Mpi base implementation #31
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Cuda staging
rozukke
requested changes
Jul 6, 2024
Comment on lines
+17
to
+22
| n_gpus=$(shell nvidia-smi --query-gpu=name --format=csv,noheader | wc -l); \ | ||
| mpirun --oversubscribe -np $$n_gpus ./speed_gpu ./weights_and_biases.txt ./tensors 100000 | ||
|
|
||
| test: build | ||
| ./speed_gpu ./weights_and_biases.txt ./tensors 1000000 | ||
| n_gpus=$(shell nvidia-smi --query-gpu=name --format=csv,noheader | wc -l); \ | ||
| mpirun --oversubscribe -np $$n_gpus ./speed_gpu ./weights_and_biases.txt ./tensors 1000000 |
There was a problem hiding this comment.
We do not need to oversubscribe on the final build.
| @@ -1,6 +1,7 @@ | |||
| #include "matrix.cuh" | |||
| #include <dirent.h> | |||
| #include <iostream> | |||
| @@ -156,11 +157,23 @@ __global__ void infer(float* d_inputs, int* d_results, matrix** d_weights, matri | |||
| } | |||
|
|
|||
| int main(int argc, char* argv[]) { | |||
| MPI_Init(&argc, &argv); | |||
| int TotalProcess, ProcessId; | |||
Comment on lines
+223
to
+224
| int local_input_count = input_count / TotalProcess + (ProcessId < (input_count % TotalProcess) ? 1 : 0); | ||
| int start_idx = ProcessId * (input_count / TotalProcess) + std::min(ProcessId, input_count % TotalProcess); |
There was a problem hiding this comment.
Why is there weird capitalised naming and C++ functions
Comment on lines
+233
to
+245
| int counter = 0; | ||
| while ((entry = readdir(dir)) != NULL) { | ||
| if (entry->d_type == DT_REG) { | ||
| strcpy(file_num_str, entry->d_name); | ||
| file_num_str[strlen(entry->d_name) - 7] = '\0'; | ||
| file_num = atoi(entry->d_name); | ||
| strcpy(file_name, directory_path); | ||
| strcat(file_name, "/"); | ||
| strcat(file_name, entry->d_name); | ||
| read_tensor((float*)&inputs[(file_num - 1) * 225], file_name); | ||
| if (file_num >= start_idx + 1 && file_num <= start_idx + local_input_count) { | ||
| strcpy(file_name, directory_path); | ||
| strcat(file_name, "/"); | ||
| strcat(file_name, entry->d_name); | ||
| read_tensor(&inputs[counter * 225], file_name); | ||
| counter++; | ||
| } |
There was a problem hiding this comment.
Do not divide inputs between gpus, divide inferences
Comment on lines
-275
to
+262
| for (int i = 0; i < input_count; i++) { | ||
| for (int i = 0; i < local_input_count; i++) { | ||
| infer<<<BLOCKS, THREADS_PER_BLOCK>>>(d_inputs, d_results, d_weights, d_biases, it_num, i); | ||
| err = cudaGetLastError(); | ||
| if (err != cudaSuccess) { | ||
| printf("CUDA error: %s\n", cudaGetErrorString(err)); | ||
| } | ||
| CUDA_CHECK(cudaGetLastError()); |
There was a problem hiding this comment.
Rearrange to divide more healthily
Comment on lines
+291
to
+292
| printf("Process %d - Total: %lu us\n", ProcessId, | ||
| (stop.tv_sec - start.tv_sec) * 1000000 + stop.tv_usec - start.tv_usec); |
There was a problem hiding this comment.
Also want timing from the root process for how long the whole run takes
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Not tested with actual 8 gpu. feel free to pr for changes