Skip to content Skip to navigation Skip to collection information

Connexions

You are here: Home » Content » Computing the fast Fourier transform on SIMD microprocessors » Appendix 3 - FFTs with vectorized loops

Navigation

Recently Viewed

This feature requires Javascript to be enabled.
 

Appendix 3 - FFTs with vectorized loops

Module by: Anthony Blake. E-mail the author

This Appendix contains source code listings corresponding to the vectorized FFT implementations in Implementation details.

Listing 1: Radix-2 FFT with vectorized loops
#include <math.h>
#include <complex.h>
#include <stdio.h>
#include <stdlib.h>
#include <xmmintrin.h>
 
typedef complex float data_t;
 
#define W(N,k) (cexp(-2.0f * M_PI * I * (float)(k) / (float)(N)))
 
data_t **LUT;
 
void ditfft2(data_t *in, data_t *out, int log2stride, int stride, int N) {
	if(N == 2) {
		out[0]   = in[0] + in[stride];
		out[N/2] = in[0] - in[stride];
	}else if(N == 4){
		ditfft2(in, out, log2stride+1, stride << 1, N >> 1);
		ditfft2(in+stride, out+N/2, log2stride+1, stride << 1, N >> 1);
	
    data_t temp0 = out[0] + out[2];
    data_t temp1 = out[0] - out[2];
    data_t temp2 = out[1] - I*out[3];
    data_t temp3 = out[1] + I*out[3];
    if(log2stride) {
			out[0] = creal(temp0) + creal(temp2)*I;
			out[1] = creal(temp1) + creal(temp3)*I;
			out[2] = cimag(temp0) + cimag(temp2)*I;
			out[3] = cimag(temp1) + cimag(temp3)*I;
		}else{
			out[0] = temp0;
			out[2] = temp1;
			out[1] = temp2;
			out[3] = temp3;
		}
	}else if(!log2stride){
		ditfft2(in, out, log2stride+1, stride << 1, N >> 1);
		ditfft2(in+stride, out+N/2, log2stride+1, stride << 1, N >> 1);
 
		int k;
		for(k=0;k<N/2;k+=4) {
			__m128 Ok_re = _mm_load_ps((float *)&out[k+N/2]);
			__m128 Ok_im = _mm_load_ps((float *)&out[k+N/2+2]);
			__m128 w_re = _mm_load_ps((float *)&LUT[log2stride][k]);
			__m128 w_im = _mm_load_ps((float *)&LUT[log2stride][k+2]);
			__m128 Ek_re = _mm_load_ps((float *)&out[k]);
			__m128 Ek_im = _mm_load_ps((float *)&out[k+2]);
			__m128 wOk_re = _mm_sub_ps(_mm_mul_ps(Ok_re,w_re),_mm_mul_ps(Ok_im,w_im));
			__m128 wOk_im = _mm_add_ps(_mm_mul_ps(Ok_re,w_im),_mm_mul_ps(Ok_im,w_re));
			__m128 out0_re = _mm_add_ps(Ek_re, wOk_re);
			__m128 out0_im = _mm_add_ps(Ek_im, wOk_im);
			__m128 out1_re = _mm_sub_ps(Ek_re, wOk_re);
			__m128 out1_im = _mm_sub_ps(Ek_im, wOk_im);
			_mm_store_ps((float *)(out+k), _mm_unpacklo_ps(out0_re, out0_im));
			_mm_store_ps((float *)(out+k+2), _mm_unpackhi_ps(out0_re, out0_im));
			_mm_store_ps((float *)(out+k+N/2), _mm_unpacklo_ps(out1_re, out1_im));
			_mm_store_ps((float *)(out+k+N/2+2), _mm_unpackhi_ps(out1_re, out1_im));
		}
	}else{
		ditfft2(in, out, log2stride+1, stride << 1, N >> 1);
		ditfft2(in+stride, out+N/2, log2stride+1, stride << 1, N >> 1);
		
