0

I have a very Complicated symbolic function in Matlab and I want to evaluate it on GPU.

As I searched, for evaluation, we need to use gpuArray but when I try to evaluate function on GPU matlab using this code : 

joint_pdf_function_handle = @(x1,x2)sqrt(1.99e+2).*1.0./pi.^2.*exp(2.4e+1./2.5e+1).*exp(sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x1).*2.0).*(sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x1).*2.0).*4.925125628140704e+1-sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x2).*2.0).*4.974874371859296e+1).*(-1.0./2.0)+(sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x2).*2.0).*(sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x1).*2.0).*4.974874371859296e+1-sqrt(2.0).*erfcinv(integral(@(x)(1.0./sqrt(x.^2+9.0).*besselk(1,sqrt(x.^2+9.0).*(4.0./2.5e+1)).*7.757157130525889e-1)./pi,-Inf,x2).*2.0).*4.925125628140704e+1))./2.0).*1.0./sqrt(x1.^2+9.0).*1.0./sqrt(x2.^2+9.0).*besselk(1,sqrt(x1.^2+9.0).*(4.0./2.5e+1)).*besselk(1,sqrt(x2.^2+9.0).*(4.0./2.5e+1)).*1.157788944723618e-1
x1_space = gpuArray.linspace(-10,10,25)  ;
x2_space = gpuArray.linspace(-10,10,25) ; 
[X Y] = meshgrid(x1_space,x2_space) ; 
F = arrayfun(joint_pdf_function_handle ,X, Y)  ;

then I get following error : 

The function 'symengine' cannot be executed because it has resolved to the P-code file:
C:\Program Files\Polyspace\R2019a\toolbox\symbolic\symbolic\symengine.p

the thing is if I run code on CPU like this, it works fine : 

x1_space = linspace(-10,10,25)  ;
x2_space = linspace(-10,10,25) ; 
[X Y] = meshgrid(x1_space,x2_space) ; 
F = arrayfun(joint_pdf_function_handle ,X, Y)  ;

I would appreciate any help you can provide.

Captain
  • 193
  • 1
  • 10
  • You are using symbolic math, which won’t run on the GPU. Not sure which of your functions does this, though. Try splitting up your function and running parts separately, see which ones cause errors. – Cris Luengo Sep 02 '19 at 18:22
  • You should also try reducing code duplication inside your large function by defining anonymous functions for parts of it that occur multiple times. – Cris Luengo Sep 02 '19 at 18:23

0 Answers0