AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

Update docs and --help menu to reflect new vcopy usage

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


[ROCm/rocprofiler-compute commit: f5b8cff990]
This commit is contained in:
colramos-amd
2023-11-20 14:45:00 -06:00
orang tua c7e7b0df14
melakukan f6c49cda2d
3 mengubah file dengan 156 tambahan dan 163 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocprofiler-compute/src/argparser.py
+5 -5
Melihat File
@@ -52,11 +52,11 @@ def omniarg_parser(parser, omniperf_home, omniperf_version):
\n\n-------------------------------------------------------------------------------
\nExamples:
\n\tomniperf profile -n vcopy_all -- ./vcopy 1048576 256
\n\tomniperf profile -n vcopy_SPI_TCC -b SQ TCC -- ./vcopy 1048576 256
\n\tomniperf profile -n vcopy_kernel -k vecCopy -- ./vcopy 1048576 256
\n\tomniperf profile -n vcopy_disp -d 0 -- ./vcopy 1048576 256
\n\tomniperf profile -n vcopy_roof --roof-only -- ./vcopy 1048576 256
\n\tomniperf profile -n vcopy_all -- ./vcopy -n 1048576 -b 256
\n\tomniperf profile -n vcopy_SPI_TCC -b SQ TCC -- ./vcopy -n 1048576 -b 256
\n\tomniperf profile -n vcopy_kernel -k vecCopy -- ./vcopy -n 1048576 -b 256
\n\tomniperf profile -n vcopy_disp -d 0 -- ./vcopy -n 1048576 -b 256
\n\tomniperf profile -n vcopy_roof --roof-only -- ./vcopy -n 1048576 -b 256
\n-------------------------------------------------------------------------------\n
""",
prog="tool",
projects/rocprofiler-compute/src/docs/getting_started.md
+1 -1
Melihat File
@@ -14,7 +14,7 @@
To collect the default set of data for all kernels in the target application, launch, e.g.:
```shell
$ omniperf profile -n vcopy_data -- ./vcopy 1048576 256
$ omniperf profile -n vcopy_data -- ./vcopy -n 1048576 -b 256
```
The app runs, each kernel is launched, and profiling results are generated. By default, results are written to (e.g.,) ./workloads/vcopy_data (configurable via the `-n` argument). To collect all requested profile information, it may be required to replay kernels multiple times.
projects/rocprofiler-compute/src/docs/profiling.md
+150 -157
Melihat File
@@ -23,7 +23,7 @@ the MI200 platform.
$ hipcc vcopy.cpp -o vcopy
$ ls
vcopy vcopy.cpp
$ ./vcopy 1048576 256
$ ./vcopy -n 1048576 -b 256
Finished allocating vectors on the CPU
Finished allocating vectors on the GPU
Finished copying vectors to the GPU
@@ -42,44 +42,43 @@ The *omniperf* script, availible through the [Omniperf](https://github.com/AMDRe
**omniperf help:**
```shell-session
$ omniperf profile --help
ROC Profiler: /usr/bin/rocprof
usage:
usage:
omniperf profile --name <workload_name> [profile options] [roofline options] -- <profile_cmd>
-------------------------------------------------------------------------------
Examples:
omniperf profile -n vcopy_all -- ./vcopy 1048576 256
omniperf profile -n vcopy_SPI_TCC -b SQ TCC -- ./vcopy 1048576 256
omniperf profile -n vcopy_kernel -k vecCopy -- ./vcopy 1048576 256
omniperf profile -n vcopy_disp -d 0 -- ./vcopy 1048576 256
omniperf profile -n vcopy_roof --roof-only -- ./vcopy 1048576 256
omniperf profile -n vcopy_all -- ./vcopy -n 1048576 -b 256
omniperf profile -n vcopy_SPI_TCC -b SQ TCC -- ./vcopy -n 1048576 -b 256
omniperf profile -n vcopy_kernel -k vecCopy -- ./vcopy -n 1048576 -b 256
omniperf profile -n vcopy_disp -d 0 -- ./vcopy -n 1048576 -b 256
omniperf profile -n vcopy_roof --roof-only -- ./vcopy -n 1048576 -b 256
-------------------------------------------------------------------------------
Help:
-h, --help show this help message and exit
-h, --help show this help message and exit
General Options:
-v, --version show program's version number and exit
-V, --verbose Increase output verbosity
-v, --version show program's version number and exit
-V, --verbose Increase output verbosity
Profile Options:
-n , --name Assign a name to workload.
