e0,0x02e0,
0x0300,0x0300,0x0300,0x0300,0x0320,0x0320,0x0320,0x0320,
0x0340,0x0340,0x0340,0x0340,0x0360,0x0360,0x0360,0x0360,
0x0380,0x0380,0x0380,0x0380,0x03a0,0x03a0,0x03a0,0x03a0,
0x03c0,0x03c0,0x03c0,0x03c0,0x03e0,0x03e0,0x03e0,0x03e0,
0x0400,0x0400,0x0400,0x0400,0x0420,0x0420,0x0420,0x0420,
0x0440,0x0440,0x0440,0x0440,0x0460,0x0460,0x0460,0x0460,
0x0480,0x0480,0x0480,0x0480,0x04a0,0x04a0,0x04a0,0x04a0,
0x04c0,0x04c0,0x04c0,0x04c0,0x04e0,0x04e0,0x04e0,0x04e0,
0x0500,0x0500,0x0500,0x0500,0x0520,0x0520,0x0520,0x0520,
0x0540,0x0540,0x0540,0x0540,0x0560,0x0560,0x0560,0x0560,
0x0580,0x0580,0x0580,0x0580,0x05a0,0x05a0,0x05a0,0x05a0,
0x05c0,0x05c0,0x05c0,0x05c0,0x05e0,0x05e0,0x05e0,0x05e0,
0x0600,0x0600,0x0600,0x0600,0x0620,0x0620,0x0620,0x0620,
0x0640,0x0640,0x0640,0x0640,0x0660,0x0660,0x0660,0x0660,
0x0680,0x0680,0x0680,0x0680,0x06a0,0x06a0,0x06a0,0x06a0,
0x06c0,0x06c0,0x06c0,0x06c0,0x06e0,0x06e0,0x06e0,0x06e0,
0x0700,0x0700,0x0700,0x0700,0x0720,0x0720,0x0720,0x0720,
0x0740,0x0740,0x0740,0x0740,0x0760,0x0760,0x0760,0x0760,
0x0780,0x0780,0x0780,0x0780,0x07a0,0x07a0,0x07a0,0x07a0,
0x07c0,0x07c0,0x07c0,0x07c0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0, // 256
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,0x07e0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 512
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 768
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
//
// Clip and convert blue value into 5-bits for RGB565
//
static const uint16_t usRangeTableB[] = {0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000, // 0
0x0001,0x0001,0x0001,0x0001,0x0001,0x0001,0x0001,0x0001,
0x0002,0x0002,0x0002,0x0002,0x0002,0x0002,0x0002,0x0002,
0x0003,0x0003,0x0003,0x0003,0x0003,0x0003,0x0003,0x0003,
0x0004,0x0004,0x0004,0x0004,0x0004,0x0004,0x0004,0x0004,
0x0005,0x0005,0x0005,0x0005,0x0005,0x0005,0x0005,0x0005,
0x0006,0x0006,0x0006,0x0006,0x0006,0x0006,0x0006,0x0006,
0x0007,0x0007,0x0007,0x0007,0x0007,0x0007,0x0007,0x0007,
0x0008,0x0008,0x0008,0x0008,0x0008,0x0008,0x0008,0x0008,
0x0009,0x0009,0x0009,0x0009,0x0009,0x0009,0x0009,0x0009,
0x000a,0x000a,0x000a,0x000a,0x000a,0x000a,0x000a,0x000a,
0x000b,0x000b,0x000b,0x000b,0x000b,0x000b,0x000b,0x000b,
0x000c,0x000c,0x000c,0x000c,0x000c,0x000c,0x000c,0x000c,
0x000d,0x000d,0x000d,0x000d,0x000d,0x000d,0x000d,0x000d,
0x000e,0x000e,0x000e,0x000e,0x000e,0x000e,0x000e,0x000e,
0x000f,0x000f,0x000f,0x000f,0x000f,0x000f,0x000f,0x000f,
0x0010,0x0010,0x0010,0x0010,0x0010,0x0010,0x0010,0x0010,
0x0011,0x0011,0x0011,0x0011,0x0011,0x0011,0x0011,0x0011,
0x0012,0x0012,0x0012,0x0012,0x0012,0x0012,0x0012,0x0012,
0x0013,0x0013,0x0013,0x0013,0x0013,0x0013,0x0013,0x0013,
0x0014,0x0014,0x0014,0x0014,0x0014,0x0014,0x0014,0x0014,
0x0015,0x0015,0x0015,0x0015,0x0015,0x0015,0x0015,0x0015,
0x0016,0x0016,0x0016,0x0016,0x0016,0x0016,0x0016,0x0016,
0x0017,0x0017,0x0017,0x0017,0x0017,0x0017,0x0017,0x0017,
0x0018,0x0018,0x0018,0x0018,0x0018,0x0018,0x0018,0x0018,
0x0019,0x0019,0x0019,0x0019,0x0019,0x0019,0x0019,0x0019,
0x001a,0x001a,0x001a,0x001a,0x001a,0x001a,0x001a,0x001a,
0x001b,0x001b,0x001b,0x001b,0x001b,0x001b,0x001b,0x001b,
0x001c,0x001c,0x001c,0x001c,0x001c,0x001c,0x001c,0x001c,
0x001d,0x001d,0x001d,0x001d,0x001d,0x001d,0x001d,0x001d,
0x001e,0x001e,0x001e,0x001e,0x001e,0x001e,0x001e,0x001e,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f, // 256
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,0x001f,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 512
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, // 768
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
#if defined (__MACH__) || defined( __LINUX__ ) || defined( __MCUXPRESSO )
//
// API for C
//
//
// Memory initialization
//
int JPEG_openRAM(JPEGIMAGE *pJPEG, uint8_t *pData, int iDataSize, JPEG_DRAW_CALLBACK *pfnDraw)
{
memset(pJPEG, 0, sizeof(JPEGIMAGE));
pJPEG->ucMemType = JPEG_MEM_RAM;
pJPEG->pfnRead = readRAM;
pJPEG->pfnSeek = seekMem;
pJPEG->pfnDraw = pfnDraw;
pJPEG->pfnOpen = NULL;
pJPEG->pfnClose = NULL;
pJPEG->JPEGFile.iSize = iDataSize;
pJPEG->JPEGFile.pData = pData;
pJPEG->iMaxMCUs = 1000; // set to an unnaturally high value to start
return JPEGInit(pJPEG);
} /* JPEG_openRAM() */
//
// File initialization
//
int JPEG_openFile(JPEGIMAGE *pJPEG, const char *szFilename, JPEG_DRAW_CALLBACK *pfnDraw)
{
memset(pJPEG, 0, sizeof(JPEGIMAGE));
pJPEG->ucMemType = JPEG_MEM_RAM;
pJPEG->pfnRead = readFile;
pJPEG->pfnSeek = seekFile;
pJPEG->pfnDraw = pfnDraw;
pJPEG->pfnOpen = NULL;
pJPEG->pfnClose = closeFile;
pJPEG->iMaxMCUs = 1000; // set to an unnaturally high value to start
pJPEG->JPEGFile.fHandle = fopen(szFilename, "r+b");
if (pJPEG->JPEGFile.fHandle == NULL)
return 0;
fseek((FILE *)pJPEG->JPEGFile.fHandle, 0, SEEK_END);
pJPEG->JPEGFile.iSize = (int)ftell((FILE *)pJPEG->JPEGFile.fHandle);
fseek((FILE *)pJPEG->JPEGFile.fHandle, 0, SEEK_SET);
return JPEGInit(pJPEG);
} /* JPEG_openFile() */
int JPEG_getLastError(JPEGIMAGE *pJPEG)
{
return pJPEG->iError;
} /* JPEG_getLastError() */
int JPEG_getWidth(JPEGIMAGE *pJPEG)
{
return pJPEG->iWidth;
} /* JPEG_getWidth() */
int JPEG_getHeight(JPEGIMAGE *pJPEG)
{
return pJPEG->iHeight;
} /* JPEG_getHeight() */
int JPEG_getOrientation(JPEGIMAGE *pJPEG)
{
return (int)pJPEG->ucOrientation;
} /* JPEG_getOrientation() */
int JPEG_getBpp(JPEGIMAGE *pJPEG)
{
return (int)pJPEG->ucBpp;
} /* JPEG_getBpp() */
int JPEG_getSubSample(JPEGIMAGE *pJPEG)
{
return (int)pJPEG->ucSubSample;
} /* JPEG_getSubSample() */
int JPEG_hasThumb(JPEGIMAGE *pJPEG)
{
return (int)pJPEG->ucHasThumb;
} /* JPEG_hasThumb() */
int JPEG_getThumbWidth(JPEGIMAGE *pJPEG)
{
return pJPEG->iThumbWidth;
} /* JPEG_getThumbWidth() */
int JPEG_getThumbHeight(JPEGIMAGE *pJPEG)
{
return pJPEG->iThumbHeight;
} /* JPEG_getThumbHeight() */
void JPEG_setPixelType(JPEGIMAGE *pJPEG, int iType)
{
pJPEG->ucPixelType = (uint8_t)iType;
} /* JPEG_setPixelType() */
void JPEG_setMaxOutputSize(JPEGIMAGE *pJPEG, int iMaxMCUs)
{
if (iMaxMCUs < 1)
iMaxMCUs = 1; // don't allow invalid value
pJPEG->iMaxMCUs = iMaxMCUs;
} /* JPEG_setMaxOutputSize() */
int JPEG_decode(JPEGIMAGE *pJPEG, int x, int y, int iOptions)
{
pJPEG->iXOffset = x;
pJPEG->iYOffset = y;
pJPEG->iOptions = iOptions;
return DecodeJPEG(pJPEG);
} /* JPEG_decode() */
int JPEG_decodeDither(JPEGIMAGE *pJPEG, uint8_t *pDither, int iOptions)
{
pJPEG->iOptions = iOptions;
pJPEG->pDitherBuffer = pDither;
return DecodeJPEG(pJPEG);
} /* JPEG_decodeDither() */
void JPEG_close(JPEGIMAGE *pJPEG)
{
if (pJPEG->pfnClose)
(*pJPEG->pfnClose)(pJPEG->JPEGFile.fHandle);
} /* JPEG_close() */
#endif // !__cplusplus
//
// Validate/adjust the requested crop area to land on MCU boundaries
// (expand in all directions if needed)
//
void JPEG_setCropArea(JPEGIMAGE *pJPEG, int x, int y, int w, int h)
{
int mcuCX=0, mcuCY=0;
if (x < 0)
x = 0;
if (y < 0)
y = 0;
switch (pJPEG->ucSubSample) // set up the parameters for the different subsampling options
{
case 0x00: // fake value to handle grayscale
case 0x01: // fake value to handle sRGB/CMYK
case 0x11:
mcuCX = mcuCY = 8;
break;
case 0x12:
mcuCX = 8;
mcuCY = 16;
break;
case 0x21:
mcuCX = 16;
mcuCY = 8;
break;
case 0x22:
mcuCX = mcuCY = 16;
break;
default: // to suppress compiler warning
break;
}
if (w & (mcuCX-1)) {
w &= ~(mcuCX-1);
w += mcuCX;
}
if (h & (mcuCY-1)) {
h &= ~(mcuCY-1);
h += mcuCY;
}
if (x > pJPEG->iWidth-mcuCX) x = pJPEG->iWidth-mcuCX;
if (y > pJPEG->iHeight-mcuCY) y = pJPEG->iHeight-mcuCY;
if (x + w > pJPEG->iWidth) w = pJPEG->iWidth - mcuCX;
if (y + h > pJPEG->iHeight) h = pJPEG->iHeight - mcuCY;
x &= ~(mcuCX-1);
y &= ~(mcuCY-1);
pJPEG->iCropX = x; pJPEG->iCropY = y;
pJPEG->iCropCX = w; pJPEG->iCropCY = h;
} /* JPEG_setCropArea() */
void JPEG_getCropArea(JPEGIMAGE *pJPEG, int *x, int *y, int *w, int *h)
{
*x = pJPEG->iCropX; *y = pJPEG->iCropY;
*w = pJPEG->iCropCX; *h = pJPEG->iCropCY;
} /* JPEG_getCropArea() */
void JPEG_setFramebuffer(JPEGIMAGE *pJPEG, void *pFramebuffer)
{
pJPEG->pFramebuffer = pFramebuffer;
} /* JPEG_setFramebuffer() */
//
// Helper functions for memory based images
//
static int32_t readRAM(JPEGFILE *pFile, uint8_t *pBuf, int32_t iLen)
{
int32_t iBytesRead;
iBytesRead = iLen;
if ((pFile->iSize - pFile->iPos) < iLen)
iBytesRead = pFile->iSize - pFile->iPos;
if (iBytesRead <= 0)
return 0;
memcpy(pBuf, &pFile->pData[pFile->iPos], iBytesRead);
pFile->iPos += iBytesRead;
return iBytesRead;
} /* readRAM() */
static int32_t readFLASH(JPEGFILE *pFile, uint8_t *pBuf, int32_t iLen)
{
int32_t iBytesRead;
iBytesRead = iLen;
if ((pFile->iSize - pFile->iPos) < iLen)
iBytesRead = pFile->iSize - pFile->iPos;
if (iBytesRead <= 0)
return 0;
memcpy_P(pBuf, &pFile->pData[pFile->iPos], iBytesRead);
pFile->iPos += iBytesRead;
return iBytesRead;
} /* readFLASH() */
static int32_t seekMem(JPEGFILE *pFile, int32_t iPosition)
{
if (iPosition < 0) iPosition = 0;
else if (iPosition >= pFile->iSize) iPosition = pFile->iSize-1;
pFile->iPos = iPosition;
return iPosition;
} /* seekMem() */
#if defined (__MACH__) || defined( __LINUX__ ) || defined( __MCUXPRESSO )
static void closeFile(void *handle)
{
fclose((FILE *)handle);
} /* closeFile() */
static int32_t seekFile(JPEGFILE *pFile, int32_t iPosition)
{
if (iPosition < 0) iPosition = 0;
else if (iPosition >= pFile->iSize) iPosition = pFile->iSize-1;
pFile->iPos = iPosition;
fseek((FILE *)pFile->fHandle, iPosition, SEEK_SET);
return iPosition;
} /* seekFile() */
static int32_t readFile(JPEGFILE *pFile, uint8_t *pBuf, int32_t iLen)
{
int32_t iBytesRead;
iBytesRead = iLen;
if ((pFile->iSize - pFile->iPos) < iLen)
iBytesRead = pFile->iSize - pFile->iPos;
if (iBytesRead <= 0)
return 0;
iBytesRead = (int)fread(pBuf, 1, iBytesRead, (FILE *)pFile->fHandle);
pFile->iPos += iBytesRead;
return iBytesRead;
} /* readFile() */
#endif // __LINUX__
//
// The following functions are written in plain C and have no
// 3rd party dependencies, not even the C runtime library
//
//
// Initialize a JPEG file and callback access from a file on SD or memory
// returns 1 for success, 0 for failure
// Fills in the basic image info fields of the JPEGIMAGE structure
//
static int JPEGInit(JPEGIMAGE *pJPEG)
{
return JPEGParseInfo(pJPEG, 0); // gather info for image
} /* JPEGInit() */
//
// Unpack the Huffman tables
//
static int JPEGGetHuffTables(uint8_t *pBuf, int iLen, JPEGIMAGE *pJPEG)
{
int i, j, iOffset, iTableOffset;
uint8_t ucTable, *pHuffVals;
iOffset = 0;
pHuffVals = (uint8_t *)pJPEG->usPixels; // temp holding area to save RAM
while (iLen > 17) // while there are tables to copy (we may have combined more than 1 table together)
{
ucTable = pBuf[iOffset++]; // get table index
if (ucTable & 0x10) // convert AC offset of 0x10 into offset of 4
ucTable ^= 0x14;
pJPEG->ucHuffTableUsed |= (1 << ucTable); // mark this table as being defined
if (ucTable <= 7) // tables are 0-3, AC+DC
{
iTableOffset = ucTable * HUFF_TABLEN;
j = 0; // total bits
for (i=0; i<16; i++)
{
j += pBuf[iOffset];
pHuffVals[iTableOffset+i] = pBuf[iOffset++];
}
iLen -= 17; // subtract length of bit lengths
if (j == 0 || j > 256 || j > iLen) // bogus bit lengths
{
return -1;
}
iTableOffset += 16;
for (i=0; i<j; i++)
{ // copy huffman table
pHuffVals[iTableOffset+i] = pBuf[iOffset++];
}
iLen -= j;
}
}
return 0;
} /* JPEGGetHuffTables() */
#ifdef FUTURE
//
// Create 11-bit lookup tables for some images where it doesn't work
// for 10-bit tables
//
static int JPEGMakeHuffTables_Slow(JPEGIMAGE *pJPEG, int bThumbnail)
{
int code, repeat, count, codestart;
int j;
int iLen, iTable;
unsigned short *pTable, *pShort, *pLong;
unsigned char *pucTable, *pucShort, *pucLong;
uint32_t ul, *pLongTable;
int iBitNum; // current code bit length
int cc; // code
unsigned char *p, *pBits, ucCode;
int iMaxLength, iMaxMask;
pJPEG->b11Bit = 1; // indicate we're using the bigger A/C decode tables
// first do DC components (up to 4 tables of 12-bit codes)
// we can save time and memory for the DC codes by knowing that there exist short codes (<= 6 bits)
// and long codes (>6 bits, but the first 5 bits are 1's). This allows us to create 2 tables: a 6-bit and 7 or 8-bit
// to handle any DC codes
iMaxLength = 12; // assume DC codes can be 12-bits
iMaxMask = 0x7f; // lower 7 bits after truncate 5 leading 1's
if (pJPEG->ucMode == 0xc3) // create 13-bit tables for lossless mode
{
iMaxLength = 13;
iMaxMask = 0xff;
}
for (iTable = 0; iTable < 2; iTable++)
{
if (pJPEG->ucHuffTableUsed & (1<<iTable))
{
// pJPEG->huffdcFast[iTable] = (int *)PILIOAlloc(0x180); // short table = 128 bytes, long table = 256 bytes
pucShort = (unsigned char *)&pJPEG->ucHuffDC[iTable*DC_TABLE_SIZE];
// pJPEG->huffdc[iTable] = pJPEG->huffdcFast[iTable] + 0x20; // 0x20 longs = 128 bytes
pucLong = (unsigned char *)&pJPEG->ucHuffDC[iTable*DC_TABLE_SIZE + 128];
pBits = &pJPEG->ucHuffVals[iTable * HUFF_TABLEN];
p = pBits;
p += 16; // point to bit data
cc = 0; // start with a code of 0
for (iBitNum = 1; iBitNum <= 16; iBitNum++)
{
iLen = *pBits++; // get number of codes for this bit length
if (iBitNum > iMaxLength && iLen > 0) // we can't handle codes longer a certain length
{
return -1;
}
while (iLen)
{
// if (iBitNum > 6) // do long table
if ((cc >> (iBitNum-5)) == 0x1f) // first 5 bits are 1 - use long table
{
count = iMaxLength - iBitNum;
codestart = cc << count;
pucTable = &pucLong[codestart & iMaxMask]; // use lower 7/8 bits of code
}
else // do short table
{
count = 6 - iBitNum;
if (count < 0)
return -1; // DEBUG - something went wrong
codestart = cc << count;
pucTable = &pucShort[codestart];
}
ucCode = *p++; // get actual huffman code
if (ucCode == 16 && pJPEG->ucMode == 0xc3) // lossless mode
{
// in lossless mode, this code won't fit in 4 bits, so save it's length in the next slot
ucCode = 255;
pucLong[256] = (unsigned char)iBitNum;
}
// does precalculating the DC value save time on ARM?
#ifndef USE_ARM_ASM
if (ucCode != 0 && (ucCode + iBitNum) <= 6 && pJPEG->ucMode != 0xc2) // we can fit the magnitude value in the code lookup (not for progressive)
{
int k, iLoop;
unsigned char ucCoeff;
unsigned char *d = &pucTable[512];
unsigned char ucMag = ucCode;
ucCode |= ((iBitNum+ucCode) << 4); // add magnitude bits to length
repeat = 1<<ucMag;
iLoop = 1<<(count-ucMag);
for (j=0; j<repeat; j++)
{ // calcuate the magnitude coeff already
if (j & 1<<(ucMag-1)) // positive number
ucCoeff = (unsigned char)j;
else // negative number
ucCoeff = (unsigned char)(j - ((1<<ucMag)-1));
for (k=0; k<iLoop; k++)
{
*d++ = ucCoeff;
} // for k
} // for j
}
#endif
else
{
ucCode |= (iBitNum << 4);
}
if (count) // do it as dwords to save time
{
repeat = (1<<count);
memset(pucTable, ucCode, repeat);
// pLongTable = (uint32_t *)pTable;
// repeat = 1 << (count-2); // store as dwords (/4)
// ul = code | (code << 16);
// for (j=0; j<repeat; j++)
// *pLongTable++ = ul;
}
else
{
pucTable[0] = ucCode;
}
cc++;
iLen--;
}
cc <<= 1;
}
} // if table defined
}
// now do AC components (up to 2 tables of 16-bit codes)
// We split the codes into a short table (9 bits or less) and a long table (first 5 bits are 1)
for (iTable = 0; iTable < 2; iTable++)
{
if (pJPEG->ucHuffTableUsed & (1<<(iTable+4))) // if this table is defined
{
pBits = &pJPEG->ucHuffVals[(iTable+4) * HUFF_TABLEN];
p = pBits;
p += 16; // point to bit data
pShort = &pJPEG->usHuffAC[iTable*HUFF11SIZE];
pLong = &pJPEG->usHuffAC[iTable*HUFF11SIZE + 1024]; // long codes start here
cc = 0; // start with a code of 0
// construct the decode table
for (iBitNum = 1; iBitNum <= 16; iBitNum++)
{
iLen = *pBits++; // get number of codes for this bit length
while (iLen)
{
if ((cc >> (iBitNum-4)) == 0xf) // first 4 bits are 1 - use long table
{
count = 16 - iBitNum;
codestart = cc << count;
pTable = &pLong[codestart & 0xfff]; // use lower 12 bits of code
}
else
{
count = 12 - iBitNum;
if (count < 0) // a 13-bit? code - that doesn't fit our optimized scheme, see if we can do a bigger table version
{
return -1; // DEBUG - fatal error, we currently don't support it
}
codestart = cc << count;
pTable = &pShort[codestart]; // 11 bits or shorter
}
code = *p++; // get actual huffman code
if (bThumbnail && code != 0) // add "extra" bits to code length since we skip these codes
{
// get rid of extra bits in code and add increment (1) for AC index
code = ((iBitNum+(code & 0xf)) << 8) | ((code >> 4)+1);
}
else
{
code |= (iBitNum << 8);
}
if (count) // do it as dwords to save time
{
repeat = 1 << (count-1); // store as dwords (/2)
ul = code | (code << 16);
pLongTable = (uint32_t *)pTable;
for (j=0; j<repeat; j++)
*pLongTable++ = ul;
}
else
{
pTable[0] = (unsigned short)code;
}
cc++;
iLen--;
}
cc <<= 1;
} // for each bit length
} // if table defined
}
return 0;
} /* JPEGMakeHuffTables_Slow() */
#endif // FUTURE
//
// Expand the Huffman tables for fast decoding
// returns 1 for success, 0 for failure
//
static int JPEGMakeHuffTables(JPEGIMAGE *pJPEG, int bThumbnail)
{
int code, repeat, count, codestart;
int j;
int iLen, iTable;
uint16_t *pTable, *pShort, *pLong;
uint8_t *pHuffVals, *pucTable, *pucShort, *pucLong;
uint32_t ul, *pLongTable;
int iBitNum; // current code bit length
int cc; // code
uint8_t *p, *pBits, ucCode;
int iMaxLength, iMaxMask;
int iTablesUsed;
iTablesUsed = 0;
pHuffVals = (uint8_t *)pJPEG->usPixels;
for (j=0; j<4; j++)
{
if (pJPEG->ucHuffTableUsed & (1 << j))
iTablesUsed++;
}
// first do DC components (up to 4 tables of 12-bit codes)
// we can save time and memory for the DC codes by knowing that there exist short codes (<= 6 bits)
// and long codes (>6 bits, but the first 5 bits are 1's). This allows us to create 2 tables: a 6-bit and 7 or 8-bit
// to handle any DC codes
iMaxLength = 12; // assume DC codes can be 12-bits
iMaxMask = 0x7f; // lower 7 bits after truncate 5 leading 1's
for (iTable = 0; iTable < 4; iTable++)
{
if (pJPEG->ucHuffTableUsed & (1 << iTable))
{
// pJPEG->huffdcFast[iTable] = (int *)PILIOAlloc(0x180); // short table = 128 bytes, long table = 256 bytes
pucShort = &pJPEG->ucHuffDC[iTable*DC_TABLE_SIZE];
// pJPEG->huffdc[iTable] = pJPEG->huffdcFast[iTable] + 0x20; // 0x20 longs = 128 bytes
pucLong = &pJPEG->ucHuffDC[iTable*DC_TABLE_SIZE + 128];
pBits = &pHuffVals[iTable * HUFF_TABLEN];
p = pBits;
p += 16; // point to bit data
cc = 0; // start with a code of 0
for (iBitNum = 1; iBitNum <= 16; iBitNum++)
{
iLen = *pBits++; // get number of codes for this bit length
if (iBitNum > iMaxLength && iLen > 0) // we can't handle codes longer a certain length
{
return 0;
}
while (iLen)
{
// if (iBitNum > 6) // do long table
if ((cc >> (iBitNum-5)) == 0x1f) // first 5 bits are 1 - use long table
{
count = iMaxLength - iBitNum;
codestart = cc << count;
pucTable = &pucLong[codestart & iMaxMask]; // use lower 7/8 bits of code
}
else // do short table
{
count = 6 - iBitNum;
if (count < 0)
return 0; // DEBUG - something went wrong
codestart = cc << count;
pucTable = &pucShort[codestart];
}
ucCode = *p++; // get actual huffman code
// does precalculating the DC value save time on ARM?
