[Doc] Fix URLs (ROCmSoftwarePlatform -> ROCm) in the doc, comments, a…

…nd code. + more (#2597)

* Update URLs (ROCmSoftwarePlatform -> ROCm) in the documentation and comments in the source code.

* (2) Update URLs (ROCmSoftwarePlatform -> ROCm) in the documentation and comments in the source code.

* Fix incorrect link

* Fix links

CMakeLists.txt

@@ -341,7 +341,7 @@ if(MIOPEN_USE_MLIR)
         find_library(LIBMLIRMIOPEN MLIRMIOpen REQUIRED)
         if(NOT LIBMLIRMIOPEN)
             message(FATAL_ERROR "library libMLIRMIOpen not found, please reinstall dependencies. \
-            Refer to https://github.com/ROCmSoftwarePlatform/MIOpen#installing-the-dependencies")
+            Refer to https://github.com/ROCm/MIOpen#installing-the-dependencies")
         else()
             message(STATUS "Build with library libMLIRMIOpen: " ${LIBMLIRMIOPEN})
             set(rocMLIR_VERSION 0.0.1)

README.md

@@ -1,13 +1,13 @@
 # MIOpen
 
 AMD's library for high performance machine learning primitives.
-Sources and binaries can be found at [MIOpen's GitHub site](https://github.com/ROCmSoftwarePlatform/MIOpen).
+Sources and binaries can be found at [MIOpen's GitHub site](https://github.com/ROCm/MIOpen).
 The latest released documentation can be read online [here](https://rocm.docs.amd.com/projects/MIOpen/en/latest/index.html).
 
-MIOpen supports two programming models
+MIOpen supports two programming models, or backends:
 
-1. [HIP](https://github.com/ROCm-Developer-Tools/HIP) (Primary Support).
-2. OpenCL.
+1. [HIP](https://github.com/ROCm-Developer-Tools/HIP)
+2. OpenCL (deprecated).
 
 ## Documentation
 
@@ -44,11 +44,11 @@ python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
 * [SQLite3](https://sqlite.org/index.html) - reading and writing performance database
 * lbzip2 - multi-threaded compress or decompress utility
 * [MIOpenTENSILE](https://github.com/ROCmSoftwarePlatform/MIOpenTensile) - users can enable this library using the cmake configuration flag`-DMIOPEN_USE_MIOPENTENSILE=On`. (deprecated after ROCm 5.1.1)
-* [rocBLAS](https://github.com/ROCmSoftwarePlatform/rocBLAS) - AMD library for Basic Linear Algebra Subprograms (BLAS) on the ROCm platform.
-  * Minimum version branch for pre-ROCm 3.5 [master-rocm-2.10](https://github.com/ROCmSoftwarePlatform/rocBLAS/tree/master-rocm-2.10)
-  * Minimum version branch for post-ROCm 3.5 [master-rocm-3.5](https://github.com/ROCmSoftwarePlatform/rocBLAS/releases/tag/rocm-3.5.0)
-* [MLIR](https://github.com/ROCmSoftwarePlatform/llvm-project-mlir) - (Multi-Level Intermediate Representation) with its MIOpen dialect to support and complement kernel development.
-* [Composable Kernel](https://github.com/ROCmSoftwarePlatform/composable_kernel) - C++ templated device library for GEMM-like and reduction-like operators.
+* [rocBLAS](https://github.com/ROCm/rocBLAS) - AMD library for Basic Linear Algebra Subprograms (BLAS) on the ROCm platform.
+  * Minimum version branch for pre-ROCm 3.5 [master-rocm-2.10](https://github.com/ROCm/rocBLAS/tree/master-rocm-2.10)
+  * Minimum version branch for post-ROCm 3.5 [master-rocm-3.5](https://github.com/ROCm/rocBLAS/releases/tag/rocm-3.5.0)
+* [MLIR](https://github.com/ROCm/llvm-project-mlir) - (Multi-Level Intermediate Representation) with its MIOpen dialect to support and complement kernel development.
+* [Composable Kernel](https://github.com/ROCm/composable_kernel) - C++ templated device library for GEMM-like and reduction-like operators.
 