		int k;
		for(k=0;k<N/2;k+=4) {
			__m128 Ok_re = _mm_load_ps((float *)&out[k+N/2]);
			__m128 Ok_im = _mm_load_ps((float *)&out[k+N/2+2]);
			__m128 w_re = _mm_load_ps((float *)&LUT[log2stride][k]);
			__m128 w_im = _mm_load_ps((float *)&LUT[log2stride][k+2]);
			__m128 Ek_re = _mm_load_ps((float *)&out[k]);
			__m128 Ek_im = _mm_load_ps((float *)&out[k+2]);
			__m128 wOk_re = _mm_sub_ps(_mm_mul_ps(Ok_re,w_re),_mm_mul_ps(Ok_im,w_im));
			__m128 wOk_im = _mm_add_ps(_mm_mul_ps(Ok_re,w_im),_mm_mul_ps(Ok_im,w_re));
			_mm_store_ps((float *)(out+k), _mm_add_ps(Ek_re, wOk_re));
			_mm_store_ps((float *)(out+k+2), _mm_add_ps(Ek_im, wOk_im));
			_mm_store_ps((float *)(out+k+N/2), _mm_sub_ps(Ek_re, wOk_re));
			_mm_store_ps((float *)(out+k+N/2+2), _mm_sub_ps(Ek_im, wOk_im));
		}
	}
}
 
void fft_init(int N) {
	int i;
	
	#define log2(x) ((int)(log(x)/log(2)))
	
	int n_luts = log2(N)-2;
	LUT = malloc(n_luts * sizeof(data_t *));
	for(i=0;i<n_luts;i++) {
		int n = N / pow(2,i);
		LUT[i] = _mm_malloc(n/2 * sizeof(data_t), 16);
		
		
		int j;		
		for(j=0;j<n/2;j+=4) {
			data_t w[4];
			int k;
			for(k=0;k<4;k++) w[k] = W(n,j+k);
 
			LUT[i][j]   = creal(w[0]) + creal(w[1])*I;
			LUT[i][j+1] = creal(w[2]) + creal(w[3])*I;
			LUT[i][j+2] = cimag(w[0]) + cimag(w[1])*I;
			LUT[i][j+3] = cimag(w[2]) + cimag(w[3])*I;
		}
	}
	
}
Listing 2: Vectorized math functions for split-radix implementations
typedef struct _reg_t {
	__m128 re, im;
} reg_t;
 
static inline reg_t MUL(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_sub_ps(_mm_mul_ps(a.re,b.re),_mm_mul_ps(a.im,b.im));
	r.im = _mm_add_ps(_mm_mul_ps(a.re,b.im),_mm_mul_ps(a.im,b.re));
	return r;
}
static inline reg_t MULJ(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_add_ps(_mm_mul_ps(a.re,b.re),_mm_mul_ps(a.im,b.im));
	r.im = _mm_sub_ps(_mm_mul_ps(a.im,b.re),_mm_mul_ps(a.re,b.im));
	return r;
}
 
static inline reg_t ADD(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_add_ps(a.re,b.re);
	r.im = _mm_add_ps(a.im,b.im);
	return r;
}
static inline reg_t SUB(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_sub_ps(a.re,b.re);
	r.im = _mm_sub_ps(a.im,b.im);
	return r;
}
static inline reg_t ADD_I(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_sub_ps(a.re,b.im);
	r.im = _mm_add_ps(a.im,b.re);
	return r;
}
static inline reg_t SUB_I(reg_t a, reg_t b) {
	reg_t r;
	r.re = _mm_add_ps(a.re,b.im);
	r.im = _mm_sub_ps(a.im,b.re);
	return r;
}
 
static inline reg_t LOAD(float *a) {
	reg_t r;
	r.re = _mm_load_ps(a);
	r.im = _mm_load_ps(a+4);
	return r;
}
static inline void STORE(float *a, reg_t r) {
	_mm_store_ps(a, r.re);
	_mm_store_ps(a+4, r.im);
}
static inline void STOREIL(float *a, reg_t r) {
	_mm_store_ps(a, _mm_unpacklo_ps(r.re, r.im));
	_mm_store_ps(a+4, _mm_unpackhi_ps(r.re, r.im));
}
Listing 3: Split-radix FFT with vectorized loops
#include <math.h>
#include <complex.h>
#include <stdio.h>
#include <stdlib.h>
#include <xmmintrin.h>
 
typedef complex float data_t;
 
#define W(N,k) (cexp(-2.0f * M_PI * I * (float)(k) / (float)(N)))
data_t **LUT1;
data_t **LUT3;
 