#ifndef USE_ARM_ASM
if (ucCode != 0 && (ucCode + iBitNum) <= 6 && pJPEG->ucMode != 0xc2) // we can fit the magnitude value in the code lookup (not for progressive)
{
int k, iLoop;
unsigned char ucCoeff;
unsigned char *d = &pucTable[512];
unsigned char ucMag = ucCode;
ucCode |= ((iBitNum+ucCode) << 4); // add magnitude bits to length
repeat = 1<<ucMag;
iLoop = 1<<(count-ucMag);
for (j=0; j<repeat; j++)
{ // calcuate the magnitude coeff already
if (j & 1<<(ucMag-1)) // positive number
ucCoeff = (unsigned char)j;
else // negative number
ucCoeff = (unsigned char)(j - ((1<<ucMag)-1));
for (k=0; k<iLoop; k++)
{
*d++ = ucCoeff;
} // for k
} // for j
}
#endif
else
{
ucCode |= (iBitNum << 4);
}
if (count) // do it as dwords to save time
{
repeat = (1<<count);
memset(pucTable, ucCode, repeat);
// pLongTable = (uint32_t *)pTable;
// repeat = 1 << (count-2); // store as dwords (/4)
// ul = code | (code << 16);
// for (j=0; j<repeat; j++)
// *pLongTable++ = ul;
}
else
{
pucTable[0] = ucCode;
}
cc++;
iLen--;
}
cc <<= 1;
}
} // if table defined
}
// now do AC components (up to 4 tables of 16-bit codes)
// We split the codes into a short table (9 bits or less) and a long table (first 5 bits are 1)
for (iTable = 0; iTable < 4; iTable++)
{
if (pJPEG->ucHuffTableUsed & (1 << (iTable+4))) // if this table is defined
{
pBits = &pHuffVals[(iTable+4) * HUFF_TABLEN];
p = pBits;
p += 16; // point to bit data
if (iTable * HUFF11SIZE >= sizeof(pJPEG->usHuffAC) / 2)
return 0;
pShort = &pJPEG->usHuffAC[iTable*HUFF11SIZE];
pLong = &pJPEG->usHuffAC[iTable*HUFF11SIZE + 1024];
cc = 0; // start with a code of 0
// construct the decode table
for (iBitNum = 1; iBitNum <= 16; iBitNum++)
{
iLen = *pBits++; // get number of codes for this bit length
while (iLen)
{
if ((cc >> (iBitNum-6)) == 0x3f) // first 6 bits are 1 - use long table
{
count = 16 - iBitNum;
codestart = cc << count;
pTable = &pLong[codestart & 0x3ff]; // use lower 10 bits of code
}
else
{
count = 10 - iBitNum;
if (count < 0) // an 11/12-bit? code - that doesn't fit our optimized scheme, see if we can do a bigger table version
{
if (count == -1 && iTablesUsed <= 4) // we need to create "slow" tables
{ // DEBUG
// j = JPEGMakeHuffTables_Slow(pJPEG, bThumbnail);
return 0;
}
else
return 0; // DEBUG - fatal error, more than 2 big tables we currently don't support
}
codestart = cc << count;
pTable = &pShort[codestart]; // 10 bits or shorter
}
code = *p++; // get actual huffman code
if (bThumbnail && code != 0) // add "extra" bits to code length since we skip these codes
{
// get rid of extra bits in code and add increment (1) for AC index
code = ((iBitNum+(code & 0xf)) << 8) | ((code >> 4)+1);
}
#ifdef BOGUS // precalculating the AC coeff makes it run slightly slower
else if ((code & 0xf) != 0 && (code + iBitNum) <= 10) // we can fit the magnitude value + huffman code in a single read
{
int k, iLoop;
unsigned short usCoeff;
unsigned short *d = &pTable[4096]; // use unused table slots 2+3 for extra coeff data
unsigned char ucMag = (unsigned char)(code & 0xf);
code |= ((iBitNum + (code & 0xf)) << 8); // add magnitude bits to length
repeat = 1<<ucMag;
iLoop = 1<<(count-ucMag);
for (j=0; j<repeat; j++)
{ // calcuate the magnitude coeff already
if (j & 1<<(ucMag-1)) // positive number
usCoeff = (unsigned short)j;
else // negative number
usCoeff = (unsigned short)(j - ((1<<ucMag)-1));
for (k=0; k<iLoop; k++)
{
*d++ = usCoeff;
} // for k
} // for j
}
#endif
else
{
code |= (iBitNum << 8);
}
if (count) // do it as dwords to save time
{
repeat = 1 << (count-1); // store as dwords (/2)
ul = code | (code << 16);
pLongTable = (uint32_t *)pTable;
for (j=0; j<repeat; j++)
*pLongTable++ = ul;
}
else
{
pTable[0] = (unsigned short)code;
}
cc++;
iLen--;
}
cc <<= 1;
} // for each bit length
} // if table defined
}
return 1;
} /* JPEGMakeHuffTables() */
//
// TIFFSHORT
// read a 16-bit unsigned integer from the given pointer
// and interpret the data as big endian (Motorola) or little endian (Intel)
//
static uint16_t TIFFSHORT(unsigned char *p, int bMotorola)
{
unsigned short s;
if (bMotorola)
s = *p * 0x100 + *(p+1); // big endian (AKA Motorola byte order)
else
s = *p + *(p+1)*0x100; // little endian (AKA Intel byte order)
return s;
} /* TIFFSHORT() */
//
// TIFFLONG
// read a 32-bit unsigned integer from the given pointer
// and interpret the data as big endian (Motorola) or little endian (Intel)
//
static uint32_t TIFFLONG(unsigned char *p, int bMotorola)
{
uint32_t l;
if (bMotorola)
l = *p * 0x1000000 + *(p+1) * 0x10000 + *(p+2) * 0x100 + *(p+3); // big endian
else
l = *p + *(p+1) * 0x100 + *(p+2) * 0x10000 + *(p+3) * 0x1000000; // little endian
return l;
} /* TIFFLONG() */
//
// TIFFVALUE
// read an integer value encoded in a TIFF TAG (12-byte structure)
// and interpret the data as big endian (Motorola) or little endian (Intel)
//
static int TIFFVALUE(unsigned char *p, int bMotorola)
{
int i, iType;
iType = TIFFSHORT(p+2, bMotorola);
/* If pointer to a list of items, must be a long */
if (TIFFSHORT(p+4, bMotorola) > 1)
{
iType = 4;
}
switch (iType)
{
case 3: /* Short */
i = TIFFSHORT(p+8, bMotorola);
break;
case 4: /* Long */
case 7: // undefined (treat it as a long since it's usually a multibyte buffer)
i = TIFFLONG(p+8, bMotorola);
break;
case 6: // signed byte
i = (signed char)p[8];
break;
case 2: /* ASCII */
case 5: /* Unsigned Rational */
case 10: /* Signed Rational */
i = TIFFLONG(p+8, bMotorola);
break;
default: /* to suppress compiler warning */
i = 0;
break;
}
return i;
} /* TIFFVALUE() */
static void GetTIFFInfo(JPEGIMAGE *pPage, int bMotorola, int iOffset)
{
int iTag, iTagCount, i;
uint8_t *cBuf = pPage->ucFileBuf;
iTagCount = TIFFSHORT(&cBuf[iOffset], bMotorola); /* Number of tags in this dir */
if (iTagCount < 1 || iTagCount > 256) // invalid tag count
return; /* Bad header info */
/*--- Search the TIFF tags ---*/
for (i=0; i<iTagCount; i++)
{
unsigned char *p = &cBuf[iOffset + (i*12) +2];
iTag = TIFFSHORT(p, bMotorola); /* current tag value */
if (iTag == 274) // orientation tag
{
pPage->ucOrientation = TIFFVALUE(p, bMotorola);
}
else if (iTag == 256) // width of thumbnail
{
pPage->iThumbWidth = TIFFVALUE(p, bMotorola);
}
else if (iTag == 257) // height of thumbnail
{
pPage->iThumbHeight = TIFFVALUE(p, bMotorola);
}
else if (iTag == 513) // offset to JPEG data
{
pPage->iThumbData = TIFFVALUE(p, bMotorola);
}
}
} /* GetTIFFInfo() */
static int JPEGGetSOS(JPEGIMAGE *pJPEG, int *iOff)
{
int16_t sLen;
int iOffset = *iOff;
int i, j;
uint8_t uc,c,cc;
uint8_t *buf = pJPEG->ucFileBuf;
sLen = MOTOSHORT(&buf[iOffset]);
iOffset += 2;
// Assume no components in this scan
for (i=0; i<4; i++)
pJPEG->JPCI[i].component_needed = 0;
uc = buf[iOffset++]; // get number of components
pJPEG->ucComponentsInScan = uc;
sLen -= 3;
if (uc < 1 || uc > MAX_COMPS_IN_SCAN || sLen != (uc*2+3)) // check length of data packet
return 1; // error
for (i=0; i<uc; i++)
{
cc = buf[iOffset++];
c = buf[iOffset++];
sLen -= 2;
for (j=0; j<4; j++) // search for component id
{
if (pJPEG->JPCI[j].component_id == cc)
break;
}
if (j == 4) // error, not found
return 1;
if ((c & 0xf) > 3 || (c & 0xf0) > 0x30)
return 1; // bogus table numbers
pJPEG->JPCI[j].dc_tbl_no = c >> 4;
pJPEG->JPCI[j].ac_tbl_no = c & 0xf;
pJPEG->JPCI[j].component_needed = 1; // mark this component as being included in the scan
}
pJPEG->iScanStart = buf[iOffset++]; // Get the scan start (or lossless predictor) for this scan
pJPEG->iScanEnd = buf[iOffset++]; // Get the scan end for this scan
c = buf[iOffset++]; // successive approximation bits
pJPEG->cApproxBitsLow = c & 0xf; // also point transform in lossless mode
pJPEG->cApproxBitsHigh = c >> 4;
*iOff = iOffset;
return 0;
} /* JPEGGetSOS() */
//
// Remove markers from the data stream to allow faster decode
// Stuffed zeros and restart interval markers aren't needed to properly decode
// the data, but they make reading VLC data slower, so I pull them out first
//
static int JPEGFilter(uint8_t *pBuf, uint8_t *d, int iLen, uint8_t *bFF)
{
#ifdef HAS_SSE
__m128i xmmIn, xmmOut;
__m128i xmmFF = _mm_cmpeq_epi8(xmmIn, xmmIn);
#endif // HAS_SSE
#ifdef HAS_NEON
uint8x16_t u816FF = vdupq_n_u8(0xff);
uint8x16_t u816In, u816Out;
#ifdef OLD_NEON
uint8x8_t u88Merged;
uint32x2_t u322merged;
#endif // OLD_NEON
#endif // HAS_NEON
unsigned char c, *s, *pEnd, *pStart;
pStart = d;
s = pBuf;
pEnd = &s[iLen-1]; // stop just shy of the end to not miss a final marker/stuffed 0
if (*bFF) // last byte was a FF, check the next one
{
if (s[0] == 0) // stuffed 0, keep the FF
*d++ = 0xff;
s++;
*bFF = 0;
}
#ifdef HAS_SSE
while (s < pEnd-16)
{
xmmIn = _mm_loadu_si128((__m128i*)s);
xmmOut = _mm_cmpeq_epi8(xmmFF, xmmIn); // any FF's in these 16 bytes?
if (_mm_movemask_epi8(xmmOut) == 0) // no FF's, just copy this block
{
_mm_storeu_si128((__m128i*)d, xmmIn);
s += 16;
d += 16;
}
else
{
int i = 16; // do these 16 bytes the slow way
while (i) {
c = *d++ = *s++;
if (c == 0xff) { // marker or stuffed zeros?
if (s[0] != 0) { // it's a marker, skip both
d--;
}
s++; // for stuffed 0's, store the FF, skip the 00
} // found FF
i--;
} // while processing the 16 "slow" bytes
}
} // while SSE filtering
#endif // HAS_SSE
#ifdef HAS_NEON
while (s < pEnd - 16)
{
u816In = vld1q_u8(s);
u816Out = vceqq_u8(u816FF, u816In); // any FF's in these 16 bytes?
#ifdef OLD_NEON
u88Merged = vpadd_u8(vget_high_u8(u816Out), vget_low_u8(u816Out));
u322merged = vpadd_u32 (vreinterpret_u32_u8(u88Merged), vreinterpret_u32_u8(u88Merged));
if (vget_lane_u32 (u322merged, 0) == 0) // no FF's, just copy this block
#else
if (vaddvq_u8(u816Out) == 0) // any byte != 0 means FFs
#endif
{
vst1q_u8(d, u816In);
s += 16;
d += 16;
}
else
{
int i = 16; // do these 16 bytes the slow way
while (i) {
c = *d++ = *s++;
if (c == 0xff) { // marker or stuffed zeros?
if (s[0] != 0) { // it's a marker, skip both
d--;
}
s++; // for stuffed 0's, store the FF, skip the 00
} // found FF
i--;
} // while processing the 16 "slow" bytes
} // if need to remove stuffed FF's or markers
} // while processing buffer with SIMD
#endif // HAS_NEON
while (s < pEnd)
{
c = *d++ = *s++;
if (c == 0xff) // marker or stuffed zeros?
{
if (s[0] != 0) // it's a marker, skip both
{
d--;
}
s++; // for stuffed 0's, store the FF, skip the 00
}
}
if (s == pEnd) // need to test the last byte
{
c = s[0];
if (c == 0xff) // last byte is FF, take care of it next time through
*bFF = 1; // take care of it next time through
else
*d++ = c; // nope, just store it
}
return (int)(d-pStart); // filtered output length
} /* JPEGFilter() */
//
// Read and filter more VLC data for decoding
//
static void JPEGGetMoreData(JPEGIMAGE *pPage)
{
int iDelta = pPage->iVLCSize - pPage->iVLCOff;
// printf("Getting more data...size=%d, off=%d\n", pPage->iVLCSize, pPage->iVLCOff);
// move any existing data down
if (iDelta >= (JPEG_FILE_BUF_SIZE-64) || iDelta < 0)
return; // buffer is already full; no need to read more data
if (pPage->iVLCOff != 0)
{
memcpy(pPage->ucFileBuf, &pPage->ucFileBuf[pPage->iVLCOff], pPage->iVLCSize - pPage->iVLCOff);
pPage->iVLCSize -= pPage->iVLCOff;
pPage->iVLCOff = 0;
pPage->bb.pBuf = pPage->ucFileBuf; // reset VLC source pointer too
}
if (pPage->JPEGFile.iPos < pPage->JPEGFile.iSize && pPage->iVLCSize < JPEG_FILE_BUF_SIZE-64)
{
int i;
// Try to read enough to fill the buffer
i = (*pPage->pfnRead)(&pPage->JPEGFile, &pPage->ucFileBuf[pPage->iVLCSize], JPEG_FILE_BUF_SIZE - pPage->iVLCSize); // max length we can read
// Filter out the markers
pPage->iVLCSize += JPEGFilter(&pPage->ucFileBuf[pPage->iVLCSize], &pPage->ucFileBuf[pPage->iVLCSize], i, &pPage->ucFF);
}
} /* JPEGGetMoreData() */
//
// Parse the JPEG header, gather necessary info to decode the image
// Returns 1 for success, 0 for failure
//
static int JPEGParseInfo(JPEGIMAGE *pPage, int bExtractThumb)
{
int iBytesRead;
int i, iOffset, iTableOffset;
uint8_t ucTable, *s = pPage->ucFileBuf;
uint16_t usMarker, usLen = 0;
int iFilePos = 0;
pPage->pFramebuffer = NULL; // this must be set AFTER calling this function
// make sure usPixels is 16-byte aligned for S3 SIMD (and possibly others)
i = (int)(int64_t)pPage->usUnalignedPixels;
i &= 15;
if (i == 0) i = 16; // already 16-byte aligned
pPage->usPixels = &pPage->usUnalignedPixels[(16-i)>>1];
// do the same for the MCU buffers
i = (int)(int64_t)pPage->sUnalignedMCUs;
i &= 15;
if (i == 0) i = 16;
pPage->sMCUs = &pPage->sUnalignedMCUs[(16-i)>>1];
if (bExtractThumb) // seek to the start of the thumbnail image
{
iFilePos = pPage->iThumbData;
(*pPage->pfnSeek)(&pPage->JPEGFile, iFilePos);
}
iBytesRead = (*pPage->pfnRead)(&pPage->JPEGFile, s, JPEG_FILE_BUF_SIZE);
if (iBytesRead < 256) // a JPEG file this tiny? probably bad
{
pPage->iError = JPEG_INVALID_FILE;
return 0;
}
iFilePos += iBytesRead;
if (MOTOSHORT(pPage->ucFileBuf) != 0xffd8)
{
pPage->iError = JPEG_INVALID_FILE;
return 0; // not a JPEG file
}
iOffset = 2; /* Start at offset of first marker */
usMarker = 0; /* Search for SOFx (start of frame) marker */
while (usMarker != 0xffda && iOffset < pPage->JPEGFile.iSize)
{
if (iOffset >= JPEG_FILE_BUF_SIZE/2) // too close to the end, read more data
{
// Do we need to seek first?
if (iOffset >= JPEG_FILE_BUF_SIZE)
{
iFilePos += (iOffset - iBytesRead);
iOffset = 0;
(*pPage->pfnSeek)(&pPage->JPEGFile, iFilePos);
iBytesRead = 0; // throw away any old data
}
// move existing bytes down
if (iOffset)
{
memcpy(pPage->ucFileBuf, &pPage->ucFileBuf[iOffset], iBytesRead - iOffset);
iBytesRead -= iOffset;
iOffset = 0;
}
i = (*pPage->pfnRead)(&pPage->JPEGFile, &pPage->ucFileBuf[iBytesRead], JPEG_FILE_BUF_SIZE-iBytesRead);
iFilePos += i;
iBytesRead += i;
}
usMarker = MOTOSHORT(&s[iOffset]);
iOffset += 2;
usLen = MOTOSHORT(&s[iOffset]); // marker length
if (usMarker < 0xffc0 || usMarker == 0xffff) // invalid marker, could be generated by "Arles Image Web Page Creator" or Accusoft
{
iOffset++;
continue; // skip 1 byte and try to resync
}
switch (usMarker)
{
case 0xffc1:
case 0xffc2:
case 0xffc3:
pPage->iError = JPEG_UNSUPPORTED_FEATURE;
return 0; // currently unsupported modes
case 0xffe1: // App1 (EXIF?)
if (s[iOffset+2] == 'E' && s[iOffset+3] == 'x' && (s[iOffset+8] == 'M' || s[iOffset+8] == 'I')) // the EXIF data we want
{
int bMotorola, IFD, iTagCount;
pPage->iEXIF = iFilePos - iBytesRead + iOffset + 8; // start of TIFF file
// Get the orientation value (if present)
bMotorola = (s[iOffset+8] == 'M');
IFD = TIFFLONG(&s[iOffset+12], bMotorola);
iTagCount = TIFFSHORT(&s[iOffset+16], bMotorola);
GetTIFFInfo(pPage, bMotorola, IFD+iOffset+8);
// The second IFD defines the thumbnail (if present)
if (iTagCount >= 1 && iTagCount < 32) // valid number of tags for EXIF data 'page'
{
// point to next IFD
IFD += (12 * iTagCount) + 2;
IFD = TIFFLONG(&s[IFD + iOffset + 8], bMotorola);
if (IFD != 0) // Thumbnail present?
{
pPage->ucHasThumb = 1;
GetTIFFInfo(pPage, bMotorola, IFD+iOffset+8); // info for second 'page' of TIFF
pPage->iThumbData += iOffset + 8; // absolute offset in the file
}
}
}
break;
case 0xffc0: // SOFx - start of frame
pPage->ucMode = (uint8_t)usMarker;
pPage->ucBpp = s[iOffset+2]; // bits per sample
pPage->iCropX = pPage->iCropY = 0; // initialize crop rectangle to full image size
pPage->iCropCY = pPage->iHeight = MOTOSHORT(&s[iOffset+3]);
pPage->iCropCX = pPage->iWidth = MOTOSHORT(&s[iOffset+5]);
pPage->ucNumComponents = s[iOffset+7];
pPage->ucBpp = pPage->ucBpp * pPage->ucNumComponents; /* Bpp = number of components * bits per sample */
if (pPage->ucNumComponents == 1)
pPage->ucSubSample = 0; // use this to differentiate from color 1:1
else
{
usLen -= 8;
iOffset += 8;
// pPage->ucSubSample = s[iOffset+9]; // subsampling option for the second color component
for (i=0; i<pPage->ucNumComponents; i++)
{
uint8_t ucSamp;
pPage->JPCI[i].component_id = s[iOffset++];
pPage->JPCI[i].component_index = (unsigned char)i;
ucSamp = s[iOffset++]; // get the h+v sampling factor
if (i == 0) // Y component?
pPage->ucSubSample = ucSamp;
// pPage->JPCI[i].h_samp_factor = ucSamp >> 4;
// pPage->JPCI[i].v_samp_factor = ucSamp & 0xf;
pPage->JPCI[i].quant_tbl_no = s[iOffset++]; // quantization table number
if (pPage->JPCI[i].quant_tbl_no > 3)
{
pPage->iError = JPEG_DECODE_ERROR;
return 0; // error
}
usLen -= 3;
}
}
break;
case 0xffdd: // Restart Interval
if (usLen == 4)
pPage->iResInterval = MOTOSHORT(&s[iOffset+2]);
break;
case 0xffc4: /* M_DHT */ // get Huffman tables
iOffset += 2; // skip length
usLen -= 2; // subtract length length
if (JPEGGetHuffTables(&s[iOffset], usLen, pPage) != 0) // bad tables?
