Merge branch '1.3.x' of https://github.com/AMDComputeLibraries/MLOpen …

…into 1.3.x

README.md

@@ -18,7 +18,17 @@ AMD's library for high peformance machine learning primitives. MIOpen supports t
 * [Boost](http://www.boost.org/) at least version 1.58
   * MIOpen uses `boost-system` and `boost-filesystem` packages to enable persistent [kernel cache](https://github.com/ROCmSoftwarePlatform/MIOpen/blob/master/doc/src/cache.md)
 
-Instructions to install the above dependencies are present in this [section](#installing-the-dependencies).
+## Installing the dependencies
+
+The dependencies can be installed with the `install_deps.cmake`, script: `cmake -P install_deps.cmake`
+
+
+This will install by default to `/usr/local` but it can be installed in another location with `--prefix` argument:
+```
+cmake -P install_deps.cmake --prefix /some/local/dir
+```
+
+Instructions to manually install all the dependencies on Ubuntu v16 are present in this [section](#installing-the-dependencies-manually).
 
 ## Installing MIOpen with pre-built packages
 
@@ -167,19 +177,7 @@ Also, githooks can be installed to format the code per-commit:
 ./.githooks/install
 ```
 
-## Installing the dependencies
-
-The dependencies can be installed with the `install_deps.cmake`, script:
-
-```
-cmake -P install_deps.cmake
-```
-
-This will install by default to `/usr/local` but it can be installed in another location with `--prefix` argument:
-
-```
-cmake -P install_deps.cmake --prefix /some/local/dir
-```
+## Installing the dependencies manually
 
 If Ubuntu v16 is used then the `OpenSSL` and `Boost` packages can also be installed by:
 ```
@@ -188,3 +186,15 @@ sudo apt-get install libboost-dev
 sudo apt-get install libboost-system-dev
 sudo apt-get install libboost-filesystem-dev
 ```
+
+`half` header needs to be installed from [here](http://half.sourceforge.net/). 
+
+## Using docker
+
+The easiest way is to use docker. You can build the top-level docker file:
+
+    docker build -t miopen .
+
+Then to enter the developement environment use `docker run`:
+
+    docker run --device='/dev/kfd' --device='/dev/dri' -v=`pwd`:/data -w /data --group-add video -it miopen

doc/src/cache.md

@@ -6,7 +6,7 @@ MIOpen will cache binary kernels to disk, so they don't need to be compiled the
 Clear the cache
 ---------------
 
-The cache can be cleared by simply deleting the cache directory(ie `$HOME/.cache/miopen`). This should only be needed for development purposes or to free disk space. The cache does not need to be cleared when upgrading MIOpen.
+The cache can be cleared by simply deleting the cache directory (i.e., `$HOME/.cache/miopen`). This should only be needed for development purposes or to free disk space. The cache does not need to be cleared when upgrading MIOpen.
 
 Disabling the cache
 -------------------

doc/src/perfdatabase.md

@@ -11,6 +11,9 @@ MIOpen performs Exhaustive Search only if explicitly requested via MIOpen API an
 
 The optimized solution found during the successful Search process is written into the PerfDb for future re-use. That is why MIOpen will not Search for optimized solution more than once for a given problem in this mode.
 
+See documentation about miopenFind*() API calls for more info on how Search can be explicitly requested.
+
+
 **DB_UPDATE (2)**
 
 Similar to NONE, but Search will NOT be skipped if PerfDb contains relevant record. If Search is requested via MIOpen API, then MIOpen will perform the Search and update PerfDb.
@@ -34,3 +37,15 @@ Note: This mode is intended for tuning the MIOpen installation. When MIOpen is i
 **DB_CLEAN (5)**
 
 MIOpen removes relevant records from the PerfDb instead of just reading and using those. Search is blocked, even if explicitly requested.
+
+## MIOPEN_FIND_ENFORCE_SCOPE
+
+This variable allows to limit the scope of `MIOPEN_FIND_ENFORCE`, so that only forward, backward data or backward weights convolutions will be affected. Both symbolic and numeric values are supported, as shown below.
+
+**ALL (1)** `MIOPEN_FIND_ENFORCE` affects all convolutions. This is the default.
+
+**CONV_FWD (2)** `MIOPEN_FIND_ENFORCE` affects only Forward convolutions.
+
+**CONV_BWD (3)** `MIOPEN_FIND_ENFORCE` affects only Backward Data convolutions.
+
+**CONV_WRW (4)** `MIOPEN_FIND_ENFORCE` affects only Backward With Regard to Weights (a.k.a WRW) convolutions.

doc/src/releasenotes.md

@@ -1,6 +1,39 @@
 
 ## MIOpen Release notes
 
+### 03/30/2018 [ 1.3.0 ]
+
+Notes: 
+
+- Performance improvements for RNNs
+- Performance improvements for convolutions using 1x1 filters
+- Performance improvement for Batch Normalization
+- This release adds preliminary fp16 support for Inference using CNNs
+- Bug fixes for various components of MIOpen
+
+Changes:
+
+- Added 2 new API for RNNs: miopenGetRNNLayerParamOffset and miopenGetRNNLayerBiasOffset
+- Added support for uninitialized hidden states and nullptr outputs in RNNs
+- Added support for Set and Scale operations for strided tensors with dimensions 1 to 5
+- Added multi-thread and multi-process support for the performance database
+- Improved performance for OpTensor
+- Fixed bug in convolutions for backward bias
+- Fixed logic issues in get and set layer functions and related w_supertensor test
+- Fixed hang in batch norm with batch sizes greater than 256
+
+Known Issues:
+
+- RNNs do not support fp16
+- Training with CNNs does not support fp16
+
+
+### 03/08/2018 [ 1.2.1 ]
+
+Notes:
+
+- This release adds support for ROCm 1.7.1.
+
 
 ### 12/15/2017 [ 1.2.0 ]
 

doc/src/rnn.rst

@@ -51,7 +51,6 @@ miopenGetRNNDescriptor
 .. doxygenfunction::  miopenGetRNNDescriptor
 
 
-
 miopenDestroyRNNDescriptor
 --------------------------
 
@@ -135,6 +134,17 @@ miopenSetRNNLayerBias
 
 .. doxygenfunction::  miopenSetRNNLayerBias
 
+miopenGetRNNLayerParamOffset
+----------------------------
+
+.. doxygenfunction::  miopenGetRNNLayerParamOffset
+
+
+miopenGetRNNLayerBiasOffset
+---------------------------
+
+.. doxygenfunction::  miopenGetRNNLayerBiasOffset
+
 miopenRNNForwardTraining
 ------------------------