Conversation
…k instead of max Signed-off-by: Akshay Tondak <aktondak@amd.com>
aktondak
requested review from
AShivangi,
chvamshi-xilinx and
rozumx
as code owners
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
|
clang-tidy review says "All clean, LGTM! 👍" |
| run.start(); | ||
| run.wait2(); | ||
| } | ||
| uint64_t clock_mhz = get_clock(dev); |
There was a problem hiding this comment.
getting clock value from PMF before starting the test should be returning a very low MHz as clocks will not ramp up. this would result in some non consistent tops and low tops
There was a problem hiding this comment.
get_clock() should be getting dynamic clock number from PMF
There was a problem hiding this comment.
The get_clock() api has been updated to query max clock for both strx and npu3. We'll change this once the DPM level querying is available from firmware
|
clang-tidy review says "All clean, LGTM! 👍" |
| if (res.type != xrt_core::query::xrt_resource_raw::resource_type::npu_clk_max) | ||
| continue; | ||
| npu_hclock = res.data_uint64; | ||
| } |
There was a problem hiding this comment.
add the error message
if (npu_hclock == 0) {
XBValidateUtils::logger(ptree, "Error", "NPU H-clock is 0");
ptree.put("status", XBValidateUtils::test_token_failed);
return;
}
|
clang-tidy review says "All clean, LGTM! 👍" |
1 similar comment
|
clang-tidy review says "All clean, LGTM! 👍" |
5 participants
Problem solved by the commit
This PR updates the Gemm test to query max clock for npu3 platform.
Bug / issue (if any) fixed, which PR introduced the bug, how it was discovered
https://jira.xilinx.com/browse/AIESW-10440
https://jira.xilinx.com/browse/AIESW-24842
How problem was solved, alternative solutions (if any) and why they were rejected
This PR creates a common query API between strix and NPU to query max clocks for both strx and npu3. This PR also now allows to run benchmarks in different pmode other that performance and turbo. The default is still performance.
Risks (if any) associated the changes in the commit
Might change some existing behavior.
What has been tested and how, request additional testing if necessary
Tested on Linux strix machine. Following results were seen when run on different pmodes (in sudo mode)
Documentation impact (if any)
None