IA307

GPU Programming for Deep Learning

Programmation GPU pour l'apprentissage

Goran Frehse, ENSTA Paris

05/02/2023:

Course: Linear Classification
Lab: Implementing Essential Matrix Operations (GRADED!, submit before Fri 9/2/2023 at 16:30)
Parallelizing sums : see slides 37ff on dot product here (simply omit the product part)

9/02/2023 12:30-15:30:

Lab: Linear Classification (GRADED!, submit before Mon 26/02/2023 at 12:30)

compare with results on Tensorflow here.

(no quizz today)

Erratum : COLAB has changed the directory structure. If you get an error message 

fatal error: helper_cuda.h: No such file or directory

then you need to update the include directory in the command line. Instead of 

!nvcc -g -G -I /usr/local/cuda/samples/common/inc/ ... (broken)

you need to download the files with

  !git clone https://github.com/NVIDIA/cuda-samples.git

and then use

!nvcc -g -G -I cuda-samples/Common/ ... (should work)


Erratum : Please replace

int fmatrix_size(fmatrix mat) {

           fmatrix_assert(mat); // bad: throws if data not allocated
  return fmatrix_elements(mat) * sizeof(mat.data[0]);  // bad: throws if data not allocated

}

   by

int fmatrix_size(fmatrix mat) {

      return fmatrix_elements(mat) * sizeof(float); // ok, always works

}

26/02/2023:

Course: Using fmatrix in Python

    From linear classifcation to neural networks (link to Caltech CS 179)

     Quizz (submit before 26/02/2022 12h30)


1/03/2023:

Lab: Neural Networks (ungraded)



Extra material:

FAQ for Colab and CUDA