coffeebeforearch / cuda_programming Goto Github PK

How do I modify code to accommodate arrays of arbitrary length instead of powers of 2 in sumReduction?

Please note that when I attempt to run your ~/1d_constant_memory/convolution.cu with CUDA compilation tools, release 7.5, V7.5.17 on Ubuntu 16.04, I am getting these errors.

~/Resources/Github/cuda_programming/convolution/1d_constant_memory $ nvcc convolution.cu -o convolution.x 
~/Resources/Github/cuda_programming/convolution/1d_constant_memory $ ./convolution.x 
convolution.x: convolution.cu:58: void verify_result(int*, int*, int*, int): Assertion `temp == result[i]' failed.
aborted (core dumped)

Do you have any suggestions? TIA.

Hi, I am using the following system configuration:

• Windows 10
• Visual Studio 2019 Community
• Cuda 10.2
• Nvidia Nsight Compute 2019.5.0
• Nvidia RTX 2060 GPU (Turing Architecture)

I am following your tutorials on YouTube and used the file alignment_matrix_mul.cu, in three configuartions:

• No transpose (just as we were doing it before)
• Transpose a matrix (temp_sum += a[k * n + row] * b[col + n * k];)
• Transpose b matrix (temp_sum += a[k + n * row] * b[col * n + k];)

We would expect that the GPU would perform best when we transpose matrix a, as the memory accesses for each thread are coalesced in this way, but the profiling shows that it performs better when I transpose matrix b.

The only thing that I am doing different here is that I am using Nsight Compute as a separate application to profile the built binary from Visual Studio and not the inbuilt extension. I am also attaching the performance images I got:

• No Transpose: https://drive.google.com/file/d/18-l8W3csIjCRRoxgASsWevjRIV9hINXp/view?usp=sharing
• Transpose a matrix: https://drive.google.com/file/d/1rPwMpalSwfVpZ8-jBpO3ROL1R7POAzRt/view?usp=sharing
• Transpose b matrix: https://drive.google.com/file/d/1WHIQBRRk1KjJk5MXVUc4AopGzqWPDwFh/view?usp=sharing

I have double checked the transpositions and this is what I get. Can there be any other bottleneck causing these results? i.e. the cost of fetching multiple elements for the loop (index k) overpowers the coalesced access?

Hi @CoffeeBeforeArch,

I'm not sure where else to put this, as I don't have a twitter account and don't plan on getting one + verified just to get in touch.

Is there any chance you do consulting? If so, would you be open to being hired for a few hours to help me optimize a specific kernel?

Thanks very much,
Ian

I am using

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

running the example return :

vectorAdd.cu:24: void verify_result(std::vector<int>&, std::vector<int>&, std::vector<int>&): Assertion `c[i] == a[i] + b[i]' failed.
[1]    49753 abort (core dumped)  ./prog

if I print few first elements of a, b, c after running kernel and copy data from device to host it returns all elements of c is zero.

The environment:
Windows 10
VS 2017
CUDA 10.1

The snapshot of this error:
https://drive.google.com/open?id=1V-Mv2xk9Leny3GEUYsMkBWBHumWastxU

However,
run the same code in Linux environment, there is no error.

Any help is appreciated.

Shouldn't the second call be <<<1, GRID_SIZE>>> instead of <<<1, TB_SIZE>>>? I think GRID_SIZE is the number of partial sums.

I found that a TB_SIZE and SH_SIZE= 128, won't give the expected 65536 result.

Hi nick, i'm trying to profile the matrix multiplication cuda code, it is the same as your naive matrix multiplication code with nvidia Nsight, i tried with 1<<10 and it worked, and i tried with 1<<11 and the profiler didn't catch the kernel launch. i have nvidia GTX 960M gpu.
so is it the problem with my gpu capability or there is something else wrong?
thanks in advance.

where is the CUDA Crash Course (v3) series ? i can't find it on youtube😢

Hi I'm trying to test my GPGPU Sim build and I'm following the blog https://coffeebeforearch.github.io/2020/03/30/gpgpu-sim-1.html
On following the instructions, I'm getting
mmul: mmul.cu:42: void verify_result(std::vector&, std::vector&, std::vector&, int): Assertion `tmp == c[i * N + j]' failed.
Aborted (core dumped)

My configuration-
intel i5-8265U
8 GB DDR4 RAM
Ubuntu 20.04 LTS
gcc 10.2
Cuda build version - 11.2
Please help

I can only find a link for CUDA Crash Course and CUDA Crash Course (v2). Is there a link for the CUDA Crash Course (v3) somewhere?

Thank you so much for the invaluable content. That is extremely helpful.

Hey Nick,
I am getting an error in / Boundary Check
if (tid < N) c[tid] = a[tid] + b[tid];
but when i corrected according to vs suggestion to / Boundary Check
if (tid < N) c[tid] == a[tid] + b[tid];
then it says warning #174-D: expression has no effect
1> if (tid < N) c[tid] == a[tid] + b[tid];
However, after this warning, it showed "completed successfully" .
I used the VS-2022 community. Can you please explain what happened, I am still confused

Also, when I compile the same program with nvcc on WSL2-ubuntu 20.2 it says
vector-add.cu(26): warning #174-D: expression has no effect
but the compilation was done and on running the output file give this -
vector-add: vector-add.cu:33: void verify_result(std::vector&, std::vector&, std::vector&): Assertion `c[i] == a[i] + b[i]' failed.
Aborted

Can you please help me out with the same

Regards
Pronod