#include "vecmath.h"
 
void splitfft(data_t *in, data_t *out,
					int log2stride, int stride, int N) {
	
	if(N == 1) {
		out[0] = in[0];
	}else if(N == 2) {
		out[0] = in[0] + in[stride];
		out[1] = in[0] - in[stride];
	}else if(N == 4) {
		splitfft(in, out, log2stride+1, stride << 1, N >> 1);
		splitfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  splitfft(in+3*stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
	
			data_t temp0 = out[0]  + (out[2] + out[3]);
			data_t temp1 = out[0]  - (out[2] + out[3]);
			data_t temp2 = out[1] - I*(out[2] - out[3]);
			data_t temp3 = out[1] + I*(out[2] - out[3]);
    if(log2stride) {
  		out[0] = creal(temp0) + creal(temp2)*I;
  		out[1] = creal(temp1) + creal(temp3)*I;
  		out[2] = cimag(temp0) + cimag(temp2)*I;
  		out[3] = cimag(temp1) + cimag(temp3)*I;
  	}else{
			out[0] = temp0;
			out[2] = temp1;
			out[1] = temp2;
			out[3] = temp3;
  	}
	
	}else if(N == 8) {
		splitfft(in, out, log2stride+1, stride << 1, N >> 1);
		splitfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  splitfft(in+3*stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
	
		data_t o[8];
		{
			data_t Uk  = creal(out[0]) + creal(out[2])*I;
			data_t Zk  = out[4];
			data_t Uk2  = creal(out[1]) + creal(out[3])*I;
			data_t Zdk = out[6];
			
			o[0] = Uk  + (Zk + Zdk);
			o[4] = Uk  - (Zk + Zdk);
			o[2] = Uk2 - I*(Zk - Zdk);
			o[6] = Uk2 + I*(Zk - Zdk);
		}
		{
			data_t Uk = cimag(out[0]) + cimag(out[2])*I;
			data_t Zk  = out[5];
			data_t Uk2 = cimag(out[1]) + cimag(out[3])*I;
			data_t Zdk = out[7];
			data_t w1 = LUT1[log2stride][1];
			data_t w3 = LUT3[log2stride][1];
			
			o[1] = Uk  + (w1*Zk + w3*Zdk);
			o[5] = Uk  - (w1*Zk + w3*Zdk);
			o[3] = Uk2 - I*(w1*Zk - w3*Zdk);
			o[7] = Uk2 + I*(w1*Zk - w3*Zdk);
		}
		if(log2stride) {
			out[0] = creal(o[0]) + creal(o[1])*I;
			out[1] = creal(o[2]) + creal(o[3])*I;
			out[2] = cimag(o[0]) + cimag(o[1])*I;
			out[3] = cimag(o[2]) + cimag(o[3])*I;
			out[4] = creal(o[4]) + creal(o[5])*I;
			out[5] = creal(o[6]) + creal(o[7])*I;
			out[6] = cimag(o[4]) + cimag(o[5])*I;
			out[7] = cimag(o[6]) + cimag(o[7])*I;
		}else{
			int i;
			for(i=0;i<8;i++) out[i] = o[i];
		}
	}else if(!log2stride){
 
		splitfft(in, out, log2stride+1, stride << 1, N >> 1);
		splitfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  splitfft(in+3*stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
		int k;	
		for(k=0;k<N/4;k+=4) {
			reg_t Uk  = LOAD((float *)&out[k]);
			reg_t Zk  = LOAD((float *)&out[k+N/2]);
			reg_t Uk2 = LOAD((float *)&out[k+N/4]);
			reg_t Zdk = LOAD((float *)&out[k+3*N/4]);
			reg_t w1 = LOAD((float *)&LUT1[log2stride][k]);
			reg_t w3 = LOAD((float *)&LUT3[log2stride][k]);
			
			reg_t w3Zdk = MUL(w3, Zdk);
			reg_t w1Zk = MUL(w1, Zk);
			reg_t sum = ADD(w1Zk, w3Zdk);
			reg_t dif = SUB(w1Zk, w3Zdk);
			
