parallel_for with range<3> iterates over wider range than specified

[baranovsky]

Hi,

I've been developing with DPC++, and recently trying to change version from 2023  to the latest 2024.0.1.

The output of my software is corrupted because of the following phenomena, which may not be intended.

In the attached program,

• parallel_for(range<1>().... works fine
• but parallel_for(range<3>()... iterates  over wider range than specified.

I'm wondering if this is a bug or something is missing ...

Attachments:

main.txt: this file is originally 'main.cpp'

Makefile.txt: this file is originally 'Makefile'

My development environment:

OS: Ubuntu 22.04 LTS

oneAPI: 2024.0.1 with oneapi-for-nvidia-gpus-2024.0.1-cuda-12.0-linux.sh

CPU: Intel Core i9-14900KF

GPU: NVIDIA GeForce RTX 4090

CUDA: 12.2 (driver version: 535.146.02)

Thank you in advance for your help.

[baranovsky]

I've noticed that no result is shown in my post.

In my environment, the console output looks like

> 32767 should be 32767
> 49151 should be 32767

In parallel_for(range<1>(x * y * z), where x = y = z = 32, 

the maximum value returned by item.get_linear_id() is 32767,

which is identical to 32 * 32 * 32 - 1.

This seems correct.

On the other hand, in parallel_for(range<3>(x, y, z),

the maximum value is 49151, which exceeds (32 * 32 * 32 - 1)

This seems to be too large.