Commit Graph

593 Commits

Author SHA1 Message Date
benrichard-amd 033922ecc0 Improve perfmon coalescing
Signed-off-by: benrichard-amd <ben.richard@amd.com>


[ROCm/rocprofiler-compute commit: c93fead779]
2024-08-09 09:46:42 -04:00
coleramos425 244ae98968 Comply to Python formatting
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 013bf218d0]
2024-08-09 09:46:42 -04:00
coleramos425 3f954de70d Add detail to Omniperf logs
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 3d5493a3ff]
2024-08-09 09:46:42 -04:00
coleramos425 7cc89dbeff Split rocprofv2 cmd args prior to subprocess call (#347)
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 19a229a9f2]
2024-08-09 09:46:42 -04:00
coleramos425 7de6a2f997 Fix bug in abs diff calculation for analysis output
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: fa90752d0c]
2024-08-09 09:46:42 -04:00
Ben Richard 21b23a08d6 Work around crash when profiling multi-process/multi-GPU application (#376)
* Fix crash in multi-GPU scenario

Exclude -o option when invoking rocprof so that each rocprof process
writes to a different .csv file. Concatenate into a single .csv file
when finished.

Signed-off-by: benrichard-amd <ben.richard@amd.com>

* Only combine csv files when using rocprofv2

rocprofv1 does not have separate csv files

Signed-off-by: benrichard-amd <ben.richard@amd.com>

* Fix indices in combined CSV file

Use ignore_index flag to ensure there are no duplicate indices.

Signed-off-by: benrichard-amd <ben.richard@amd.com>

* Fix Dispatch_ID column and remove unnamed column

-Pandas was inserting an unnamed column (index column)
-Overwrite the Dispatch_ID column so that every row is unique, starting at 0
-Remove fixup_rocprofv2_dispatch_ids as no longer needed

Signed-off-by: benrichard-amd <ben.richard@amd.com>

* Fix code formatting

Signed-off-by: benrichard-amd <ben.richard@amd.com>

* Fix code formatting (for real this time)

Signed-off-by: benrichard-amd <ben.richard@amd.com>

---------

Signed-off-by: benrichard-amd <ben.richard@amd.com>

[ROCm/rocprofiler-compute commit: 69e5c32d52]
2024-08-09 09:46:42 -04:00
coleramos425 68387da116 Comply to formatting
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 7046ea15bd]
2024-06-03 13:47:45 -05:00
coleramos425 67c4075950 Detection of MI308X and hardcode mclk to address bug in rocm-smi
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 1d19ae9483]
2024-06-03 13:47:45 -05:00
coleramos425 48c7833d07 Re-implementing HBM stack / XCD info for incoming product sku
Co-authored-by: Nicholas Curtis <nicholas.curtis@amd.com>
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 295b344646]
2024-06-03 13:47:45 -05:00
coleramos425 23cbe3f595 Create dedicated subdirectory in perfmon configs for archs supporting roofline
Separate subdirs allows us to target different roofline counters for different archs (i.e. MI300 vs MI200)

Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 49371cacec]
2024-05-31 16:09:58 -05:00
Karl W. Schulz dd374ea0ef remove use of distutils package entirely to avoid future deprecation
issues

Signed-off-by: Karl W. Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: 7a01f499d7]
2024-05-28 15:34:00 -05:00
Karl W. Schulz 7288bf34f4 updated approach for runtime dependency check that does use "pkg_resources"
which will reportedly be deprecated at some point in the future.

Signed-off-by: Karl W. Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: aca084f41c]
2024-05-28 15:34:00 -05:00
coleramos425 782ea3099b Wrap text displayed in 'Top Dispatch' table for neatness
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: c6cfa9cc26]
2024-05-28 15:34:00 -05:00
Nicholas Curtis 8bd72673e8 Add fix for case where we pass a single 'nan' value to to_avg
This is triggered by doing e.g., analyze -p <whatever> -k <kernel> -n per_kernel -b 17 18
Manifests as e.g.:

```
  ERROR [analysis] 'float' object has no attribute 'empty'
```

because of:

https://github.com/ROCm/omniperf/blob/f3a0360a5640e4539fac534f73876f51ddd268d0/src/utils/parser.py#L135