{
pPage->iError = JPEG_DECODE_ERROR;
return 0; // error
}
break;
case 0xffdb: /* M_DQT */
/* Get the quantization tables */
/* first byte has PPPPNNNN where P = precision and N = table number 0-3 */
iOffset += 2; // skip length
usLen -= 2; // subtract length length
while (usLen > 0)
{
ucTable = s[iOffset++]; // table number
if ((ucTable & 0xf) > 3) // invalid table number
{
pPage->iError = JPEG_DECODE_ERROR;
return 0;
}
iTableOffset = (ucTable & 0xf) * DCTSIZE;
if (ucTable & 0xf0) // if word precision
{
for (i=0; i<DCTSIZE; i++)
{
pPage->sQuantTable[i+iTableOffset] = MOTOSHORT(&s[iOffset]);
iOffset += 2;
}
usLen -= (DCTSIZE*2 + 1);
}
else // byte precision
{
for (i=0; i<DCTSIZE; i++)
{
pPage->sQuantTable[i+iTableOffset] = (unsigned short)s[iOffset++];
}
usLen -= (DCTSIZE + 1);
}
}
break;
} // switch on JPEG marker
iOffset += usLen;
} // while
if (usMarker == 0xffda) // start of image
{
if (pPage->ucBpp != 8) // need to match up table IDs
{
iOffset -= usLen;
JPEGGetSOS(pPage, &iOffset); // get Start-Of-Scan info for decoding
}
if (!JPEGMakeHuffTables(pPage, 0)) //int bThumbnail) DEBUG
{
pPage->iError = JPEG_UNSUPPORTED_FEATURE;
return 0;
}
// Now the offset points to the start of compressed data
i = JPEGFilter(&pPage->ucFileBuf[iOffset], pPage->ucFileBuf, iBytesRead-iOffset, &pPage->ucFF);
pPage->iVLCOff = 0;
pPage->iVLCSize = i;
JPEGGetMoreData(pPage); // read more VLC data
return 1;
}
pPage->iError = JPEG_DECODE_ERROR;
return 0;
} /* JPEGParseInfo() */
//
// Fix and reorder the quantization table for faster decoding.*
//
static void JPEGFixQuantD(JPEGIMAGE *pJPEG)
{
int iTable, iTableOffset;
signed short sTemp[DCTSIZE];
int i;
uint16_t *p;
for (iTable=0; iTable<pJPEG->ucNumComponents; iTable++)
{
iTableOffset = iTable * DCTSIZE;
p = (uint16_t *)&pJPEG->sQuantTable[iTableOffset];
for (i=0; i<DCTSIZE; i++)
sTemp[i] = p[cZigZag[i]];
memcpy(&pJPEG->sQuantTable[iTableOffset], sTemp, DCTSIZE*sizeof(short)); // copy back to original spot
// Prescale for DCT multiplication
p = (uint16_t *)&pJPEG->sQuantTable[iTableOffset];
for (i=0; i<DCTSIZE; i++)
{
p[i] = (uint16_t)((p[i] * iScaleBits[i]) >> 12);
}
}
} /* JPEGFixQuantD() */
//
// Decode the DC and 2-63 AC coefficients of the current DCT block
// For 1/4 and 1/8 scaled images, we don't store most of the AC values since we
// won't use them. For skipped MCUs (outside crop area), we don't decode any AC values
//
static int JPEGDecodeMCU(JPEGIMAGE *pJPEG, int iMCU, int *iDCPredictor)
{
my_ulong ulCode, ulTemp;
uint8_t *pZig;
signed char cCoeff;
unsigned short *pFast;
unsigned char ucHuff, *pucFast;
uint32_t usHuff; // this prevents an unnecessary & 65535 for shorts
uint32_t ulBitOff;
my_ulong ulBits; // local copies to allow compiler to use register vars
uint8_t *pBuf, *pEnd, *pEnd2;
signed short *pMCU = &pJPEG->sMCUs[iMCU];
uint16_t u16MCUFlags;
#define MIN_DCT_THRESHOLD 8
ulBitOff = pJPEG->bb.ulBitOff;
ulBits = pJPEG->bb.ulBits;
pBuf = pJPEG->bb.pBuf;
if (ulBitOff > (REGISTER_WIDTH-17)) { // need to get more data
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
if (iMCU < 0) { // skip this block (cropped, or grayscale output from color)
pEnd2 = (uint8_t *)&cZigZag2[1]; // we only capture the DC value
} else if (pJPEG->iOptions & (JPEG_SCALE_QUARTER | JPEG_SCALE_EIGHTH)) { // reduced size DCT
pMCU[1] = pMCU[8] = pMCU[9] = 0;
pEnd2 = (uint8_t *)&cZigZag2[5]; // we only need to store the 4 elements we care about
} else { // decode all the AC coefficients
memset(pMCU, 0, 64*sizeof(short)); // pre-fill with zero since we may skip coefficients
pEnd2 = (uint8_t *)&cZigZag2[64];
}
u16MCUFlags = 0;
pZig = (unsigned char *)&cZigZag2[1];
pEnd = (unsigned char *)&cZigZag2[64];
// get the DC component
pucFast = &pJPEG->ucHuffDC[pJPEG->ucDCTable * DC_TABLE_SIZE];
ulCode = (ulBits >> (REGISTER_WIDTH - 12 - ulBitOff)) & 0xfff; // get as lower 12 bits
if (ulCode >= 0xf80) // it's a long code
ulCode = (ulCode & 0xff); // point to long table and trim to 7-bits + 0x80 offset into long table
else
ulCode >>= 6; // it's a short code, use first 6 bits only
ucHuff = pucFast[ulCode];
cCoeff = (signed char)pucFast[ulCode+512]; // get pre-calculated extra bits for "small" values
if (ucHuff == 0) // invalid code
return -1;
ulBitOff += (ucHuff >> 4); // add the Huffman length
ucHuff &= 0xf; // get the actual code (SSSS)
if (ucHuff) // if there is a change to the DC value
{ // get the 'extra' bits
if (cCoeff)
{
(*iDCPredictor) += cCoeff;
}
else
{
if (ulBitOff > (REGISTER_WIDTH - 17)) // need to get more data
{
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
ulCode = ulBits << ulBitOff;
ulTemp = ~(my_ulong)(((my_long)ulCode)>>(REGISTER_WIDTH-1)); // slide sign bit across other 63/31 bits
ulCode >>= (REGISTER_WIDTH - ucHuff);
ulCode -= ulTemp>>(REGISTER_WIDTH-ucHuff);
ulBitOff += ucHuff; // add bit length
(*iDCPredictor) += (int)ulCode;
}
}
if (iMCU >= 0) { // non-skipped block
pMCU[0] = (short)*iDCPredictor; // store in MCU[0]
}
if (pJPEG->ucACTable > 1) // unsupported
return -1;
// Now get the other 63 AC coefficients
pFast = &pJPEG->usHuffAC[pJPEG->ucACTable * HUFF11SIZE];
if (pJPEG->b11Bit) // 11-bit "slow" tables used
{
// if (pJPEG->pHuffACFast == pJPEG->huffacFast[1]) // second table
// pFast = &pJPEG->ucAltHuff[0];
while (pZig < pEnd)
{
if (ulBitOff >(REGISTER_WIDTH - 17)) // need to get more data
{
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
ulCode = (ulBits >> (REGISTER_WIDTH - 16 - ulBitOff)) & 0xffff; // get as lower 16 bits
if (ulCode >= 0xf000) // first 4 bits = 1, use long table
ulCode = (ulCode & 0x1fff);
else
ulCode >>= 4; // use lower 12 bits (short table)
usHuff = pFast[ulCode];
if (usHuff == 0) // invalid code
return -1;
ulBitOff += (usHuff >> 8); // add length
usHuff &= 0xff; // get code (RRRR/SSSS)
if (usHuff == 0) // no more AC components
{
goto mcu_done;
}
pZig += (usHuff >> 4); // get the skip amount (RRRR)
usHuff &= 0xf; // get (SSSS) - extra length
if (pZig < pEnd2 && usHuff)
{
ulCode = ulBits << ulBitOff;
ulTemp = ~(my_ulong) (((my_long) ulCode) >> (REGISTER_WIDTH-1)); // slide sign bit across other 63 bits
ulCode >>= (REGISTER_WIDTH - usHuff);
ulCode -= ulTemp >> (REGISTER_WIDTH - usHuff);
u16MCUFlags |= 1<<(*pZig & 7); // keep track of occupied columns
u16MCUFlags |= *pZig << 8; // for testing occupied rows
pMCU[*pZig] = (signed short)ulCode; // store AC coefficient (already reordered)
}
ulBitOff += usHuff; // add (SSSS) extra length
pZig++;
if (ulBitOff > (REGISTER_WIDTH - 17)) // need to get more data
{
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
} // while
}
else // 10-bit "fast" tables used
{
while (pZig < pEnd)
{
if (ulBitOff >(REGISTER_WIDTH - 17)) // need to get more data
{
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
ulCode = (ulBits >> (REGISTER_WIDTH - 16 - ulBitOff)) & 0xffff; // get as lower 16 bits
if (ulCode >= 0xfc00) // first 6 bits = 1, use long table
ulCode = (ulCode & 0x7ff); // (ulCode & 0x3ff) + 0x400;
else
ulCode >>= 6; // use lower 10 bits (short table)
usHuff = pFast[ulCode];
if (usHuff == 0) // invalid code
return -1;
ulBitOff += (usHuff >> 8); // add length
usHuff &= 0xff; // get code (RRRR/SSSS)
if (usHuff == 0) // no more AC components
{
goto mcu_done;
}
pZig += (usHuff >> 4); // get the skip amount (RRRR)
usHuff &= 0xf; // get (SSSS) - extra length
if (pZig < pEnd2 && usHuff)
{
ulCode = ulBits << ulBitOff;
ulTemp = ~(my_ulong) (((my_long) ulCode) >> (REGISTER_WIDTH-1)); // slide sign bit across other 63 bits
ulCode >>= (REGISTER_WIDTH - usHuff);
ulCode -= ulTemp >> (REGISTER_WIDTH - usHuff);
u16MCUFlags |= 1<<(*pZig & 7); // keep track of occupied columns
u16MCUFlags |= *pZig << 8; // for testing occupied rows
pMCU[*pZig] = (signed short)ulCode; // store AC coefficient (already reordered)
}
ulBitOff += usHuff; // add (SSSS) extra length
pZig++;
if (ulBitOff >(REGISTER_WIDTH - 17)) // need to get more data
{
pBuf += (ulBitOff >> 3);
ulBitOff &= 7;
ulBits = MOTOLONG(pBuf);
}
} // while
} // 10-bit tables
mcu_done:
pJPEG->bb.pBuf = pBuf;
pJPEG->iVLCOff = (int)(pBuf - pJPEG->ucFileBuf);
pJPEG->bb.ulBitOff = ulBitOff;
pJPEG->bb.ulBits = ulBits;
pJPEG->u16MCUFlags = u16MCUFlags;
return 0;
} /* JPEGDecodeMCU() */
//
// Inverse DCT
//
static void JPEGIDCT(JPEGIMAGE *pJPEG, int iMCUOffset, int iQuantTable)
{
int iRow;
unsigned char ucColMask;
int iCol;
signed int tmp6,tmp7,tmp10,tmp11,tmp12,tmp13;
signed int z5,z10,z11,z12,z13;
signed int tmp0,tmp1,tmp2,tmp3,tmp4,tmp5;
signed short *pQuant;
unsigned char *pOutput;
uint16_t u16MCUFlags;
int16_t *pMCUSrc = &pJPEG->sMCUs[iMCUOffset];
#ifdef HAS_SSE
__m128i mmxRow0, mmxRow1, mmxRow2, mmxRow3, mmxRow4, mmxRow5, mmxRow6, mmxRow7;
__m128i mmxTemp, mmxTemp0, mmxTemp1, mmxTemp2, mmxTemp3, mmxTemp4, mmxTemp5, mmxTemp6, mmxTemp7, mmxTemp10, mmxTemp11, mmxTemp12, mmxTemp13;
__m128i mmxZ5, mmxZ10, mmxZ11, mmxZ12, mmxZ13;
#endif // HAS_SSE
#ifdef HAS_NEON
int16x8_t mmxRow0, mmxRow1, mmxRow2, mmxRow3, mmxRow4, mmxRow5, mmxRow6, mmxRow7;
int16x8_t mmxTemp, mmxTemp0, mmxTemp1, mmxTemp2, mmxTemp3, mmxTemp4, mmxTemp5, mmxTemp6, mmxTemp7, mmxTemp10, mmxTemp11, mmxTemp12, mmxTemp13;
int16x8_t mmxZ5, mmxZ10, mmxZ11, mmxZ12, mmxZ13;
#endif // HAS_NEON
u16MCUFlags = pJPEG->u16MCUFlags;
// my shortcut method appears to violate patent 20020080052
// but the patent is invalidated by prior art:
// http://netilium.org/~mad/dtj/DTJ/DTJK04/
pQuant = &pJPEG->sQuantTable[iQuantTable * DCTSIZE];
if (pJPEG->iOptions & JPEG_SCALE_QUARTER) // special case
{
/* Column 0 */
tmp4 = pMCUSrc[0] * pQuant[0];
tmp5 = pMCUSrc[8] * pQuant[8];
tmp0 = tmp4 + tmp5;
tmp2 = tmp4 - tmp5;
/* Column 1 */
tmp4 = pMCUSrc[1] * pQuant[1];
tmp5 = pMCUSrc[9] * pQuant[9];
tmp1 = tmp4 + tmp5;
tmp3 = tmp4 - tmp5;
/* Pass 2: process 2 rows, store into output array. */
/* Row 0 */
pOutput = (unsigned char *)pMCUSrc; // store output pixels back into MCU
pOutput[0] = ucRangeTable[(((tmp0 + tmp1)>>5) & 0x3ff)];
pOutput[1] = ucRangeTable[(((tmp0 - tmp1)>>5) & 0x3ff)];
/* Row 1 */
pOutput[2] = ucRangeTable[(((tmp2 + tmp3)>>5) & 0x3ff)];
pOutput[3] = ucRangeTable[(((tmp2 - tmp3)>>5) & 0x3ff)];
return;
}
#ifdef HAS_SSE // SSE2 version
// Columns first
// even part
if ((u16MCUFlags & 0x2000) == 0) // rows 4-7 are not populated, simpler calculations
{
// even part
mmxTemp10 = _mm_loadu_si128((__m128i *)&pMCUSrc[0]); // row 0
mmxTemp1 = _mm_loadu_si128((__m128i *)&pMCUSrc[16]); // row 2
mmxTemp = _mm_loadu_si128((__m128i *)&pQuant[0]);
mmxTemp2 = _mm_loadu_si128((__m128i *)&pQuant[16]);
mmxTemp10 = _mm_mullo_epi16(mmxTemp10, mmxTemp); // dequant row 0
mmxTemp1 = _mm_mullo_epi16(mmxTemp1, mmxTemp2); // dequant row 2
mmxTemp = _mm_loadu_si128((__m128i *)&s0414[0]); // 0.414
mmxTemp12 = _mm_mulhi_epi16(_mm_slli_epi16(mmxTemp1, 2), mmxTemp); // tmp12 = ((tmp1*106)>>8)
mmxTemp0 = _mm_add_epi16(mmxTemp10, mmxTemp1); // 0+2
mmxTemp3 = _mm_sub_epi16(mmxTemp10, mmxTemp1); // 0-2
mmxTemp1 = _mm_add_epi16(mmxTemp10, mmxTemp12); // 10+12
mmxTemp2 = _mm_sub_epi16(mmxTemp10, mmxTemp12); // 10-12
// odd part
mmxTemp4 = _mm_loadu_si128((__m128i *)&pMCUSrc[8]); // row 1
mmxTemp5 = _mm_loadu_si128((__m128i *)&pMCUSrc[24]); // row 3
mmxTemp = _mm_loadu_si128((__m128i *)&pQuant[8]);
mmxTemp11 = _mm_loadu_si128((__m128i *)&pQuant[24]);
mmxTemp4 = _mm_mullo_epi16(mmxTemp4, mmxTemp); // dequant row 1
mmxTemp5 = _mm_mullo_epi16(mmxTemp5, mmxTemp11); // dequant row 3
mmxTemp7 = _mm_add_epi16(mmxTemp4, mmxTemp5); // tmp7 = tmp4 + tmp5
mmxTemp = _mm_loadu_si128((__m128i *)&s1414[0]); // load 1.414213562 constant
mmxTemp11 = _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(mmxTemp4, mmxTemp5), 2), mmxTemp); // tmp11 = (((tmp4-tmp5)*362)>>8)
mmxTemp = _mm_loadu_si128((__m128i *)&s1847[0]); // 1.8477
mmxZ5 = _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(mmxTemp4, mmxTemp5), 2), mmxTemp); // z5 = (((tmp4-tmp5)*473)>>8)
mmxTemp = _mm_loadu_si128((__m128i*)&sp2613[0]); // positive 2.6131259
mmxTemp12 = _mm_mulhi_epi16(_mm_slli_epi16(mmxTemp5, 2), mmxTemp); // tmp12 = ((-tmp5 * -669)>>8) + z5
// can't make that constant without overflowing, so double it after
mmxTemp12 = _mm_add_epi16(mmxTemp12, mmxTemp12);
mmxTemp12 = _mm_add_epi16(mmxTemp12, mmxZ5);
mmxTemp6 = _mm_sub_epi16(mmxTemp12, mmxTemp7); // tmp6 = tmp12 - tmp7
mmxTemp5 = _mm_sub_epi16(mmxTemp11, mmxTemp6); // tmp5 = tmp11 - tmp6
mmxTemp = _mm_loadu_si128((__m128i *)&s1082[0]); // 1.08239
mmxTemp10 = _mm_sub_epi16(_mm_mulhi_epi16(_mm_slli_epi16(mmxTemp4, 2), mmxTemp), mmxZ5); // tmp10 = ((tmp4 * 277)>>8) - z5
mmxTemp4 = _mm_add_epi16(mmxTemp10, mmxTemp5); // tmp4 = tmp10 + tmp5
}
else // need to do full calculation
{
// even part
mmxTemp0 = _mm_loadu_si128((__m128i *)&pMCUSrc[0]); // get row 0
mmxTemp2 = _mm_loadu_si128((__m128i *)&pMCUSrc[32]); // get row 4
mmxTemp10 = _mm_loadu_si128((__m128i *)&pQuant[0]);
mmxTemp11 = _mm_loadu_si128((__m128i *)&pQuant[32]);
mmxTemp0 = _mm_mullo_epi16(mmxTemp0, mmxTemp10); // dequant row 0
mmxTemp2 = _mm_mullo_epi16(mmxTemp2, mmxTemp11); // dequant row 4
mmxTemp10 = _mm_add_epi16(mmxTemp0, mmxTemp2); // 0+4
mmxTemp11 = _mm_sub_epi16(mmxTemp0, mmxTemp2); // 0-4
mmxTemp1 = _mm_loadu_si128((__m128i *)&pMCUSrc[16]); // get row 2
mmxTemp3 = _mm_loadu_si128((__m128i *)&pMCUSrc[48]); // get row 6
mmxTemp = _mm_loadu_si128((__m128i *)&pQuant[16]);
mmxTemp12 = _mm_loadu_si128((__m128i *)&pQuant[48]);
mmxTemp1 = _mm_mullo_epi16(mmxTemp1, mmxTemp); // dequant row 2
mmxTemp3 = _mm_mullo_epi16(mmxTemp3, mmxTemp12); // dequant row 6
mmxTemp13 = _mm_add_epi16(mmxTemp1, mmxTemp3); // 1+3
mmxTemp = _mm_loadu_si128((__m128i *)&s1414[0]); // load 1.414213562 constant
mmxTemp12 = _mm_sub_epi16(_mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(mmxTemp1,mmxTemp3),2), mmxTemp), mmxTemp13); // tmp12 = (((tmp1 - tmp3) * 1.414) - tmp13;
mmxTemp0 = _mm_add_epi16(mmxTemp10, mmxTemp13); // tmp0 = tmp10 + tmp13
mmxTemp3 = _mm_sub_epi16(mmxTemp10, mmxTemp13); // tmp3 = tmp10 - tmp13
mmxTemp1 = _mm_add_epi16(mmxTemp11, mmxTemp12); // tmp1 = tmp11 + tmp12
mmxTemp2 = _mm_sub_epi16(mmxTemp11, mmxTemp12); // tmp2 = tmp11 - tmp12
// odd part
mmxTemp5 = _mm_loadu_si128((__m128i *)&pMCUSrc[24]); // get row 3
mmxTemp6 = _mm_loadu_si128((__m128i *)&pMCUSrc[40]); // get row 5
mmxTemp10 = _mm_loadu_si128((__m128i *)&pQuant[24]);
mmxTemp11 = _mm_loadu_si128((__m128i *)&pQuant[40]);
mmxTemp5 = _mm_mullo_epi16(mmxTemp5, mmxTemp10); // dequant row 3
mmxTemp6 = _mm_mullo_epi16(mmxTemp6, mmxTemp11); // dequant row 5
mmxZ13 = _mm_add_epi16(mmxTemp6, mmxTemp5); // z13 = tmp6 + tmp5;
mmxZ10 = _mm_sub_epi16(mmxTemp6, mmxTemp5); // z10 = tmp6 - tmp5;
mmxTemp4 = _mm_loadu_si128((__m128i *)&pMCUSrc[8]); // get row 1
mmxTemp7 = _mm_loadu_si128((__m128i *)&pMCUSrc[56]); // get row 7
mmxTemp10 = _mm_loadu_si128((__m128i *)&pQuant[8]);
mmxTemp11 = _mm_loadu_si128((__m128i *)&pQuant[56]);
mmxTemp4 = _mm_mullo_epi16(mmxTemp4, mmxTemp10); // dequant row 1
mmxTemp7 = _mm_mullo_epi16(mmxTemp7, mmxTemp11); // dequant row 7