-p , --path Specify path to save workload.
(DEFAULT: /home/colramos/GitHub/omniperf/workloads/<name>)
(DEFAULT: /home/colramos/GitHub/omniperf-pub/workloads/<name>)
-k [ ...], --kernel [ ...] Kernel filtering.
-d [ ...], --dispatch [ ...] Dispatch ID filtering.
-b [ ...], --ipblocks [ ...] IP block filtering:
SQ
SQC
@@ -90,16 +89,19 @@ Profile Options:
SPI
CPC
CPF
-d [ ...], --dispatch [ ...] Dispatch ID filtering.
--join-type Choose how to join rocprof runs: (DEFAULT: grid)
kernel (i.e. By unique kernel name dispatches)
grid (i.e. By unique kernel name + grid size dispatches)
--no-roof Profile without collecting roofline data.
-- [ ...] Provide command for profiling after double dash.
--kernel-verbose Specify Kernel Name verbose level 1-5. Lower the level, shorter the kernel name. (DEFAULT: 2) (DISABLE: 5)
Standalone Roofline Options:
--roof-only Profile roofline data only.
--sort Overlay top kernels or top dispatches: (DEFAULT: kernels)
kernels
dispatches
-m , --mem-level Filter by memory level: (DEFAULT: ALL)
-m [ ...], --mem-level [ ...] Filter by memory level: (DEFAULT: ALL)
HBM
L2
vL1D
@@ -113,36 +115,40 @@ The following sample command profiles the *vcopy* workload.
**vcopy profiling:**
```shell-session
$ omniperf profile --name vcopy -- ./vcopy 1048576 256
Resolving rocprof
ROC Profiler: /usr/bin/rocprof
ROC Profiler: /opt/rocm-5.2.1/bin/rocprof
Execution mode = profile
___ _ __
/ _ \ _ __ ___ _ __ (_)_ __ ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_
| |_| | | | | | | | | | | |_) | __/ | | _|
\___/|_| |_| |_|_| |_|_| .__/ \___|_| |_|
|_|
-------------
Profile only
-------------
omniperf ver: 1.0.8-PR1
Path: /home/colramos/GitHub/omniperf-pub/workloads
Target: mi200
Command: /home/colramos/vcopy 1048576 256
Kernel Selection: None
Dispatch Selection: None
SoC = mi200
Profiler choice = rocprofv1
omniperf ver: 1.0.10
Path: /home/colramos/omniperf/workloads/vcopy_test/mi200
Target: mi200
Command: ./vcopy -n 1048576 -b 256
Kernel Selection: None
Dispatch Selection: None
IP Blocks: All
Log: /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/log.txt
KernelName verbose: 2
/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt
RPL: on '230411_165021' from '/opt/rocm-5.2.1' in '/home/colramos/GitHub/omniperf-pub'
RPL: profiling '""/home/colramos/vcopy 1048576 256""'
RPL: input file '/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt'
RPL: output dir '/tmp/rpl_data_230411_165021_26406'
RPL: result dir '/tmp/rpl_data_230411_165021_26406/input0_results_230411_165021'
Finished allocating vectors on the CPU
ROCProfiler: input from "/tmp/rpl_data_230411_165021_26406/input0.xml"
