Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Altera_Forum
Honored Contributor I
908 Views

writing and reading the same pipe in one kernel

Hi  

I want to write and read the same pipe in one kernel like this: 

 

channel float c0 __attribute__((depth(2))); 

__kernel void testkernel() { 

 

for(int i = 0; i < n; i ++){ 

page_ranks = read_channel_intel(c0); //read from the channel 

 

for(int i = 0; i < n; i ++){ 

//do some computation of the page_rank and generage new_rank 

write_channel_intel(c0, new_rank); //write new_rank to this pipe 

 

 

and this kernel is being run in multiple iterations, each time it will read from last iteration's result and compute new rank and use the new rank as the new input for the next iteration. 

When I compile this, I get the error of "Compiler Error: Multiple writes to channel c0", this is the only line I got in the log.  

Any suggestions on how to deal with this case using pipe or how to resolve this bug?
0 Kudos
3 Replies
Altera_Forum
Honored Contributor I
24 Views

Are you unrolling your loops? Post your complete kernel since the snippet you have posted should compile correctly.

Altera_Forum
Honored Contributor I
24 Views

 

--- Quote Start ---  

Are you unrolling your loops? Post your complete kernel since the snippet you have posted should compile correctly. 

--- Quote End ---  

 

Below is my kernel code: 

# define M 100 

channel float c0 __attribute__((depth(2))); 

 

 

__kernel void mapreduce_page_rank(__global float* restrict page_ranks, 

int n, 

const __global int* restrict pages, 

const __global unsigned int* restrict noutlinks){ 

 

 

int i, j, t; 

float new_rank; 

float d_factor = 0.85; 

float cur; 

float maps; 

float outbound_rank; 

float new_rank_copies[M]; 

 

 

# pragma unroll 

for (i = 0; i < M; i ++){ 

new_rank_copies = 0; 

 

 

for(i=0; i<n; ++i){ 

new_rank = 0.0; 

for(j=0; j<n; ++j){ 

outbound_rank = page_ranks[j] / (double)noutlinks[j]; 

maps = pages[i*n+j] * outbound_rank; 

cur = new_rank_copies[m-1] + maps; 

# pragma unroll 

for(int c = m-1; c > 0; c--){ 

new_rank_copies[c] = new_rank_copies[c - 1]; 

new_rank_copies[0] = cur; 

# pragma unroll 

for(int j = 0; j < m; j ++){ 

new_rank += new_rank_copies[j]; 

new_rank_copies[j] = 0; 

 

 

new_rank = ((1-d_factor)/n)+(d_factor*new_rank); 

write_channel_altera(c0, new_rank); 

 

 

 

 

__kernel void produce_page_rank(__global float* restrict page_ranks, 

int n, 

const __global int* restrict pages, 

const __global unsigned int* restrict noutlinks){ 

 

 

int i, j, t; 

float new_rank; 

float d_factor = 0.85; 

float cur; 

float maps; 

float outbound_rank; 

float new_rank_copies[m]; 

 

 

# pragma unroll 

for (i = 0; i < m; i ++){ 

new_rank_copies = 0; 

 

 

for(int i = 0; i < n; i ++){ 

page_ranks[i] = read_channel_altera(c0); 

 

 

for(i=0; i<n; ++i){ 

new_rank = 0.0; 

for(j=0; j<n; ++j){ 

outbound_rank = page_ranks[j] / (double)noutlinks[j]; 

maps = pages[i*n+j] * outbound_rank; 

cur = new_rank_copies[M-1] + maps; 

# pragma unroll 

for(int c = M-1; c > 0; c--){ 

new_rank_copies[c] = new_rank_copies[c - 1]; 

new_rank_copies[0] = cur; 

# pragma unroll 

for(int j = 0; j < M; j ++){ 

new_rank += new_rank_copies[j]; 

new_rank_copies[j] = 0; 

 

 

new_rank = ((1-d_factor)/n)+(d_factor*new_rank); 

write_channel_altera(c0, new_rank); 

}
Altera_Forum
Honored Contributor I
24 Views

Well, you are clearly writing to the same channel twice, once in your first kernel and the second time in your second kernel. This is not allowed since each channel works as a FIFO with one read point and one write point. However, I think the newer versions of the compiler (v17 or v17.1) allow multiple channel call sites but I haven't tried it myself.

Reply