mmxZ11 = _mm_add_epi16(mmxTemp4, mmxTemp7); // z11 = tmp4 + tmp7;
mmxZ12 = _mm_sub_epi16(mmxTemp4, mmxTemp7); // z12 = tmp4 - tmp7;
mmxTemp7 = _mm_add_epi16(mmxZ11, mmxZ13); // tmp7 = z11 + z13;
mmxTemp11 = _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(mmxZ11, mmxZ13),2), mmxTemp); // tmp11 = ((z11 - z13) * 1.1414);
mmxTemp = _mm_loadu_si128((__m128i *)&s1847[0]); // 1.8477
mmxZ5 = _mm_mulhi_epi16(_mm_slli_epi16(_mm_add_epi16(mmxZ10, mmxZ12),2), mmxTemp); // z5 = ((z10+z12)*1.8477);
mmxTemp = _mm_loadu_si128((__m128i *)&s2613[0]); // -2.6131259
mmxTemp12 = _mm_mulhi_epi16(_mm_slli_epi16(mmxZ10,2), mmxTemp); // tmp12 = (z10 * -2.6131259) + z5;
// can't make that constant without overflowing, so double it after
mmxTemp12 = _mm_add_epi16(mmxTemp12, mmxTemp12);
mmxTemp12 = _mm_add_epi16(mmxTemp12, mmxZ5);
mmxTemp = _mm_loadu_si128((__m128i *)&s1082[0]); // 1.08239
mmxTemp6 = _mm_sub_epi16(mmxTemp12, mmxTemp7); // tmp6 = tmp12 - tmp7
mmxTemp5 = _mm_sub_epi16(mmxTemp11, mmxTemp6); // tmp5 = tmp11 - tmp6
mmxTemp10 = _mm_sub_epi16(_mm_mulhi_epi16(_mm_slli_epi16(mmxZ12,2), mmxTemp), mmxZ5); // tmp10 = (z12 * 1.08239) - z5;
mmxTemp4 = _mm_add_epi16(mmxTemp10, mmxTemp5); // tmp4 = tmp10 + tmp5;
}
mmxRow0 = _mm_add_epi16(mmxTemp0, mmxTemp7); // row 0
_mm_storeu_si128((__m128i *)&pMCUSrc[0], mmxRow0);
mmxRow1 = _mm_add_epi16(mmxTemp1, mmxTemp6); // row 1
_mm_storeu_si128((__m128i *)&pMCUSrc[8], mmxRow1);
mmxRow2 = _mm_add_epi16(mmxTemp2, mmxTemp5); // row 2
_mm_storeu_si128((__m128i *)&pMCUSrc[16], mmxRow2);
mmxRow3 = _mm_sub_epi16(mmxTemp3, mmxTemp4); // row 3
_mm_storeu_si128((__m128i *)&pMCUSrc[24], mmxRow3);
mmxRow4 = _mm_add_epi16(mmxTemp3, mmxTemp4); // row 4
_mm_storeu_si128((__m128i *)&pMCUSrc[32], mmxRow4);
mmxRow5 = _mm_sub_epi16(mmxTemp2, mmxTemp5); // row 5
_mm_storeu_si128((__m128i *)&pMCUSrc[40], mmxRow5);
mmxRow6 = _mm_sub_epi16(mmxTemp1, mmxTemp6); // row 6
_mm_storeu_si128((__m128i *)&pMCUSrc[48], mmxRow6);
mmxRow7 = _mm_sub_epi16(mmxTemp0, mmxTemp7); // row 7
_mm_storeu_si128((__m128i *)&pMCUSrc[56], mmxRow7);
#endif // HAS_SSE
#ifdef HAS_NEON
if ((u16MCUFlags & 0x2000) == 0) // rows 4-7 are not populated, simpler calculations
{
// even part
mmxTemp10 = vld1q_s16(&pMCUSrc[0]); // row 0
mmxTemp1 = vld1q_s16(&pMCUSrc[16]); // row 2
mmxTemp = vld1q_s16(&pQuant[0]);
mmxTemp2 = vld1q_s16(&pQuant[16]);
mmxTemp10 = vmulq_s16(mmxTemp10, mmxTemp); // dequant row 0
mmxTemp1 = vmulq_s16(mmxTemp1, mmxTemp2); // dequant row 2
mmxTemp = vld1q_s16(&s0414[0]); // 0.414
mmxTemp12 = vqdmulhq_s16(vshlq_n_s16(mmxTemp1, 2), mmxTemp); // tmp12 = ((tmp1*106)>>8)
mmxTemp0 = vaddq_s16(mmxTemp10, mmxTemp1); // 0+2
mmxTemp3 = vsubq_s16(mmxTemp10, mmxTemp1); // 0-2
mmxTemp1 = vaddq_s16(mmxTemp10, mmxTemp12); // 10+12
mmxTemp2 = vsubq_s16(mmxTemp10, mmxTemp12); // 10-12
// odd part
mmxTemp4 = vld1q_s16(&pMCUSrc[8]); // row 1
mmxTemp5 = vld1q_s16(&pMCUSrc[24]); // row 3
mmxTemp = vld1q_s16(&pQuant[8]);
mmxTemp11 = vld1q_s16(&pQuant[24]);
mmxTemp4 = vmulq_s16(mmxTemp4, mmxTemp); // dequant row 1
mmxTemp5 = vmulq_s16(mmxTemp5, mmxTemp11); // dequant row 3
mmxTemp7 = vaddq_s16(mmxTemp4, mmxTemp5); // tmp7 = tmp4 + tmp5
mmxTemp = vld1q_s16(&s1414[0]); // load 1.414213562 constant
mmxTemp11 = vqdmulhq_s16(vshlq_n_s16(vsubq_s16(mmxTemp4, mmxTemp5), 2), mmxTemp); // tmp11 = (((tmp4-tmp5)*362)>>8)
mmxTemp = vld1q_s16(&s1847[0]); // 1.8477
mmxZ5 = vqdmulhq_s16(vshlq_n_s16(vsubq_s16(mmxTemp4, mmxTemp5), 2), mmxTemp); // z5 = (((tmp4-tmp5)*473)>>8)
mmxTemp = vld1q_s16(&sp2613[0]); // positive 2.6131259
mmxTemp12 = vqdmulhq_s16(vshlq_n_s16(mmxTemp5, 2), mmxTemp); // tmp12 = ((-tmp5 * -669)>>8) + z5
// can't make that constant without overflowing, so double it after
mmxTemp12 = vaddq_s16(mmxTemp12, mmxTemp12);
mmxTemp12 = vaddq_s16(mmxTemp12, mmxZ5);
mmxTemp6 = vsubq_s16(mmxTemp12, mmxTemp7); // tmp6 = tmp12 - tmp7
mmxTemp5 = vsubq_s16(mmxTemp11, mmxTemp6); // tmp5 = tmp11 - tmp6
mmxTemp = vld1q_s16(&s1082[0]); // 1.08239
mmxTemp10 = vsubq_s16(vqdmulhq_s16(vshlq_n_s16(mmxTemp4, 2), mmxTemp), mmxZ5); // tmp10 = ((tmp4 * 277)>>8) - z5
mmxTemp4 = vaddq_s16(mmxTemp10, mmxTemp5); // tmp4 = tmp10 + tmp5
}
else // need to do full calculation
{
// even part
mmxTemp0 = vld1q_s16(&pMCUSrc[0]); // get row 0
mmxTemp2 = vld1q_s16(&pMCUSrc[32]); // get row 4
mmxTemp10 = vld1q_s16(&pQuant[0]);
mmxTemp11 = vld1q_s16(&pQuant[32]);
mmxTemp0 = vmulq_s16(mmxTemp0, mmxTemp10); // dequant row 0
mmxTemp2 = vmulq_s16(mmxTemp2, mmxTemp11); // dequant row 4
mmxTemp10 = vaddq_s16(mmxTemp0, mmxTemp2); // 0+4
mmxTemp11 = vsubq_s16(mmxTemp0, mmxTemp2); // 0-4
mmxTemp1 = vld1q_s16(&pMCUSrc[16]); // get row 2
mmxTemp3 = vld1q_s16(&pMCUSrc[48]); // get row 6
mmxTemp = vld1q_s16(&pQuant[16]);
mmxTemp12 = vld1q_s16(&pQuant[48]);
mmxTemp1 = vmulq_s16(mmxTemp1, mmxTemp); // dequant row 2
mmxTemp3 = vmulq_s16(mmxTemp3, mmxTemp12); // dequant row 6
mmxTemp13 = vaddq_s16(mmxTemp1, mmxTemp3); // 1+3
mmxTemp = vld1q_s16(&s1414[0]); // load 1.414213562 constant
mmxTemp12 = vsubq_s16(vqdmulhq_s16(vshlq_n_s16(vsubq_s16(mmxTemp1,mmxTemp3),2), mmxTemp), mmxTemp13); // tmp12 = (((tmp1 - tmp3) * 1.414) - tmp13;
mmxTemp0 = vaddq_s16(mmxTemp10, mmxTemp13); // tmp0 = tmp10 + tmp13
mmxTemp3 = vsubq_s16(mmxTemp10, mmxTemp13); // tmp3 = tmp10 - tmp13
mmxTemp1 = vaddq_s16(mmxTemp11, mmxTemp12); // tmp1 = tmp11 + tmp12
mmxTemp2 = vsubq_s16(mmxTemp11, mmxTemp12); // tmp2 = tmp11 - tmp12
// odd part
mmxTemp5 = vld1q_s16(&pMCUSrc[24]); // get row 3
mmxTemp6 = vld1q_s16(&pMCUSrc[40]); // get row 5
mmxTemp10 = vld1q_s16(&pQuant[24]);
mmxTemp11 = vld1q_s16(&pQuant[40]);
mmxTemp5 = vmulq_s16(mmxTemp5, mmxTemp10); // dequant row 3
mmxTemp6 = vmulq_s16(mmxTemp6, mmxTemp11); // dequant row 5
mmxZ13 = vaddq_s16(mmxTemp6, mmxTemp5); // z13 = tmp6 + tmp5;
mmxZ10 = vsubq_s16(mmxTemp6, mmxTemp5); // z10 = tmp6 - tmp5;
mmxTemp4 = vld1q_s16(&pMCUSrc[8]); // get row 1
mmxTemp7 = vld1q_s16(&pMCUSrc[56]); // get row 7
mmxTemp10 = vld1q_s16(&pQuant[8]);
mmxTemp11 = vld1q_s16(&pQuant[56]);
mmxTemp4 = vmulq_s16(mmxTemp4, mmxTemp10); // dequant row 1
mmxTemp7 = vmulq_s16(mmxTemp7, mmxTemp11); // dequant row 7
mmxZ11 = vaddq_s16(mmxTemp4, mmxTemp7); // z11 = tmp4 + tmp7;
mmxZ12 = vsubq_s16(mmxTemp4, mmxTemp7); // z12 = tmp4 - tmp7;
mmxTemp7 = vaddq_s16(mmxZ11, mmxZ13); // tmp7 = z11 + z13;
mmxTemp11 = vqdmulhq_s16(vshlq_n_s16(vsubq_s16(mmxZ11, mmxZ13),2), mmxTemp); // tmp11 = ((z11 - z13) * 1.1414);
mmxTemp = vld1q_s16(&s1847[0]); // 1.8477
mmxZ5 = vqdmulhq_s16(vshlq_n_s16(vaddq_s16(mmxZ10, mmxZ12),2), mmxTemp); // z5 = ((z10+z12)*1.8477);
mmxTemp = vld1q_s16(&s2613[0]); // -2.6131259
mmxTemp12 = vqdmulhq_s16(vshlq_n_s16(mmxZ10,2), mmxTemp); // tmp12 = (z10 * -2.6131259) + z5;
// can't make that constant without overflowing, so double it after
mmxTemp12 = vaddq_s16(mmxTemp12, mmxTemp12);
mmxTemp12 = vaddq_s16(mmxTemp12, mmxZ5);
mmxTemp = vld1q_s16(&s1082[0]); // 1.08239
mmxTemp6 = vsubq_s16(mmxTemp12, mmxTemp7); // tmp6 = tmp12 - tmp7
mmxTemp5 = vsubq_s16(mmxTemp11, mmxTemp6); // tmp5 = tmp11 - tmp6
mmxTemp10 = vsubq_s16(vqdmulhq_s16(vshlq_n_s16(mmxZ12,2), mmxTemp), mmxZ5); // tmp10 = (z12 * 1.08239) - z5;
mmxTemp4 = vaddq_s16(mmxTemp10, mmxTemp5); // tmp4 = tmp10 + tmp5;
}
mmxRow0 = vaddq_s16(mmxTemp0, mmxTemp7); // row 0
vst1q_s16(&pMCUSrc[0], mmxRow0);
mmxRow1 = vaddq_s16(mmxTemp1, mmxTemp6); // row 1
vst1q_s16(&pMCUSrc[8], mmxRow1);
mmxRow2 = vaddq_s16(mmxTemp2, mmxTemp5); // row 2
vst1q_s16(&pMCUSrc[16], mmxRow2);
mmxRow3 = vsubq_s16(mmxTemp3, mmxTemp4); // row 3
vst1q_s16(&pMCUSrc[24], mmxRow3);
mmxRow4 = vaddq_s16(mmxTemp3, mmxTemp4); // row 4
vst1q_s16(&pMCUSrc[32], mmxRow4);
mmxRow5 = vsubq_s16(mmxTemp2, mmxTemp5); // row 5
vst1q_s16(&pMCUSrc[40], mmxRow5);
mmxRow6 = vsubq_s16(mmxTemp1, mmxTemp6); // row 6
vst1q_s16(&pMCUSrc[48], mmxRow6);
mmxRow7 = vsubq_s16(mmxTemp0, mmxTemp7); // row 7
vst1q_s16(&pMCUSrc[56], mmxRow7);
#endif // HAS_NEON
#if !defined (HAS_SSE) && !defined(HAS_NEON)
// do columns first
u16MCUFlags |= 1; // column 0 must always be calculated
for (iCol = 0; iCol < 8 && u16MCUFlags; iCol++)
{
if (u16MCUFlags & (1<<iCol)) // column has data in it
{
u16MCUFlags &= ~(1<<iCol); // unmark the col after done
if ((u16MCUFlags & 0x2000) == 0) // simpler calculations if only half populated
{
// even part
tmp10 = pMCUSrc[iCol] * pQuant[iCol];
tmp1 = pMCUSrc[iCol+16] * pQuant[iCol+16]; // get 2nd row
tmp12 = ((tmp1*106)>>8); // used to be 362 - 1 (256)
tmp0 = tmp10 + tmp1;
tmp3 = tmp10 - tmp1;
tmp1 = tmp10 + tmp12;
tmp2 = tmp10 - tmp12;
// odd part
tmp4 = pMCUSrc[iCol+8] * pQuant[iCol+8]; // get 1st row
tmp5 = pMCUSrc[iCol+24];
if (tmp5) // this value is usually 0
{
tmp5 *= pQuant[iCol+24]; // get 3rd row
tmp7 = tmp4 + tmp5;
tmp11 = (((tmp4 - tmp5) * 362) >> 8); // 362>>8 = 1.414213562
z5 = (((tmp4-tmp5) * 473) >> 8); // 473>>8 = 1.8477
tmp12 = ((-tmp5 * -669)>>8) + z5; // -669>>8 = -2.6131259
tmp6 = tmp12 - tmp7;
tmp5 = tmp11 - tmp6;
tmp10 = ((tmp4 * 277)>>8) - z5; // 277>>8 = 1.08239
tmp4 = tmp10 + tmp5;
}
else // simpler case when we only have 1 odd row to calculate
{
tmp7 = tmp4;
tmp5 = (145*tmp4) >> 8;
tmp6 = (217*tmp4) >> 8;
tmp4 = (-51*tmp4) >> 8;
}
pMCUSrc[iCol] = (short)(tmp0 + tmp7); // row0
pMCUSrc[iCol+8] = (short)(tmp1 + tmp6); // row 1
pMCUSrc[iCol+16] = (short)(tmp2 + tmp5); // row 2
pMCUSrc[iCol+24] = (short)(tmp3 - tmp4); // row 3
pMCUSrc[iCol+32] = (short)(tmp3 + tmp4); // row 4
pMCUSrc[iCol+40] = (short)(tmp2 - tmp5); // row 5
pMCUSrc[iCol+48] = (short)(tmp1 - tmp6); // row 6
pMCUSrc[iCol+56] = (short)(tmp0 - tmp7); // row 7
}
else // need to do full column calculation
{
// even part
tmp0 = pMCUSrc[iCol] * pQuant[iCol];
tmp2 = pMCUSrc[iCol+32]; // get 4th row
if (tmp2) // 4th row is most likely 0
{
tmp2 = tmp2 * pQuant[iCol+32];
tmp10 = tmp0 + tmp2;
tmp11 = tmp0 - tmp2;
}
else
{
tmp10 = tmp11 = tmp0;
}
tmp1 = pMCUSrc[iCol+16] * pQuant[iCol+16]; // get 2nd row
tmp3 = pMCUSrc[iCol+48]; // get 6th row
if (tmp3) // 6th row is most likely 0
{
tmp3 = tmp3 * pQuant[iCol+48];
tmp13 = tmp1 + tmp3;
tmp12 = (((tmp1 - tmp3) * 362) >> 8) - tmp13; // 362>>8 = 1.414213562
}
else
{
tmp13 = tmp1;
tmp12 = ((tmp1*362)>>8) - tmp1;
}
tmp0 = tmp10 + tmp13;
tmp3 = tmp10 - tmp13;
tmp1 = tmp11 + tmp12;
tmp2 = tmp11 - tmp12;
// odd part
tmp5 = pMCUSrc[iCol+24] * pQuant[iCol+24]; // get 3rd row
tmp6 = pMCUSrc[iCol+40]; // get 5th row
if (tmp6) // very likely that row 5 = 0
{
tmp6 = tmp6 * pQuant[iCol+40];
z13 = tmp6 + tmp5;
z10 = tmp6 - tmp5;
}
else
{
z13 = tmp5;
z10 = -tmp5;
}
tmp4 = pMCUSrc[iCol+8] * pQuant[iCol+8]; // get 1st row
tmp7 = pMCUSrc[iCol+56]; // get 7th row
if (tmp7) // very likely that row 7 = 0
{
tmp7 = tmp7 * pQuant[iCol+56];
z11 = tmp4 + tmp7;
z12 = tmp4 - tmp7;
}
else
{
z11 = z12 = tmp4;
}
tmp7 = z11 + z13;
tmp11 = (((z11 - z13) * 362) >> 8); // 362>>8 = 1.414213562
z5 = (((z10 + z12) * 473) >> 8); // 473>>8 = 1.8477
tmp12 = ((z10 * -669)>>8) + z5; // -669>>8 = -2.6131259
tmp6 = tmp12 - tmp7;
tmp5 = tmp11 - tmp6;
tmp10 = ((z12 * 277)>>8) - z5; // 277>>8 = 1.08239
tmp4 = tmp10 + tmp5;
pMCUSrc[iCol] = (short)(tmp0 + tmp7); // row0
pMCUSrc[iCol+8] = (short)(tmp1 + tmp6); // row 1
pMCUSrc[iCol+16] = (short)(tmp2 + tmp5); // row 2
pMCUSrc[iCol+24] = (short)(tmp3 - tmp4); // row 3
pMCUSrc[iCol+32] = (short)(tmp3 + tmp4); // row 4
pMCUSrc[iCol+40] = (short)(tmp2 - tmp5); // row 5
pMCUSrc[iCol+48] = (short)(tmp1 - tmp6); // row 6
pMCUSrc[iCol+56] = (short)(tmp0 - tmp7); // row 7
} // full calculation needed
} // if column has data in it
} // for each column
#endif // NO SIMD
// now do rows
u16MCUFlags = pJPEG->u16MCUFlags;
pOutput = (unsigned char *)pMCUSrc; // store output pixels back into MCU
for (iRow=0; iRow<64; iRow+=8) // all rows must be calculated
{
// even part
if ((u16MCUFlags & 0xf0) == 0) // quick and dirty calculation (right 4 columns are all 0's)
{
if ((u16MCUFlags & 0xfc) == 0) // very likely case (1 or 2 columns occupied)
{
// even part
tmp0 = tmp1 = tmp2 = tmp3 = pMCUSrc[iRow+0];
// odd part
tmp7 = pMCUSrc[iRow+1];
tmp6 = (tmp7 * 217)>>8; // * 0.8477
tmp5 = (tmp7 * 145)>>8; // * 0.5663
tmp4 = -((tmp7 * 51)>>8); // * -0.199
}
else
{
tmp10 = pMCUSrc[iRow+0];
tmp13 = pMCUSrc[iRow+2];
tmp12 = ((tmp13 * 106)>>8); // 2-6 * 1.414
tmp0 = tmp10 + tmp13;
tmp3 = tmp10 - tmp13;
tmp1 = tmp10 + tmp12;
tmp2 = tmp10 - tmp12;
// odd part
z13 = pMCUSrc[iRow+3];
z11 = pMCUSrc[iRow+1];
tmp7 = z11 + z13;
tmp11 = ((z11 - z13)*362)>>8; // * 1.414
z5 = ((z11 - z13)*473)>>8; // * 1.8477
tmp10 = ((z11*277)>>8) - z5; // * 1.08239
tmp12 = ((z13*669)>>8) + z5; // * 2.61312
tmp6 = tmp12 - tmp7;
tmp5 = tmp11 - tmp6;
tmp4 = tmp10 + tmp5;
}
}
else // need to do the full calculation
{
tmp10 = pMCUSrc[iRow+0] + pMCUSrc[iRow+4];
tmp11 = pMCUSrc[iRow+0] - pMCUSrc[iRow+4];
tmp13 = pMCUSrc[iRow+2] + pMCUSrc[iRow+6];
tmp12 = (((pMCUSrc[iRow+2] - pMCUSrc[iRow+6]) * 362)>>8) - tmp13; // 2-6 * 1.414
tmp0 = tmp10 + tmp13;
tmp3 = tmp10 - tmp13;
tmp1 = tmp11 + tmp12;
tmp2 = tmp11 - tmp12;
// odd part
z13 = pMCUSrc[iRow+5] + pMCUSrc[iRow+3];
z10 = pMCUSrc[iRow+5] - pMCUSrc[iRow+3];
z11 = pMCUSrc[iRow+1] + pMCUSrc[iRow+7];
z12 = pMCUSrc[iRow+1] - pMCUSrc[iRow+7];
tmp7 = z11 + z13;
tmp11 = ((z11 - z13)*362)>>8; // * 1.414
z5 = ((z10 + z12)*473)>>8; // * 1.8477
tmp10 = ((z12*277)>>8) - z5; // * 1.08239
tmp12 = ((z10*-669)>>8) + z5; // * 2.61312
tmp6 = tmp12 - tmp7;
tmp5 = tmp11 - tmp6;
tmp4 = tmp10 + tmp5;
}
// final output stage - scale down and range limit
#ifdef HAS_SIMD
{
uint32_t ul, ulOut;
const uint32_t ulAdj = 0x800080;
ulOut = __SSAT16((((tmp0+tmp7)>>5) & 0xffff) | (((tmp2+tmp5)>>5)<<16), 8);
ulOut = __SADD16(ulOut, ulAdj); // adjust
ul = __SSAT16((((tmp1+tmp6)>>5) & 0xffff) | (((tmp3-tmp4)>>5)<<16), 8);
ul = __SADD16(ul, ulAdj); // adjust
ulOut |= (ul << 8); // combine 4 outputs
*(uint32_t *)pOutput = ulOut; // store first 4
ulOut = __SSAT16((((tmp3+tmp4)>>5) & 0xffff) | (((tmp1-tmp6)>>5)<<16), 8);
ulOut = __SADD16(ulOut, ulAdj); // adjust
ul = __SSAT16((((tmp2-tmp5)>>5) & 0xffff) | (((tmp0-tmp7)>>5)<<16), 8);
ul = __SADD16(ul, ulAdj); // adjust
ulOut |= (ul << 8); // combine 4 outputs
*(uint32_t *)&pOutput[4] = ulOut; // store second 4
}
#else
// I've tried various things to speed this up, but it always seems to take the same amount of time
#ifdef HAS_NEON
{
int16x4_t L_in_16x4, R_in_16x4, L_out, R_out;
int8x8_t LR_out_8x8;
int16x8_t LR_out;
L_in_16x4 = vdup_n_s16(tmp0); // suppresses warning of setting lane 0 of uninitialized var
L_in_16x4 = vset_lane_s16(tmp1, L_in_16x4, 1);
L_in_16x4 = vset_lane_s16(tmp2, L_in_16x4, 2);
L_in_16x4 = vset_lane_s16(tmp3, L_in_16x4, 3);
R_in_16x4 = vdup_n_s16(tmp7);
R_in_16x4 = vset_lane_s16(tmp6, R_in_16x4, 1);
R_in_16x4 = vset_lane_s16(tmp5, R_in_16x4, 2);
R_in_16x4 = vset_lane_s16(-tmp4, R_in_16x4, 3);
L_out = vadd_s16(L_in_16x4, R_in_16x4); // tmp0 + tmp7, tmp1 + tmp6, ...
R_out = vsub_s16(L_in_16x4, R_in_16x4); // tmp0 - tmp7, tmp1 - tmp6, ...
