experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Docfix: Document usage of --config=opt for SIMD instruction sets. (#7211)

Browse Source 
* Docfix: Document usage of --config=opt for SIMD instruction sets.
This commit is contained in:
gunan 2017-02-02 09:27:52 -08:00 committed by Martin Wicke
parent 85235cbcb4
commit b47dc70e54
14 changed files with 31 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
configure vendored
View File

@ -76,7 +76,8 @@ done
## Set up architecture-dependent optimization flags. ## Set up architecture-dependent optimization flags.
if [ -z "$CC_OPT_FLAGS" ]; then if [ -z "$CC_OPT_FLAGS" ]; then
default_cc_opt_flags="-march=native" default_cc_opt_flags="-march=native"
read -p "Please specify optimization flags to use during compilation [Default is $default_cc_opt_flags]: " CC_OPT_FLAGS read -p "Please specify optimization flags to use during compilation when bazel option "\
"\"--config=opt\" is specified [Default is $default_cc_opt_flags]: " CC_OPT_FLAGS
if [ -z "$CC_OPT_FLAGS" ]; then if [ -z "$CC_OPT_FLAGS" ]; then
CC_OPT_FLAGS=$default_cc_opt_flags CC_OPT_FLAGS=$default_cc_opt_flags
fi fi

4
tensorflow/core/kernels/BUILD
View File

@ -7,9 +7,9 @@
# append "_gpu" to the test name to invoke the GPU benchmarks. Example: # append "_gpu" to the test name to invoke the GPU benchmarks. Example:
# #
# # for CPU tests # # for CPU tests
# $ bazel test -c opt --copt=-mavx //third_party/tensorflow/core/kernels:my_op_test # $ bazel test --config opt //third_party/tensorflow/core/kernels:my_op_test
# # for GPU benchmarks # # for GPU benchmarks
# $ bazel run -c opt --copt=-mavx --config=cuda //third_party/tensorflow/core/kernels:my_op_test_gpu -- --benchmarks=.. # $ bazel run --config opt --config=cuda //third_party/tensorflow/core/kernels:my_op_test_gpu -- --benchmarks=..
# #
package(default_visibility = ["//visibility:public"]) package(default_visibility = ["//visibility:public"])

2
tensorflow/examples/how_tos/reading_data/fully_connected_preloaded.py
View File

@ -17,7 +17,7 @@
Run using bazel: Run using bazel:
bazel run -c opt \ bazel run --config opt \
<...>/tensorflow/examples/how_tos/reading_data:fully_connected_preloaded <...>/tensorflow/examples/how_tos/reading_data:fully_connected_preloaded
or, if installed via pip: or, if installed via pip:

2
tensorflow/examples/how_tos/reading_data/fully_connected_preloaded_var.py
View File

@ -17,7 +17,7 @@
Run using bazel: Run using bazel:
bazel run -c opt \ bazel run --config opt \
<...>/tensorflow/examples/how_tos/reading_data:fully_connected_preloaded_var <...>/tensorflow/examples/how_tos/reading_data:fully_connected_preloaded_var
or, if installed via pip: or, if installed via pip:

2
tensorflow/examples/multibox_detector/README.md
View File

@ -32,7 +32,7 @@ normalized from 0-1 in left top right bottom order.
To build it, run this command: To build it, run this command:
```bash ```bash
$ bazel build -c opt tensorflow/examples/multibox_detector/... $ bazel build --config opt tensorflow/examples/multibox_detector/...
``` ```
That should build a binary executable that you can then run like this: That should build a binary executable that you can then run like this:

27
tensorflow/g3doc/get_started/os_setup.md
View File

@ -856,13 +856,13 @@ default and if you want to limit RAM usage you can add `--local_resources
2048,.5,1.0` while invoking bazel. 2048,.5,1.0` while invoking bazel.
```bash ```bash
$ bazel build -c opt //tensorflow/tools/pip_package:build_pip_package $ bazel build --config opt //tensorflow/tools/pip_package:build_pip_package
# To build with support for CUDA: # To build with support for CUDA:
$ bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package $ bazel build --config opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
# Alternatively, to build with support for OpenCL: # Alternatively, to build with support for OpenCL (Experimental):
$ bazel build -c opt --config=sycl //tensorflow/tools/pip_package:build_pip_package $ bazel build --config opt --config=sycl //tensorflow/tools/pip_package:build_pip_package
$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg $ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
@ -873,20 +873,21 @@ $ sudo pip install /tmp/tensorflow_pkg/tensorflow-0.12.1-py2-none-any.whl
## Optimizing CPU performance ## Optimizing CPU performance
To be compatible with as wide a range of machines as possible, TensorFlow To be compatible with as wide a range of machines as possible, TensorFlow
defaults to only using SSE4.1 SIMD instructions on x86 machines. Most modern PCs defaults to only using SSE4 SIMD instructions. Most modern computers support
and Macs support more advanced instructions, so if you're building a binary more advanced instructions. So if you're building a binary that you'll only
that you'll only be running on your own machine, you can enable these by using be running on your own machine, you can enable these by using `-march=native`
`--copt=-march=native` in your bazel build command. For example: for optimization options when running `configure`. Then you can build your
optimized binaries with the following command:
``` bash ``` bash
$ bazel build --copt=-march=native -c opt //tensorflow/tools/pip_package:build_pip_package $ bazel build --config opt //tensorflow/tools/pip_package:build_pip_package
``` ```
If you are distributing a binary but know the capabilities of the machines If you are distributing a binary but know the capabilities of the machines
you'll be running on, you can manually choose the right instructions with you'll be running on, you can manually choose the right instructions with
something like `--copt=-march=avx`. You may also want to enable multiple something like `-march=avx`. You may also want to enable multiple
features using several arguments, for example features using several arguments, for example
`--copt=-mavx2 --copt=-mfma`. `-mavx2,-mfma`.
If you run a binary built using SIMD instructions on a machine that doesn't If you run a binary built using SIMD instructions on a machine that doesn't
support them, you'll see an illegal instruction error when that code is support them, you'll see an illegal instruction error when that code is
@ -902,10 +903,10 @@ system directories, run the following commands inside the TensorFlow root
directory: directory:
```bash ```bash
bazel build -c opt //tensorflow/tools/pip_package:build_pip_package bazel build --config opt //tensorflow/tools/pip_package:build_pip_package
# To build with GPU support: # To build with GPU support:
bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package bazel build --config opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
mkdir _python_build mkdir _python_build
cd _python_build cd _python_build

