More Offloading Code Hits GCC Mainline For Both HSA & NVPTX
For those following GCC's offloading capabilities to devices like GPUs, more work continued being mainline this week. We are onto stage 3 development of GCC 7 but items that were still being reviewed at that time are still being allowed to land. It looks like in 2017 we may finally see more GCC support come to reality when it comes to AMD HSA support and OpenMP / OpenACC offloading to NVIDIA GPUs.
Some of the latest work to talk about on top of our other recent GCC 7 articles include:
- The libgomp changes for supporting OpenMP offloading to NVPTX for in turn being piped to the NVIDIA proprietary driver for GPU offloading.
- The middle-end changes for OpenMP offloading to NVPTX. (Just over a week ago were the back-end changes needed.)
- A lot more HSA code was merged to trunk.
Will be nice when all these GPU accelerator efforts for GCC (and Clang) have finally become a fully-working and supported reality. More details on the GCC offloading support via this Wiki page. GCC 7.1 is due out in early 2017 as the first stable release of GCC 7.
Some of the latest work to talk about on top of our other recent GCC 7 articles include:
- The libgomp changes for supporting OpenMP offloading to NVPTX for in turn being piped to the NVIDIA proprietary driver for GPU offloading.
- The middle-end changes for OpenMP offloading to NVPTX. (Just over a week ago were the back-end changes needed.)
- A lot more HSA code was merged to trunk.
Will be nice when all these GPU accelerator efforts for GCC (and Clang) have finally become a fully-working and supported reality. More details on the GCC offloading support via this Wiki page. GCC 7.1 is due out in early 2017 as the first stable release of GCC 7.
1 Comment