ARM NEON 指令

    在初学NDK时，接触到 HelloNeon例程，了解到 Neon是ARMv7-AR 系列中引入的并行模块，可以让你同时操作8个16位数据或4个32位数据，在信号处理，图像处理，视频编解码优化方面有很高的应用价值。在本文中罗列一些信息，供以后参考。

NEON 汇编指令一览 

http://infocenter.arm.com/help/basic/help.jsp?topic=/com.arm.doc.dui0204ic/CJAJIIGG.html

ARM 体系结构参考手册 ARMv7-A 和 ARMv7-R 版

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0406c/index.html

HelloNeon中FIR滤波器的C语言实现

/* this is a FIR filter implemented in C */
static void
fir_filter_c(short *output, const short* input, const short* kernel, int width, int kernelSize)
{
    int  offset = -kernelSize/2;
    int  nn;
    for (nn = 0; nn < width; nn++) {
        int sum = 0;
        int mm;
        for (mm = 0; mm < kernelSize; mm++) {
            sum += kernel[mm]*input[nn+offset+mm];
        }
        output[nn] = (short)((sum + 0x8000) >> 16);
    }
}

HelloNeon中FIR滤波器的Neon实现 （在调用前需要确定芯片支持NEON指令，具体见例程）

void
fir_filter_neon_intrinsics(short *output, const short* input, const short* kernel, int width, int kernelSize)
{
   int nn, offset = -kernelSize/2;

   for (nn = 0; nn < width; nn++)
   {
        int mm, sum = 0;
        int32x4_t sum_vec = vdupq_n_s32(0);
        for(mm = 0; mm < kernelSize/4; mm++)
        {
            int16x4_t  kernel_vec = vld1_s16(kernel + mm*4);
            int16x4_t  input_vec = vld1_s16(input + (nn+offset+mm*4));
            sum_vec = vmlal_s16(sum_vec, kernel_vec, input_vec);
        }

        sum += vgetq_lane_s32(sum_vec, 0);
        sum += vgetq_lane_s32(sum_vec, 1);
        sum += vgetq_lane_s32(sum_vec, 2);
        sum += vgetq_lane_s32(sum_vec, 3);

        if(kernelSize & 3)
        {
            for(mm = kernelSize - (kernelSize & 3); mm < kernelSize; mm++)
                sum += kernel[mm] * input[nn+offset+mm];
        }

        output[nn] = (short)((sum + 0x8000) >> 16);
    }
}

    经过一天的学习，将上面的代码中内循环部分改为了汇编代码，性能有一点点的提高。

		int sum, mm, sum_buf[4]={1,4,9,16};

		asm(
				"MOV r6, %1 \n"
				"MOV r3, %3 \n"
				"ADD r2, %2, %4 \n"
				"ADD r5, %5, r2, LSL #1 \n"
				"BIC r3, r3, #3 \n"
				"MOV r2, #0 \n"
				"VDUP.16 Q8, r2 \n"
			"loop_mm:\n"
				"VLD1.16 {D0}, [r6]! \n"
				"VLD1.16 {D4}, [r5]! \n"
				"VMLAL.S16 Q8, D0, D4 \n"
				"ADD r2, r2, #4 \n"
				"CMP r2, r3 \n"
				"BNE loop_mm \n"
				"MOV r2, %0 \n"
				"VST1.32 {D16-D17}, [r2]"
				:
				:"r"(sum_buf),"r"(kernel),"r"(nn),"r"(kernelSize),"r"(offset),"r"(input)
				:"memory","r1","r2","r3","r4","r5","r6"
		   );

		sum = sum_buf[0]+sum_buf[1]+sum_buf[2]+sum_buf[3];

    输出结果可能如下，下面这个图的数据是我ps出来的，由于台式机不支持NEON指令（除非你开发用的电脑是用ARM芯片 :-P），模拟器上输出的实际数据要么更慢，要么报告Not an ARMv7 CPU !