r46mht's recent timeline updates r46mht's recent timeline updates |
r46mhtV2EX member #656376, joined on 2023-10-25 21:46:50 +08:00Today's activity rank 4597 |
| 加密货币是否会导致婚姻内被爆金币 生活 • r46mht • Apr 29, 2025 • Lastly replied by yesha | 19 |
r46mht's recent replies
5h 36m ago Replied to a topic by Radeon › 吐槽 › 中文互联网最好笑的事情之一是和某个西方人不是朋友，非要用名字而不是姓称呼对方，比如把 Amodei 说成 Dario，把 Sutskever 说成 Ilya。而且几乎人人这样说，包括有海外经验的播客博主。也是服了 |
科比布莱恩特是不是不应该叫科比，应该叫牢大 lol
Sep 7 Replied to a topic by studyingss › 生活 › 我发现一句对人生帮助巨大的 prompt， GPT 给我的回答好到我打算拿来当座右铭 |
人体呼吸的核心驱动力来自于排除二氧化碳而非补充氧气，因此高空缺氧发生时，身体并不会产生窒息感警报，而会在无痛且不知不觉中失去意识。
Jul 6 Replied to a topic by rungo › Python › 为什么这段 cuda 的并行前缀算法不会数据竞争 |
理论上可能会数据竞争，实际上数据竞争不太可能。如果你刻意的让一个 thread 等一等再算，就有可能看到 undfined 的结果：
```
__global__ void naive_prefix(double *vec, double *out)
{
__shared__ double sum_buf[1024];
int tid = threadIdx.x;
sum_buf[tid] = vec[tid];
int iter = 1;
for (int i=0; i < 10; i++)
{
__syncthreads();
// ARTIFICIAL DIVERGENCE:
// Force even-numbered threads to stall for 500,000 clock cycles.
// Odd-numbered threads will blow past this and execute the math below.
if (tid % 2 == 0) {
long long start = clock64();
while(clock64() - start < 500000) {}
}
if (tid >= iter )
{
// The odd threads will update sum_buf before the even threads even reach this line.
// When the even threads finally wake up, they will read corrupted, already-updated data.
sum_buf[tid]= sum_buf[tid] + sum_buf[tid - iter];
}
iter *= 2;
}
__syncthreads();
out[tid] = sum_buf[tid];
}
```
我猜 shared_memory 写入要比从 cache 读慢的多的多，并且 GPU thread 之间不像 CPU 一样有 cache coherence 的，所以直接读基本上不可能读到其他 thread 写入的结果。
```
__global__ void naive_prefix(double *vec, double *out)
{
__shared__ double sum_buf[1024];
int tid = threadIdx.x;
sum_buf[tid] = vec[tid];
int iter = 1;
for (int i=0; i < 10; i++)
{
__syncthreads();
// ARTIFICIAL DIVERGENCE:
// Force even-numbered threads to stall for 500,000 clock cycles.
// Odd-numbered threads will blow past this and execute the math below.
if (tid % 2 == 0) {
long long start = clock64();
while(clock64() - start < 500000) {}
}
if (tid >= iter )
{
// The odd threads will update sum_buf before the even threads even reach this line.
// When the even threads finally wake up, they will read corrupted, already-updated data.
sum_buf[tid]= sum_buf[tid] + sum_buf[tid - iter];
}
iter *= 2;
}
__syncthreads();
out[tid] = sum_buf[tid];
}
```
我猜 shared_memory 写入要比从 cache 读慢的多的多，并且 GPU thread 之间不像 CPU 一样有 cache coherence 的，所以直接读基本上不可能读到其他 thread 写入的结果。
Jun 29 Replied to a topic by liu11onepoint › 户外运动 › 有人喜欢露营看日出吗？ |
其实还是更喜欢露营睡大觉 lol
Jun 8 Replied to a topic by zhaoyeye › 问与答 › 不要使用国产手机登陆推特 |
刑不可知 则威不可测
Apr 16 Replied to a topic by yujianfei › Claude › CLaude 说是后面要实名了 |
Claude 打击违规用户跟我精通中文的尼日利亚天才程序员 Musa 有什么关系
Select Language