Archives Discussions

[pavandsp]

Hi

I am using a lookup table in the kernel and defined in the kernel itself.It seems the kernel is using rightly this table in CPU mode but not in GPU mode...shouldn't we define  a tables like this  in kernel in GPU...or we should send the table as an arg to the kernel.

I defined the table something like this is .cl file

const unsigned char lookup_table[256]={.....};

I am getting weired results in GPU mode when accessing the table.

Please confirm

[nou]

i use lookup table too and i do not have any problem.

[pavandsp]

Hi nou,

I am still facing the issue in GPU mode ..,please find the part of the kernel code where lookup table is used.

                output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 0] = lookup_table[output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 0]];
                output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 1] = lookup_table[output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 1]];
                output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 2] = lookup_table[output[(ty - 1) * 3 * lineLen + (tx - 1) * 3 + 2]];

I am not sure why this is not working properly and guess there is some issue with the lookup_table definition and usage in kernel.

[kbrafford]

Out of curiosity, does the CL compiler do common subexpression elimination?  Does the expression ((ty - 1) * 3 * lineLen + (tx - 1) * 3) get handled efficiently?

[davibu]

Originally posted by: pavandsp

const unsigned char lookup_table[256]={.....};

Shouldn't be "constant" instead of "const" ?