R_out = vrev64_s16(R_out); // flip order of 4-7
LR_out = vcombine_s16(L_out, R_out);
LR_out = vaddq_s16(LR_out, vdupq_n_s16(0x80 << 5)); // adjust output +0x80
LR_out_8x8 = vqshrun_n_s16(LR_out, 5); // shift, narrow and clip to 0-255
vst1_u8(pOutput, LR_out_8x8);
}
#else
pOutput[0] = ucRangeTable[(((tmp0 + tmp7)>>5) & 0x3ff)];
pOutput[1] = ucRangeTable[(((tmp1 + tmp6)>>5) & 0x3ff)];
pOutput[2] = ucRangeTable[(((tmp2 + tmp5)>>5) & 0x3ff)];
pOutput[3] = ucRangeTable[(((tmp3 - tmp4)>>5) & 0x3ff)];
pOutput[4] = ucRangeTable[(((tmp3 + tmp4)>>5) & 0x3ff)];
pOutput[5] = ucRangeTable[(((tmp2 - tmp5)>>5) & 0x3ff)];
pOutput[6] = ucRangeTable[(((tmp1 - tmp6)>>5) & 0x3ff)];
pOutput[7] = ucRangeTable[(((tmp0 - tmp7)>>5) & 0x3ff)];
#endif // !HAS_NEON
#endif
pOutput += 8;
} // for each row
} /* JPEGIDCT() */
static void JPEGPutMCU8BitGray(JPEGIMAGE *pJPEG, int x, int iPitch)
{
int i, j, xcount, ycount;
uint8_t *pDest, *pSrc = (uint8_t *)&pJPEG->sMCUs[0];
if (pJPEG->pDitherBuffer)
pDest = &pJPEG->pDitherBuffer[x];
else
pDest = (uint8_t *)&pJPEG->usPixels[x/2];
if (pJPEG->ucSubSample <= 0x11) // single Y
{
if (pJPEG->iOptions & JPEG_SCALE_HALF) // special handling of 1/2 size (pixel averaging)
{
int pix;
for (i=0; i<4; i++)
{
for (j=0; j<4; j++)
{
pix = (pSrc[0] + pSrc[1] + pSrc[8] + pSrc[9] + 2) >> 2; // average 2x2 block
pDest[j] = (uint8_t)pix;
pSrc += 2;
}
pSrc += 8; // skip extra line
pDest += iPitch;
}
return;
}
xcount = ycount = 8; // debug
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
xcount = ycount = 2;
else if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
xcount = ycount = 1;
for (i=0; i<ycount; i++) // do up to 8 rows
{
for (j=0; j<xcount; j++)
*pDest++ = *pSrc++;
pDest -= xcount;
pDest += iPitch; // next line
}
return;
} // single Y source
if (pJPEG->ucSubSample == 0x21) // stacked horizontally
{
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
// only 2 pixels emitted
pDest[0] = pSrc[0];
pDest[1] = pSrc[128];
return;
} /* 1/8 */
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (i=0; i<4; i++)
{
for (j=0; j<4; j++)
{
int pix;
pix = (pSrc[j*2] + pSrc[j*2+1] + pSrc[j*2 + 8] + pSrc[j*2 + 9] + 2) >> 2;
pDest[j] = (uint8_t)pix;
pix = (pSrc[j*2 + 128] + pSrc[j*2+129] + pSrc[j*2 + 136] + pSrc[j*2 + 137] + 2) >> 2;
pDest[j+4] = (uint8_t)pix;
}
pSrc += 16;
pDest += iPitch;
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{
// each MCU contributes a 2x2 block
pDest[0] = pSrc[0]; // Y0
pDest[1] = pSrc[1];
pDest[iPitch] = pSrc[2];
pDest[iPitch+1] = pSrc[3];
pDest[2] = pSrc[128]; // Y`
pDest[3] = pSrc[129];
pDest[iPitch+2] = pSrc[130];
pDest[iPitch+3] = pSrc[131];
return;
}
for (i=0; i<8; i++)
{
for (j=0; j<8; j++)
{
pDest[j] = pSrc[j];
pDest[j+8] = pSrc[128 + j];
}
pSrc += 8;
pDest += iPitch;
}
} // 0x21
if (pJPEG->ucSubSample == 0x12) // stacked vertically
{
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
// only 2 pixels emitted
pDest[0] = pSrc[0];
pDest[iPitch] = pSrc[128];
return;
} /* 1/8 */
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (i=0; i<4; i++)
{
for (j=0; j<4; j++)
{
int pix;
pix = (pSrc[j*2] + pSrc[j*2+1] + pSrc[j*2 + 8] + pSrc[j*2 + 9] + 2) >> 2;
pDest[j] = (uint8_t)pix;
pix = (pSrc[j*2 + 128] + pSrc[j*2+129] + pSrc[j*2 + 136] + pSrc[j*2 + 137] + 2) >> 2;
pDest[4*iPitch+j] = (uint8_t)pix;
}
pSrc += 16;
pDest += iPitch;
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{
// each MCU contributes a 2x2 block
pDest[0] = pSrc[0]; // Y0
pDest[1] = pSrc[1];
pDest[iPitch] = pSrc[2];
pDest[iPitch+1] = pSrc[3];
pDest[iPitch*2] = pSrc[128]; // Y`
pDest[iPitch*2+1] = pSrc[129];
pDest[iPitch*3] = pSrc[130];
pDest[iPitch*3+1] = pSrc[131];
return;
}
for (i=0; i<8; i++)
{
for (j=0; j<8; j++)
{
pDest[j] = pSrc[j];
pDest[8*iPitch + j] = pSrc[128 + j];
}
pSrc += 8;
pDest += iPitch;
}
} // 0x12
if (pJPEG->ucSubSample == 0x22)
{
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
// each MCU contributes 1 pixel
pDest[0] = pSrc[0]; // Y0
pDest[1] = pSrc[128]; // Y1
pDest[iPitch] = pSrc[256]; // Y2
pDest[iPitch + 1] = pSrc[384]; // Y3
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{
// each MCU contributes 2x2 pixels
pDest[0] = pSrc[0]; // Y0
pDest[1] = pSrc[1];
pDest[iPitch] = pSrc[2];
pDest[iPitch+1] = pSrc[3];
pDest[2] = pSrc[128]; // Y1
pDest[3] = pSrc[129];
pDest[iPitch+2] = pSrc[130];
pDest[iPitch+3] = pSrc[131];
pDest[iPitch*2] = pSrc[256]; // Y2
pDest[iPitch*2+1] = pSrc[257];
pDest[iPitch*3] = pSrc[258];
pDest[iPitch*3+1] = pSrc[259];
pDest[iPitch*2+2] = pSrc[384]; // Y3
pDest[iPitch*2+3] = pSrc[385];
pDest[iPitch*3+2] = pSrc[386];
pDest[iPitch*3+3] = pSrc[387];
return;
}
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (i=0; i<4; i++)
{
for (j=0; j<4; j++)
{
int pix;
pix = (pSrc[j*2] + pSrc[j*2+1] + pSrc[j*2 + 8] + pSrc[j*2 + 9] + 2) >> 2;
pDest[j] = (uint8_t)pix; // Y0
pix = (pSrc[j*2+128] + pSrc[j*2+129] + pSrc[j*2 + 136] + pSrc[j*2 + 137] + 2) >> 2;
pDest[j+4] = (uint8_t)pix; // Y1
pix = (pSrc[j*2+256] + pSrc[j*2+257] + pSrc[j*2 + 264] + pSrc[j*2 + 265] + 2) >> 2;
pDest[iPitch*4 + j] = (uint8_t)pix; // Y2
pix = (pSrc[j*2+384] + pSrc[j*2+385] + pSrc[j*2 + 392] + pSrc[j*2 + 393] + 2) >> 2;
pDest[iPitch*4 + j + 4] = (uint8_t)pix; // Y3
}
pSrc += 16;
pDest += iPitch;
}
return;
}
#ifdef ALLOWS_UNALIGNED
for (i=0; i<8; i++)
{
*(uint32_t *)pDest = *(uint32_t *)pSrc; // Y0
*(uint32_t *)&pDest[4] = *(uint32_t *)&pSrc[4]; // Y0
*(uint32_t *)&pDest[8] = *(uint32_t *)&pSrc[128]; // Y1
*(uint32_t *)&pDest[12] = *(uint32_t *)&pSrc[132]; // Y1
*(uint32_t *)&pDest[iPitch*8] = *(uint32_t *)&pSrc[256]; // Y2
*(uint32_t *)&pDest[(iPitch*8)+4] = *(uint32_t *)&pSrc[260]; // Y2
*(uint32_t *)&pDest[(iPitch*8) + 8] = *(uint32_t *)&pSrc[384]; // Y3
*(uint32_t *)&pDest[(iPitch*8) + 12] = *(uint32_t *)&pSrc[388]; // Y3
pSrc += 8;
pDest += iPitch;
}
#else
for (i=0; i<8; i++)
{
for (j=0; j<8; j++)
{
pDest[j] = pSrc[j]; // Y0
pDest[j+8] = pSrc[j+128]; // Y1
pDest[iPitch*8 + j] = pSrc[j+256]; // Y2
pDest[iPitch*8 + j + 8] = pSrc[j + 384]; // Y3
}
pSrc += 8;
pDest += iPitch;
}
#endif
} // 0x22
} /* JPEGMPutMCU8BitGray() */
static void JPEGPutMCUGray(JPEGIMAGE *pJPEG, int x, int iPitch)
{
uint16_t *usDest = (uint16_t *)&pJPEG->usPixels[x];
int i, j, xcount, ycount;
uint8_t *pSrc = (uint8_t *)&pJPEG->sMCUs[0];
if (pJPEG->iOptions & JPEG_SCALE_HALF) // special handling of 1/2 size (pixel averaging)
{
int pix;
for (i=0; i<4; i++)
{
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
for (j=0; j<4; j++)
{
pix = (pSrc[0] + pSrc[1] + pSrc[8] + pSrc[9] + 2) >> 2; // average 2x2 block
usDest[j] = usGrayTo565[pix];
pSrc += 2;
}
}
else
{
for (j=0; j<4; j++)
{
pix = (pSrc[0] + pSrc[1] + pSrc[8] + pSrc[9] + 2) >> 2; // average 2x2 block
usDest[j] = __builtin_bswap16(usGrayTo565[pix]);
pSrc += 2;
}
}
pSrc += 8; // skip extra line
usDest += iPitch;
}
return;
}
xcount = ycount = 8; // debug
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
xcount = ycount = 2;
else if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
xcount = ycount = 1;
for (i=0; i<ycount; i++) // do up to 8 rows
{
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
for (j=0; j<xcount; j++)
*usDest++ = usGrayTo565[*pSrc++];
}
else
{
for (j=0; j<xcount; j++)
*usDest++ = __builtin_bswap16(usGrayTo565[*pSrc++]);
}
usDest -= xcount;
usDest += iPitch; // next line
}
} /* JPEGPutMCUGray() */
static void JPEGPixelLE(uint16_t *pDest, int iY, int iCb, int iCr)
{
//
// Cortex-M4/M7 has some SIMD instructions which can shave a few cycles
// off of this function (e.g. Teensy, Arduino Nano 33 BLE, Portenta, etc)
//
#ifdef HAS_SIMD
uint32_t ulPixel;
uint32_t ulCbCr;
uint32_t ulTmp = 0xfa7f /*-1409*/ | 0xf4930000 /*(-2925 << 16)*/; // for green calc
iCb -= 0x80; iCr -= 0x80;
ulCbCr = (iCb | (iCr << 16));
ulPixel = __SMLAD(ulCbCr, ulTmp, iY) >> 14; // G
ulPixel = __USAT16(ulPixel, 6) << 5; // range limit to 6 bits
ulTmp = __SMLAD(7258, iCb, iY) >> 15; // Blue
ulTmp = __USAT16(ulTmp, 5); // range limit to 5 bits
ulPixel |= ulTmp; // now we have G + B
ulTmp = __SMLAD(5742, iCr, iY) >> 15; // Red
ulTmp = __USAT16(ulTmp, 5); // range limit to 5 bits
ulPixel |= (ulTmp << 11); // now we have R + G + B
pDest[0] = (uint16_t)ulPixel;
#else
int iCBB, iCBG, iCRG, iCRR;
unsigned short usPixel;
iCBB = 7258 * (iCb-0x80);
iCBG = -1409 * (iCb-0x80);
iCRG = -2925 * (iCr-0x80);
iCRR = 5742 * (iCr-0x80);
usPixel = usRangeTableB[((iCBB + iY) >> 12) & 0x3ff]; // blue pixel
usPixel |= usRangeTableG[((iCBG + iCRG + iY) >> 12) & 0x3ff]; // green pixel
usPixel |= usRangeTableR[((iCRR + iY) >> 12) & 0x3ff]; // red pixel
pDest[0] = usPixel;
#endif
} /* JPEGPixelLE() */
static void JPEGPixelBE(uint16_t *pDest, int iY, int iCb, int iCr)
{
int iCBB, iCBG, iCRG, iCRR;
unsigned short usPixel;
iCBB = 7258 * (iCb-0x80);
iCBG = -1409 * (iCb-0x80);
iCRG = -2925 * (iCr-0x80);
iCRR = 5742 * (iCr-0x80);
usPixel = usRangeTableB[((iCBB + iY) >> 12) & 0x3ff]; // blue pixel
usPixel |= usRangeTableG[((iCBG + iCRG + iY) >> 12) & 0x3ff]; // green pixel
usPixel |= usRangeTableR[((iCRR + iY) >> 12) & 0x3ff]; // red pixel
pDest[0] = __builtin_bswap16(usPixel);
} /* JPEGPixelBE() */
static void JPEGPixelRGB(uint32_t *pDest, int iY, int iCb, int iCr)
{
int iCBB, iCBG, iCRG, iCRR;
uint32_t u32Pixel;
int32_t i32;
iCBB = 7258 * (iCb-0x80);
iCBG = -1409 * (iCb-0x80);
iCRG = -2925 * (iCr-0x80);
iCRR = 5742 * (iCr-0x80);
u32Pixel = 0xff000000; // Alpha = 0xff
i32 = ((iCBB + iY) >> 12);
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel |= (uint32_t)i32; // blue
i32 = ((iCBG + iCRG + iY) >> 12); // green pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel |= (uint32_t)(i32 << 8);
i32 = ((iCRR + iY) >> 12); // red pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel |= (uint32_t)(i32 << 16);
pDest[0] = u32Pixel;
} /* JPEGPixelRGB() */
static void JPEGPixel2LE(uint16_t *pDest, int iY1, int iY2, int iCb, int iCr)
{
uint32_t ulPixel1, ulPixel2;
//
// Cortex-M4/M7 has some SIMD instructions which can shave a few cycles
// off of this function (e.g. Teensy, Arduino Nano 33 BLE, Portenta, etc)
//
#ifdef HAS_SIMD
uint32_t ulCbCr;
uint32_t ulTmp2 ,ulTmp = 0xfa7f /*-1409*/ | 0xf4930000 /*(-2925 << 16)*/; // for green calc
iCb -= 0x80; iCr -= 0x80;
ulCbCr = (iCb | (iCr << 16));
//ulCbCr = __SSUB16(ulCbCr, 0x00800080); // dual 16-bit subtraction
ulPixel1 = __SMLAD(ulCbCr, ulTmp, iY1) >> 14; // G for pixel 1
ulPixel2 = __SMLAD(ulCbCr, ulTmp, iY2) >> 14; // G for pixel 2
ulPixel1 |= (ulPixel2 << 16);
ulPixel1 = __USAT16(ulPixel1, 6) << 5; // range limit both to 6 bits
ulTmp = __SMLAD(7258, iCb, iY1) >> 15; // Blue 1
ulTmp2 = __SMLAD(7258, iCb, iY2) >> 15; // Blue 2
ulTmp = __USAT16(ulTmp | (ulTmp2 << 16), 5); // range limit both to 5 bits
ulPixel1 |= ulTmp; // now we have G + B
ulTmp = __SMLAD(5742, iCr, iY1) >> 15; // Red 1
ulTmp2 = __SMLAD(5742, iCr, iY2) >> 15; // Red 2
ulTmp = __USAT16(ulTmp | (ulTmp2 << 16), 5); // range limit both to 5 bits
ulPixel1 |= (ulTmp << 11); // now we have R + G + B
*(uint32_t *)&pDest[0] = ulPixel1;
#else
int iCBB, iCBG, iCRG, iCRR;
iCBB = 7258 * (iCb-0x80);
iCBG = -1409 * (iCb-0x80);
iCRG = -2925 * (iCr-0x80);
iCRR = 5742 * (iCr-0x80);
ulPixel1 = usRangeTableB[((iCBB + iY1) >> 12) & 0x3ff]; // blue pixel
ulPixel1 |= usRangeTableG[((iCBG + iCRG + iY1) >> 12) & 0x3ff]; // green pixel
ulPixel1 |= usRangeTableR[((iCRR + iY1) >> 12) & 0x3ff]; // red pixel
ulPixel2 = usRangeTableB[((iCBB + iY2) >> 12) & 0x3ff]; // blue pixel
ulPixel2 |= usRangeTableG[((iCBG + iCRG + iY2) >> 12) & 0x3ff]; // green pixel
ulPixel2 |= usRangeTableR[((iCRR + iY2) >> 12) & 0x3ff]; // red pixel
*(uint32_t *)&pDest[0] = (ulPixel1 | (ulPixel2<<16));
#endif
} /* JPEGPixel2LE() */
static void JPEGPixel2BE(uint16_t *pDest, int32_t iY1, int32_t iY2, int32_t iCb, int32_t iCr)
{
int32_t iCBB, iCBG, iCRG, iCRR;
uint32_t ulPixel1, ulPixel2;
iCBB = 7258L * (iCb-0x80);
iCBG = -1409L * (iCb-0x80);
iCRG = -2925L * (iCr-0x80);
iCRR = 5742L * (iCr-0x80);
ulPixel1 = usRangeTableB[((iCBB + iY1) >> 12) & 0x3ff]; // blue pixel
ulPixel1 |= usRangeTableG[((iCBG + iCRG + iY1) >> 12) & 0x3ff]; // green pixel
ulPixel1 |= usRangeTableR[((iCRR + iY1) >> 12) & 0x3ff]; // red pixel
ulPixel2 = usRangeTableB[((iCBB + iY2) >> 12) & 0x3ff]; // blue pixel
ulPixel2 |= usRangeTableG[((iCBG + iCRG + iY2) >> 12) & 0x3ff]; // green pixel
ulPixel2 |= usRangeTableR[((iCRR + iY2) >> 12) & 0x3ff]; // red pixel
*(uint32_t *)&pDest[0] = __builtin_bswap16(ulPixel1) | ((uint32_t)__builtin_bswap16(ulPixel2)<<16);
} /* JPEGPixel2BE() */
static void JPEGPixel2RGB(uint32_t *pDest, int32_t iY1, int32_t iY2, int32_t iCb, int32_t iCr)
{
int32_t iCBB, iCBG, iCRG, iCRR;
uint32_t u32Pixel1, u32Pixel2;
int32_t i32;
iCBB = 7258L * (iCb-0x80);
iCBG = -1409L * (iCb-0x80);
iCRG = -2925L * (iCr-0x80);
iCRR = 5742L * (iCr-0x80);
i32 = ((iCBB + iY1) >> 12); // blue pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel1 = u32Pixel2 = 0xff000000; // Alpha = 255
u32Pixel1 |= (uint32_t)i32; // blue
i32 = ((iCBG + iCRG + iY1) >> 12); // green pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel1 |= (uint32_t)(i32 << 8); // green
i32 = ((iCRR + iY1) >> 12); // red pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel1 |= (uint32_t)(i32 << 16); // red
i32 = ((iCBB + iY2) >> 12); // blue pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel2 |= (uint32_t)i32;
i32 = ((iCBG + iCRG + iY2) >> 12); // green pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel2 |= (uint32_t)(i32 << 8);
i32 = ((iCRR + iY2) >> 12); // red pixel
if (i32 < 0) i32 = 0;
else if (i32 > 255) i32 = 255;
u32Pixel2 |= (uint32_t)(i32 << 16);
pDest[0] = u32Pixel1;
pDest[1] = u32Pixel2;
} /* JPEGPixel2RGB() */
static void JPEGPutMCU11(JPEGIMAGE *pJPEG, int x, int iPitch)
{
int iCr, iCb;
signed int Y;
int iCol;
int iRow;
uint8_t *pY, *pCr, *pCb;
uint16_t *pOutput = &pJPEG->usPixels[x];
if (pJPEG->ucPixelType == RGB8888) {
pOutput += x; // 4 bytes per pixel, not 2
}
pY = (unsigned char *)&pJPEG->sMCUs[0*DCTSIZE];
pCb = (unsigned char *)&pJPEG->sMCUs[1*DCTSIZE];
pCr = (unsigned char *)&pJPEG->sMCUs[2*DCTSIZE];
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (iRow=0; iRow<4; iRow++) // up to 8 rows to do
{
for (iCol=0; iCol<4; iCol++) // up to 4x2 cols to do
{
iCr = (pCr[0] + pCr[1] + pCr[8] + pCr[9] + 2) >> 2;
iCb = (pCb[0] + pCb[1] + pCb[8] + pCb[9] + 2) >> 2;
Y = (pY[0] + pY[1] + pY[8] + pY[9]) << 10;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol, Y, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol, Y, iCb, iCr);
else
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y, iCb, iCr);
pCr += 2;
pCb += 2;
pY += 2;
} // for col
pCr += 8;
pCb += 8;
pY += 8;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
} // for row
return;
}
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH) // special case for 1/8 scaling
{
// only 4 pixels to draw, so no looping needed
iCr = pCr[0];
iCb = pCb[0];
Y = (int)(pY[0]) << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput, Y, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput, Y, iCb, iCr);
else
JPEGPixelRGB((uint32_t *)pOutput, Y, iCb, iCr);
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER) // special case for 1/4 scaling
{
// only 4 pixels to draw, so no looping needed
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelLE(pOutput, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelLE(pOutput+1, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelLE(pOutput+iPitch, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelLE(pOutput+1+iPitch, Y, iCb, iCr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelBE(pOutput, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelBE(pOutput+1, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelBE(pOutput+iPitch, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelBE(pOutput+1+iPitch, Y, iCb, iCr);
} else { // RGB8888
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelRGB((uint32_t *)pOutput, Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelRGB((uint32_t *)&pOutput[2], Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*2], Y, iCb, iCr);
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelRGB((uint32_t *)&pOutput[2+iPitch*2], Y, iCb, iCr);
}
return;
}
// full size
#ifdef ESP32S3_SIMD
if (pJPEG->ucPixelType == RGB8888) iPitch *= 2;
for (iRow=0; iRow<8; iRow++) {
s3_ycbcr_convert_444(pY, pCb, pCr, pOutput, i16_Consts, pJPEG->ucPixelType);
pCb += 8; pCr += 8; pY += 8; pOutput += iPitch;
}
return;
#endif // ESP32S3_SIMD
#ifdef HAS_SSE
// SSE2 version
// R = Y + 1.40200 * Cr
// G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
// B = Y - 0.22800 * Cb + Cb + Cb
if (pJPEG->ucPixelType == RGB8888) {
__m128i mmxY, mmxCr, mmxCb, mmxTemp;
__m128i mmxTemp2, mmxR, mmxG, mmxB;
iPitch *= 2; // points to 32-bit values, not 16-bit
mmxTemp2 = _mm_cmpeq_epi16(_mm_setzero_si128(), _mm_setzero_si128()); // fix Cr/Cb values by subtracting 0x80
mmxTemp2 = _mm_slli_epi16 (mmxTemp2, 15); // now has 0x8000, 0x8000...
for (iRow=0; iRow<8; iRow++) { // do 8 rows
mmxCr = _mm_loadl_epi64((__m128i *)pCr); // load 1 row of Cr
mmxCb = _mm_loadl_epi64((__m128i *)pCb); // load 1 row of Cb
mmxY = _mm_loadl_epi64((__m128i *)pY); // load 1 row of Y
pCr += 8;
pCb += 8;
pY += 8;
mmxCr = _mm_unpacklo_epi8 (_mm_setzero_si128(), mmxCr); // zero-extend 8 Cr values to 16-bits
mmxCb = _mm_unpacklo_epi8 (_mm_setzero_si128(), mmxCb); // zero-extend 8 Cb values to 16-bits
mmxY = _mm_unpacklo_epi8 (mmxY, _mm_setzero_si128()); // zero-extend 8 Y values to 16-bits
mmxCr = _mm_add_epi16(mmxCr, mmxTemp2); // subtract 0x80
mmxCb = _mm_add_epi16(mmxCb, mmxTemp2); // subtract 0x80
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxTemp = _mm_loadu_si128((__m128i *)&s1402[0]); // load the 1.402 constant
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxTemp); // almost ready with R
mmxR = _mm_add_epi16(mmxTemp, mmxY); // now we have 8 R values
mmxTemp = _mm_loadu_si128((__m128i *)&s0714[0]);
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxTemp); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, mmxTemp);
mmxTemp = _mm_loadu_si128((__m128i *)&s0344[0]); // Y -= 0.34414*Cb
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxTemp);
mmxG = _mm_add_epi16(mmxG, mmxTemp); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_loadu_si128((__m128i *)&s1772[0]); // B = Y - 1.772*Cb
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxTemp);
mmxB = _mm_add_epi16(mmxY, mmxTemp); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_cmpeq_epi16(mmxTemp, mmxTemp); // Alpha set to FFFF
mmxCr = _mm_unpacklo_epi8(mmxB, mmxG); // interleave 8 B's and 8 G's
mmxCb = _mm_unpacklo_epi8(mmxR, mmxTemp); // interlave 8 R's and 8 A's
mmxTemp = _mm_unpacklo_epi16(mmxCr, mmxCb); // interleave 4 BG's and 4 RA's
mmxCr = _mm_unpackhi_epi16(mmxCr, mmxCb); // interleave 4 BG's and 4 RA's
// _mm_stream_si128((__m128i*)pOutput, mmxTemp);
// _mm_stream_si128((__m128i*)(pOutput+8), mmxCr);
_mm_storeu_si128((__m128i *)pOutput, mmxTemp); // write 4 RGBA pixels
_mm_storeu_si128((__m128i *)(pOutput+8), mmxCr); // write 4 RGBA pixels
pOutput += iPitch;
} // for each row
return;
} else { // 16-bpp
__m128i mmxY, mmxCr, mmxCb, mmxTemp;
__m128i mmxTemp2, mmxR, mmxG, mmxB;
for (iRow=0; iRow<8; iRow++) { // do 8 rows
mmxCr = _mm_loadl_epi64((__m128i *)pCr); // load 1 row of Cr
mmxCb = _mm_loadl_epi64((__m128i *)pCb); // load 1 row of Cb
mmxY = _mm_loadl_epi64((__m128i *)pY); // load 1 row of Y
pCr += 8;
pCb += 8;
pY += 8;
mmxTemp = _mm_cmpeq_epi16(mmxY, mmxY); // fix Cr/Cb values by subtracting 0x80
mmxTemp = _mm_slli_epi16 (mmxTemp, 15); // now has 0x8000, 0x8000...