 ## Installing MIOpen with pre-built packages
 
@@ -84,7 +84,7 @@ The script `utils/install_precompiled_kernels.sh` provided as part of MIOpen aut
 
 The above script depends on the *rocminfo* package to query the GPU architecture.
 
-More info can be found [here](https://github.com/ROCmSoftwarePlatform/MIOpen/blob/develop/docs/cache.md#installing-pre-compiled-kernels).
+More info can be found [here](https://github.com/ROCm/MIOpen/blob/develop/docs/cache.md#installing-pre-compiled-kernels).
 
 ## Installing the dependencies
 
@@ -104,7 +104,7 @@ cmake -P install_deps.cmake --minimum --prefix /root/MIOpen/install_dir
 
 This prefix can used to specify the dependency path during the configuration phase using the `CMAKE_PREFIX_PATH`.
 
-* MIOpen's HIP backend uses [rocBLAS](https://github.com/ROCmSoftwarePlatform/rocBLAS) by default. Users can install rocBLAS minimum release by using `apt-get install rocblas`. To disable using rocBLAS set the configuration flag `-DMIOPEN_USE_ROCBLAS=Off`. rocBLAS is *not* available for the OpenCL backend.
+* MIOpen's HIP backend uses [rocBLAS](https://github.com/ROCm/rocBLAS) by default. Users can install rocBLAS minimum release by using `apt-get install rocblas`. To disable using rocBLAS set the configuration flag `-DMIOPEN_USE_ROCBLAS=Off`. rocBLAS is *not* available for the OpenCL backend.
 
 * MIOpen's OpenCL backend uses [MIOpenGEMM](https://github.com/ROCmSoftwarePlatform/MIOpenGEMM) by default. Users can install MIOpenGEMM minimum release by using `apt-get install miopengemm`.
 
@@ -212,7 +212,7 @@ This will install the library to the `CMAKE_INSTALL_PREFIX` path that was set.
 
 ## Building the driver
 
-MIOpen provides an [application-driver](https://github.com/ROCmSoftwarePlatform/MIOpen/tree/master/driver) which can be used to execute any one particular layer in isolation and measure performance and verification of the library.
+MIOpen provides an [application-driver](https://github.com/ROCm/MIOpen/tree/master/driver) which can be used to execute any one particular layer in isolation and measure performance and verification of the library.
 
 The driver can be built using the `MIOpenDriver` target:
 
@@ -338,5 +338,5 @@ MIOpen's paper is freely available and can be accessed on arXiv:
 ## Porting from cuDNN to MIOpen
 
 The [porting
-guide](https://github.com/ROCmSoftwarePlatform/MIOpen/tree/develop/docs/MIOpen_Porting_Guide.md)
+guide](https://github.com/ROCm/MIOpen/tree/develop/docs/MIOpen_Porting_Guide.md)
 highlights the key differences between the current cuDNN and MIOpen APIs.

docs/DebugAndLogging.md

@@ -172,14 +172,14 @@ Additionally, using environment variable "MIOPEN_GEMM_ENFORCE_BACKEND", can over
 both MIOpenGEMM and rocBlas depending on the input configuration:
 
 * `MIOPEN_GEMM_ENFORCE_BACKEND=1`, use rocBLAS if enabled
-* `MIOPEN_GEMM_ENFORCE_BACKEND=2`, use MIOpenGEMM for FP32, use rocBLAS for FP16 if enabled
+* `MIOPEN_GEMM_ENFORCE_BACKEND=2`, reserved
 * `MIOPEN_GEMM_ENFORCE_BACKEND=3`, no gemm will be called
-* `MIOPEN_GEMM_ENFORCE_BACKEND=4`, use MIOpenTensile for FP32, use rocBLAS for FP16 if enabled
+* `MIOPEN_GEMM_ENFORCE_BACKEND=4`, reserved
 * `MIOPEN_GEMM_ENFORCE_BACKEND=<any other value>`, use default behavior
 
 To disable using rocBlas entirely, set the configuration flag `-DMIOPEN_USE_ROCBLAS=Off` during MIOpen configuration.
 
