Re: LLVM offload compress

[Xuanteng Huang <xuanteng.huang@outlook.com>]

Hi Cory,

> On Sep 19, 2024, at 04:51, Cordell Bloor <cgmb@slerp.xyz> wrote:
> 
> They will be decompressed by the HIP runtime when needed. As GPU code tends to be very repetitive, it's likely that this will result in a space savings >90% for many of the largest ROCm libraries.

This sounds promising. What I’m curious about is the moment the decompression happens, and the overhead to the end-to-end latency. Does it mean that the compressed GPU kernels should be decompressed first before their launch to GPU? Does it happen every time or for the first time? Or the decompression happens at the time when the package gets installed?

Best,
Xuanteng