			STOREIL((float *)&out[k], ADD(Uk, sum));
			STOREIL((float *)&out[k+N/2], SUB(Uk, sum));
			STOREIL((float *)&out[k+N/4], SUB_I(Uk2, dif));
			STOREIL((float *)&out[k+3*N/4], ADD_I(Uk2, dif));
		}
 
	}else{
		splitfft(in, out, log2stride+1, stride << 1, N >> 1);
		splitfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  splitfft(in+3*stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
		
		int k;
		for(k=0;k<N/4;k+=4) {
			reg_t Uk  = LOAD((float *)&out[k]);
			reg_t Zk  = LOAD((float *)&out[k+N/2]);
			reg_t Uk2 = LOAD((float *)&out[k+N/4]);
			reg_t Zdk = LOAD((float *)&out[k+3*N/4]);
			reg_t w1 = LOAD((float *)&LUT1[log2stride][k]);
			reg_t w3 = LOAD((float *)&LUT3[log2stride][k]);
			
			reg_t w3Zdk = MUL(w3, Zdk);
			reg_t w1Zk = MUL(w1, Zk);
			reg_t sum = ADD(w1Zk, w3Zdk);
			reg_t dif = SUB(w1Zk, w3Zdk);
			
			STORE((float *)&out[k], ADD(Uk, sum));
			STORE((float *)&out[k+N/2], SUB(Uk, sum));
			STORE((float *)&out[k+N/4], SUB_I(Uk2, dif));
			STORE((float *)&out[k+3*N/4], ADD_I(Uk2, dif));
		}
	}
}
 
void fft_init(int N) {
	#define log2(x) ((int)(log(x)/log(2)))
	
	int n_luts = log2(N)-1;
	LUT1 = malloc(n_luts * sizeof(data_t *));
	LUT3 = malloc(n_luts * sizeof(data_t *));
	int i;
	for(i=0;i<n_luts;i++) {
		int n = N / pow(2,i);
		LUT1[i] = _mm_malloc(n/4 * sizeof(data_t),16);
		LUT3[i] = _mm_malloc(n/4 * sizeof(data_t),16);
		
		if(n == 8) {
			int j;
			for(j=0;j<n/4;j++) {
				LUT1[i][j] = W(n,j);
				LUT3[i][j] = W(n,3*j);
			}
		}else{
			int j;		
			for(j=0;j<n/4;j+=4) {
				data_t w1[4], w3[4];
				int k;
				for(k=0;k<4;k++) w1[k] = W(n,j+k);
				for(k=0;k<4;k++) w3[k] = W(n,3*(j+k));
 
				LUT1[i][j]   = creal(w1[0]) + creal(w1[1])*I;
				LUT1[i][j+1] = creal(w1[2]) + creal(w1[3])*I;
				LUT1[i][j+2] = cimag(w1[0]) + cimag(w1[1])*I;
				LUT1[i][j+3] = cimag(w1[2]) + cimag(w1[3])*I;
				LUT3[i][j]   = creal(w3[0]) + creal(w3[1])*I;
				LUT3[i][j+1] = creal(w3[2]) + creal(w3[3])*I;
				LUT3[i][j+2] = cimag(w3[0]) + cimag(w3[1])*I;
				LUT3[i][j+3] = cimag(w3[2]) + cimag(w3[3])*I;
			}
		}
	}
}
 
Listing 4: Conjugate-pair FFT with vectorized loops
#include <math.h>
#include <complex.h>
#include <stdio.h>
#include <stdlib.h>
#include <xmmintrin.h>
 
typedef complex float data_t;
 
#define W(N,k) (cexp(-2.0f * M_PI * I * (float)(k) / (float)(N)))
data_t **LUT1;
 