-More information on logging with rocBlas can be found [here](https://github.com/ROCmSoftwarePlatform/rocBLAS/wiki/5.Logging).
+More information on logging with rocBlas can be found [here](https://github.com/ROCm/rocBLAS/wiki/5.Logging).
 
 
 ## Numerical Checking

docs/Getting_Started_FusionAPI.md

@@ -3,7 +3,7 @@ Fusion API: Getting Started
 ## Introduction
 Increasing depth of deep learning networks necessitate the need for novel mechanisms to improve performance on GPUs. One mechanism to achieve higher efficiency is to _fuse_ separate kernels into a single kernel to reduce off-chip memory access and avoid kernel launch overhead. This document outlines the addition of a Fusion API to the MIOpen library. The fusion API would allow users to specify operators that they wants to fuse in a single kernel, compile it and then launch the kernel. While not all combinations might be supported by the library, the API is flexible enough to allow the specification of many operations in any order from a finite set of supported operations. The API provides a mechanism to report unsupported combinations.
 
-A complete example of the Fusion API in the context of MIOpen is given [here](https://github.com/ROCmSoftwarePlatform/MIOpenExamples/tree/master/fusion). We will use code from the example project as we go along. The example project creates a fusion plan to merge the convolution, bias and activation operations. For a list of supported fusion operations and associated constraints please refer to the [Supported Fusions](#supported-fusions) section. The example depicts bare-bones code without any error checking or even populating the tensors with meaningful data in the interest of simplicity.
+A complete example of the Fusion API in the context of MIOpen is given [here](https://github.com/ROCm/MIOpenExamples/tree/master/fusion). We will use code from the example project as we go along. The example project creates a fusion plan to merge the convolution, bias and activation operations. For a list of supported fusion operations and associated constraints please refer to the [Supported Fusions](#supported-fusions) section. The example depicts bare-bones code without any error checking or even populating the tensors with meaningful data in the interest of simplicity.
 
 The following list outlines the steps required 
 
@@ -48,7 +48,7 @@ The fusion API introduces the notion of **operators** which represent different
 
 Notice that _Bias_ is a separate operator, although it is typically only used with convolution. This list is expected to grow as support for more operators is added to the API, moreover, operators for backward passes are in the works as well.
 
-The fusion API provides calls for the creation of the supported operators, here we would describe the process for the convolution operator, details for other operators may be found in the [miopen header file](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/fusion.html) 
+The fusion API provides calls for the creation of the supported operators, here we would describe the process for the convolution operator, details for other operators may be found in the [miopen header file](https://rocm.docs.amd.com/projects/MIOpen/en/latest/fusion.html) 
 
 Once the fusion plan descriptor is created, two or more operators can be added to it by using the individual operator creation API calls. Creation of an operator might fail if the API does not support the fusion of the operations being added and report back immediately to the user. For our example we need to add the Convolution, Bias and Activation operations to our freshly minted fusion plan. This is done using the following calls for the Convolution, Bias and Activation operations respectively:
 