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxCr = _mm_unpacklo_epi8 (mmxTemp2, mmxCr); // zero-extend 8 Cr values to 16-bits
mmxCb = _mm_unpacklo_epi8 (mmxTemp2, mmxCb); // zero-extend 8 Cb values to 16-bits
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxCr = _mm_add_epi16(mmxCr, mmxTemp); // subtract 0x80
mmxCb = _mm_add_epi16(mmxCb, mmxTemp); // subtract 0x80
mmxTemp = _mm_loadu_si128((__m128i *)&s1402[0]); // load the 1.402 constant
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxTemp); // almost ready with R
mmxR = _mm_add_epi16(mmxTemp, mmxY); // now we have 8 R values
mmxTemp = _mm_loadu_si128((__m128i *)&s0714[0]);
mmxR = _mm_srai_epi16(mmxR, 4);
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxTemp); // Y-0.71414*Cr
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxG = _mm_add_epi16(mmxY, mmxTemp);
mmxTemp = _mm_loadu_si128((__m128i *)&s0344[0]); // Y -= 0.34414*Cb
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxTemp);
mmxG = _mm_add_epi16(mmxG, mmxTemp); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_loadu_si128((__m128i *)&s1772[0]); // B = Y - 1.772*Cb
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxTemp);
mmxB = _mm_add_epi16(mmxY, mmxTemp); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxTemp = _mm_setzero_si128(); // interleave with 0 to get back to 16-bit values
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxR = _mm_unpacklo_epi8(mmxR, mmxTemp); // zero-extend to 16-bits again
mmxB = _mm_unpacklo_epi8(mmxB, mmxTemp);
mmxG = _mm_unpacklo_epi8(mmxG, mmxTemp);
mmxR = _mm_srli_epi16(mmxR, 3); // reduce to 5-bits
mmxR = _mm_slli_epi16(mmxR, 11); // set in proper position
mmxB = _mm_srli_epi16(mmxB, 3); // reduce to 5-bits
mmxG = _mm_srli_epi16(mmxG, 2); // reduce to 6-bits
mmxG = _mm_slli_epi16(mmxG, 5); // set in proper position
mmxTemp = _mm_or_si128(mmxR, mmxG); // R+G
mmxTemp = _mm_or_si128(mmxTemp, mmxB); // R+G+B
_mm_storeu_si128((__m128i *)pOutput, mmxTemp); // write 8 RGB565 pixels
pOutput += iPitch;
} // for each row
return;
}
#endif // HAS_SSE
for (iRow=0; iRow<8; iRow++) // up to 8 rows to do
{
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
for (iCol=0; iCol<8; iCol++) // up to 4x2 cols to do
{
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelLE(pOutput+iCol, Y, iCb, iCr);
} // for col
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
for (iCol=0; iCol<8; iCol++) // up to 4x2 cols to do
{
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelBE(pOutput+iCol, Y, iCb, iCr);
} // for col
} else { // RGB888
for (iCol=0; iCol<8; iCol++) // up to 4x2 cols to do
{
iCr = *pCr++;
iCb = *pCb++;
Y = (int)(*pY++) << 12;
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y, iCb, iCr);
} // for col
}
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
} // for row
} /* JPEGPutMCU11() */
static void JPEGPutMCU22(JPEGIMAGE *pJPEG, int x, int iPitch)
{
uint32_t Cr,Cb;
signed int Y1, Y2, Y3, Y4;
int iRow, iCol, iXCount1, iXCount2, iYCount;
unsigned char *pY, *pCr, *pCb;
int bUseOdd1, bUseOdd2; // special case where 24bpp odd sized image can clobber first column
uint16_t *pOutput = &pJPEG->usPixels[x];
if (pJPEG->ucPixelType == RGB8888) {
pOutput += x; // 4 bytes per pixel, not 2
}
pY = (unsigned char *)&pJPEG->sMCUs[0*DCTSIZE];
pCb = (unsigned char *)&pJPEG->sMCUs[4*DCTSIZE];
pCr = (unsigned char *)&pJPEG->sMCUs[5*DCTSIZE];
if (pJPEG->iOptions & JPEG_SCALE_HALF) // special handling of 1/2 size (pixel averaging)
{
for (iRow=0; iRow<4; iRow++) // 16x16 becomes 8x8 of 2x2 pixels
{
for (iCol=0; iCol<4; iCol++)
{
Y1 = (pY[iCol*2] + pY[iCol*2+1] + pY[iCol*2+8] + pY[iCol*2+9]) << 10;
Cb = pCb[iCol];
Cr = pCr[iCol];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol, Y1, Cb, Cr); // top left
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y1, Cb, Cr);
Y1 = (pY[iCol*2+(DCTSIZE*2)] + pY[iCol*2+1+(DCTSIZE*2)] + pY[iCol*2+8+(DCTSIZE*2)] + pY[iCol*2+9+(DCTSIZE*2)]) << 10;
Cb = pCb[iCol+4];
Cr = pCr[iCol+4];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol+4, Y1, Cb, Cr); // top right
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol+4, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[(iCol*2)+8], Y1, Cb, Cr);
Y1 = (pY[iCol*2+(DCTSIZE*4)] + pY[iCol*2+1+(DCTSIZE*4)] + pY[iCol*2+8+(DCTSIZE*4)] + pY[iCol*2+9+(DCTSIZE*4)]) << 10;
Cb = pCb[iCol+32];
Cr = pCr[iCol+32];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol+iPitch*4, Y1, Cb, Cr); // bottom left
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol+iPitch*4, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[(iCol+iPitch*4)*2], Y1, Cb, Cr);
Y1 = (pY[iCol*2+(DCTSIZE*6)] + pY[iCol*2+1+(DCTSIZE*6)] + pY[iCol*2+8+(DCTSIZE*6)] + pY[iCol*2+9+(DCTSIZE*6)]) << 10;
Cb = pCb[iCol+32+4];
Cr = pCr[iCol+32+4];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol+4+iPitch*4, Y1, Cb, Cr); // bottom right
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol+4+iPitch*4, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[(iCol+4+iPitch*4)*2], Y1, Cb, Cr);
}
pY += 16;
pCb += 8;
pCr += 8;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
Y1 = pY[0] << 12; // scale to level of conversion table
Cb = pCb[0];
Cr = pCr[0];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput, Y1, Cb, Cr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)pOutput, Y1, Cb, Cr);
// top right block
Y1 = pY[DCTSIZE*2] << 12; // scale to level of conversion table
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput + 1, Y1, Cb, Cr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput + 1, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[2], Y1, Cb, Cr);
// bottom left block
Y1 = pY[DCTSIZE*4] << 12; // scale to level of conversion table
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iPitch, Y1, Cb, Cr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iPitch, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*2], Y1, Cb, Cr);
// bottom right block
Y1 = pY[DCTSIZE*6] << 12; // scale to level of conversion table
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+ 1 + iPitch, Y1, Cb, Cr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+ 1 + iPitch, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[2 + iPitch*2], Y1, Cb, Cr);
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER) // special case of 1/4
{
for (iRow=0; iRow<2; iRow++)
{
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
for (iCol=0; iCol<2; iCol++)
{
// top left block
Y1 = pY[iCol] << 12; // scale to level of conversion table
Cb = pCb[0];
Cr = pCr[0];
JPEGPixelLE(pOutput + iCol, Y1, Cb, Cr);
// top right block
Y1 = pY[iCol+(DCTSIZE*2)] << 12; // scale to level of conversion table
Cb = pCb[1];
Cr = pCr[1];
JPEGPixelLE(pOutput + 2+iCol, Y1, Cb, Cr);
// bottom left block
Y1 = pY[iCol+DCTSIZE*4] << 12; // scale to level of conversion table
Cb = pCb[2];
Cr = pCr[2];
JPEGPixelLE(pOutput+iPitch*2 + iCol, Y1, Cb, Cr);
// bottom right block
Y1 = pY[iCol+DCTSIZE*6] << 12; // scale to level of conversion table
Cb = pCb[3];
Cr = pCr[3];
JPEGPixelLE(pOutput+iPitch*2 + 2+iCol, Y1, Cb, Cr);
} // for each column
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
for (iCol=0; iCol<2; iCol++)
{
// top left block
Y1 = pY[iCol] << 12; // scale to level of conversion table
Cb = pCb[0];
Cr = pCr[0];
JPEGPixelBE(pOutput + iCol, Y1, Cb, Cr);
// top right block
Y1 = pY[iCol+(DCTSIZE*2)] << 12; // scale to level of conversion table
Cb = pCb[1];
Cr = pCr[1];
JPEGPixelBE(pOutput + 2+iCol, Y1, Cb, Cr);
// bottom left block
Y1 = pY[iCol+DCTSIZE*4] << 12; // scale to level of conversion table
Cb = pCb[2];
Cr = pCr[2];
JPEGPixelBE(pOutput+iPitch*2 + iCol, Y1, Cb, Cr);
// bottom right block
Y1 = pY[iCol+DCTSIZE*6] << 12; // scale to level of conversion table
Cb = pCb[3];
Cr = pCr[3];
JPEGPixelBE(pOutput+iPitch*2 + 2+iCol, Y1, Cb, Cr);
} // for each column
} else { // RGB8888
for (iCol=0; iCol<2; iCol++)
{
// top left block
Y1 = pY[iCol] << 12; // scale to level of conversion table
Cb = pCb[0];
Cr = pCr[0];
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y1, Cb, Cr);
// top right block
Y1 = pY[iCol+(DCTSIZE*2)] << 12; // scale to level of conversion table
Cb = pCb[1];
Cr = pCr[1];
JPEGPixelRGB((uint32_t *)&pOutput[(2+iCol)*2], Y1, Cb, Cr);
// bottom left block
Y1 = pY[iCol+DCTSIZE*4] << 12; // scale to level of conversion table
Cb = pCb[2];
Cr = pCr[2];
JPEGPixelRGB((uint32_t *)&pOutput[(iPitch*2 + iCol)*2], Y1, Cb, Cr);
// bottom right block
Y1 = pY[iCol+DCTSIZE*6] << 12; // scale to level of conversion table
Cb = pCb[3];
Cr = pCr[3];
JPEGPixelRGB((uint32_t *)&pOutput[(iPitch*2 + 2+iCol)*2], Y1, Cb, Cr);
} // for each column
}
pY += 2; // skip 1 line of source pixels
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
}
return;
}
// full size
#ifdef ESP32S3_SIMD
if (pJPEG->ucPixelType == RGB8888) iPitch *= 2;
for (iRow=0; iRow<4; iRow++) { // top L+R, 4 pairs of lines x 16 pixels
// each call converts 16 pixels
s3_ycbcr_convert_420(pY, pCb, pCr, pOutput, i16_Consts, pJPEG->ucPixelType);
s3_ycbcr_convert_420(pY+8, pCb, pCr, pOutput+iPitch, i16_Consts, pJPEG->ucPixelType);
pCb += 8; pCr += 8; pY += 16; pOutput += iPitch*2;
}
pY += (256 - 64);
for (iRow=0; iRow<4; iRow++) { // bottom L+R
s3_ycbcr_convert_420(pY, pCb, pCr, pOutput, i16_Consts, pJPEG->ucPixelType);
s3_ycbcr_convert_420(pY+8, pCb, pCr, pOutput+iPitch, i16_Consts, pJPEG->ucPixelType);
pCb += 8; pCr += 8; pY += 16; pOutput += iPitch*2;
}
return;
#endif // ESP32S3_SIMD
#ifdef HAS_NEON
if (pJPEG->ucPixelType == RGB8888) {
int8x8_t i88Cr, i88Cb;
uint8x16_t u816YL, u816YR;
int16x8_t i168Cr, i168Cb, i168Y, i168Temp;
int16x4_t i164Constants;
int16x8_t i168R, i168G, i168B;
uint8x8_t u88R, u88G, u88B, u88A;
int16x8x2_t i168Crx2, i168Cbx2;
uint8x8x4_t u884Hack;
i164Constants = vld1_s16(&sYCCRGBConstants[0]); // 4 different constants used for "lane" multiplications by scalar
u88A = vdup_n_u8(0xff); // Alpha set to FF
for (iRow=0; iRow<8; iRow++) { // do 8 rows
i88Cr = vld1_s8((const int8_t *)pCr); // load 1 row of Cr
i88Cb = vld1_s8((const int8_t *) pCb); // load 1 row of Cb
u816YL = vld1q_u8(pY); // load 2 rows of Y (left block)
u816YR = vld1q_u8(pY+128); // load 2 rows of Y (right block)
// top left block
i168Temp = vdupq_n_s16((int16_t)0x8000); // fix Cr/Cb values by subtracting 0x80
i168Cr = vshll_n_s8(i88Cr, 8); // widen 8 Cr values and shift left 8
i168Cb = vshll_n_s8(i88Cb, 8); // widen 8 Cb values and shift left 8
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_low_u8(u816YL), 4)); // widen and x16 to put on par with Cr/Cb values
i168Cr = vsubq_s16(i168Cr, i168Temp); // fix Cr/Cb (-0x80)
i168Cb = vsubq_s16(i168Cb, i168Temp);
i168Crx2 = vzipq_s16(i168Cr, i168Cr); // double elements in horizonal direction
i168Cbx2 = vzipq_s16(i168Cb, i168Cb);
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_low_u8(u816YR), 4)); // widen and x16 to put on par with Cr/Cb values (right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
// ugly hack due to bug in GCC of vst4 intrinsics
u884Hack.val[0] = u88B;
u884Hack.val[1] = u88G;
u884Hack.val[2] = u88R;
u884Hack.val[3] = u88A;
vst4_u8((uint8_t *)pOutput, u884Hack);
// top right block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_high_u8(u816YL), 4)); // widen and x16 to put on par with Cr/Cb values (right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
// ugly hack due to bug in GCC of vst4 intrinsics
u884Hack.val[0] = u88B;
u884Hack.val[1] = u88G;
u884Hack.val[2] = u88R;
u884Hack.val[3] = u88A;
vst4_u8((uint8_t *)(pOutput+16), u884Hack);
// bottom left block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_high_u8(u816YR), 4)); // widen and x16 to put on par with Cr/Cb values (bottom right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
// ugly hack due to bug in GCC of vst4 intrinsics
u884Hack.val[0] = u88B;
u884Hack.val[1] = u88G;
u884Hack.val[2] = u88R;
u884Hack.val[3] = u88A;
vst4_u8((uint8_t *)(pOutput+iPitch*2), u884Hack);
// bottom right block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
// ugly hack due to bug in GCC of vst4 intrinsics
u884Hack.val[0] = u88B;
u884Hack.val[1] = u88G;
u884Hack.val[2] = u88R;
u884Hack.val[3] = u88A;
vst4_u8((uint8_t *)(pOutput+iPitch*2+16), u884Hack);
pCr += 8;
pCb += 8;
if (iRow == 3) // bottom 4 rows Y values are in 2 other MCUs
pY += 16 + 192; // skip to other 2 Y blocks
else
pY += 16;
pOutput += 4*iPitch;
} // for each row
return; // 32bpp
} else { // 16bpp
int8x8_t i88Cr, i88Cb;
uint8x16_t u816YL, u816YR;
int16x8_t i168Cr, i168Cb, i168Y, i168Temp;
int16x4_t i164Constants;
int16x8x2_t i168Crx2, i168Cbx2;
int16x8_t i168R, i168G, i168B;
uint8x8_t u88R, u88G, u88B;
uint16x8_t u168Temp, u168Temp2;
uint8_t ucPixelType = pJPEG->ucPixelType;
i164Constants = vld1_s16(&sYCCRGBConstants[0]); // 4 different constants used for "lane" multiplications by scalar
for (iRow=0; iRow<8; iRow++) { // do 8 rows
i88Cr = vld1_s8((const int8_t *) pCr); // load 1 row of Cr
i88Cb = vld1_s8((const int8_t *) pCb); // load 1 row of Cb
u816YL = vld1q_u8(pY); // load 2 rows of Y (left block)
u816YR = vld1q_u8(pY+128); // load 2 rows of Y (right block)
// top left block
i168Temp = vdupq_n_s16((int16_t) 0x8000); // fix Cr/Cb values by subtracting 0x80
i168Cr = vshll_n_s8(i88Cr, 8); // widen 8 Cr values and shift left 8
i168Cb = vshll_n_s8(i88Cb, 8); // widen 8 Cb values and shift left 8
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_low_u8(u816YL), 4)); // widen and x16 to put on par with Cr/Cb values
i168Cr = vsubq_s16(i168Cr, i168Temp); // fix Cr/Cb (-0x80)
i168Cb = vsubq_s16(i168Cb, i168Temp);
i168Crx2 = vzipq_s16(i168Cr, i168Cr); // double elements in horizonal direction
i168Cbx2 = vzipq_s16(i168Cb, i168Cb);
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_low_u8(u816YR), 4)); // widen and x16 to put on par with Cr/Cb values (right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
u168Temp = vshll_n_u8(u88R, 8); // place red in upper part of 16-bit words
u168Temp2 = vshll_n_u8(u88G, 8); // shift green elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 5); // shift green elements right and insert red elements
u168Temp2 = vshll_n_u8(u88B, 8); // shift blue elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 11); // shift blue elements right and insert
if (ucPixelType == RGB565_BIG_ENDIAN) { // reverse the bytes
u168Temp = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(u168Temp)));
}
vst1q_u16((uint16_t *)pOutput, u168Temp); // top left block
// top right block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_high_u8(u816YL), 4)); // widen and x16 to put on par with Cr/Cb values (right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
u168Temp = vshll_n_u8(u88R, 8); // place red in upper part of 16-bit words
u168Temp2 = vshll_n_u8(u88G, 8); // shift green elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 5); // shift green elements right and insert red elements
u168Temp2 = vshll_n_u8(u88B, 8); // shift blue elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 11); // shift blue elements right and insert
if (ucPixelType == RGB565_BIG_ENDIAN) { // reverse the bytes
u168Temp = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(u168Temp)));
}
vst1q_u16((uint16_t *)(pOutput+8), u168Temp); // top right block
// bottom left block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[0], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[0], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
i168Y = vreinterpretq_s16_u16(vshll_n_u8(vget_high_u8(u816YR), 4)); // widen and x16 to put on par with Cr/Cb values (bottom right block)
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
u168Temp = vshll_n_u8(u88R, 8); // place red in upper part of 16-bit words
u168Temp2 = vshll_n_u8(u88G, 8); // shift green elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 5); // shift green elements right and insert red elements
u168Temp2 = vshll_n_u8(u88B, 8); // shift blue elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 11); // shift blue elements right and insert
if (ucPixelType == RGB565_BIG_ENDIAN) { // reverse the bytes
u168Temp = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(u168Temp)));
}
vst1q_u16((uint16_t *)(pOutput+iPitch), u168Temp); // bottom left block
// bottom right block
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 0); // Cr x 1.402
i168R = vaddq_s16(i168Temp, i168Y); // now we have 8 R values
i168Temp = vqdmulhq_lane_s16(i168Crx2.val[1], i164Constants, 1); // Cr x -0.71414
u88R = vqshrun_n_s16(i168R, 4); // narrow and saturate to 8-bit unsigned
i168G = vaddq_s16(i168Y, i168Temp);
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 2); // Cb x -0.34414
i168G = vaddq_s16(i168G, i168Temp); // now we have 8 G values
u88G = vqrshrun_n_s16(i168G, 4); // shift right, narrow and saturate to 8-bit unsigned
i168Temp = vqdmulhq_lane_s16(i168Cbx2.val[1], i164Constants, 3); // Cb x -1.772
i168B = vaddq_s16(i168Y, i168Temp); // now we have 8 B values
u88B = vqrshrun_n_s16(i168B, 4); // shift right, narrow and saturate to 8-bit unsigned
u168Temp = vshll_n_u8(u88R, 8); // place red in upper part of 16-bit words
u168Temp2 = vshll_n_u8(u88G, 8); // shift green elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 5); // shift green elements right and insert red elements
u168Temp2 = vshll_n_u8(u88B, 8); // shift blue elements to top of 16-bit words
u168Temp = vsriq_n_u16(u168Temp, u168Temp2, 11); // shift blue elements right and insert
if (ucPixelType == RGB565_BIG_ENDIAN) { // reverse the bytes
u168Temp = vreinterpretq_u16_u8(vrev16q_u8(vreinterpretq_u8_u16(u168Temp)));
}
vst1q_u16((uint16_t *)(pOutput+iPitch+8), u168Temp); // bottom right block
// advance to next pair of lines
pCr += 8;
pCb += 8;
if (iRow == 3) // bottom 4 rows Y values are in 2 other MCUs
pY += 16 + 192; // skip to other 2 Y blocks
else
pY += 16;
pOutput += iPitch*2;
} // for each row
return;
} // 16bpp
#endif // HAS_NEON
#ifdef HAS_SSE
// SSE2 version
// R = Y + 1.40200 * Cr
// G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr
// B = Y - 0.22800 * Cb + Cb + Cb
if (pJPEG->ucPixelType == RGB8888) {
__m128i mmxY, mmxCr, mmxCb, mmxTemp;
__m128i mmxTemp2, mmxR, mmxG, mmxB;
__m128i mmxConst1402, mmxConst0714, mmxConst0344, mmxConst1772;
mmxConst0344 = _mm_load_si128((__m128i *)&s0344[0]);
mmxConst1402 = _mm_load_si128((__m128i *)&s1402[0]);
mmxConst0714 = _mm_load_si128((__m128i *)&s0714[0]);
mmxConst1772 = _mm_load_si128((__m128i *)&s1772[0]);
iPitch *= 2; // destination is 32-bit values, not 16
for (iRow = 0; iRow<8; iRow++) // do 8 pairs of rows in 4 quadrants
{
// left block
mmxCr = _mm_loadl_epi64((__m128i *)pCr); // load 1 row of Cr
mmxCb = _mm_loadl_epi64((__m128i *)pCb); // load 1 row of Cb
mmxY = _mm_loadl_epi64((__m128i *)pY); // load 1 row of Y (top left block)
mmxTemp = _mm_cmpeq_epi16(mmxY, mmxY); // fix Cr/Cb values by subtracting 0x80
mmxTemp = _mm_slli_epi16 (mmxTemp, 15); // now has 0x8000, 0x8000...
