Why "Pinned host memory" is uncached for iGPU?

[Shaquille]

hello, nv experts:
I found “Pinned host memory” is uncached for iGPU from this document: CUDA for Tegra

it is cached for host, but, it is uncached for iGPU
I think it will be better if it can both cached for host and iGPU.
what reason cause it cached for host and uncached for iGPU?
I cannot understand it.
I think there is some hardware constraint for this feature. but, I don’t know what is this hardware constraint.

Is there anyone would like to teach me?

[AastaLLL]

Hi,

It is related to the I/O coherency issue.
Please find more info in the section 3.1 on the same document.

Thanks.