Instead, we first check whether numpy thinks the whole array is nan's, and bail early if so

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>


[ROCm/rocprofiler-compute commit: 047d7771f3]
2024-05-28 15:34:00 -05:00
Nicholas Curtis f7da503e6f handle unspecified case
Signed-off-by: Nicholas Curtis <nicurtis@amd.com>


[ROCm/rocprofiler-compute commit: 1f584c1612]
2024-05-28 15:34:00 -05:00
Nick Curtis 562e36a5c3 fix formatting
Signed-off-by: Nick Curtis <nicholas.curtis@amd.com>


[ROCm/rocprofiler-compute commit: 5579beeed5]
2024-05-28 15:34:00 -05:00
Nick Curtis fbb384bd2f Add ability to overide arch when name missing in rocminfo
Signed-off-by: Nick Curtis <nicholas.curtis@amd.com>


[ROCm/rocprofiler-compute commit: 989dd3b7ae]
2024-05-28 15:34:00 -05:00
coleramos425 219914fa9e Format CMake and Python
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 1f370c9fe7]
2024-05-10 09:07:40 -06:00
coleramos425 e6b2b03f32 Add support for --quiet flag to roofline
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: dba868973b]
2024-05-10 09:07:40 -06:00
coleramos425 1517617004 Add docs for --quiet mode and update README
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 3ab51735b5]
2024-05-10 09:07:40 -06:00
coleramos425 156305f391 Update docs for new Grafana reorg
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 8bca70a6ef]
2024-05-10 09:07:40 -06:00
coleramos425 5fd6022d9a Update from https://github.com/ROCm/mibench/commit/b704bd3ec439f8cbece6713852fcafc855c5b07e
[ROCm/rocprofiler-compute commit: 519bcb9b3e]
2024-05-10 09:07:40 -06:00
Karl W. Schulz f3a0360a56 Adding a top-level runtime python dependency checker. Goal is to
provide a kinder error message in the case where python dependencies
are not available locally. This is motivated for future execution by
users who are running from rocm-based binary packaging instead of using
normal cmake build system which would have verified the dependencies.

Signed-off-by: Karl W. Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: d1ee2ec870]
2024-05-03 15:26:27 -05:00
coleramos425 9a0d5f0663 Replace deprecated roofline warning with logging helper function
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 7d34e80567]
2024-04-25 18:43:20 +00:00
coleramos425 efcf9001d7 Add TCC_TOO_MANY_EA_WRREQS_STALL to gfx940 input configs (#349)
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 0fc620ce79]
2024-04-25 18:22:00 +00:00
Karl W Schulz 2653dc850b additional mod needed to support roofline binaries potentially
executing from two different locations

Signed-off-by: Karl W Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: b5011ff0ae]
2024-04-22 09:00:18 -05:00
Karl W Schulz ab7db47eb6 update logic to detect roofline binaries in two alternate paths
depending on whether user is running within local clone or from form
install.

Signed-off-by: Karl W Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: 093a4511ee]
2024-04-21 14:53:14 -05:00
Karl W Schulz c7d3c861be update logic to detect VERSION file to accommodate rocm packaging;
check two locations to cover case where user is running within local
git clone directly or alternatively, from package install.

Signed-off-by: Karl W Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: 3c562588ff]
2024-04-21 14:53:14 -05:00
Karl W Schulz a99b74810c fix execution error when OMNIPERF_COLOR env is set; update coloring to
support four modes:

(0) - no coloring and no loglevel delimiters
(1) - colored loglevel delimiters
(2) - non-colored loglevel delimiters
(3) - fully colored messages for all levels besides INFO