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxCr = _mm_unpacklo_epi8 (mmxTemp2, mmxCr); // zero-extend 8 Cr values to 16-bits
mmxCb = _mm_unpacklo_epi8 (mmxTemp2, mmxCb); // zero-extend 8 Cb values to 16-bits (left shifted 8)
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxCr = _mm_add_epi16(mmxCr, mmxTemp); // subtract 0x80
mmxCb = _mm_add_epi16(mmxCb, mmxTemp); // subtract 0x80
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpacklo_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344);
mmxG = _mm_add_epi16(mmxG, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772);
mmxB = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_cmpeq_epi16(mmxTemp, mmxTemp); // Alpha set to FFFF
mmxY = _mm_unpacklo_epi8(mmxB, mmxG); // interleave 8 B's and 8 G's
mmxTemp2 = _mm_unpacklo_epi8(mmxR, mmxTemp); // interlave 8 R's and 8 A's
mmxTemp = _mm_unpacklo_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
mmxTemp2 = _mm_unpackhi_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
// store first row of pair
_mm_storeu_si128((__m128i *)pOutput, mmxTemp); // write 4 RGBA pixels
_mm_storeu_si128((__m128i *)(pOutput+8), mmxTemp2); // write 4 RGBA pixels
// second row of left block
mmxY = _mm_loadl_epi64((__m128i *)(pY+8)); // load 1 row of Y (top left block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxR = _mm_add_epi16(_mm_unpacklo_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344); // Y -= 0.34414*Cb
mmxG = _mm_add_epi16(mmxG, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772); // B = Y - 1.772*Cb
mmxB = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_cmpeq_epi16(mmxTemp, mmxTemp); // Alpha set to FFFF
mmxY = _mm_unpacklo_epi8(mmxB, mmxG); // interleave 8 B's and 8 G's
mmxTemp2 = _mm_unpacklo_epi8(mmxR, mmxTemp); // interlave 8 R's and 8 A's
mmxTemp = _mm_unpacklo_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
mmxTemp2 = _mm_unpackhi_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
// store second row of pair
_mm_storeu_si128((__m128i *)(pOutput+iPitch), mmxTemp); // write 4 RGBA pixels
_mm_storeu_si128((__m128i *)(pOutput+iPitch+8), mmxTemp2); // write 4 RGBA pixels
// right block
mmxY = _mm_loadl_epi64((__m128i *)(pY+128)); // load 1 row of Y (right block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpackhi_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344);
mmxG = _mm_add_epi16(mmxG, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772);
mmxB = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_cmpeq_epi16(mmxTemp, mmxTemp); // Alpha set to FFFF
mmxY = _mm_unpacklo_epi8(mmxB, mmxG); // interleave 8 B's and 8 G's
mmxTemp2 = _mm_unpacklo_epi8(mmxR, mmxTemp); // interlave 8 R's and 8 A's
mmxTemp = _mm_unpacklo_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
mmxTemp2 = _mm_unpackhi_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
// store first row of right block
_mm_storeu_si128((__m128i *)(pOutput+16), mmxTemp); // write 4 RGBA pixels
_mm_storeu_si128((__m128i *)(pOutput+24), mmxTemp2); // write 4 RGBA pixels
// prepare second row of right block
mmxY = _mm_loadl_epi64((__m128i *)(pY+136)); // load 1 row of Y (right block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpackhi_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344);
mmxG = _mm_add_epi16(mmxG, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772);
mmxB = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_cmpeq_epi16(mmxTemp, mmxTemp); // Alpha set to FFFF
mmxY = _mm_unpacklo_epi8(mmxB, mmxG); // interleave 8 B's and 8 G's
mmxTemp2 = _mm_unpacklo_epi8(mmxR, mmxTemp); // interlave 8 R's and 8 A's
mmxTemp = _mm_unpacklo_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
mmxTemp2 = _mm_unpackhi_epi16(mmxY, mmxTemp2); // interleave 4 BG's and 4 RA's
// store second row of right block
_mm_storeu_si128((__m128i *)(pOutput+iPitch+16), mmxTemp); // write 4 RGBA pixels
_mm_storeu_si128((__m128i *)(pOutput+iPitch+24), mmxTemp2); // write 4 RGBA pixels
pOutput += iPitch*2;
pCr += 8;
pCb += 8;
if (iRow == 3) // bottom 4 rows Y values are in 2 other MCUs
pY += 16 + 192; // skip to other 2 Y blocks
else
pY += 16;
} // for each row
return;
} else { // 16-bit pixels
__m128i mmxY, mmxCr, mmxCb, mmxTemp;
__m128i mmxTemp2, mmxR, mmxG, mmxB;
__m128i mmxConst1402, mmxConst0714, mmxConst0344, mmxConst1772;
mmxConst0344 = _mm_load_si128((__m128i *)&s0344[0]);
mmxConst1402 = _mm_load_si128((__m128i *)&s1402[0]);
mmxConst0714 = _mm_load_si128((__m128i *)&s0714[0]);
mmxConst1772 = _mm_load_si128((__m128i *)&s1772[0]);
for (iRow = 0; iRow<8; iRow++) // do 8 pairs of rows in 4 quadrants
{
// left block
mmxCr = _mm_loadl_epi64((__m128i *)pCr); // load 1 row of Cr
mmxCb = _mm_loadl_epi64((__m128i *)pCb); // load 1 row of Cb
mmxY = _mm_loadl_epi64((__m128i *)pY); // load 1 row of Y (top left block)
mmxTemp = _mm_cmpeq_epi16(mmxY, mmxY); // fix Cr/Cb values by subtracting 0x80
mmxTemp = _mm_slli_epi16 (mmxTemp, 7); // now has 0xff80, 0xff80...
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxCr = _mm_unpacklo_epi8 (mmxCr, mmxTemp2); // zero-extend 8 Cr values to 16-bits
mmxCb = _mm_unpacklo_epi8 (mmxCb, mmxTemp2); // zero-extend 8 Cb values to 16-bits
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxCr = _mm_add_epi16(mmxCr, mmxTemp); // subtract 0x80
mmxCb = _mm_add_epi16(mmxCb, mmxTemp); // subtract 0x80
mmxCr = _mm_slli_epi16(mmxCr, 8); // put in top half of 16-bits
mmxCb = _mm_slli_epi16(mmxCb, 8); // put in top half of 16-bits
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpacklo_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344); // Y -= 0.34414*Cb
mmxG = _mm_add_epi16(mmxG, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772); // B = Y - 1.772*Cb
mmxB = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_setzero_si128(); // interleave with 0 to get back to 16-bit values
mmxR = _mm_unpacklo_epi8(mmxR, mmxTemp); // zero-extend to 16-bits again
mmxB = _mm_unpacklo_epi8(mmxB, mmxTemp);
mmxG = _mm_unpacklo_epi8(mmxG, mmxTemp);
mmxR = _mm_srli_epi16(mmxR, 3); // reduce to 5-bits
mmxR = _mm_slli_epi16(mmxR, 11); // set in proper position
mmxB = _mm_srli_epi16(mmxB, 3); // reduce to 5-bits
mmxG = _mm_srli_epi16(mmxG, 2); // reduce to 6-bits
mmxG = _mm_slli_epi16(mmxG, 5); // set in proper position
mmxTemp = _mm_or_si128(mmxR, mmxG); // R+G
mmxTemp = _mm_or_si128(mmxTemp, mmxB); // R+G+B
// store first row of pair
_mm_storeu_si128((__m128i *)pOutput, mmxTemp); // write 8 RGB565 pixels
// second row of left block
mmxY = _mm_loadl_epi64((__m128i *)(pY+8)); // load 1 row of Y (top left block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxR = _mm_add_epi16(_mm_unpacklo_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344);
mmxG = _mm_add_epi16(mmxG, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772); // B = Y - 1.772*Cb
mmxB = _mm_add_epi16(mmxY, _mm_unpacklo_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_setzero_si128(); // interleave with 0 to get back to 16-bit values
mmxR = _mm_unpacklo_epi8(mmxR, mmxTemp); // zero-extend to 16-bits again
mmxB = _mm_unpacklo_epi8(mmxB, mmxTemp);
mmxG = _mm_unpacklo_epi8(mmxG, mmxTemp);
mmxR = _mm_srli_epi16(mmxR, 3); // reduce to 5-bits
mmxR = _mm_slli_epi16(mmxR, 11); // set in proper position
mmxB = _mm_srli_epi16(mmxB, 3); // reduce to 5-bits
mmxG = _mm_srli_epi16(mmxG, 2); // reduce to 6-bits
mmxG = _mm_slli_epi16(mmxG, 5); // set in proper position
mmxTemp = _mm_or_si128(mmxR, mmxG); // R+G
mmxTemp = _mm_or_si128(mmxTemp, mmxB); // R+G+B
// store second row of pair
_mm_storeu_si128((__m128i *)(pOutput+iPitch), mmxTemp); // write 8 RGB565 pixels
// right block
mmxY = _mm_loadl_epi64((__m128i *)(pY+128)); // load 1 row of Y (right block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpackhi_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344); // Y -= 0.34414*Cb
mmxG = _mm_add_epi16(mmxG, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772); // B = Y - 1.772*Cb
mmxB = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_setzero_si128(); // interleave with 0 to get back to 16-bit values
mmxR = _mm_unpacklo_epi8(mmxR, mmxTemp); // zero-extend to 16-bits again
mmxB = _mm_unpacklo_epi8(mmxB, mmxTemp);
mmxG = _mm_unpacklo_epi8(mmxG, mmxTemp);
mmxR = _mm_srli_epi16(mmxR, 3); // reduce to 5-bits
mmxR = _mm_slli_epi16(mmxR, 11); // set in proper position
mmxB = _mm_srli_epi16(mmxB, 3); // reduce to 5-bits
mmxG = _mm_srli_epi16(mmxG, 2); // reduce to 6-bits
mmxG = _mm_slli_epi16(mmxG, 5); // set in proper position
mmxTemp = _mm_or_si128(mmxR, mmxG); // R+G
mmxTemp = _mm_or_si128(mmxTemp, mmxB); // R+G+B
// store first row of right block
_mm_storeu_si128((__m128i *)(pOutput+16), mmxTemp); // write 8 RGB565 pixels
// prepare second row of right block
mmxY = _mm_loadl_epi64((__m128i *)(pY+136)); // load 1 row of Y (right block)
mmxTemp2 = _mm_setzero_si128(); // zero it to use to set upper bits to 0
mmxY = _mm_unpacklo_epi8 (mmxY, mmxTemp2); // zero-extend 8 Y values to 16-bits
mmxY = _mm_slli_epi16(mmxY, 4); // x16 to put on par with Cr/Cb values
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst1402); // almost ready with R
mmxR = _mm_add_epi16(_mm_unpackhi_epi16(mmxTemp, mmxTemp), mmxY); // now we have 8 R values
mmxR = _mm_srai_epi16(mmxR, 4);
mmxR = _mm_packus_epi16 (mmxR, mmxR); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCr, mmxConst0714); // Y-0.71414*Cr
mmxG = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp));
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst0344);
mmxG = _mm_add_epi16(mmxG, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 G values
mmxG = _mm_srai_epi16(mmxG, 4);
mmxG = _mm_packus_epi16 (mmxG, mmxG); // i16->u8 bit and saturate
mmxTemp = _mm_mulhi_epi16(mmxCb, mmxConst1772); // B = Y - 1.772*Cb
mmxB = _mm_add_epi16(mmxY, _mm_unpackhi_epi16(mmxTemp, mmxTemp)); // now we have 8 B values
mmxB = _mm_srai_epi16(mmxB, 4);
mmxB = _mm_packus_epi16 (mmxB, mmxB); // i16->u8 bit and saturate
mmxTemp = _mm_setzero_si128(); // interleave with 0 to get back to 16-bit values
mmxR = _mm_unpacklo_epi8(mmxR, mmxTemp); // zero-extend to 16-bits again
mmxB = _mm_unpacklo_epi8(mmxB, mmxTemp);
mmxG = _mm_unpacklo_epi8(mmxG, mmxTemp);
mmxR = _mm_srli_epi16(mmxR, 3); // reduce to 5-bits
mmxR = _mm_slli_epi16(mmxR, 11); // set in proper position
mmxB = _mm_srli_epi16(mmxB, 3); // reduce to 5-bits
mmxG = _mm_srli_epi16(mmxG, 2); // reduce to 6-bits
mmxG = _mm_slli_epi16(mmxG, 5); // set in proper position
mmxTemp = _mm_or_si128(mmxR, mmxG); // R+G
mmxTemp = _mm_or_si128(mmxTemp, mmxB); // R+G+B
// store second row of right block
_mm_storeu_si128((__m128i *)(pOutput+16+iPitch), mmxTemp); // write 8 RGB565 pixels
pOutput += iPitch*2;
pCr += 8;
pCb += 8;
if (iRow == 3) // bottom 4 rows Y values are in 2 other MCUs
pY += 16 + 192; // skip to other 2 Y blocks
else
pY += 16;
} // for each row
return;
} // 16bpp
#endif // HAS_SSE
/* Reference C code */
/* Convert YCC pixels into RGB pixels and store in output image */
iYCount = 4;
bUseOdd1 = bUseOdd2 = 1; // assume odd column can be used
if ((x+15) >= pJPEG->iWidth)
{
iCol = (((pJPEG->iWidth & 15)+1) >> 1);
if (iCol >= 4)
{
iXCount1 = 4;
iXCount2 = iCol-4;
if (pJPEG->iWidth & 1 && (iXCount2 * 2) + 8 + (x * 16) > pJPEG->iWidth)
bUseOdd2 = 0;
}
else
{
iXCount1 = iCol;
iXCount2 = 0;
if (pJPEG->iWidth & 1 && (iXCount1 * 2) + (x * 16) > pJPEG->iWidth)
bUseOdd1 = 0;
}
}
else
iXCount1 = iXCount2 = 4;
for (iRow=0; iRow<iYCount; iRow++) // up to 4 rows to do
{
for (iCol=0; iCol<iXCount1; iCol++) // up to 4 cols to do
{
// for top left block
Y1 = pY[iCol*2];
Y2 = pY[iCol*2+1];
Y3 = pY[iCol*2+8];
Y4 = pY[iCol*2+9];
Y1 <<= 12; // scale to level of conversion table
Y2 <<= 12;
Y3 <<= 12;
Y4 <<= 12;
Cb = pCb[iCol];
Cr = pCr[iCol];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2LE(pOutput + (iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2LE(pOutput+iPitch + (iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelLE(pOutput + (iCol<<1), Y1, Cb, Cr);
JPEGPixelLE(pOutput+iPitch + (iCol<<1), Y3, Cb, Cr);
}
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2BE(pOutput + (iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2BE(pOutput+iPitch + (iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelBE(pOutput + (iCol<<1), Y1, Cb, Cr);
JPEGPixelBE(pOutput+iPitch + (iCol<<1), Y3, Cb, Cr);
}
} else { // RGB8888
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2RGB((uint32_t *)&pOutput[iCol<<2], Y1, Y2, Cb, Cr);
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch + (iCol<<1))], Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelRGB((uint32_t *)&pOutput[iCol<<2], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[(iPitch + (iCol<<1))*2], Y3, Cb, Cr);
}
} // RGB8888
// for top right block
if (iCol < iXCount2)
{
Y1 = pY[iCol*2+DCTSIZE*2];
Y2 = pY[iCol*2+1+DCTSIZE*2];
Y3 = pY[iCol*2+8+DCTSIZE*2];
Y4 = pY[iCol*2+9+DCTSIZE*2];
Y1 <<= 12; // scale to level of conversion table
Y2 <<= 12;
Y3 <<= 12;
Y4 <<= 12;
Cb = pCb[iCol+4];
Cr = pCr[iCol+4];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2LE(pOutput + 8+(iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2LE(pOutput+iPitch + 8+(iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelLE(pOutput+ 8+(iCol<<1), Y1, Cb, Cr);
JPEGPixelLE(pOutput+iPitch+ 8+(iCol<<1), Y3, Cb, Cr);
}
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2BE(pOutput + 8+(iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2BE(pOutput+iPitch + 8+(iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelBE(pOutput+ 8+(iCol<<1), Y1, Cb, Cr);
JPEGPixelBE(pOutput+iPitch+ 8+(iCol<<1), Y3, Cb, Cr);
}
} else { // RGB8888
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2RGB((uint32_t *)&pOutput[16+(iCol<<2)], Y1, Y2, Cb, Cr);
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch + 8+(iCol<<1))], Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelRGB((uint32_t *)&pOutput[16+(iCol<<2)], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch+ 8+(iCol<<1))], Y3, Cb, Cr);
}
} // RGB8888
}
// for bottom left block
Y1 = pY[iCol*2+DCTSIZE*4];
Y2 = pY[iCol*2+1+DCTSIZE*4];
Y3 = pY[iCol*2+8+DCTSIZE*4];
Y4 = pY[iCol*2+9+DCTSIZE*4];
Y1 <<= 12; // scale to level of conversion table
Y2 <<= 12;
Y3 <<= 12;
Y4 <<= 12;
Cb = pCb[iCol+32];
Cr = pCr[iCol+32];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2LE(pOutput+iPitch*8+ (iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2LE(pOutput+iPitch*9+ (iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelLE(pOutput+iPitch*8+ (iCol<<1), Y1, Cb, Cr);
JPEGPixelLE(pOutput+iPitch*9+ (iCol<<1), Y3, Cb, Cr);
}
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2BE(pOutput+iPitch*8+ (iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2BE(pOutput+iPitch*9+ (iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelBE(pOutput+iPitch*8+ (iCol<<1), Y1, Cb, Cr);
JPEGPixelBE(pOutput+iPitch*9+ (iCol<<1), Y3, Cb, Cr);
}
} else { // RGB8888
if (bUseOdd1 || iCol != (iXCount1-1)) // only render if it won't go off the right edge
{
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch*8+ (iCol<<1))], Y1, Y2, Cb, Cr);
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch*9+ (iCol<<1))], Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch*8+(iCol<<1))], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch*9+ (iCol<<1))], Y3, Cb, Cr);
}
} // RGB8888
// for bottom right block
if (iCol < iXCount2)
{
Y1 = pY[iCol*2+DCTSIZE*6];
Y2 = pY[iCol*2+1+DCTSIZE*6];
Y3 = pY[iCol*2+8+DCTSIZE*6];
Y4 = pY[iCol*2+9+DCTSIZE*6];
Y1 <<= 12; // scale to level of conversion table
Y2 <<= 12;
Y3 <<= 12;
Y4 <<= 12;
Cb = pCb[iCol+36];
Cr = pCr[iCol+36];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2LE(pOutput+iPitch*8+ 8+(iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2LE(pOutput+iPitch*9+ 8+(iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelLE(pOutput+iPitch*8+ 8+(iCol<<1), Y1, Cb, Cr);
JPEGPixelLE(pOutput+iPitch*9+ 8+(iCol<<1), Y3, Cb, Cr);
}
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2BE(pOutput+iPitch*8+ 8+(iCol<<1), Y1, Y2, Cb, Cr);
JPEGPixel2BE(pOutput+iPitch*9+ 8+(iCol<<1), Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelBE(pOutput+iPitch*8+ 8+(iCol<<1), Y1, Cb, Cr);
JPEGPixelBE(pOutput+iPitch*9+ 8+(iCol<<1), Y3, Cb, Cr);
}
} else { // RGB8888
if (bUseOdd2 || iCol != (iXCount2-1)) // only render if it won't go off the right edge
{
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch*8+ 8+(iCol<<1))], Y1, Y2, Cb, Cr);
JPEGPixel2RGB((uint32_t *)&pOutput[2*(iPitch*9+ 8+(iCol<<1))], Y3, Y4, Cb, Cr);
}
else
{
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch*8+ 8+(iCol<<1))], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch*9+ 8+(iCol<<1))], Y3, Cb, Cr);
}
} // RGB8888
}
} // for each column
pY += 16; // skip to next line of source pixels
pCb += 8;
pCr += 8;
pOutput += iPitch*2;
if (pJPEG->ucPixelType == RGB8888) {
pOutput += iPitch*2;
}
}
} /* JPEGPutMCU22() */
static void JPEGPutMCU12(JPEGIMAGE *pJPEG, int x, int iPitch)
{
uint32_t Cr,Cb;
signed int Y1, Y2;
int iRow, iCol, iXCount, iYCount;
uint8_t *pY, *pCr, *pCb;
uint16_t *pOutput = &pJPEG->usPixels[x];
if (pJPEG->ucPixelType == RGB8888) {
pOutput += x; // 4 bytes per pixel, not 2
}
pY = (uint8_t *)&pJPEG->sMCUs[0*DCTSIZE];
pCb = (uint8_t *)&pJPEG->sMCUs[2*DCTSIZE];
pCr = (uint8_t *)&pJPEG->sMCUs[3*DCTSIZE];
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (iRow=0; iRow<8; iRow++)
{
for (iCol=0; iCol<4; iCol++)
{
Y1 = (pY[0] + pY[1] + pY[8] + pY[9]) << 10;
Cb = (pCb[0] + pCb[1] + 1) >> 1;
Cr = (pCr[0] + pCr[1] + 1) >> 1;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol, Y1, Cb, Cr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol, Y1, Cb, Cr);
else
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y1, Cb, Cr);
pCb += 2;
pCr += 2;
pY += 2;
}
pY += 8;
if (iRow == 3) // skip to next Y MCU block
pY += 64;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
Y1 = pY[0] << 12;
Y2 = pY[DCTSIZE*2] << 12;
Cb = pCb[0];
Cr = pCr[0];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput, Y1, Cb, Cr);
JPEGPixelLE(pOutput + iPitch, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput, Y1, Cb, Cr);
JPEGPixelBE(pOutput + iPitch, Y2, Cb, Cr);
}
else { // RGB8888
JPEGPixelRGB((uint32_t *)pOutput, Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*2], Y2, Cb, Cr);
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{ // draw a 2x4 block
Y1 = pY[0] << 12;
Y2 = pY[2] << 12;
Cb = pCb[0];
Cr = pCr[0];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput, Y1, Cb, Cr);
JPEGPixelLE(pOutput + iPitch, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput, Y1, Cb, Cr);
JPEGPixelBE(pOutput + iPitch, Y2, Cb, Cr);
} else { // RGB8888
JPEGPixelRGB((uint32_t *)&pOutput, Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*2], Y2, Cb, Cr);
