Layer/Routine Operation Commands¶
Layer and routine related SoftNeuro commands. It is possible to verify routine information per layer, such as available devices and parameters.
plugins¶
Manages plugins for additional routines. User can view the list of installed plugins, install/uninstall plugins, and view the list of all avalable plugins with their status.
Usage
softneuro plugins [--help] [-i PLUGIN|auto] [-u PLUGIN|all] [-s]
Flags
| Flag | Description |
|---|---|
| -h, --help | Shows the command help. |
| -i, --install PLUGIN | Installs a plugin PLUGIN. If auto is provieded, it installs all appropriate plugins automatically. |
| -u, --uninstall PLUGIN | Uninstall a plugin PLUGIN. If all is provided, it uninstalls all plugins. |
| -s, --status | Lists available plugins and their status. The meaning of staus symbols is shown below. |
| Symbol | Status |
|---|---|
I |
Installed successfully |
C |
Installed but conflicted with other plugins |
? |
Installed but may not be loaded on run time |
|
Not installed |
Example
Running without any options, you can view the installed plugins.
$ softneuro plugins
plugin_avx2 AVX routines.
plugin_opencl OpenCL routines.
You may check if other plugins are available. The following result shows plugin_cuda is avalable.
$ softneuro plugins -s
I plugin_avx2
I plugin_opencl
plugin_cuda
Then you can install plugin_cuda by the following command.
$ softneuro plugin -i plugin_cuda
`plugin_cuda` is installed.
layers¶
List all available layers.
Usage
usage: softneuro layers [--help]
Flags
| Flag | Description |
|---|---|
| -h, --help | Shows the command help. |
Example
$ softneuro layers
adaptive_unit_scale
add
ave_pool2
batch_norm
box_filter
:
lparams¶
Shows the available paramters and their default values for a given layer.
Usage
usage: softneuro lparams [--help] LAYER
Arguments
| Argument | Description |
|---|---|
| LAYER | Layer to have its parameters shown. |
Flags
| Flag | Description |
|---|---|
| -h, --help | Shows the command help. |
Example
$ softneuro lparams conv2
{'padding': 'valid', 'pads': 0, 'dilations': 1, 'strides': 1, 'has_relu': false, 'relu_max_value': -1.0}
routines¶
Shows the available execution routines in the current environment. Shows the available execution schemas in the current environment.
Usage
usage: softneuro routines [--help] LAYER
Arguments
| Argument | Description |
|---|---|
| LAYER | Layer to have its routines shown. Shows schemas when no layer was set. |
Flags
| Flag | Description |
|---|---|
| -h, --help | Shows the command help. |
Example
$ softneuro routines conv2
cpu/naive
cpu/naive/wg2
cpu:chf/naive
cpu:qint8:pt_sym/naive
cpu:qint8:pt_asym/naive
cpu:qint8:pc_sym/naive
cpu:qint8:pc_asym/naive
cpu:qint8/naive
cpu/wg2
:
rparams¶
Show a recipe for routine parameters.
Usage
usage: softneuro rparams [--help] LAYER
Arguments
| Argument | Description |
|---|---|
| LAYER | Layer that has the routine for which parameters will be shown. |
| ROUTINE | Routine to have its parameters shown. |
Flags
| Flag | Description |
|---|---|
| -h, --help | Shows the command help. |
Example
$ softneuro rparams conv2 cpu/wg3x3_nxn_avx
{'cache': [8192, 16384, 32768], 'task_unit': [4, 8, 16], 'tile_size': [2, 4, 6, 8]}
test¶
Does a test run of the given layer.
Usage
usage: softneuro test [-o OUTPUT]... [--weight WNAME TENSOR]...[--thread NTHREADS]
[--affinity MASK[@IDS]]
[--params PARAMS][--routine ROUTINE] [--rparams RPARAMS]
[--loop NLOOPS] [--help]
LAYER [INPUT]...
Arguments
| Argument | Description |
|---|---|
| LAYER | Layer to be tested. |
| INPUT | Input numpy file or tensor specification. |
Flags
| Flag | Description |
|---|---|
| -o OUTPUT, --output OUTPUT | Filename for the output numpy file. Must be set. |
| --weight WNAME TENSOR | Insert TENSOR data into the WNAME weight tensor. TENSOR can be a numpy file or tensor specification. If weights aren't set they are initialized to random numbers in [-1, 1]. |
| --thread NTHREADS | How many threads to be used on execution. Defaults to the amount of CPU cores. |
| --affinity MASK[@THREAD_INDICES] | Use the affinity mask given by MASK on the threads given byTHREAD_INDICES. MASK should be a little endian hexadecimal (0x..), binary (0b..), or decimal number. If THREAD_INDICES isn't set all threads will use the given mask. For more information on THREAD_INDICES use the softneuro help thread_indices command. |
| --params PARAMS | Layer parameters in JSON format. Parameters can be checked with the lparams command. |
| -r ROUTINE, --routine ROUTINE | Set the routine to be executed. Available routines can be checked with the routines command. Defaults to the best available. |
| --rparams RPARAMS | Routine parameters in JSON format. Parameters can be cheked with the rparams command. |
| -l NLOOPS, --loop NLOOPS | Amount of times to run the test. Useful for benchmarking. |
| -h, --help | Shows the command help. |
Example
$ softneuro test -o out.npy --weight kernel "shape=3x3x128x128" --weight bias "shape=128" --params '{"padding": "valid", "stride": 2}' --routine "cpu/fast" conv2 "shape=1x64x64x128"
---------------------------------
Statistics
---------------------------------
FUNCTION AVE(usec) MIN(usec) MAX(usec) #CALL
Dnn_forward() 2,348 2,348 2,348 1
Used memory: 9,924,608 Bytes