Signed-off-by: Karl W Schulz <karl.schulz@amd.com>


[ROCm/rocprofiler-compute commit: 65967658e9]
2024-04-04 14:44:51 -05:00
coleramos425 b6e857a4ee Define a README for /src subdir
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 9403dce667]
2024-04-01 14:30:21 -05:00
coleramos425 fc315248c7 Reorganizing docs runner and setting archive subdir for old docs
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: aac471c0fa]
2024-04-01 14:30:21 -05:00
coleramos425 6287e6dad3 Remove hardcoded URLs from docs in favor of relative links
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 9c93449cc7]
2024-04-01 14:30:21 -05:00
coleramos425 55ebe7059c Move dispatch id patch to proper util func in utils. Enable in rocprofv2 post-processing
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: e0556f32ab]
2024-04-01 14:30:21 -05:00
Nick Curtis 22b596dc32 implement rocprofv2 workaround for dispatch ids (#336)
* implement rocprofv2 workaround for dispatch ids

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>

* formatting

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>

---------

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>
Co-authored-by: Nicholas Curtis <nicurtis@amd.com>

[ROCm/rocprofiler-compute commit: a1017b68e9]
2024-04-01 14:30:21 -05:00
coleramos425 22e5069fc5 Adding documentation for global command line options
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 1a3bdad90a]
2024-04-01 14:30:21 -05:00
colramos-amd 5d15f4d6fa Restore OMNIPERF_COLOR global to disable default log coloring
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 6cc8f0154f]
2024-04-01 14:30:21 -05:00
colramos-amd 091e796643 Extending log coloring to message text. Enable by default.
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: b1d0b3905c]
2024-04-01 14:30:21 -05:00
colramos-amd dc13c30d74 Remove superfluous logging statement
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 78c48eaed5]
2024-04-01 14:30:21 -05:00
colramos-amd e170da2c39 Throw more helpful warning on missing csv in analyze mode (#313)
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: f47b8d2547]
2024-04-01 14:30:21 -05:00
coleramos425 e1470d7a38 Register TRACE loglevel ahead of parse_args to avoid error in --specs printout
Signed-off-by: coleramos425 <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 706afa7f37]
2024-04-01 14:30:21 -05:00
Cole Ramos fa02efc962 Revert #333 deterministic ordering (#337)
* Revert "formatting"

This reverts commit e387561791898467371b6219921f10447d3486b3.

* Revert "Implement deterministic ordering of perfmon"

This reverts commit 2a24600a37c8002838145962a957cea447555b0c.

[ROCm/rocprofiler-compute commit: f7394bf272]
2024-04-01 14:30:21 -05:00
Nicholas Curtis ed86d5c64c formatting
Signed-off-by: Nicholas Curtis <nicurtis@amd.com>


[ROCm/rocprofiler-compute commit: f6f6165379]
2024-04-01 14:30:21 -05:00
Nicholas Curtis 3ff4117c4e Implement deterministic ordering of perfmon
Signed-off-by: Nicholas Curtis <nicurtis@amd.com>


[ROCm/rocprofiler-compute commit: 97af96e435]
2024-04-01 14:30:21 -05:00
colramos-amd e55b9b1d34 Adding redirect to help button in standalone GUI
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 3928f36d5e]
2024-04-01 14:30:21 -05:00
colramos-amd fad337db06 Removing unused modules
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 33e763178e]
2024-04-01 14:30:21 -05:00
colramos-amd 8a9123d722 Documentation enhancements and removing watermark
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 3d8b748924]
2024-04-01 14:30:21 -05:00
colramos-amd 4a1293b5d7 Adding comments to roofline and changing output file name for specificity
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 73ac6c9561]
2024-04-01 14:30:21 -05:00
colramos-amd da0986846e Implementing docs feedback with typos and minor enhancements
Co-authored-by: Jose Santos <josantos@amd.com>
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: cf36fb7fbf]
2024-04-01 14:30:21 -05:00
colramos-amd 033c4e8fc5 Correct typo in Mi100 metric defs for missing SQ_INSTS_VMEM
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 6cfdca93b9]
2024-04-01 14:30:21 -05:00
colramos-amd 2c8f4de5e4 Responding to first round of Ben R.'s docs feedback.\n\nThis round includes fixes for comments up to the 'Performance Model' section. I will need to work with our documentation group to respond to those higher level comments.
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: 20670da2b7]
2024-04-01 14:30:21 -05:00