Release Briefing 5.1.0

Content/ReleaseBriefings/5_1/Section_BackendUpdates.tex

@@ -0,0 +1,57 @@
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge Backend Updates}
+
+  \vspace{10pt}
+
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge CUDA}
+
+  \vspace{10pt}
+
+\end{frame}
+
+\begin{frame}[fragile]{Features and Fixes for \texttt{CUDA}}
+
+  \begin{itemize}
+    \item Added support for \texttt{Kokkos\_ARCH\_BLACKWELL103} for B300 GPUs (requires \texttt{CUDA} $\geq 12.8$).
+    \item Fixed building with \texttt{clang} when enabling both \texttt{CUDA} and \texttt{OpenMP}.
+    \item Added support for the flags \texttt{-Ofc} and \texttt{--fdevice-time-trace} to \texttt{nvcc\_wrapper} (Useful for getting insights into compile time).
+  \end{itemize}
+
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge HIP}
+
+  \vspace{10pt}
+
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{Features and Fixes for \texttt{HIP}}
+
+  \begin{itemize}
+    \item Added support for \texttt{Kokkos\_ARCH\_AMD\_GFX950} for MI350 and MI355 (requires \texttt{ROCm} $\geq 7.0$ recommended $\geq 7.1$)
+    \item Added support for \texttt{bhalf\_t} (in \texttt{Kokkos::Experimental}).
+    \item Added math functions for \texttt{half\_t} and \texttt{bhalf\_t} instead of casting to \texttt{float}.
+	\item Fixed race conditions in HIP \texttt{parallel\_scan} when running on MI300A
+  	\begin{itemize}
+		\item Added an extra synchronization at the end of \texttt{parallel\_scan} implementation.
+		\item Seems to be a compiler bug on MI300A.
+		\item Extra overhead is negligible.
+  	\end{itemize}
+  \end{itemize}
+
+\end{frame}

Content/ReleaseBriefings/5_1/Section_BreakingChanges.tex

@@ -0,0 +1,57 @@
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge Breaking Changes}
+
+  \vspace{10pt}
+
+\end{frame}
+
+%==========================================================================
+
+% Examples
+
+% note: always keep the [fragile] for your frames!
+
+\begin{frame}[fragile]{Breaking Changes}
+ \begin{itemize}
+   \item Kokkos now detects and reports an error if an execution space is outside its valid lifetime scope, i.e., before \texttt{initialize()} or after \texttt{finalize()}.
+     \begin{itemize}
+      \item An example has been added to show users how to migrate from static execution spaces to comply with the new lifetime scope enforcement.
+     \end{itemize}
+   \item \texttt{Kokkos::TeamPolicy} now validates the \texttt{vector\_length} argument at construction time and aborts with a descriptive error if it is out of range, making misconfiguration bugs easier to catch.
+     \begin{itemize}
+       \item \texttt{Kokkos::TeamPolicy} now requires \texttt{vector\_length} to be a power of two.
+     \end{itemize}
+   \item \texttt{ScatterView} is no longer move constructible or move assignable.
+ \end{itemize}
+\end{frame}
+
+\begin{frame}[fragile]{Breaking Changes on OpenMP and OpenMPTarget}
+ \begin{itemize}
+   \item Kokkos now warns when an OpenMP execution space instance is created inside an active OpenMP parallel region.
+   \item The OpenMPTarget backend has been removed from Kokkos.
+ \end{itemize}
+\end{frame}
+
+\begin{frame}[fragile]{Example code}
+   \begin{code}[keywords={std}]
+       #include <iostream>
+
+       int main() {
+           std::cout << "hello world\n";
+       }
+   \end{code}
+\end{frame}
+
+% \begin{frame}[fragile]{Example table}
+%    \begin{center}
+%        \begin{tabular}{l|l}
+%            a & b \\\hline
+%            c & d
+%        \end{tabular}
+%    \end{center}
+% \end{frame}
+
+%==========================================================================