@@ -77,7 +77,7 @@ miopenCreateOpBiasForward(fusePlanDesc, &biasOp, bias.desc);
 miopenCreateOpActivationForward(fusePlanDesc, &activOp, miopenActivationRELU);
 ```
 
-It may be noted that `conv_desc` is the regular MIOpen Convolution descriptor and is created in the standard way before it is referenced here. For more details on creating and setting the convolution descriptor please refer to the example code as well as the [MIOpen documentation](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/convolution.html). In the above snippet `weights.desc` refers to the `miopenTensorDescriptor_t` for the convolution operations and `bias.desc` refers to the object of the same type for the bias operation. The order of insertion of operators indicates the order in which the operations would be performed on the data. Therefore, the above code implies that the convolution operation would be the first operation to execute on the incoming data, followed by the bias and activation operations. 
+It may be noted that `conv_desc` is the regular MIOpen Convolution descriptor and is created in the standard way before it is referenced here. For more details on creating and setting the convolution descriptor please refer to the example code as well as the [MIOpen documentation](https://rocm.docs.amd.com/projects/MIOpen/en/latest/convolution.html). In the above snippet `weights.desc` refers to the `miopenTensorDescriptor_t` for the convolution operations and `bias.desc` refers to the object of the same type for the bias operation. The order of insertion of operators indicates the order in which the operations would be performed on the data. Therefore, the above code implies that the convolution operation would be the first operation to execute on the incoming data, followed by the bias and activation operations. 
 
 During this process, it is important that the returned codes be checked to make sure that the operations as well as their order is supported. The operator insertion might fail for a number of reasons such as unsupported sequence of operations, unsupported dimensions of the input or in case of convolution unsupported dimensions for the filters. In the above example, these aspects are ignored for the sake of simplicity.
 

docs/cache.md

@@ -15,7 +15,7 @@ The are several ways to disable the cache. This is generally useful for developm
 
 Updating MIOpen and removing the cache
 --------------------------------------
-For MIOpen version 2.3 and earlier, if the compiler changes, or the user modifies the kernels then the cache must be deleted for the MIOpen version in use; e.g., `rm -rf $HOME/.cache/miopen/<miopen-version-number>`. More information about the cache can be found [here](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/cache.html).
+For MIOpen version 2.3 and earlier, if the compiler changes, or the user modifies the kernels then the cache must be deleted for the MIOpen version in use; e.g., `rm -rf $HOME/.cache/miopen/<miopen-version-number>`. More information about the cache can be found [here](https://rocm.docs.amd.com/projects/MIOpen/en/latest/cache.html).
 
 For MIOpen version 2.4 and later, MIOpen's kernel cache directory is versioned so that users' cached kernels will not collide when upgrading from earlier version.
 
@@ -30,4 +30,4 @@ If MIOpen kernels package is not installed, or if we do not deliver the kernels
 
 The performance degradation mentioned in the warning only affects the network start-up time (aka "initial iteration time") and thus can be safely ignored.
 
-Please refer to the MIOpen installation instructions: [installing MIOpen kernels package](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/install.html#installing-miopen-kernels-package) for guidance on installing the MIOpen kernels package.
+Please refer to the MIOpen installation instructions: [installing MIOpen kernels package](https://rocm.docs.amd.com/projects/MIOpen/en/latest/install.html#installing-miopen-kernels-package) for guidance on installing the MIOpen kernels package.

docs/driver.md

@@ -1,9 +1,9 @@
 ## Building the driver
 
-MIOpen provides an [application-driver](https://github.com/ROCmSoftwarePlatform/MIOpen/tree/master/driver) which can be used to execute any one particular layer in isolation and measure performance and verification of the library. 
+MIOpen provides an [application-driver](https://github.com/ROCm/MIOpen/tree/master/driver) which can be used to execute any one particular layer in isolation and measure performance and verification of the library. 
 
 The driver can be built using the `MIOpenDriver` target:
 
 ` cmake --build . --config Release --target MIOpenDriver ` **OR** ` make MIOpenDriver `
 
-Documentation on how to run the driver is [here](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/driver.html). 
+Documentation on how to run the driver is [here](https://rocm.docs.amd.com/projects/MIOpen/en/latest/driver.html). 

docs/find_and_immediate.md

