fengling has joined #m-labs
_rht has quit [Quit: Connection closed for inactivity]
fengling has quit [Ping timeout: 240 seconds]
fengling has joined #m-labs
<
mithro>
Have you guys seen cocotb? If so what do you think?
kuldeep has quit [Ping timeout: 244 seconds]
FelixVi has quit []
kuldeep has joined #m-labs
FelixVi has joined #m-labs
Gurty has quit [Ping timeout: 276 seconds]
stekern_ is now known as stekern
mumptai has joined #m-labs
evilspirit has joined #m-labs
_rht has joined #m-labs
evilspirit has quit [Ping timeout: 246 seconds]
evilspirit has joined #m-labs
evilspirit has quit [Ping timeout: 250 seconds]
FelixVi has quit [Read error: Connection reset by peer]
<
whitequark>
amazing. yesterday I wasted a few hours porting the backend to LLVM 3.8
<
whitequark>
the reason was a /GDB bug/. this is new to me. I've never encountered those before.
<
whitequark>
lolwhat
<
whitequark>
CC vsnprintf.o
<
whitequark>
implement EmitVAArg
<
whitequark>
UNREACHABLE executed at ../tools/clang/lib/CodeGen/TargetInfo.cpp:5923!
<
whitequark>
how on earth could vsnprintf have possibly worked?
<
whitequark>
clang for or1k lacks a va_arg implementation
*completely*
evilspirit has joined #m-labs
evilspirit has quit [Ping timeout: 260 seconds]
<
whitequark>
wow, the OR1K ABI is vastly underspecified
<
whitequark>
annnnnnnd we're on 3.8.1, which is the latest released LLVM
<
GitHub102>
conda-recipes/master 2202738 whitequark: llvm-or1k: update to 3.8.1.
<
GitHub102>
conda-recipes/master b7726b3 whitequark: llvmlite-artiq: update to LLVM 3.8.
mumptai has quit [Remote host closed the connection]
<
whitequark>
bb-m-labs: force build --props=package=llvm-or1k conda-all
<
bb-m-labs>
build forced [ETA 1h26m31s]
<
bb-m-labs>
I'll give a shout when the build finishes
MiW has joined #m-labs
FabM has quit [Ping timeout: 250 seconds]
_rht has quit [Excess Flood]
_rht has joined #m-labs
fengling has quit [Quit: WeeChat 1.4]
FabM has joined #m-labs
Gurty has joined #m-labs
<
whitequark>
bb-m-labs: force build --props=package=llvm-or1k conda-win32
<
bb-m-labs>
build forced [ETA 53m37s]
<
bb-m-labs>
I'll give a shout when the build finishes
<
whitequark>
bb-m-labs: force build --props=package=llvm-or1k conda-win64
<
bb-m-labs>
build forced [ETA 1h06m21s]
<
bb-m-labs>
I'll give a shout when the build finishes
<
whitequark>
bb-m-labs: force build --props=package=llvmlite-artiq conda-all
<
bb-m-labs>
build forced [ETA 1h26m31s]
<
bb-m-labs>
I'll give a shout when the build finishes
<
whitequark>
rjo: verbose assembly is in.
<
GitHub198>
artiq/master d9dd7df whitequark: compiler: add analysis passes from TargetMachine....
<
GitHub198>
artiq/master df8018f whitequark: compiler: emit verbose assembly via ARTIQ_DUMP_ASM.
<
GitHub198>
artiq/master 6a34a75 whitequark: conda: require llvmlite-artiq built for LLVM 3.8.
<
mithro>
So mocking out timestamps in files, /dev/urandom and any time function seems to make ISE deterministic....
<
rjo>
mithro: just email their contact address and describe what and why.
<
rjo>
whitequark: yay
<
whitequark>
hahaha, upgrade to 3.8 pessimized pulse_rate_dds
sandeepkr has joined #m-labs
_rht has quit [Quit: Connection closed for inactivity]
sandeepkr has quit [Ping timeout: 252 seconds]
sb0 has quit [Ping timeout: 268 seconds]
FabM has quit [Quit: ChatZilla 0.9.92 [Firefox 45.0.1/20160315153207]]
sandeepkr has joined #m-labs
sb0 has joined #m-labs
<
whitequark>
it's cute that Daniel thinks migrating to C++ will scare us. well, not me, anyway :)
* whitequark
goes back to debugging LLVM
sandeepkr has quit [Quit: Leaving]
<
rjo>
maybe this implies that ASM and IR are also forbidden.
<
whitequark>
and machine code too? :
<
whitequark>
I know a guy who was writing a C interpreter in HDL. maybe we write one of those to run ARTIQ.
<
whitequark>
(no, I don't know why)
evilspirit has joined #m-labs
sandeepkr has joined #m-labs
evilspirit has quit [Ping timeout: 260 seconds]
kuldeep has quit [Ping timeout: 276 seconds]
kuldeep has joined #m-labs
<
cr1901_modern>
Does this mean all the HDL synthesis code needs to be rewritten as well?
<
cr1901_modern>
Or just the ARTIQ code (and spawning python processes to build the SoC's is fine)?
Gurty has quit [Ping timeout: 276 seconds]
<
whitequark>
rjo: okay, so... the pessimization of pulse_rate_dds was intended.
<
whitequark>
the reason was, in essence, the fact that !dereferenceable is loosely defined
<
whitequark>
!dereferenceable as intended in LLVM is meant to be conditional: when code motion is performed, the attribute can lose its validity
<
whitequark>
whereas, since we're memory-safe, our dereferenceability is unconditional.
<
whitequark>
until that's fixed, it doesn't really make sense to bother with, ah, pretty much any other optimization
<
whitequark>
since attribute loads in loops are pervasive, and they're blocked on that.
Gurty has joined #m-labs
sb0 has quit [Ping timeout: 244 seconds]