Content/ReleaseBriefings/5_1/Section_BugFixes.tex

@@ -0,0 +1,84 @@
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge Bug Fixes}
+
+  \vspace{10pt}
+
+\end{frame}
+
+%==========================================================================
+
+% Examples
+
+% note: always keep the [fragile] for your frames!
+
+%\begin{frame}[fragile]{Example list}
+%  \begin{itemize}
+%      \item Item 1
+%      \item Item 2 with some \texttt{code}
+%      \begin{itemize}
+%        \item Sub-item 2.1
+%        \item Sub-item 2.2
+%      \end{itemize}
+%  \end{itemize}
+%\end{frame}
+
+%\begin{frame}[fragile]{Example code}
+%    \begin{code}[keywords={std}]
+%        #include <iostream>
+%
+%        int main() {
+%            std::cout << "hello world\n";
+%        }
+%    \end{code}
+%\end{frame}
+
+%\begin{frame}[fragile]{Example table}
+%    \begin{center}
+%        \begin{tabular}{l|l}
+%            a & b \\\hline
+%            c & d
+%        \end{tabular}
+%    \end{center}
+%\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{General}
+  \begin{itemize}
+    \item Fix \texttt{reduction\_identity} with the \texttt{BAnd} reducer
+    \item Update \texttt{team\_fan\_\{in|out\}} member functions of \texttt{ThreadsExecTeamMember} not to call host-only functions on the device
+    \item Ensure that execution space instances fence when they are destructed
+    \item Correctly identify GCC and LLVM Clang on macOS
+  \end{itemize}
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{Restrict host atomics to sizes that are \texttt{lock-free}}
+  The atomics for integers were not restricted on the sizes for which they are \texttt{lock-free} (except with \texttt{MSVC}).
+ \begin{itemize}
+   \item If you used 128bit atomics on integers the linker needed to link \texttt{latomic}.
+   \item \textbf{Problem}: Linking \texttt{latomic} by anyone outside of Kokkos' own build system results in \textbf{unsafe} atomic operations.
+   \item Kokkos' atomics on integers with 128 bit or more use our internal lock-tables. Linking \texttt{libatomic} is \textbf{neither} necessary \textbf{nor} recommended.
+ \end{itemize}
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{Conformance with the C++ standard}
+  \begin{itemize}
+    \item Add missing constexpr specifiers on \texttt{conj()}, and for the \texttt{real()} and \texttt{imag()} non-member functions taking complex numbers
+    \item Make overloads of \texttt{isnormal} compliant with std
+  \end{itemize}
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{SIMD}
+  \begin{itemize}
+    \item Use intrinsics when calling \texttt{min} and \texttt{max} on simd vectors of integral types
+  \end{itemize}
+\end{frame}

Content/ReleaseBriefings/5_1/Section_BuildSystemUpdates.tex

@@ -0,0 +1,20 @@
+%==========================================================================
+
+\begin{frame}[fragile]
+
+  {\Huge Build System Updates}
+
+  \vspace{10pt}
+
+\end{frame}
+
+%==========================================================================
+
+\begin{frame}[fragile]{Integtation of \texttt{ROCm} and \texttt{HIP} in \texttt{CMake}}
+  Kokkos now uses \texttt{find\_package(HIP)} to find \texttt{ROCm}. The search includes the path in the environment variable \texttt{ROCM\_PATH}.
+  \begin{itemize}
+    \item Allows us to get the \texttt{ROCm} version independently of the compiler version.
+    \item \textbf{Problem}: \texttt{find\_package(HIP)} runs autodetection of GPU architectures if \texttt{GPU\_TARGETS} is not defined as \texttt{CMake} variable.
+    \item We warn if the architecture in \texttt{GPU\_TARGETS} does not match the device architecture enabled in Kokkos. This can happen e.g. when cross-compiling.
+  \end{itemize}
+\end{frame}