ROCm Standard Parallelism Runtime Implementation

This repository contains:

1. The runtime component that enables ROCm to support GPU accelerated computation via C++ Standard Algorithms. This is an implementation detail and thus intended to be transparently composed with compiler and low-level runtime components. More specifically, since it is meant to be implicitly included by the compiler, and forwards various calls to the HIP and rocThrust libraries, its standalone uses are limited. For detailed documentation on ROCm GPU acceleration for Standard C++ Algorithms, please see the RFC and the patch set.
2. A set of diffs, under the data folder, which capture the changes needed to enable Standard Algorithm offload in LLVM (this is a pre-requisite), as well as in a number of client apps;
3. A slowly growing set of minimalistic performance studies, under the docs folder, intended to give a sense of the possible performance benefits / pitfalls associated with offload;
4. A minimal but slowly growing set of unit tests.

Algorithm Support Status

Allocation / Deallocation Interposition Status