Closed zjin-lcf closed 3 years ago
Can't reproduce, the test works for me:
$ ./quicksort 1 16 16
--------------------------------------------------------------------
Allocating array size of 256
Sorting the regular way...
Time to sort: 0.008231 ms
quicksort on the cpu:
Time to sort: 0.008392 ms
verifying: true
Sorting with GPU quicksort:
Time to sort: 3.09003 ms
verifying: true
Number of failures: 0 out of 1
Average Time: 3.09003 ms
Average speedup over CPU quicksort: 0.00271583
Average speedup over CPU std::sort: 0.00266373
-------done--------------------------------------------------------
This is with Intel Skylake GPU:
Platform Name Intel(R) OpenCL HD Graphics
Number of devices 1
Device Name Intel(R) Graphics [0x1912]
Device Vendor Intel(R) Corporation
Device Vendor ID 0x8086
Device Version OpenCL 3.0 NEO
Driver Version 20.51.18762
Device OpenCL C Version OpenCL C 3.0
Which hardware do you use, is it CPU or GPU ? and which driver version ? It seems your driver has a problem with the generated SPIR-V.
It is a Gen8 GPU:
Platform Name Intel(R) OpenCL HD Graphics Number of devices 1 Device Name Intel(R) Graphics [0x162a] Device Vendor Intel(R) Corporation Device Vendor ID 0x8086 Device Version OpenCL 3.0 NEO Driver Version 20.51.18762 Device OpenCL C Version OpenCL C 3.0
The reproducer is https://github.com/zjin-lcf/oneAPI-DirectProgramming/tree/master/quicksort-hip