Siarhei Siamashka

Yes, please update the patch. Also it would be great if tinymembench used this workaround, but still showed a warning to the users, encouraging them to fix the problem on...

Yeah, the plot thickens...

There was one more kernel patch - https://patchwork.kernel.org/patch/9144983/ I think that we need to ping kernel developers to clarify its status.

Sigh. I don't feel like discussing it with the kernel developers again. But the least invasive solution is to just take your patch to treat 'asimd' as an alias for...

Just run `sunxi-fel spl uart0-helloworld-sdboot.sunxi` on your computer (after switching the device in FEL mode and connecting a USB cable). There are wiki pages, which explain the process in more...

Yes, overlapped_blt_neon is mostly optimized for reading from uncached sources. In addition, because it performs reading using aligned 32 byte chunks, it can access memory before and after the source...

> As it turns out, the slow code path in Xorg also happens for ShmPutImage. It is triggered > when the width of the image to be copied to the...

Thanks, pushed with the following tweaks: ``` --- a/src/sunxi_x_g2d.c +++ b/src/sunxi_x_g2d.c @@ -215,7 +215,7 @@ xCopyArea(DrawablePtr pSrcDrawable, * The following function is adapted from xserver/fb/fbPutImage.c. */ -void xPutImage(DrawablePtr pDrawable, +static...

Though this issue is still open until it it gets resolved in Xorg :-) Having duct-tape band aid code is not very good.

How did you generate the ``configure`` script?