#include "vecmath.h"
 
data_t *base;
int TN;
 
void conjfft(data_t *in, data_t *out,
					int log2stride, int stride, int N) {
	
	if(N == 1) {
		if(in < base) in += TN;
		out[0] = in[0];
	}else if(N == 2) {
		data_t *i0 = in, *i1 = in + stride;
		if(i0 < base) i0 += TN;
		if(i1 < base) i1 += TN;
		out[0]   = *i0 + *i1;
		out[N/2] = *i0 - *i1;
	}else if(N == 4) {
		conjfft(in, out, log2stride+1, stride << 1, N >> 1);
		conjfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  conjfft(in-stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
	
			data_t temp0 = out[0]  + (out[2] + out[3]);
			data_t temp1 = out[0]  - (out[2] + out[3]);
			data_t temp2 = out[1] - I*(out[2] - out[3]);
			data_t temp3 = out[1] + I*(out[2] - out[3]);
    if(log2stride) {
  		out[0] = creal(temp0) + creal(temp2)*I;
  		out[1] = creal(temp1) + creal(temp3)*I;
  		out[2] = cimag(temp0) + cimag(temp2)*I;
  		out[3] = cimag(temp1) + cimag(temp3)*I;
  	}else{
			out[0] = temp0;
			out[2] = temp1;
			out[1] = temp2;
			out[3] = temp3;
  	}
	
	}else if(N == 8) {
		conjfft(in, out, log2stride+1, stride << 1, N >> 1);
		conjfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  conjfft(in-stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
	
		data_t o[8];
		{
			data_t Uk  = creal(out[0]) + creal(out[2])*I;
			data_t Zk  = out[4];
			data_t Uk2  = creal(out[1]) + creal(out[3])*I;
			data_t Zdk = out[6];
			
			o[0] = Uk  + (Zk + Zdk);
			o[4] = Uk  - (Zk + Zdk);
			o[2] = Uk2 - I*(Zk - Zdk);
			o[6] = Uk2 + I*(Zk - Zdk);
		}
		{
			data_t Uk = cimag(out[0]) + cimag(out[2])*I;
			data_t Zk  = out[5];
			data_t Uk2 = cimag(out[1]) + cimag(out[3])*I;
			data_t Zdk = out[7];
			data_t w1 = LUT1[log2stride][1];
			
			o[1] = Uk  + (w1*Zk + conj(w1)*Zdk);
			o[5] = Uk  - (w1*Zk + conj(w1)*Zdk);
			o[3] = Uk2 - I*(w1*Zk - conj(w1)*Zdk);
			o[7] = Uk2 + I*(w1*Zk - conj(w1)*Zdk);
		}
		if(log2stride) {
			out[0] = creal(o[0]) + creal(o[1])*I;
			out[1] = creal(o[2]) + creal(o[3])*I;
			out[2] = cimag(o[0]) + cimag(o[1])*I;
			out[3] = cimag(o[2]) + cimag(o[3])*I;
			out[4] = creal(o[4]) + creal(o[5])*I;
			out[5] = creal(o[6]) + creal(o[7])*I;
			out[6] = cimag(o[4]) + cimag(o[5])*I;
			out[7] = cimag(o[6]) + cimag(o[7])*I;
		}else{
			int i;
			for(i=0;i<8;i++) out[i] = o[i];
		}
	}else if(!log2stride){
 
		conjfft(in, out, log2stride+1, stride << 1, N >> 1);
		conjfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  conjfft(in-stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
		int k;	
		for(k=0;k<N/4;k+=4) {
			reg_t Uk  = LOAD((float *)&out[k]);
			reg_t Zk  = LOAD((float *)&out[k+N/2]);
			reg_t Uk2 = LOAD((float *)&out[k+N/4]);
			reg_t Zdk = LOAD((float *)&out[k+3*N/4]);
			reg_t w1 = LOAD((float *)&LUT1[log2stride][k]);
			
			reg_t w3Zdk = MULJ(Zdk, w1);
			reg_t w1Zk = MUL(w1, Zk);
			reg_t sum = ADD(w1Zk, w3Zdk);
			reg_t dif = SUB(w1Zk, w3Zdk);
			
			STOREIL((float *)&out[k], ADD(Uk, sum));
			STOREIL((float *)&out[k+N/2], SUB(Uk, sum));
			STOREIL((float *)&out[k+N/4], SUB_I(Uk2, dif));
			STOREIL((float *)&out[k+3*N/4], ADD_I(Uk2, dif));
		}
 
