Hi In the following code, I would like to collect the SM IDs that have executed the kernel. The kernel is a simple addition one and I have written this code: __device__ uint get_smid(void) { uint ret; asm("mov.u32 %0, %smid;" : "=r"(ret) ); return ret; } __global__ void simpleAdd(fl…

That is no surprise, is it? nsmid is simply an upper bound for smid. they are not equivalent. This code will print a histogram of used sm ids. #include <iostream> #include <map> __device__ int get_smid(void) { int ret; asm("mov.u32 %0, %smid;" : "=r"(ret) ); return ret; } __global__…

Collecting busy SM IDs

Accelerated Computing CUDA CUDA Programming and Performance

mahmood.nt March 12, 2024, 12:59pm 5

OK Thank you very much.
I also found an answer to the first question with an example in this page

gist.github.com

https://gist.github.com/allanmac/4751080

smid.cu

#include <stdio.h>

//
//
//

#define DEVICE_INTRINSIC_QUALIFIERS   __device__ __forceinline__

DEVICE_INTRINSIC_QUALIFIERS
unsigned int

This file has been truncated. show original

SM Used by a particular program

Topic		Replies	Views
Counting values --- what's wrong? CUDA first timer's naive code~ help please! CUDA Programming and Performance	2	984	March 22, 2009
Illegal memory access was encountered when launch some kernels in a loop CUDA Programming and Performance	4	1510	October 12, 2021
how to add CUDA Programming and Performance	11	3927	May 27, 2008
Best way to implement element-wise add kernel CUDA Programming and Performance	3	125	August 24, 2024
How to reach HBM Peak bandwidth performance CUDA Programming and Performance	2	858	November 30, 2021
First CUDA trial but unexpected behavior CUDA Programming and Performance	4	1003	March 29, 2012
Some metric set and section are not enable Nsight Compute cuda , ubuntu	5	1426	January 16, 2024
memory access error CUDA Programming and Performance	11	1397	January 12, 2013
can you give me sample code for atomicAdd()? CUDA Programming and Performance	9	48110	June 5, 2009
Kernel functions as parameters? CUDA Programming and Performance	5	5243	February 11, 2011

Collecting busy SM IDs

Related topics