}
Y1 = pY[1] << 12;
Y2 = pY[3] << 12;
Cb = pCb[1];
Cr = pCr[1];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput + 1, Y1, Cb, Cr);
JPEGPixelLE(pOutput + 1 + iPitch, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput + 1, Y1, Cb, Cr);
JPEGPixelBE(pOutput + 1 + iPitch, Y2, Cb, Cr);
} else { // RGB8888
JPEGPixelRGB((uint32_t *)&pOutput[2], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[(1 + iPitch)*2], Y2, Cb, Cr);
}
pY += DCTSIZE*2; // next Y block below
Y1 = pY[0] << 12;
Y2 = pY[2] << 12;
Cb = pCb[2];
Cr = pCr[2];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput + iPitch*2, Y1, Cb, Cr);
JPEGPixelLE(pOutput + iPitch*3, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput + iPitch*2, Y1, Cb, Cr);
JPEGPixelBE(pOutput + iPitch*3, Y2, Cb, Cr);
} else { // RGB8888
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*4], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[iPitch*6], Y2, Cb, Cr);
}
Y1 = pY[1] << 12;
Y2 = pY[3] << 12;
Cb = pCb[3];
Cr = pCr[3];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput + 1 + iPitch*2, Y1, Cb, Cr);
JPEGPixelLE(pOutput + 1 + iPitch*3, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput + 1 + iPitch*2, Y1, Cb, Cr);
JPEGPixelBE(pOutput + 1 + iPitch*3, Y2, Cb, Cr);
}
else { // RGB8888
JPEGPixelRGB((uint32_t *)&pOutput[2 + iPitch*4], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[2 + iPitch*6], Y2, Cb, Cr);
}
return;
}
// full size
/* Convert YCC pixels into RGB pixels and store in output image */
iYCount = 16;
iXCount = 8;
for (iRow=0; iRow<iYCount; iRow+=2) // up to 16 rows to do
{
for (iCol=0; iCol<iXCount; iCol++) // up to 8 cols to do
{
Y1 = pY[iCol];
Y2 = pY[iCol+8];
Y1 <<= 12; // scale to level of conversion table
Y2 <<= 12;
Cb = pCb[iCol];
Cr = pCr[iCol];
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
{
JPEGPixelLE(pOutput + iCol, Y1, Cb, Cr);
JPEGPixelLE(pOutput + iPitch + iCol, Y2, Cb, Cr);
}
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
{
JPEGPixelBE(pOutput + iCol, Y1, Cb, Cr);
JPEGPixelBE(pOutput + iPitch + iCol, Y2, Cb, Cr);
} else { // RGB8888
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y1, Cb, Cr);
JPEGPixelRGB((uint32_t *)&pOutput[2*(iPitch + iCol)], Y2, Cb, Cr);
} // RGB8888
}
pY += 16; // skip to next 2 lines of source pixels
if (iRow == 6) // next MCU block, skip ahead to correct spot
pY += (128-64);
pCb += 8;
pCr += 8;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*4 : iPitch*2; // next 2 lines of dest pixels
}
} /* JPEGPutMCU12() */
static void JPEGPutMCU21(JPEGIMAGE *pJPEG, int x, int iPitch)
{
int iCr, iCb;
signed int Y1, Y2;
int iCol;
int iRow;
uint8_t *pY, *pCr, *pCb;
uint16_t *pOutput = &pJPEG->usPixels[x];
if (pJPEG->ucPixelType == RGB8888) {
pOutput += x; // 4 bytes per pixel, not 2
}
pY = (uint8_t *)&pJPEG->sMCUs[0*DCTSIZE];
pCb = (uint8_t *)&pJPEG->sMCUs[2*DCTSIZE];
pCr = (uint8_t *)&pJPEG->sMCUs[3*DCTSIZE];
if (pJPEG->iOptions & JPEG_SCALE_HALF)
{
for (iRow=0; iRow<4; iRow++)
{
for (iCol=0; iCol<4; iCol++)
{ // left block
iCr = (pCr[0] + pCr[8] + 1) >> 1;
iCb = (pCb[0] + pCb[8] + 1) >> 1;
Y1 = (signed int)(pY[0] + pY[1] + pY[8] + pY[9]) << 10;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol, Y1, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol, Y1, iCb, iCr);
else
JPEGPixelRGB((uint32_t *)&pOutput[iCol*2], Y1, iCb, iCr);
// right block
iCr = (pCr[4] + pCr[12] + 1) >> 1;
iCb = (pCb[4] + pCb[12] + 1) >> 1;
Y1 = (signed int)(pY[128] + pY[129] + pY[136] + pY[137]) << 10;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixelLE(pOutput+iCol+4, Y1, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixelBE(pOutput+iCol+4, Y1, iCb, iCr);
else
JPEGPixelRGB((uint32_t *)&pOutput[(iCol+4)*2], Y1, iCb, iCr);
pCb++;
pCr++;
pY += 2;
}
pCb += 12;
pCr += 12;
pY += 8;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
}
return;
}
if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{ // draw 2 pixels
iCr = pCr[0];
iCb = pCb[0];
Y1 = (signed int)(pY[0]) << 12;
Y2 = (signed int)(pY[DCTSIZE*2]) << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput, Y1, Y2, iCb, iCr);
else
JPEGPixel2RGB((uint32_t *)pOutput, Y1, Y2, iCb, iCr);
return;
}
if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{ // draw 4x2 pixels
// top left
iCr = pCr[0];
iCb = pCb[0];
Y1 = (signed int)(pY[0]) << 12;
Y2 = (signed int)(pY[1]) << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput, Y1, Y2, iCb, iCr);
else
JPEGPixel2RGB((uint32_t *)pOutput, Y1, Y2, iCb, iCr);
// top right
iCr = pCr[1];
iCb = pCb[1];
Y1 = (signed int)pY[DCTSIZE*2] << 12;
Y2 = (signed int)pY[DCTSIZE*2+1] << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput + 2, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput + 2, Y1, Y2, iCb, iCr);
else
JPEGPixel2RGB((uint32_t *)&pOutput[4], Y1, Y2, iCb, iCr);
// bottom left
iCr = pCr[2];
iCb = pCb[2];
Y1 = (signed int)(pY[2]) << 12;
Y2 = (signed int)(pY[3]) << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput + iPitch, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput + iPitch, Y1, Y2, iCb, iCr);
else
JPEGPixel2RGB((uint32_t *)&pOutput[iPitch*2], Y1, Y2, iCb, iCr);
// bottom right
iCr = pCr[3];
iCb = pCb[3];
Y1 = (signed int)pY[DCTSIZE*2+2] << 12;
Y2 = (signed int)pY[DCTSIZE*2+3] << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput + iPitch + 2, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput + iPitch + 2, Y1, Y2, iCb, iCr);
else
JPEGPixel2RGB((uint32_t *)&pOutput[(iPitch+2)*2], Y1, Y2, iCb, iCr);
return;
}
// Full size
/* Convert YCC pixels into RGB pixels and store in output image */
for (iRow=0; iRow<8; iRow++) // up to 8 rows to do
{
for (iCol=0; iCol<4; iCol++) // up to 4x2 cols to do
{ // left block
iCr = *pCr++;
iCb = *pCb++;
Y1 = (signed int)(*pY++) << 12;
Y2 = (signed int)(*pY++) << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput + iCol*2, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput + iCol*2, Y1, Y2, iCb, iCr);
else // RGB8888
JPEGPixel2RGB((uint32_t *)&pOutput[iCol*4], Y1, Y2, iCb, iCr);
// right block
iCr = pCr[3];
iCb = pCb[3];
Y1 = (signed int)pY[126] << 12;
Y2 = (signed int)pY[127] << 12;
if (pJPEG->ucPixelType == RGB565_LITTLE_ENDIAN)
JPEGPixel2LE(pOutput + 8 + iCol*2, Y1, Y2, iCb, iCr);
else if (pJPEG->ucPixelType == RGB565_BIG_ENDIAN)
JPEGPixel2BE(pOutput + 8 + iCol*2, Y1, Y2, iCb, iCr);
else // RGB8888
JPEGPixel2RGB((uint32_t *)&pOutput[16 + iCol*4], Y1, Y2, iCb, iCr);
} // for col
pCb += 4;
pCr += 4;
pOutput += (pJPEG->ucPixelType == RGB8888) ? iPitch*2 : iPitch;
} // for row
} /* JPEGPutMCU21() */
// Dither the 8-bit gray pixels into 1, 2, or 4-bit gray
static void JPEGDither(JPEGIMAGE *pJPEG, int iWidth, int iHeight)
{
int x, y, xmask=0, iDestPitch=0;
int32_t cNew, lFErr, v=0, h;
int32_t e1,e2,e3,e4;
uint8_t cOut, ucPixelType; // forward errors for gray
uint8_t *pSrc, *pDest, *errors, *pErrors=NULL, *d, *pPixels; // destination 8bpp image
uint8_t pixelmask=0, shift=0;
ucPixelType = pJPEG->ucPixelType;
errors = (uint8_t *)pJPEG->usPixels; // plenty of space here
errors[0] = errors[1] = errors[2] = 0;
pDest = pSrc = pJPEG->pDitherBuffer; // write the new pixels over the original
switch (ucPixelType)
{
case FOUR_BIT_DITHERED:
iDestPitch = (iWidth+1)/2;
pixelmask = 0xf0;
shift = 4;
xmask = 1;
break;
case TWO_BIT_DITHERED:
iDestPitch = (iWidth+3)/4;
pixelmask = 0xc0;
shift = 2;
xmask = 3;
break;
case ONE_BIT_DITHERED:
iDestPitch = (iWidth+7)/8;
pixelmask = 0x80;
shift = 1;
xmask = 7;
break;
}
for (y=0; y<iHeight; y++)
{
pPixels = &pSrc[y * iWidth];
d = &pDest[y * iDestPitch];
pErrors = &errors[1]; // point to second pixel to avoid boundary check
lFErr = 0;
cOut = 0;
for (x=0; x<iWidth; x++)
{
cNew = *pPixels++; // get grayscale uint8_t pixel
// add forward error
cNew += lFErr;
if (cNew > 255) cNew = 255; // clip to uint8_t
cOut <<= shift; // pack new pixels into a byte
cOut |= (cNew >> (8-shift)); // keep top N bits
if ((x & xmask) == xmask) // store it when the byte is full
{
*d++ = cOut;
cOut = 0;
}
// calculate the Floyd-Steinberg error for this pixel
v = cNew - (cNew & pixelmask); // new error for N-bit gray output (always positive)
h = v >> 1;
e1 = (7*h)>>3; // 7/16
e2 = h - e1; // 1/16
e3 = (5*h) >> 3; // 5/16
e4 = h - e3; // 3/16
// distribute error to neighbors
lFErr = e1 + pErrors[1];
pErrors[1] = (uint8_t)e2;
pErrors[0] += e3;
pErrors[-1] += e4;
pErrors++;
} // for x
} // for y
} /* JPEGDither() */
//
// Decode the image
// returns 0 for error, 1 for success
//
static int DecodeJPEG(JPEGIMAGE *pJPEG)
{
int cx, cy, x, y, mcuCX, mcuCY;
int iLum0, iLum1, iLum2, iLum3, iCr, iCb;
signed int iDCPred0, iDCPred1, iDCPred2;
int i, iQuant1, iQuant2, iQuant3, iErr;
int iSkipMask, bSkipRow;
uint8_t c;
int iMCUCount, xoff, iPitch, bThumbnail = 0;
int bContinue = 1; // early exit if the DRAW callback wants to stop
uint32_t l, *pl;
unsigned char cDCTable0, cACTable0, cDCTable1, cACTable1, cDCTable2, cACTable2;
JPEGDRAW jd;
int iMaxFill = 16, iScaleShift = 0;
// Requested the Exif thumbnail
if (pJPEG->iOptions & JPEG_EXIF_THUMBNAIL)
{
if (pJPEG->iThumbData == 0 || pJPEG->iThumbWidth == 0) // doesn't exist
{
pJPEG->iError = JPEG_INVALID_PARAMETER;
return 0;
}
if (!JPEGParseInfo(pJPEG, 1)) // parse the embedded thumbnail file header
return 0; // something went wrong
}
// Fast downscaling options
if (pJPEG->iOptions & JPEG_SCALE_HALF)
iScaleShift = 1;
else if (pJPEG->iOptions & JPEG_SCALE_QUARTER)
{
iScaleShift = 2;
iMaxFill = 1;
}
else if (pJPEG->iOptions & JPEG_SCALE_EIGHTH)
{
iScaleShift = 3;
iMaxFill = 1;
bThumbnail = 1;
}
// reorder and fix the quantization table for decoding
JPEGFixQuantD(pJPEG);
pJPEG->bb.ulBits = MOTOLONG(&pJPEG->ucFileBuf[0]); // preload first 4/8 bytes
pJPEG->bb.pBuf = pJPEG->ucFileBuf;
pJPEG->bb.ulBitOff = 0;
cDCTable0 = pJPEG->JPCI[0].dc_tbl_no;
cACTable0 = pJPEG->JPCI[0].ac_tbl_no;
cDCTable1 = pJPEG->JPCI[1].dc_tbl_no;
cACTable1 = pJPEG->JPCI[1].ac_tbl_no;
cDCTable2 = pJPEG->JPCI[2].dc_tbl_no;
cACTable2 = pJPEG->JPCI[2].ac_tbl_no;
iDCPred0 = iDCPred1 = iDCPred2 = mcuCX = mcuCY = 0;
switch (pJPEG->ucSubSample) // set up the parameters for the different subsampling options
{
case 0x00: // fake value to handle grayscale
case 0x01: // fake value to handle sRGB/CMYK
case 0x11:
cx = (pJPEG->iWidth + 7) >> 3; // number of MCU blocks
cy = (pJPEG->iCropY + pJPEG->iCropCY) >> 3;
iCr = MCU1;
iCb = MCU2;
mcuCX = mcuCY = 8;
break;
case 0x12:
cx = (pJPEG->iWidth + 7) >> 3; // number of MCU blocks
cy = (pJPEG->iCropY + pJPEG->iCropCY) >> 4;
iCr = MCU2;
iCb = MCU3;
mcuCX = 8;
mcuCY = 16;
break;
case 0x21:
cx = (pJPEG->iWidth + 15) >> 4; // number of MCU blocks
cy = (pJPEG->iCropY + pJPEG->iCropCY) >> 3;
iCr = MCU2;
iCb = MCU3;
mcuCX = 16;
mcuCY = 8;
break;
case 0x22:
cx = (pJPEG->iWidth + 15) >> 4; // number of MCU blocks
cy = (pJPEG->iCropY + pJPEG->iCropCY) >> 4;
iCr = MCU4;
iCb = MCU5;
mcuCX = mcuCY = 16;
break;
default: // to suppress compiler warning
cx = cy = 0;
iCr = iCb = 0;
break;
}
// Scale down the MCUs by the requested amount
mcuCX >>= iScaleShift;
mcuCY >>= iScaleShift;
iQuant1 = pJPEG->sQuantTable[pJPEG->JPCI[0].quant_tbl_no*DCTSIZE]; // DC quant values
iQuant2 = pJPEG->sQuantTable[pJPEG->JPCI[1].quant_tbl_no*DCTSIZE];
iQuant3 = pJPEG->sQuantTable[pJPEG->JPCI[2].quant_tbl_no*DCTSIZE];
// luminance values are always in these positions
iLum0 = MCU0;
iLum1 = MCU1;
iLum2 = MCU2;
iLum3 = MCU3;
iErr = 0;
pJPEG->iResCount = pJPEG->iResInterval;
// Calculate how many MCUs we can fit in the pixel buffer to maximize LCD drawing speed
iMCUCount = MAX_BUFFERED_PIXELS / (mcuCX * mcuCY);
if (pJPEG->ucPixelType == RGB8888) {
iMCUCount /= 2; // half as many will fit
}
if (pJPEG->ucPixelType == EIGHT_BIT_GRAYSCALE)
iMCUCount *= 2; // each pixel is only 1 byte
if (iMCUCount > cx)
iMCUCount = cx; // don't go wider than the image
if (iMCUCount > pJPEG->iMaxMCUs) // did the user set an upper bound on how many pixels per JPEGDraw callback?
iMCUCount = pJPEG->iMaxMCUs;
if (pJPEG->ucPixelType > EIGHT_BIT_GRAYSCALE) // dithered, override the max MCU count
iMCUCount = cx; // do the whole row
jd.iBpp = 16;
switch (pJPEG->ucPixelType)
{
case RGB8888:
jd.iBpp = 32;
break;
case EIGHT_BIT_GRAYSCALE:
jd.iBpp = 8;
break;
case FOUR_BIT_DITHERED:
jd.iBpp = 4;
break;
case TWO_BIT_DITHERED:
jd.iBpp = 2;
break;
case ONE_BIT_DITHERED:
jd.iBpp = 1;
break;
}
if (pJPEG->ucPixelType > EIGHT_BIT_GRAYSCALE)
jd.pPixels = (uint16_t *)pJPEG->pDitherBuffer;
else
jd.pPixels = pJPEG->usPixels;
jd.iHeight = mcuCY;
for (y = 0; y < cy && bContinue && iErr == 0; y++)
{
bSkipRow = (y*mcuCY < pJPEG->iCropY);
jd.x = pJPEG->iXOffset;
xoff = 0; // start of new LCD output group
if (pJPEG->pFramebuffer) { // user-supplied buffer is full width
int ty = (y * mcuCY) - pJPEG->iCropY;
iPitch = pJPEG->iCropCX; // size of cropped width
pJPEG->usPixels = (uint16_t *)pJPEG->pFramebuffer;
if (pJPEG->ucPixelType >= EIGHT_BIT_GRAYSCALE) {
pJPEG->usPixels += (ty * iPitch/2); // 1 byte per pixel
} else if (pJPEG->ucPixelType == RGB8888) {
pJPEG->usPixels += (ty * iPitch*2); // 4 bytes per pixel
} else { // 2 bytes per pixel
pJPEG->usPixels += (ty * iPitch);
}
} else { // use our internal buffer to do it a block at a time
iPitch = iMCUCount * mcuCX; // pixels per line of LCD buffer
}
for (x = 0; x < cx && bContinue && iErr == 0; x++)
{
iSkipMask = 0; // assume not skipping
if (bSkipRow || x*mcuCX < pJPEG->iCropX || x*mcuCX >= pJPEG->iCropX+pJPEG->iCropCX) {
iSkipMask = MCU_SKIP;
}
pJPEG->ucACTable = cACTable0;
pJPEG->ucDCTable = cDCTable0;
// do the first luminance component
iErr = JPEGDecodeMCU(pJPEG, iLum0 | iSkipMask, &iDCPred0);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
pl = (uint32_t *)&pJPEG->sMCUs[iLum0];
c = ucRangeTable[((iDCPred0 * iQuant1) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iLum0, pJPEG->JPCI[0].quant_tbl_no); // first quantization table
}
// do the second luminance component
if (pJPEG->ucSubSample > 0x11) // subsampling
{
iErr |= JPEGDecodeMCU(pJPEG, iLum1 | iSkipMask, &iDCPred0);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
c = ucRangeTable[((iDCPred0 * iQuant1) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
pl = (uint32_t *)&pJPEG->sMCUs[iLum1];
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iLum1, pJPEG->JPCI[0].quant_tbl_no); // first quantization table
}
if (pJPEG->ucSubSample == 0x22)
{
iErr |= JPEGDecodeMCU(pJPEG, iLum2 | iSkipMask, &iDCPred0);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
c = ucRangeTable[((iDCPred0 * iQuant1) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
pl = (uint32_t *)&pJPEG->sMCUs[iLum2];
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iLum2, pJPEG->JPCI[0].quant_tbl_no); // first quantization table
}
iErr |= JPEGDecodeMCU(pJPEG, iLum3 | iSkipMask, &iDCPred0);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
c = ucRangeTable[((iDCPred0 * iQuant1) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
pl = (uint32_t *)&pJPEG->sMCUs[iLum3];
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iLum3, pJPEG->JPCI[0].quant_tbl_no); // first quantization table
}
} // if 2:2 subsampling
} // if subsampling used
if (pJPEG->ucSubSample && pJPEG->ucNumComponents == 3) // if color (not CMYK)
{
// first chroma
pJPEG->ucACTable = cACTable1;
pJPEG->ucDCTable = cDCTable1;
if (pJPEG->ucPixelType >= EIGHT_BIT_GRAYSCALE) {
// We're not going to use the color channels, so avoid as much work as possible
iErr |= JPEGDecodeMCU(pJPEG, MCU_SKIP, &iDCPred1); // decode Cr block
iErr |= JPEGDecodeMCU(pJPEG, MCU_SKIP, &iDCPred2); // decode Cb block
} else {
iErr |= JPEGDecodeMCU(pJPEG, iCr | iSkipMask, &iDCPred1);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
c = ucRangeTable[((iDCPred1 * iQuant2) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
pl = (uint32_t *)&pJPEG->sMCUs[iCr];
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iCr, pJPEG->JPCI[1].quant_tbl_no); // second quantization table
}
// second chroma
pJPEG->ucACTable = cACTable2;
pJPEG->ucDCTable = cDCTable2;
iErr |= JPEGDecodeMCU(pJPEG, iCb | iSkipMask, &iDCPred2);
if (pJPEG->u16MCUFlags == 0 || bThumbnail) // no AC components, save some time
{
c = ucRangeTable[((iDCPred2 * iQuant3) >> 5) & 0x3ff];
l = c | ((uint32_t) c << 8) | ((uint32_t) c << 16) | ((uint32_t) c << 24);
// dct stores byte values
pl = (uint32_t *)&pJPEG->sMCUs[iCb];
for (i = 0; i<iMaxFill; i++) // 8x8 bytes = 16 longs
pl[i] = l;
}
else
{
JPEGIDCT(pJPEG, iCb, pJPEG->JPCI[2].quant_tbl_no);
}
}
} // if color components present
if (!iSkipMask) { // this MCU is not skipped
if (pJPEG->ucPixelType >= EIGHT_BIT_GRAYSCALE) {
JPEGPutMCU8BitGray(pJPEG, xoff, iPitch); // grayscale or color is being drawn as grayscale
} else {
switch (pJPEG->ucSubSample) {
case 0x00: // grayscale
JPEGPutMCUGray(pJPEG, xoff, iPitch);
break;
case 0x11:
JPEGPutMCU11(pJPEG, xoff, iPitch);
break;
case 0x12:
JPEGPutMCU12(pJPEG, xoff, iPitch);
break;
case 0x21:
JPEGPutMCU21(pJPEG, xoff, iPitch);
break;
case 0x22:
JPEGPutMCU22(pJPEG, xoff, iPitch);
break;
} // switch on color option
} // normal MCU drawing
xoff += mcuCX;
} // if not skipped
if (pJPEG->pFramebuffer == NULL && (xoff == iPitch || x == cx-1) && !bSkipRow) // time to draw
{
jd.iWidth = jd.iWidthUsed = iPitch; // width of each LCD block group
jd.pUser = pJPEG->pUser;
if (pJPEG->ucPixelType > EIGHT_BIT_GRAYSCALE) // dither to 4/2/1 bits
JPEGDither(pJPEG, cx * mcuCX, mcuCY);
if ((x+1)*mcuCX > pJPEG->iWidth) { // right edge has clipped pixels
jd.iWidthUsed = iPitch - (cx*mcuCX - pJPEG->iWidth);
} else if (jd.x + iPitch > pJPEG->iCropCX) { // not a full width
jd.iWidthUsed = pJPEG->iCropCX - jd.x;
}
jd.y = pJPEG->iYOffset + (y * mcuCY) - pJPEG->iCropY;
if ((jd.y - pJPEG->iYOffset + mcuCY) > (pJPEG->iHeight>>iScaleShift)) { // last row needs to be trimmed
jd.iHeight = (pJPEG->iHeight>>iScaleShift) - (jd.y - pJPEG->iYOffset);
}
bContinue = (*pJPEG->pfnDraw)(&jd);
jd.x += iPitch;
if ((cx - 1 - x) < iMCUCount) // change pitch for the last set of MCUs on this row
iPitch = (cx - 1 - x) * mcuCX;
xoff = 0;
}
if (pJPEG->iResInterval)
{
if (--pJPEG->iResCount == 0)
{
pJPEG->iResCount = pJPEG->iResInterval;
iDCPred0 = iDCPred1 = iDCPred2 = 0; // reset DC predictors
if (pJPEG->bb.ulBitOff & 7) // need to start at the next even byte
{
pJPEG->bb.ulBitOff += (8 - (pJPEG->bb.ulBitOff & 7)); // new restart interval starts on byte boundary
}
} // if restart interval needs to reset
} // if there is a restart interval
// See if we need to feed it more data
if (pJPEG->iVLCOff >= FILE_HIGHWATER)
JPEGGetMoreData(pJPEG); // need more 'filtered' VLC data
} // for x
} // for y
if (iErr != 0)
pJPEG->iError = JPEG_DECODE_ERROR;
return (iErr == 0);
} /* DecodeJPEG() */
Reference in New Issue View Git Blame Copy Permalink