AMD Releases AOMP 17.0-0 For Latest Radeon OpenMP Offloading Compiler

[phoronix]

Phoronix: AMD Releases AOMP 17.0-0 For Latest Radeon OpenMP Offloading Compiler

On Thursday AMD engineers released AOMP 17.0-0 as the newest version of their LLVM/Clang downstream compiler that carries their latest development patches around Radeon/Instinct OpenMP GPU/accelerator offloading support...

https://www.phoronix.com/news/AMD-AOMP-17.0-0

[binarybanana]

I wish something like this existed for Intel, too. With an iGPU you don't have to copy stuff around and latency/overhead should be low enough that offloading typical OpenMP workloads to the iGPU might actually be faster. I guess the same would go for APUs. That was kind of the original idea, after all.

[kn00tcn]

I wish something like this existed for Intel, too. With an iGPU you don't have to copy stuff around and latency/overhead should be low enough that offloading typical OpenMP workloads to the iGPU might actually be faster. I guess the same would go for APUs. That was kind of the original idea, after all.

but the igpu is so much weaker and has to talk to slow dram instead of fast vram

[Zyten]

I wish something like this existed for Intel, too. With an iGPU you don't have to copy stuff around and latency/overhead should be low enough that offloading typical OpenMP workloads to the iGPU might actually be faster. I guess the same would go for APUs. That was kind of the original idea, after all.

You can do OpenMP target offloading to Intels iGPUs. Instead of aomp, you can use their OneAPI with the option to use Spir-V for compliation.

As for the news: Great to hear that the additions for OMPT and the device tracing interface are in this release while we wait for the next full ROCm release. This means that I‘ve got at least one compiler which supports the whole OMPT standard to a somewhat complete degree.