ifquant / simd-sse-avx-neon Goto Github PK

1. __m128i

typedef union __declspec(intrin_type) _CRT_ALIGN(16) __m128i {
	__int8			m128i_i8[16];
	__int16			m128i_i16[8];
	__int32			m128i_i32[4];
	__int64			m128i_i64[2];
	unsigned __int8		m128i_u8[16];
	unsigned __int16	m128i_u16[8];
	unsigned __int32	m128i_u32[4];
	unsigned __int64	m128i_u64[2];
} __m128i;

#if !defined(_CRT_ALIGN)
	#if defined(__midl)
		#define _CRT_ALIGN(x)
	#else
		#define _CRT_ALIGN(x) __declspec(align(x))
	#endif
#endif

2. SSE2

• __m128i _mm_unpackhi_epi16 (__m128i a, __m128i b)
• __m128i _mm_unpackhi_epi32 (__m128i a, __m128i b)
• __m128i _mm_unpackhi_epi64 (__m128i a, __m128i b)
• __m128i _mm_unpackhi_epi8 (__m128i a, __m128i b)
• __m128i _mm_unpacklo_epi16 (__m128i a, __m128i b)
• __m128i _mm_unpacklo_epi32 (__m128i a, __m128i b)
• __m128i _mm_unpacklo_epi64 (__m128i a, __m128i b)
• __m128i _mm_unpacklo_epi8 (__m128i a, __m128i b)
• void _mm_storeu_si128 (__m128i* mem_addr, __m128i a)
• __m128i _mm_xor_si128 (__m128i a, __m128i b)
• __m128i _mm_or_si128 (__m128i a, __m128i b)
• __m128i _mm_and_si128 (__m128i a, __m128i b)
• __m128i _mm_slli_epi16 (__m128i a, int imm8)
• __m128i _mm_srli_epi16 (__m128i a, int imm8)
• __m128i _mm_add_epi16 (__m128i a, __m128i b)
• __m128i _mm_adds_epi16 (__m128i a, __m128i b)
  • 两个加法的不同在于，add会舍弃进位，adds会在溢出时将结果设置为最大值或最小值

#include <stdio.h>
#include <emmintrin.h>

int main (void ) {
	__m128i a = _mm_set1_epi8(0x11);
	return 0;
}

Compile with

$ gcc -march=native test.c

• FastMemcpy

• vld4_16

  uint16x4x4 vld4_16(const uint16_t *)
  Form of expected instructions: vld4.16 {d0, d1, d2, d3}, [r0]

• vld4q_u16

  uint16x8x4_t vld4q_u16 (const uint16_t *) 
  Form of expected instruction(s): vld4.16 {d0, d1, d2, d3}, [r0]

• vld4_lane_u16

 uit16x4x4_t vld4_lane_u16 (const uint16_t *, uint16x4x4_t, const int) 
 Form of expected instruction(s): vld4.16 {d0[0], d1[0], d2[0], d3[0]}, [r0]

• 从一个复数点积算法看NEON的汇编优化向量乘法加法示例

• AARCH64 ARMV8 NEON的变动
  • v8-v15在子函数调用时必须要保留，但是只保留低64bits
• Procedure Call Standard for the ARM 64-bit Architecture
• Getting to know ARM64 NEON一个向量加法示例

• Ne10

• neon-simple-guide
• NEON and VFP编程指令含义中文介绍, page 323
• Cortex-A9 NEON Media Processing Engine Revision: r4p1指令时间说明
• ARM中文社区
• ARM NEON optimization, ARM社区
  • 示例：Remove data dependencies
  • 示例：Reduce branched
  • NEON assembly and intrinsic优劣势比较
• ARM NEON programming quick reference, ARM社区
  • AMMv7-A, ARMv8-A AArch32, ARMv8-A AArch64寄存器、指令结构
  • NEON汇编示例：Assembly files or Inline assembly
• Coding for NEON - Part 1: load and stores
• Coding for NEON - Part 2: Dealing With Leftovers
• Coding for NEON - Part 4: Shifting Left and Right
• Coding for NEON - Part 5: Rearranging Vectors
• ARM NEON编程初探——一个简单的BGR888转YUV444实例详解
• ARM NEON Programmer's Reading Guide
• ARM NEON tips
• An Introduction to ARM NEON
• ARM NEON support in the ARM compiler White Paper
• ARM Compiler Version 5.06 armasm user guide指令描述更加详细，有图示
• objdump is a program for displaying various information about object files on Unix-like systems
• How to check the existence of NEON on arm?
• ARMv8 Instruction Set Overview
• ARM Assembly Local Labels
• Useful assembler directives and macros for the GNU assembler
• A64 Shift and Extend Operations: Operand Modifiers
• (vip)Accelerated AES for the ARM64 Linux kernel, AES-NEON性能评估，代码作者撰写
• makefiles CFLAGES -D, 编译宏定义