	}else{
		conjfft(in, out, log2stride+1, stride << 1, N >> 1);
		conjfft(in+stride, out+N/2, log2stride+2, stride << 2, N >> 2);
	  conjfft(in-stride, out+3*N/4, log2stride+2, stride << 2, N >> 2);
		
		int k;
		for(k=0;k<N/4;k+=4) {
			reg_t Uk  = LOAD((float *)&out[k]);
			reg_t Zk  = LOAD((float *)&out[k+N/2]);
			reg_t Uk2 = LOAD((float *)&out[k+N/4]);
			reg_t Zdk = LOAD((float *)&out[k+3*N/4]);
			reg_t w1 = LOAD((float *)&LUT1[log2stride][k]);
			
			reg_t w3Zdk = MULJ(Zdk, w1);
			reg_t w1Zk = MUL(w1, Zk);
			reg_t sum = ADD(w1Zk, w3Zdk);
			reg_t dif = SUB(w1Zk, w3Zdk);
			
			STORE((float *)&out[k], ADD(Uk, sum));
			STORE((float *)&out[k+N/2], SUB(Uk, sum));
			STORE((float *)&out[k+N/4], SUB_I(Uk2, dif));
			STORE((float *)&out[k+3*N/4], ADD_I(Uk2, dif));
		}
	}
}
 
void fft_init(int N) {
	#define log2(x) ((int)(log(x)/log(2)))
	int n_luts = log2(N)-1;
	LUT1 = malloc(n_luts * sizeof(data_t *));
	int i;
	for(i=0;i<n_luts;i++) {
		int n = N / pow(2,i);
		LUT1[i] = _mm_malloc(n/4 * sizeof(data_t),16);
		
		if(n == 8) {
			int j;
			for(j=0;j<n/4;j++) {
				LUT1[i][j] = W(n,j);
			}
		}else{
			int j;		
			for(j=0;j<n/4;j+=4) {
				data_t w1[4];
				int k;
				for(k=0;k<4;k++) w1[k] = W(n,j+k);
 
				LUT1[i][j]   = creal(w1[0]) + creal(w1[1])*I;
				LUT1[i][j+1] = creal(w1[2]) + creal(w1[3])*I;
				LUT1[i][j+2] = cimag(w1[0]) + cimag(w1[1])*I;
				LUT1[i][j+3] = cimag(w1[2]) + cimag(w1[3])*I;
			}
		}
	}
	TN = N;
 
}
 

Collection Navigation

Content actions

Download:

Collection as:

PDF | EPUB (?)

What is an EPUB file?

EPUB is an electronic book format that can be read on a variety of mobile devices.

Downloading to a reading device

For detailed instructions on how to download this content's EPUB to your specific device, click the "(?)" link.

| More downloads ...

Module as:

PDF | EPUB (?)

What is an EPUB file?

EPUB is an electronic book format that can be read on a variety of mobile devices.

Downloading to a reading device

For detailed instructions on how to download this content's EPUB to your specific device, click the "(?)" link.

| More downloads ...

Add:

Collection to:

My Favorites (?)

'My Favorites' is a special kind of lens which you can use to bookmark modules and collections. 'My Favorites' can only be seen by you, and collections saved in 'My Favorites' can remember the last module you were on. You need an account to use 'My Favorites'.

| A lens I own (?)

Definition of a lens

Lenses

A lens is a custom view of the content in the repository. You can think of it as a fancy kind of list that will let you see content through the eyes of organizations and people you trust.

What is in a lens?

Lens makers point to materials (modules and collections), creating a guide that includes their own comments and descriptive tags about the content.

Who can create a lens?

Any individual member, a community, or a respected organization.

What are tags? tag icon

Tags are descriptors added by lens makers to help label content, attaching a vocabulary that is meaningful in the context of the lens.

| External bookmarks

Module to:

My Favorites (?)

'My Favorites' is a special kind of lens which you can use to bookmark modules and collections. 'My Favorites' can only be seen by you, and collections saved in 'My Favorites' can remember the last module you were on. You need an account to use 'My Favorites'.

| A lens I own (?)

Definition of a lens

Lenses

A lens is a custom view of the content in the repository. You can think of it as a fancy kind of list that will let you see content through the eyes of organizations and people you trust.

What is in a lens?

Lens makers point to materials (modules and collections), creating a guide that includes their own comments and descriptive tags about the content.

Who can create a lens?

Any individual member, a community, or a respected organization.

What are tags? tag icon

Tags are descriptors added by lens makers to help label content, attaching a vocabulary that is meaningful in the context of the lens.

| External bookmarks