07:42 <davidlt[m]> Proof that I am not awake yet. Going to YouTube to search for Python3.12 package (instead of looking at dist-git).

07:55 <davidlt[m]> It seems that RX 6600, 6650, or even 6750 could become a good replacement for RX 570, 580 cards once it gets to ~200 USD or below.

07:56 <davidlt[m]> Finally something that could replace Polaris, but sadly RNDA/RNDA2 will not work on riscv64 IIRC.

07:57 <davidlt[m]> Rust 1.64.0 just landed. Time to look at 1.65.X and later test 1.66 beta and report if that works.

07:57 <davidlt[m]> kernel 6.0.10 is also almost done (RPMs being written out now).

08:01 <davidlt[m]> Coffee is here, just need to find a nice playlist to begin properly looking into things.

08:32 <davidlt[m]> I am actually doing 3 Rust 1.65.0 builds to speed up things.

08:32 <davidlt[m]> A few possible variations of what I expect to happen.

11:17 <davidlt[m]> So it looks like that 1.65.0 with patched 1.64.0 is compiling so far (which is a bit annoying).

11:17 <davidlt[m]> I wonder if I will be able to compile 1.66.0 with unpatched 1.65.0, or not.

11:17 <davidlt[m]> This definitely will be quite annoying.

11:20 <davidlt[m]> Ah, btw, since QEMU got support for Sv57 (5-page table) that breaks a bunch of stuff (OpenJDK, Golang, etc.).

11:21 <davidlt[m]> There is a patch to add something like: -cpu rv64,satp-mode=sv48

11:21 <davidlt[m]> Instead of QEMU going for the latest highest mode.

11:22 <davidlt[m]> In general Sv39 and Sv48 should be safe options. There are discussions how to support anything beyond that. x86_64 and aarch64 by default don't allocate anything high, and it relies on mmap hints to ask for it.

11:23 <davidlt[m]> So if you have some interesting crashes be aware.

11:23 <davidlt[m]> Does not apply to physical machines are all of them only support Sv39.

11:36 <davidlt[m]> So this kernel patch "[PATCH v8 0/3] riscv, mm: detect svnapot cpu support at runtime" is for "marking contiguous 4K pages as a non-4K page". That is support for pages beyond 4K. Probably will be required for RVA23.

11:36 <davidlt[m]> IIRC only 64K is allowed right now thus something like 16K (like on Apple M1/M2) wouldn't be valid.

11:37 <davidlt[m]> There is also (today): RISC-V V C Intrinsic API v1.0 release meeting reminder (November 28th, 2022)

11:38 <davidlt[m]> https://docs.google.com/presentation/d/1PjVehtwrT-WXfy0ZtFVvb3Yt45UOHLvjLW0CSNH7Bro/edit?usp=sharing

11:39 <davidlt[m]> There is one package in Fedora that already uses RVV intrinsic, but it cannot compile with GCC (not until GCC 13 at least).

11:39 <davidlt[m]> GCC will support v1.0, but LLVM/Clang supports v0.10.

11:43 <davidlt[m]> "[PATCH v2 0/13] Zbb string optimizations and call support in alternatives" is interesting on the kernel side:

11:43 <davidlt[m]> > on a 95 character string in a loop of 10000 iterations, the Zbb

11:43 <davidlt[m]> > In my half-scientific test-case of running the functions in question

11:43 <davidlt[m]> > variants shave off around 2/3 of the original runtime.

11:43 <davidlt[m]> https://lists.infradead.org/pipermail/linux-riscv/2022-November/022627.html

23:43 <dgilmore> davidlt[m]: I am using a radeon 6500 XT in my ampere workstation. sadly its still very rough. needed some code isolated and compiled with floating point support