@@ -155,19 +155,19 @@ to a mode where it merely activates an external profiler and does not do perform
155
155
profiling itself. In case the detection does not work, this mode can be forcibly activated
156
156
by passing `external=true` to `CUDA.@profile`.
157
157
158
- NVIDIA provides two tools for profiling CUDA applications: NSight Systems and NSight Compute
158
+ NVIDIA provides two tools for profiling CUDA applications: Nsight Systems and Nsight Compute
159
159
for respectively timeline profiling and more detailed kernel analysis. Both tools are
160
160
well-integrated with the Julia GPU packages, and make it possible to iteratively profile
161
161
without having to restart Julia.
162
162
163
163
#### NVIDIA Nsight Systems
164
164
165
- Generally speaking, the first external profiler you should use is NSight Systems, as it will
165
+ Generally speaking, the first external profiler you should use is Nsight Systems, as it will
166
166
give you a high-level overview of your application's performance characteristics. After
167
167
downloading and installing the tool (a version might have been installed alongside with the
168
168
CUDA toolkit, but it is recommended to download and install the latest version from the
169
169
NVIDIA website), you need to launch Julia from the command-line, wrapped by the `nsys`
170
- utility from NSight Systems:
170
+ utility from Nsight Systems:
171
171
172
172
```
173
173
$ nsys launch julia
@@ -209,7 +209,7 @@ You can open the resulting `.qdrep` file with `nsight-sys`:
209
209
210
210
!!! info
211
211
212
- If NSight Systems does not capture any kernel launch, even though you have used
212
+ If Nsight Systems does not capture any kernel launch, even though you have used
213
213
`CUDA.@profile`, try starting `nsys` with `--trace cuda`.
214
214
215
215
#### NVIDIA Nsight Compute
@@ -243,7 +243,7 @@ to the running session by selecting it in the list in the `Attach` pane:
243
243
244
244

245
245
246
- Note that this even works with remote systems, i.e., you can have NSight Compute connect
246
+ Note that this even works with remote systems, i.e., you can have Nsight Compute connect
247
247
over ssh to a remote system where you run Julia under `ncu`.
248
248
249
249
Once you've successfully attached to a Julia process, you will see that the tool has stopped
@@ -277,26 +277,26 @@ the API calls that have been made:
277
277

278
278
279
279
280
- #### Troubleshooting NSight Compute
280
+ #### Troubleshooting Nsight Compute
281
281
282
- If you're running into issues, make sure you're using the same version of NSight Compute on
282
+ If you're running into issues, make sure you're using the same version of Nsight Compute on
283
283
the host and the device, and make sure it's the latest version available. You do not need
284
- administrative permissions to install NSight Compute, the `runfile` downloaded from the
284
+ administrative permissions to install Nsight Compute, the `runfile` downloaded from the
285
285
NVIDIA home page can be executed as a regular user.
286
286
287
287
##### Kernel sources only report `File not found`
288
288
289
- When profiling a remote application, NSight Compute will not be able to find the sources of
289
+ When profiling a remote application, Nsight Compute will not be able to find the sources of
290
290
kernels, and instead show `File not found` errors in the Source view. Although it is
291
- possible to point NSight Compute to a local version of the remote file, it is recommended
291
+ possible to point Nsight Compute to a local version of the remote file, it is recommended
292
292
to enable "Auto-Resolve Remote Source File" in the global Profile preferences (Tools menu
293
293
-> Preferences). With that option set to "Yes", clicking the "Resolve" button will
294
294
automatically download and use the remote version of the requested source file.
295
295
296
296
##### `Could not load library "libpcre2-8`
297
297
298
- This is caused by an incompatibility between Julia and NSight Compute, and should be fixed
299
- in the latest versions of NSight Compute. If it's not possible to upgrade, the following
298
+ This is caused by an incompatibility between Julia and Nsight Compute, and should be fixed
299
+ in the latest versions of Nsight Compute. If it's not possible to upgrade, the following
300
300
workaround may help:
301
301
302
302
```
@@ -305,28 +305,28 @@ LD_LIBRARY_PATH=$(/path/to/julia -e 'println(joinpath(Sys.BINDIR, Base.LIBDIR, "
305
305
306
306
##### The Julia process is not listed in the "Attach" tab
307
307
308
- Make sure that the port that is used by NSight Compute (49152 by default) is accessible via
308
+ Make sure that the port that is used by Nsight Compute (49152 by default) is accessible via
309
309
ssh. To verify this, you can also try forwarding the port manually:
310
310
311
311
```
312
312
ssh user@host . com - L 49152 : localhost: 49152
313
313
```
314
314
315
- Then, in the "Connect to process" window of NSight Compute, add a connection to `localhost`
315
+ Then, in the "Connect to process" window of Nsight Compute, add a connection to `localhost`
316
316
instead of the remote host.
317
317
318
318
If SSH complains with `Address already in use`, that means the port is already in use. If
319
319
you're using VSCode, try closing all instances as VSCode might automatically forward the
320
- port when launching NSight Compute in a terminal within VSCode.
320
+ port when launching Nsight Compute in a terminal within VSCode.
321
321
322
- ##### Julia in NSight Compute only shows the Julia logo, not the REPL prompt
322
+ ##### Julia in Nsight Compute only shows the Julia logo, not the REPL prompt
323
323
324
- In some versions of NSight Compute, you might have to start Julia without the `--project`
324
+ In some versions of Nsight Compute, you might have to start Julia without the `--project`
325
325
option and switch the environment from inside Julia.
326
326
327
327
##### "Disconnected from the application" once I click "Resume"
328
328
329
- Make sure that everything is precompiled before starting Julia with NSight Compute,
329
+ Make sure that everything is precompiled before starting Julia with Nsight Compute,
330
330
otherwise you end up profiling the precompilation process instead of your actual
331
331
application.
332
332
@@ -348,7 +348,7 @@ The `nvidia.ko` kernel module needs to be reloaded after changing this configura
348
348
your system may require regenerating the initramfs or even a reboot. Refer to your
349
349
distribution's documentation for details.
350
350
351
- ##### NSight Compute breaks on various API calls
351
+ ##### Nsight Compute breaks on various API calls
352
352
353
353
Make sure `Break On API Error` is disabled in the `Debug` menu, as CUDA.jl purposefully
354
354
triggers some API errors as part of its normal operation.
@@ -378,7 +378,7 @@ For more details, refer to the documentation of the NVTX.jl package.
378
378
379
379
## Compiler options
380
380
381
- Some tools, like NSight Systems Compute, also make it possible to do source-level profiling.
381
+ Some tools, like Nsight Systems Compute, also make it possible to do source-level profiling.
382
382
CUDA.jl will by default emit the necessary source line information, which you can disable by
383
383
launching Julia with `-g0`. Conversely, launching with `-g2` will emit additional debug
384
384
information, which can be useful in combination with tools like `cuda-gdb`, but might hurt
0 commit comments