Current input file: /home/colramos/omniperf/workloads/vcopy_test/mi200/perfmon/pmc_perf_2.txt
RPL: on '231120_142934' from '/opt/rocm-5.2.1' in '/home/colramos/omniperf'
RPL: profiling '""./vcopy -n 1048576 -b 256""'
RPL: input file '/home/colramos/omniperf/workloads/vcopy_test/mi200/perfmon/pmc_perf_2.txt'
RPL: output dir '/tmp/rpl_data_231120_142934_262088'
RPL: result dir '/tmp/rpl_data_231120_142934_262088/input0_results_231120_142934'
ROCProfiler: input from "/tmp/rpl_data_231120_142934_262088/input0.xml"
gpu_index =
kernel =
range =
3 metrics
SQ_INSTS_SMEM, SQ_INST_LEVEL_SMEM, SQ_ACCUM_PREV_HIRES
26 metrics
SQ_INSTS_VALU_TRANS_F16, SQ_INSTS_VALU_ADD_F32, SQ_INSTS_VALU_MUL_F32, SQ_INSTS_VALU_FMA_F32, SQ_INSTS_VALU_TRANS_F32, SQ_INSTS_VALU_ADD_F64, SQ_INSTS_VALU_MUL_F64, SQ_INSTS_VALU_FMA_F64, TCP_VOLATILE_sum, TCP_TOTAL_ACCESSES_sum, TCP_TOTAL_READ_sum, TCP_TOTAL_WRITE_sum, TA_BUFFER_ATOMIC_WAVEFRONTS_sum, TA_BUFFER_TOTAL_CYCLES_sum, TD_ATOMIC_WAVEFRONT_sum, TD_STORE_WAVEFRONT_sum, SPI_RA_REQ_NO_ALLOC, SPI_RA_REQ_NO_ALLOC_CSN, CPC_CPC_STAT_STALL, CPC_UTCL1_STALL_ON_TRANSLATION, CPF_CPF_STAT_IDLE, CPF_CPF_TCIU_IDLE, TCC_REQ_sum, TCC_STREAMING_REQ_sum, TCC_HIT_sum, TCC_MISS_sum
vcopy testing on GCD 0
Finished allocating vectors on the CPU
Finished allocating vectors on the GPU
Finished copying vectors to the GPU
sw thinks it moved 1.000000 KB per wave
@@ -153,9 +159,10 @@ Finished copying the output vector from the GPU to the CPU
Releasing GPU memory
Releasing CPU memory
... ...
ROCPRofiler: 1 contexts collected, output directory /tmp/rpl_data_220527_130317_1787038/input_results_220527_130317
File 'workloads/vcopy/mi200/timestamps.csv' is generating
...
Empirical Roofline Calculation
Copyright © 2022 Advanced Micro Devices, Inc. All rights reserved.
Total detected GPU devices: 2
GPU Device 0: Profiling...
99% [||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| ]
@@ -213,7 +220,7 @@ etc. The SOC names are generated as a part of Omniperf, and do not necessarily
> Note: Additionally, you'll notice a few extra files. An SoC parameters file, *sysinfo.csv*, is created to reflect the target device settings. All profiling output is stored in *log.txt*. Roofline specific benchmark results are stored in *roofline.csv*.
```shell
```shell-session
$ ls workloads/vcopy/mi200/
total 112
drwxrwxr-x 3 colramos colramos 4096 Apr 11 16:42 .
@@ -254,39 +261,42 @@ i.e. filtering dispatch X, but dispatch X does not match your kernel name filter
One can profile a selected IP Block to speed up the profiling process. All profiling results are accumulated in the same target directory, without overwriting those for other IP blocks, hence enabling the incremental profiling and analysis.
