jsimd_mmi.h (2458B)
1 /* 2 * Loongson MMI optimizations for libjpeg-turbo 3 * 4 * Copyright (C) 2016-2018, Loongson Technology Corporation Limited, BeiJing. 5 * All Rights Reserved. 6 * Authors: ZhuChen <zhuchen@loongson.cn> 7 * CaiWanwei <caiwanwei@loongson.cn> 8 * SunZhangzhi <sunzhangzhi-cq@loongson.cn> 9 * QingfaLiu <liuqingfa-hf@loongson.cn> 10 * 11 * This software is provided 'as-is', without any express or implied 12 * warranty. In no event will the authors be held liable for any damages 13 * arising from the use of this software. 14 * 15 * Permission is granted to anyone to use this software for any purpose, 16 * including commercial applications, and to alter it and redistribute it 17 * freely, subject to the following restrictions: 18 * 19 * 1. The origin of this software must not be misrepresented; you must not 20 * claim that you wrote the original software. If you use this software 21 * in a product, an acknowledgment in the product documentation would be 22 * appreciated but is not required. 23 * 2. Altered source versions must be plainly marked as such, and must not be 24 * misrepresented as being the original software. 25 * 3. This notice may not be removed or altered from any source distribution. 26 */ 27 28 #define JPEG_INTERNALS 29 #include "../../jinclude.h" 30 #include "../../jpeglib.h" 31 #include "../../jdct.h" 32 #include "loongson-mmintrin.h" 33 34 35 /* Common code */ 36 #if defined(_ABI64) && _MIPS_SIM == _ABI64 37 # define PTR_ADDU "daddu " 38 # define PTR_SLL "dsll " 39 #else 40 # define PTR_ADDU "addu " 41 # define PTR_SLL "sll " 42 #endif 43 44 #define SIZEOF_MMWORD 8 45 #define BYTE_BIT 8 46 #define WORD_BIT 16 47 #define SCALEBITS 16 48 49 #define _uint64_set_pi8(a, b, c, d, e, f, g, h) \ 50 (((uint64_t)(uint8_t)a << 56) | \ 51 ((uint64_t)(uint8_t)b << 48) | \ 52 ((uint64_t)(uint8_t)c << 40) | \ 53 ((uint64_t)(uint8_t)d << 32) | \ 54 ((uint64_t)(uint8_t)e << 24) | \ 55 ((uint64_t)(uint8_t)f << 16) | \ 56 ((uint64_t)(uint8_t)g << 8) | \ 57 ((uint64_t)(uint8_t)h)) 58 #define _uint64_set1_pi8(a) _uint64_set_pi8(a, a, a, a, a, a, a, a) 59 #define _uint64_set_pi16(a, b, c, d) \ 60 (((uint64_t)(uint16_t)a << 48) | \ 61 ((uint64_t)(uint16_t)b << 32) | \ 62 ((uint64_t)(uint16_t)c << 16) | \ 63 ((uint64_t)(uint16_t)d)) 64 #define _uint64_set1_pi16(a) _uint64_set_pi16(a, a, a, a) 65 #define _uint64_set_pi32(a, b) \ 66 (((uint64_t)(uint32_t)a << 32) | \ 67 ((uint64_t)(uint32_t)b)) 68 69 #define get_const_value(index) (*(__m64 *)&const_value[index])