damico

import torch, grp, pwd, os, subprocess
devices = []
try:
	print("\n\nChecking ROCM support...")
	result = subprocess.run(['rocminfo'], stdout=subprocess.PIPE)
	cmd_str = result.stdout.decode('utf-8')
	cmd_split = cmd_str.split('Agent ')
	for part in cmd_split:
		item_single = part[0:1]
		item_double = part[0:2]

Chillee

import torch
import torch._inductor.config
import time

torch._inductor.config.triton.cudagraphs = False
torch.set_float32_matmul_precision('high')

def bench(f, name=None, iters=100, warmup=5, display=True, profile=False):
    for _ in range(warmup):
        f()

ConsciousMachines

// based on https://gist.github.com/kamino410/09df4ecdf37b03cbd05752a7b2e52d3a
// this adds ImGui to an application with CUDA and OpenGL. the thing is, once you use CUDA, ImGui renders very strangely. 
// after 9 hours of debugging I found that putting glBindBuffer before and after the draw call fixes this!
//glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo); // THE MAGIC LINE #1 
//glDrawPixels(WIDTH, HEIGHT, GL_RGBA, GL_UNSIGNED_BYTE, 0);
//glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);   // THE MAGIC LINE #2


#include "imgui/imgui.h" // version 1.78 and 1.60
#include "imgui/imgui_impl_glfw.h"

hayunjong83

#include <iostream>
#include <cstdio>
#include <cstdlib>
#include <helper_cuda.h>
#include <helper_string.h>

__device__ int g_uids = 0;

__device__ void print_info(int depth, int thread, int uid, int parent_uid)
{

maronsson

//
// Basic instrumentation profiler by Cherno

// Usage: include this header file somewhere in your code (eg. precompiled header), and then use like:
//
// Instrumentor::Get().BeginSession("Session Name");        // Begin session 
// {
//     InstrumentationTimer timer("Profiled Scope Name");   // Place code like this in scopes you'd like to include in profiling
//     // Code
// }

TheCherno

//
// Basic instrumentation profiler by Cherno

// Usage: include this header file somewhere in your code (eg. precompiled header), and then use like:
//
// Instrumentor::Get().BeginSession("Session Name");        // Begin session 
// {
//     InstrumentationTimer timer("Profiled Scope Name");   // Place code like this in scopes you'd like to include in profiling
//     // Code
// }

goldsborough

#include <cudnn.h>
#include <cassert>
#include <cstdlib>
#include <iostream>
#include <opencv2/opencv.hpp>

#define checkCUDNN(expression)                               \
  {                                                          \
    cudnnStatus_t status = (expression);                     \
    if (status != CUDNN_STATUS_SUCCESS) {                    \