Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

dpm.as_usm_memory(X[i]) returns unexpected value #583

Closed
densmirn opened this issue Sep 10, 2021 · 1 comment · Fixed by #584
Closed

dpm.as_usm_memory(X[i]) returns unexpected value #583

densmirn opened this issue Sep 10, 2021 · 1 comment · Fixed by #584
Assignees

Comments

@densmirn
Copy link
Contributor

Reproducer:

import dpctl
print(dpctl.__version__)
import dpctl.tensor as dpt
import dpctl.memory as dpm
X = dpt.usm_ndarray(10, dtype="|f8")
Y = np.arange(X.size, dtype=X.dtype)
X.usm_data.copy_from_host(Y.view("|u1"))
print("X =", dpm.as_usm_memory(X).copy_to_host().view(X.dtype))
print("X[0] =", dpm.as_usm_memory(X[0]).copy_to_host().view(X.dtype))
print("X[1] =", dpm.as_usm_memory(X[1]).copy_to_host().view(X.dtype))
print("X[2] =", dpm.as_usm_memory(X[2]).copy_to_host().view(X.dtype))

Output:

0.10.0dev0+44.gb60ce2a.dirty
X = [0. 1. 2. 3. 4. 5. 6. 7. 8. 9.]
X[0] = [0.]
X[1] = [2.]
X[2] = [4.]
@oleksandr-pavlyk
Copy link
Collaborator

Thanks for the report @densmirn

The sycl USM array interface produced by the slice gives the actual memory address, and also sets offset which specifies the location of this memory address from the start of the underlying allocation.

The dpm.as_memory interprets offset as saying start at this additional displacement relative to the provided pointer. The fix should not be complicated, but I need to trace through the code to understand what is the correct way to handle offset.

oleksandr-pavlyk added a commit that referenced this issue Sep 11, 2021
Sycl USM array interface should use the base pointer in the 'data'
field, rather than its own 'data'.

```
In [1]: import numpy as np

In [2]: import dpctl, dpctl.tensor as dpt, dpctl.memory as dpm

In [3]: X = dpt.usm_ndarray(10, dtype="|f8")
   ...: Y = np.arange(X.size, dtype=X.dtype)
   ...: X.usm_data.copy_from_host(Y.view("|u1"))

In [4]: print("X =", dpm.as_usm_memory(X).copy_to_host().view(X.dtype))
   ...: print("X[0] =", dpm.as_usm_memory(X[0]).copy_to_host().view(X.dtype))
   ...: print("X[1] =", dpm.as_usm_memory(X[1]).copy_to_host().view(X.dtype))
   ...: print("X[2] =", dpm.as_usm_memory(X[2]).copy_to_host().view(X.dtype))
X = [0. 1. 2. 3. 4. 5. 6. 7. 8. 9.]
X[0] = [0.]
X[1] = [1.]
X[2] = [2.]
```
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

3 participants