search

CHIP-SPV / chipStar

chipStar is a tool for compiling and running HIP/CUDA on SPIR-V via OpenCL or Level Zero APIs.
Other
157 stars 27 forks source link

Fix dependency chain issues #840

Closed pvelesko closed 2 months ago

pvelesko commented 2 months ago

TODO

pvelesko commented 2 months ago

CatchMemLeak1 is not reliable for Level Zero

╭─pvelesko@meatloaf ~/chipStar/build ‹event-tracking●›
╰─$ CHIP_LOGLEVEL=info /home/pvelesko/chipStar/build/tests/runtime/CatchMemLeak1                                                                                                                                                               130 ↵
CHIP info [TID 1795805] [1714443678.033519666] : CHIP_PLATFORM=0
CHIP info [TID 1795805] [1714443678.033548389] : CHIP_DEVICE_TYPE=default
CHIP info [TID 1795805] [1714443678.033550790] : CHIP_DEVICE=0
CHIP info [TID 1795805] [1714443678.033553594] : CHIP_BE=level0
CHIP info [TID 1795805] [1714443678.033555519] : CHIP_DUMP_SPIRV=off
CHIP info [TID 1795805] [1714443678.033557934] : CHIP_JIT_FLAGS_OVERRIDE=-cl-kernel-arg-info -cl-std=CL3.0
CHIP info [TID 1795805] [1714443678.033559900] : CHIP_L0_COLLECT_EVENTS_TIMEOUT=0
CHIP info [TID 1795805] [1714443678.033562786] : CHIP_L0_EVENT_TIMEOUT=0
CHIP info [TID 1795805] [1714443678.033564651] : CHIP_SKIP_UNINIT=off
CHIP info [TID 1795805] [1714443678.073542716] : Device: Intel(R) Arc(TM) A380 Graphics

RSSIncLag=54922
Didn't detect memory leak.
CHIP info [TID 1795805] [1714443704.556730647] : Command list reuse: N/A (No command lists requested)
CHIP info [TID 1795805] [1714443704.556741304] : Events reuse: 0%
╭─pvelesko@meatloaf ~/chipStar/build ‹event-tracking●›
╰─$
╭─pvelesko@meatloaf ~/chipStar/build ‹event-tracking●›
╰─$
╭─pvelesko@meatloaf ~/chipStar/build ‹event-tracking●›
╰─$
╭─pvelesko@meatloaf ~/chipStar/build ‹event-tracking●›
╰─$  /home/pvelesko/chipStar/build/tests/runtime/CatchMemLeak1
RSSIncLag=7459
Possible memory leak!
linehill commented 2 months ago

CatchMemLeak1 is not reliable for Level Zero

Or it’s pointing out that there is still a memory leak in the level0 backend which seems to be the case on this branch. This log suggests that the level0 is leaking memory steadily. Here is a link to the source I used to monitor the memory usage.

pvelesko commented 2 months ago

This is a draft or not ready for review yet @linehill

pvelesko commented 2 months ago

closing this in favor of two smaller PRs