codedk/cuda-by-example-source-code-for-the-book-s-examples-

CUDA by Example, written by two senior members of the CUDA software platform team, shows programmers how to employ this new technology. The authors introduce each area of CUDA development through working examples.

Hi there 👋, I'm Panos

I am a senior software engineer from Greece, passionate about programming,
Artificial Intelligence 🧠, and Startups 🚀

• 🔥 CTO with 5+ years of experience and proven leadership skills. Senior Software Engineer for 15+ years.
  
• ⚡ Exploring latest research in machine learning, metaheuristics algorithms and security.
  
• 🏖️ In my free time, I code for my Phd, solving optimization problems combining metaheuristics and ML.
  
• 🔀How to reach me:  
  
• 🧪 Check my research here:
  
  

Languages

---

Databases

---

Machine learning frameworks and libraries

---

Web stack

---

Backend frameworks

cuda-by-example-source-code-for-the-book-s-examples-'s Issues

Chapter 3, enum_gpu.cu -- Print issue

Good morning!

I have found a small issue with the printing of the CUDA devices in the enum_gpu.cu file.

The issue is: prop.totalGlobalMem is printed as a long int, with the format specifier %ld. However, by looking at the cudaDeviceProp struct, that field is of type size_t, which in turn is defined as unsigned long long.

Thus, the correct format specifier for printing this value should be %llu. I noticed it when the code reported I had negative memory on my GPU...

question about gpu hash table

https://github.com/CodedK/CUDA-by-Example-source-code-for-the-book-s-examples-/blob/master/appendix_a/hashtable_gpu.cu#L144C1-L144C61

should
HANDLE_ERROR( cudaMalloc((void**)&dev_values, SIZE) );
be
HANDLE_ERROR( cudaMalloc((void**)&dev_values, SIZE / sizeof(unsigned int) * sizeof(void*) ) );
?
after all, sizeof(unsigned int) != sizeof(void*).

Lock.h -- Doubt about __threadfence()

Good morning!

First of all, I have just finished the book, and I must say, it's a great resource! If only all books to explain a technology included examples such as in this one...

Now, I have a doubt about lock.h. In the code of this repo, after atomicCAS and before atomicExch, there is a call to a method named __threadfence(). I googled this method, and apparently it ensures that all writes to global memory before this call actually happen before all writes after the call, among all blocks. But I don't know how this works to "fix" the code. Is it because the mutex is in global memory, and its value must be updated accordingly?

Thanks a lot 😊!

This part seems weird?

CUDA-by-Example-source-code-for-the-book-s-examples-/chapter05/dot.cu

Lines 33 to 36 in 78d24d4

	while (tid < N) {
	temp += a[tid] * b[tid];
	tid += blockDim.x * gridDim.x;
	}

fatal error LNK1104: cannot open file 'glut64.lib'