zhuangh

""" 
baseline runtime(s) 0.5243263244628906
with reshape runtime (s) 0.0022399425506591797
@ cpu
=========
baseline runtime (s) 0.25386476516723633
with reshape runtime (s) 0.0008966922760009766
@ cuda:0
"""
import torch

zhuangh

// https://marketplace.visualstudio.com/items?itemName=coolchyni.beyond-debug
{
  "inputs": [
    {
      "id": "hostname",
      "description": "xxx",
      "default": "localhost",
      "type": "promptString"
    },
    {

zhuangh

# python single_gpu_ddp.py
# https://discuss.pytorch.org/t/single-machine-single-gpu-distributed-best-practices/169243
import torch
import torch.distributed as dist
import torch.nn as nn
import torch.multiprocessing as mp
from torch.nn.parallel import DistributedDataParallel as DDP
import os

def setup(rank, world_size):

zhuangh

IR module {
  tt.func public @matmul_kernel_0d1d2d3d4c5d6c7d8c(%arg0: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg1: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg2: !tt.ptr<f32> {tt.divisibility = 16 : i32}, %arg3: i32 {tt.divisibility = 16 : i32}, %arg4: i32 {tt.divisibility = 16 : i32}, %arg5: i32 {tt.divisibility = 16 : i32}) attributes {noinline = false} {
    %c16_i32 = arith.constant 16 : i32
    %c1024_i32 = arith.constant 1024 : i32
    %c0_i32 = arith.constant 0 : i32
    %cst = arith.constant dense<16> : tensor<16x16xi32>
    %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x16xf32>
    %0 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32>
    %1 = tt.expand_dims %0 {axis = 1 : i32} : (tensor<16xi32>) -> tensor<16x1xi32>
    %2 = tt.splat %arg3 : (i32) -> tensor<16x1xi32>

zhuangh

import torch
import triton
import triton.language as tl
import torch.nn.functional as F

@triton.jit
def matmul_kernel(
    a_ptr, b_ptr, c_ptr,
    stride_am, stride_ak,
    stride_bk, stride_bn,

zhuangh

import torch
import triton
import triton.language as tl
import torch.nn.functional as F
import time

@triton.jit
def add_kernel(x_ptr, y_ptr, output_ptr, N,
               BLOCK_SIZE: tl.constexpr):
    pid = tl.program_id(0)

zhuangh

import torch

# acknowledgement: https://gist.github.com/bwasti/7e4cb9bd1aaddeb09bd360b570a486b1

def cudagraph(f):
    _graphs = {}
    def f_(*args):
        key = hash(tuple(tuple(a.shape) for a in args))
        if key in _graphs:
            wrapped, *_ = _graphs[key]

zhuangh

class Vector2D {
private:
    vector<vector<int>>::iterator row, iBegin, iEnd;
    vector<int>::iterator col;
public:
    Vector2D(vector<vector<int>>& vec2d) {
        iBegin = row = vec2d.begin(); 
        iEnd = vec2d.end();
        if(vec2d.size())
            col = row->begin();

zhuangh

#include<string>
#include<iostream>
#include<vector>

using namespace std;

class ShortPalindromes{
public:
	
	string solve(const string & s, int i, int j, 

zhuangh

/**
 * Definition for an interval.
 * struct Interval {
 *     int start;
 *     int end;
 *     Interval() : start(0), end(0) {}
 *     Interval(int s, int e) : start(s), end(e) {}
 * };
 */