2
tensorflow/g3doc/how_tos/adding_an_op/index.md
View File

@ -177,7 +177,7 @@ tf_custom_op_library(
Run the following command to build `zero_out.so`. Run the following command to build `zero_out.so`.
```bash ```bash
$ bazel build -c opt //tensorflow/core/user_ops:zero_out.so $ bazel build --config opt //tensorflow/core/user_ops:zero_out.so
``` ```
> Note: > Note:

4
tensorflow/g3doc/how_tos/image_retraining/index.md
View File

@ -42,10 +42,10 @@ bazel build tensorflow/examples/image_retraining:retrain
If you have a machine which supports [the AVX instruction set](https://en.wikipedia.org/wiki/Advanced_Vector_Extensions) If you have a machine which supports [the AVX instruction set](https://en.wikipedia.org/wiki/Advanced_Vector_Extensions)
(common in x86 CPUs produced in the last few years) you can improve the running (common in x86 CPUs produced in the last few years) you can improve the running
speed of the retraining by building for that architecture, like this: speed of the retraining by building for that architecture, like this (after choosing appropriate options in `configure`):
```sh ```sh
bazel build -c opt --copt=-mavx tensorflow/examples/image_retraining:retrain bazel build --config opt tensorflow/examples/image_retraining:retrain
``` ```
The retrainer can then be run like this: The retrainer can then be run like this:

2
tensorflow/go/README.md
View File

@ -40,7 +40,7 @@ Construct and execute TensorFlow graphs in Go.
```sh ```sh
cd ${GOPATH}/src/github.com/tensorflow/tensorflow cd ${GOPATH}/src/github.com/tensorflow/tensorflow
./configure ./configure
bazel build -c opt //tensorflow:libtensorflow.so bazel build --config opt //tensorflow:libtensorflow.so
``` ```
This can take a while (tens of minutes, more if also building for GPU). This can take a while (tens of minutes, more if also building for GPU).

2
tensorflow/go/genop/generate.sh
View File

@ -30,7 +30,7 @@ then
then then
echo "Protocol buffer compiler protoc not found in PATH or in ${PROTOC}" echo "Protocol buffer compiler protoc not found in PATH or in ${PROTOC}"
echo "Perhaps build it using:" echo "Perhaps build it using:"
echo "bazel build -c opt @protobuf//:protoc" echo "bazel build --config opt @protobuf//:protoc"
exit 1 exit 1
fi fi
PROTOC=$PATH_PROTOC PROTOC=$PATH_PROTOC

2
tensorflow/java/README.md
View File

@ -40,7 +40,7 @@ Configure and build the Java Archive (JAR) and native library:
./configure ./configure
# Build the JAR and native library # Build the JAR and native library
bazel build -c opt \ bazel build --config opt \
//tensorflow/java:tensorflow \ //tensorflow/java:tensorflow \
//tensorflow/java:libtensorflow_jni //tensorflow/java:libtensorflow_jni
``` ```

2
tensorflow/tools/benchmark/README.md
View File

@ -39,7 +39,7 @@ $adb shell "/data/local/tmp/benchmark_model \
### On desktop: ### On desktop:
(1) build the binary (1) build the binary
```bash ```bash
$bazel build -c opt tensorflow/tools/benchmark:benchmark_model $bazel build --config opt tensorflow/tools/benchmark:benchmark_model
``` ```
(2) Run on your compute graph, similar to the Android case but without the need of adb shell. (2) Run on your compute graph, similar to the Android case but without the need of adb shell.

2
tensorflow/tools/lib_package/README.md
View File

@ -13,7 +13,7 @@ and [Rust](https://github.com/tensorflow/rust).
The command: The command:
```sh ```sh
bazel build -c opt //tensorflow/tools/lib_package:libtensorflow bazel build --config opt //tensorflow/tools/lib_package:libtensorflow
``` ```
produces `bazel-bin/tensorflow/tools/lib_package/libtensorflow.tar.gz`, which produces `bazel-bin/tensorflow/tools/lib_package/libtensorflow.tar.gz`, which

2
tensorflow/tools/tfprof/README.md
View File

@ -98,7 +98,7 @@ TODO(xpan): Provide graph.pbtxt, model.ckpt, tfprof_log and run_meta download.
```shell ```shell
# Build the tool. # Build the tool.
bazel build -c opt tensorflow/tools/tfprof/... bazel build --config opt tensorflow/tools/tfprof/...
# Help information, including detail 'option' instructions. # Help information, including detail 'option' instructions.
bazel-bin/tensorflow/tools/tfprof/tfprof help bazel-bin/tensorflow/tools/tfprof/tfprof help