cousteau has quit [Quit: ♫ I can't forget the day I shot that network down ♫]
shicz2 has quit [Ping timeout: 268 seconds]
shicz2 has joined #riscv
shicz2 has quit [Ping timeout: 260 seconds]
shicz2 has joined #riscv
aburgess has quit [Remote host closed the connection]
aburgess has joined #riscv
SpaceCoaster_ has joined #riscv
shicz2 has quit [Ping timeout: 268 seconds]
SpaceCoaster has quit [Ping timeout: 246 seconds]
SpaceCoaster_ is now known as SpaceCoaster
shicz2 has joined #riscv
jellydonut has quit [Read error: Connection reset by peer]
vagrantc has quit [Quit: leaving]
joev has quit [Quit: WeeChat 2.3]
joev has joined #riscv
EchelonX has quit [Quit: Leaving]
bsduser2 has joined #riscv
grubman9000 has quit [Ping timeout: 260 seconds]
kailo has joined #riscv
PyroPeter has quit [Ping timeout: 240 seconds]
PyroPeter has joined #riscv
BOKALDO has joined #riscv
riff-IRC has quit [Remote host closed the connection]
riff-IRC has joined #riscv
bgamari has quit [Ping timeout: 264 seconds]
bgamari has joined #riscv
mahmutov has joined #riscv
elastic_dog has quit [Ping timeout: 245 seconds]
elastic_dog has joined #riscv
shicz2 has quit [Ping timeout: 256 seconds]
shicz2 has joined #riscv
winterflaw has joined #riscv
JSharp has quit [Ping timeout: 264 seconds]
seds has quit [Ping timeout: 256 seconds]
JSharp has joined #riscv
adomas has quit [Ping timeout: 268 seconds]
seds has joined #riscv
adomas has joined #riscv
DynamiteDan_ has quit [Ping timeout: 256 seconds]
jcm has quit [Ping timeout: 256 seconds]
olofj_ has quit [Ping timeout: 256 seconds]
jcm has joined #riscv
olofj_ has joined #riscv
DynamiteDan_ has joined #riscv
<ssb>
is anyone working on riscv64-ilp32 ?
psydroid has joined #riscv
CarlosEDP has joined #riscv
pierce has joined #riscv
AhmedCharles[m] has joined #riscv
<sorear>
I've spent time agitating for people to keep __riscv_xlen and __SIZEOF_POINTER__ separate but I have not done explicit work on a port that would use that and I am not aware of other work in that area
jellydonut has joined #riscv
Sofia has quit [Remote host closed the connection]
Sofia has joined #riscv
DynamiteDan_ is now known as DynamiteDan
mahmutov has quit [Ping timeout: 260 seconds]
Andre_H has joined #riscv
<pabs3>
ssb: what would be the motivation for that? IIRC x32 (x86 equivalent) and the ARM equivalent were both abandoned
<ssb>
pabs3, 10% performance improvement on pointer heavy workload (e.g compilation)
<ssb>
x32 is not abandoned, it works, and it is being improved. It is just unusable on desktop due lack of support from web browsers/rust and such
<sorear>
proper support for sstatus.UXL in cores and kernels would give you most of the same benefits with less effects on random software, but that requires buyin from hardware people
<sorear>
there are some (niche within a niche) cases where you benefit from having 64 bit registers in a process limited to 32 bit addresses
pecastro has joined #riscv
cousteau has joined #riscv
* pabs3
thought compilers often use more virtual memory space than 32-bit pointers allow these days, making non-64-bit ports less viable
<ssb>
linux kernel compilation still comfortably fits in under 512 MB
<ssb>
(unlike compilation of gcc itself, needs at leas 1 GB per core these days)
<ssb>
and yes, few more years and that argument will be moot..
<cousteau>
I recently compiled gcc and the gnu toolchain for RISC-V on a virtual machine at work
<ssb>
still, 10% performance improvement is no joke
<cousteau>
it ran out of memory
<cousteau>
like, "trigger OOM Killer" kind of out of memory
<cousteau>
so yeah, kinda RAM-intensive
<pabs3>
I know Debian has hit virtual memory limits (not RAM limits) in some ports for some packages (maybe C++ with lots of templates), maybe that was only on MIPS, not sure
<sorear>
with some care you can support 4GB of memory per-process in ilp32 (this requires making the (-2GB) to 0 region user address space, which could be controversial in the kerel)
<sorear>
you will definitely have _some_ packages you have difficulty compiling, which makes this less attractive for distros that rely on a natively built gcc
<pabs3>
hmm, I guess you could use a 64-bit compiler to cross-build those packages to ilp32
<cousteau>
ssb: the way I see it, a 10% performance improvement means a jump of about 3 months, assuming performance scales with Moore's law (it probably doesn't)
<pabs3>
Debian doesn't have cross-built ports yet though, although there are folks talking about various non-standardly built ports
<cousteau>
sorear: heh, I've thought of negative memory addresses before. Sort of makes sense for 64b address spaces.
<ssb>
cousteau, I agree with that. Note that it also applies to the usage of high level languages (C vs assembly), using optimizing compilers, profile driven optimization, link time optimizations -- yet people use that, universally.
<cousteau>
alternatively, put the memory in 0xFFFF_FFFF_8000_0000..0xFFFF_FFFF_FFFF_FFFF rather than 0x0000_0000_8000_0000..0x0000_0000_FFFF_FFFF
<cousteau>
What's the status of the Zbk* extensions for crypto? I recall reading about them somewhere, but I just checked the document for bitmanip v1.0.0 and there's no mention of any Zbk extension there
<cousteau>
all I was able to find was a bunch of .adoc files in github
adjtm has quit [Remote host closed the connection]
adjtm has joined #riscv
<sorear>
cousteau: not sure what you're getting at. ilp32 needs to use 'lw' not 'lwu' for loading pointers in order to not compromise compression, which means that they get sign extended
<cousteau>
I actually can't remember how I reached that conclusion, but I think it had to do with being able to access any address with just two instructions (LUI + LD)
<cousteau>
or if you want memory to be "contiguous", it can also be beneficial to have it laid out from 0xFFFF_FFFF_8000_0000 to 0x0000_0000_7FFF_FFFF
<cousteau>
(it won't really be contiguous if you use unsigned pointers, but arithmetic will work as if it were)
<sorear>
"being able to access any address" only helps you if you're doing position-dependent code
mahmutov has joined #riscv
shicz2 has quit [Ping timeout: 245 seconds]
Sofia has quit [Remote host closed the connection]