The following example only gathers hardware counters for SQ and TCC, skipping all other IP Blocks:
```shell
$ omniperf profile --name vcopy -b SQ TCC -- ./sample/vcopy 1048576 256
Resolving rocprof
ROC Profiler: /usr/bin/rocprof
```shell-session
$ omniperf profile --name vcopy -b SQ TCC -- ./sample/vcopy -n 1048576 -b 256
ROC Profiler: /opt/rocm-5.2.1/bin/rocprof
Execution mode = profile
___ _ __
/ _ \ _ __ ___ _ __ (_)_ __ ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_
| |_| | | | | | | | | | | |_) | __/ | | _|
\___/|_| |_| |_|_| |_|_| .__/ \___|_| |_|
|_|
-------------
Profile only
-------------
omniperf ver: 1.0.8-PR1
Path: /home/colramos/GitHub/omniperf-pub/workloads
Target: mi200
Command: /home/colramos/vcopy 1048576 256
Kernel Selection: None
Dispatch Selection: None
IP Blocks: ['SQ', 'TCC']
fname: pmc_sq_perf2: Added
SoC = mi200
Profiler choice = rocprofv1
fname: pmc_cpc_perf: Skipped
fname: pmc_spi_perf: Skipped
fname: pmc_cpf_perf: Skipped
fname: pmc_tcp_perf: Skipped
fname: pmc_sq_perf4: Added
fname: pmc_tcc_perf: Added
fname: pmc_sq_perf8: Added
fname: pmc_ta_perf: Skipped
fname: pmc_sq_perf1: Added
fname: pmc_sq_perf3: Added
fname: pmc_td_perf: Skipped
fname: pmc_tcc2_perf: Skipped
fname: pmc_tcp_perf: Skipped
fname: pmc_spi_perf: Skipped
fname: pmc_sq_perf4: Added
fname: pmc_sqc_perf1: Skipped
fname: pmc_tcc_perf: Added
fname: pmc_cpf_perf: Skipped
fname: pmc_sq_perf8: Added
fname: pmc_cpc_perf: Skipped
fname: pmc_sq_perf1: Added
fname: pmc_ta_perf: Skipped
fname: pmc_sq_perf3: Added
fname: pmc_sq_perf6: Added
Log: /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/log.txt
fname: pmc_sq_perf2: Added
omniperf ver: 1.0.10
Path: /home/colramos/omniperf/vcopy_/mi200
Target: mi200
Command: ./vcopy -n 1048576 -b 256
Kernel Selection: None
Dispatch Selection: None
IP Blocks: ['sq', 'tcc']
KernelName verbose: 2
...
```
@@ -294,37 +304,28 @@ Log: /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/log.txt
Kernel filtering is based on the name of the kernel(s) you'd like to isolate. Use a kernel name substring list to isolate desired kernels.
The following example demonstrates profiling isolating the kernel matching substring "vecCopy":
```shell
$ omniperf profile --name vcopy -k vecCopy -- ./vcopy 1048576 256
Resolving rocprof
ROC Profiler: /usr/bin/rocprof
```shell-session
$ omniperf profile --name vcopy -k vecCopy -- ./vcopy -n 1048576 -b 256
ROC Profiler: /opt/rocm-5.2.1/bin/rocprof
Execution mode = profile
___ _ __
/ _ \ _ __ ___ _ __ (_)_ __ ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_
| |_| | | | | | | | | | | |_) | __/ | | _|
\___/|_| |_| |_|_| |_|_| .__/ \___|_| |_|
|_|
-------------
Profile only
-------------
omniperf ver: 1.0.8-PR1
Path: /home/colramos/GitHub/omniperf-pub/workloads
Target: mi200
Command: /home/colramos/vcopy 1048576 256
Kernel Selection: ['vecCopy']
Dispatch Selection: None
SoC = mi200
Profiler choice = rocprofv1
omniperf ver: 1.0.10
Path: /home/colramos/omniperf/workloads/vcopy/mi200
Target: mi200
Command: ./vcopy -n 1048576 -b 256
Kernel Selection: ['vecCopy']
Dispatch Selection: None
IP Blocks: All
Log: /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/log.txt
/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt
RPL: on '230411_170300' from '/opt/rocm-5.2.1' in '/home/colramos/GitHub/omniperf-pub'
RPL: profiling '""/home/colramos/vcopy 1048576 256""'
RPL: input file '/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt'
RPL: output dir '/tmp/rpl_data_230411_170300_29696'
RPL: result dir '/tmp/rpl_data_230411_170300_29696/input0_results_230411_170300'
Finished allocating vectors on the CPU
ROCProfiler: input from "/tmp/rpl_data_230411_170300_29696/input0.xml"
gpu_index =
kernel = vecCopy
... ...
KernelName verbose: 2
...
```
#### Dispatch Filtering
@@ -332,35 +333,26 @@ Dispatch filtering is based on the *global* dispatch index of kernels in a run.
