Adding bitwise reproducibility documentation (#403)

* Adding bitwise reproducibility documentation

* fix typo

* Change wording

* Grammar fix

* Add notes to individual function documentation

docs/bitwise-repro.rst

@@ -0,0 +1,16 @@
+.. meta::
+  :description: rocThrust documentation and API reference
+  :keywords: rocThrust, ROCm, API, reference, data type, support
+
+.. _bitwise-repro:
+
+******************************************
+Bitwise reproducibility
+******************************************
+
+With the exception of the following functions, all rocThrust API functions are bitwise reproducible.  That is, given identical inputs, the function will return the exact same result in repeated invocations.
+
+* scan (inclusive & exclusive)
+* scan_by_key (inclusive & exclusive)
+* reduce_by_key
+* transform_scan

docs/index.rst

@@ -24,6 +24,7 @@ The documentation is structured as follows:
   .. grid-item-card:: API reference
 
     * :ref:`data-type-support`
+    * :ref:`bitwise-repro`
     * :ref:`api-reference`
     * :ref:`genindex`
 

docs/sphinx/_toc.yml.in

@@ -12,6 +12,7 @@ subtrees:
 - caption: API reference
   entries:
   - file: data-type-support
+  - file: bitwise-repro
   - file: cpp_api
 
 - caption: About

thrust/reduce.h

@@ -351,6 +351,8 @@ template<typename InputIterator,
  *  This version of \p reduce_by_key uses the function object \c equal_to
  *  to test for equality and \c plus to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  The algorithm's execution is parallelized as determined by \p exec.
  *
  *  \param exec The execution policy to use for parallelization.
@@ -421,6 +423,8 @@ __host__ __device__
  *  This version of \p reduce_by_key uses the function object \c equal_to
  *  to test for equality and \c plus to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  \param keys_first The beginning of the input key range.
  *  \param keys_last  The end of the input key range.
  *  \param values_first The beginning of the input value range.
@@ -482,6 +486,8 @@ template<typename InputIterator1,
  *  This version of \p reduce_by_key uses the function object \c binary_pred
  *  to test for equality and \c plus to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  The algorithm's execution is parallelized as determined by \p exec.
  *
  *  \param exec The execution policy to use for parallelization.
@@ -557,6 +563,8 @@ __host__ __device__
  *  This version of \p reduce_by_key uses the function object \c binary_pred
  *  to test for equality and \c plus to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  \param keys_first The beginning of the input key range.
  *  \param keys_last  The end of the input key range.
  *  \param values_first The beginning of the input value range.
@@ -626,6 +634,8 @@ template<typename InputIterator1,
  *  This version of \p reduce_by_key uses the function object \c binary_pred
  *  to test for equality and \c binary_op to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  The algorithm's execution is parallelized as determined by \p exec.
  *
  *  \param exec The execution policy to use for parallelization.
@@ -710,6 +720,8 @@ __host__ __device__
  *  This version of \p reduce_by_key uses the function object \c binary_pred
  *  to test for equality and \c binary_op to reduce values with equal keys.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  \param keys_first The beginning of the input key range.
  *  \param keys_last  The end of the input key range.
  *  \param values_first The beginning of the input value range.

thrust/transform_scan.h

@@ -51,6 +51,8 @@ THRUST_NAMESPACE_BEGIN
  *  assigned to <tt>\*(result + 1)</tt>, and so on.  The transform scan
  *  operation is permitted to be in-place.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  The algorithm's execution is parallelized as determined by \p exec.
  *
  *  \param exec The execution policy to use for parallelization.
@@ -121,6 +123,8 @@ __host__ __device__
  *  assigned to <tt>\*(result + 1)</tt>, and so on.  The transform scan
  *  operation is permitted to be in-place.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  \param first The beginning of the input sequence.
  *  \param last The end of the input sequence.
  *  \param result The beginning of the output sequence.
@@ -181,6 +185,8 @@ template<typename InputIterator,
  *  to <tt>\*(result + 1)</tt>, and so on.  The transform scan operation is 
  *  permitted to be in-place.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  The algorithm's execution is parallelized as determined by \p exec.
  *
  *  \param exec The execution policy to use for parallelization.
@@ -255,6 +261,8 @@ __host__ __device__
  *  to <tt>\*(result + 1)</tt>, and so on.  The transform scan operation is 
  *  permitted to be in-place.
  *
+ *  Results from this function may vary from run to run depending on the inputs provided.
+ *
  *  \param first The beginning of the input sequence.
  *  \param last The end of the input sequence.
  *  \param result The beginning of the output sequence.