@@ -64,7 +64,7 @@ The results of Find() are returned in an array of `miopenConvAlgoPerf_t` structs
 
 This call sequence is executed once per session as it is inherently expensive. Of those, `miopenFindConvolution*()` is the most expensive call. It caches its own results on disk, so the subsequent calls during the same MIOpen session will execute faster. However, it is better to remember results of `miopenFindConvolution*()` in the application, as recommended above. 
 
-Internally MIOpen's Find calls will compile and benchmark a set of `solvers` contained in `miopenConvAlgoPerf_t` this is done in parallel per `miopenConvAlgorithm_t`. The level of parallelism can be controlled using an environment variable. See the debugging section [controlling parallel compilation](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/DebugAndLogging.html#controlling-parallel-compilation) for more details.
+Internally MIOpen's Find calls will compile and benchmark a set of `solvers` contained in `miopenConvAlgoPerf_t` this is done in parallel per `miopenConvAlgorithm_t`. The level of parallelism can be controlled using an environment variable. See the debugging section [controlling parallel compilation](https://rocm.docs.amd.com/projects/MIOpen/en/latest/DebugAndLogging.html#controlling-parallel-compilation) for more details.
 
 
 ## Immediate Mode API
@@ -139,7 +139,7 @@ miopenConvolutionForwardCompileSolution(handle,
 
 ## Immediate Mode Fallback
 
-The immediate mode is underpinned by the [Find-Db](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/finddb.html), however it may not contain every configuration of interest. If Find-Db encounters a database miss it has two fallback paths it can take, depending on whether the cmake variable MIOPEN_ENABLE_AI_IMMED_MODE_FALLBACK is set to ON or OFF. However, if the user requires the best possible performance they should run the Find stage at least once.
+The immediate mode is underpinned by the [Find-Db](https://rocm.docs.amd.com/projects/MIOpen/en/latest/finddb.html), however it may not contain every configuration of interest. If Find-Db encounters a database miss it has two fallback paths it can take, depending on whether the cmake variable MIOPEN_ENABLE_AI_IMMED_MODE_FALLBACK is set to ON or OFF. However, if the user requires the best possible performance they should run the Find stage at least once.
 
 ### 1. AI-based Heuristic Fallback (Default)
 
@@ -173,10 +173,10 @@ OpenCL support for immediate mode via the fallback is limited to fp32 datatypes.
 MIOpen provides a set of Find modes which are used to accelerate the Find calls. The different modes are set by using the environment variable `MIOPEN_FIND_MODE`, and setting it to one of the values:
 
 - `NORMAL`, or `1`: Normal Find: This is the full Find mode call, which will benchmark all the solvers and return a list.
-- `FAST`, or `2`: Fast Find: Checks the [Find-Db](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/finddb.html) for an entry. If there is a Find-Db hit, use that entry. If there is a miss, utilize the Immediate mode fallback. If Start-up times are expected to be faster, but worse GPU performance.
-- `HYBRID`, or `3`, or unset `MIOPEN_FIND_MODE`: Hybrid Find: Checks the [Find-Db](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/finddb.html) for an entry. If there is a Find-Db hit, use that entry. If there is a miss, use the existing Find machinery. Slower start-up times than Fast Find, but no GPU performance drop.
+- `FAST`, or `2`: Fast Find: Checks the [Find-Db](https://rocm.docs.amd.com/projects/MIOpen/en/latest/finddb.html) for an entry. If there is a Find-Db hit, use that entry. If there is a miss, utilize the Immediate mode fallback. If Start-up times are expected to be faster, but worse GPU performance.
+- `HYBRID`, or `3`, or unset `MIOPEN_FIND_MODE`: Hybrid Find: Checks the [Find-Db](https://rocm.docs.amd.com/projects/MIOpen/en/latest/finddb.html) for an entry. If there is a Find-Db hit, use that entry. If there is a miss, use the existing Find machinery. Slower start-up times than Fast Find, but no GPU performance drop.
 - `4`: This value is reserved and should not be used.
-- `DYNAMIC_HYBRID`, or `5`: Dynamic Hybrid Find: Checks the [Find-Db](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/finddb.html) for an entry. If there is a Find-Db hit, uses that entry. If there is a miss, uses the existing Find machinery with skipping non-dynamic kernels. Faster start-up times than Hybrid Find, but GPU performance may be a bit worse.
+- `DYNAMIC_HYBRID`, or `5`: Dynamic Hybrid Find: Checks the [Find-Db](https://rocm.docs.amd.com/projects/MIOpen/en/latest/finddb.html) for an entry. If there is a Find-Db hit, uses that entry. If there is a miss, uses the existing Find machinery with skipping non-dynamic kernels. Faster start-up times than Hybrid Find, but GPU performance may be a bit worse.
 
  Currently, the default Find mode is `DYNAMIC_HYBRID`. To run the full `NORMAL` Find mode, set the environment as:
  ```