The following example profiles only the 0th dispatched kernel in execution of the application:
```shell-session
$ omniperf profile --name vcopy -d 0 -- ./vcopy 1048576 256
Resolving rocprof
ROC Profiler: /usr/bin/rocprof
$ omniperf profile --name vcopy -d 0 -- ./vcopy -n 1048576 -b 256
ROC Profiler: /opt/rocm-5.2.1/bin/rocprof
Execution mode = profile
___ _ __
/ _ \ _ __ ___ _ __ (_)_ __ ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_
| |_| | | | | | | | | | | |_) | __/ | | _|
\___/|_| |_| |_|_| |_|_| .__/ \___|_| |_|
|_|
-------------
Profile only
-------------
omniperf ver: 1.0.8-PR1
Path: /home/colramos/GitHub/omniperf-pub/workloads
Target: mi200
Command: /home/colramos/vcopy 1048576 256
Kernel Selection: None
Dispatch Selection: ['0']
SoC = mi200
Profiler choice = rocprofv1
omniperf ver: 1.0.10
Path: /home/colramos/omniperf/workloads/vcopy/mi200
Target: mi200
Command: ./vcopy -n 1048576 -b 256
Kernel Selection: None
Dispatch Selection: ['0']
IP Blocks: All
Log: /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/log.txt
/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt
RPL: on '230411_170356' from '/opt/rocm-5.2.1' in '/home/colramos/GitHub/omniperf-pub'
RPL: profiling '""/home/colramos/vcopy 1048576 256""'
RPL: input file '/home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200/perfmon/SQ_INST_LEVEL_SMEM.txt'
RPL: output dir '/tmp/rpl_data_230411_170356_30314'
RPL: result dir '/tmp/rpl_data_230411_170356_30314/input0_results_230411_170356'
Finished allocating vectors on the CPU
ROCProfiler: input from "/tmp/rpl_data_230411_170356_30314/input0.xml"
gpu_index =
kernel =
range = 0
KernelName verbose: 2
...
```
@@ -383,27 +375,28 @@ Standalone Roofline Options:
#### Roofline Only
The following example demonstrates profiling roofline data only:
```shell-session
$ omniperf profile --name vcopy --roof-only -- ./vcopy 1048576 256
Resolving rocprof
ROC Profiler: /usr/bin/rocprof
$ omniperf profile --name vcopy --roof-only -- ./vcopy -n 1048576 -b 256
ROC Profiler: /opt/rocm-5.2.1/bin/rocprof
Execution mode = profile
___ _ __
/ _ \ _ __ ___ _ __ (_)_ __ ___ _ __ / _|
| | | | '_ ` _ \| '_ \| | '_ \ / _ \ '__| |_
| |_| | | | | | | | | | | |_) | __/ | | _|
\___/|_| |_| |_|_| |_|_| .__/ \___|_| |_|
|_|
--------
Roofline only
--------
Checking for roofline.csv in /home/colramos/GitHub/omniperf-pub/workloads/vcopy/mi200
No roofline data found. Generating...
Empirical Roofline Calculation
Copyright © 2022 Advanced Micro Devices, Inc. All rights reserved.
Total detected GPU devices: 4
GPU Device 0: Profiling...
99% [||||||||||||||||||||||||||||||||||||||||||||||||||||||||||| ]
... ...
Checking for roofline.csv in /home/colramos/GitHub/omniperf-pub/workloads/mix/mi200
Checking for sysinfo.csv in /home/colramos/GitHub/omniperf-pub/workloads/mix/mi200
Checking for pmc_perf.csv in /home/colramos/GitHub/omniperf-pub/workloads/mix/mi200
Empirical Roofline PDFs saved!
SoC = mi200
Profiler choice = rocprofv1
[roofline] Generating pmc_perf.csv
omniperf ver: 1.0.10
Path: /home/colramos/omniperf/workloads/vcopy_test/mi200
Target: mi200
Command: ./vcopy -n 1048576 -b 256
Kernel Selection: None
Dispatch Selection: None
IP Blocks: All
KernelName verbose: 2
...
```
An inspection of our workload output folder shows .pdf plots were generated successfully
```shell-session