#
#asahi · 2025-08-26 19:42
twbaren't the apple arm cores basically like co--- that'
#asahi · 2025-08-26 19:42
opticronCortex-M* vs Cortex-A*
#asahi · 2025-08-26 19:42
chaos_princessWdym by M and A cores?
#asahi · 2025-08-26 19:42
twbI've never really looked into how M and A cores are different. I assume it's like "the A cores have additional branch prediction vulnerabilities ^W^W^W performance optimizations"
#asahi · 2025-08-26 19:41
chaos_princess(i should check out what rtems even is some day)
#asahi · 2025-08-26 19:40
opticronit's not like the e-cores are equivalent to a stm32 :)
#asahi · 2025-08-26 19:40
twbFair enough I guess
#asahi · 2025-08-26 19:40
chaos_princessE-cores are also quite good, it isn't some atom trashheap
#asahi · 2025-08-26 19:40
twbLike it'd be more efficient mips-per-watt overall to do 4 and force them to stay only on the performance cores
#asahi · 2025-08-26 19:39
twbchaos_princess: I guess I naively thought they'd be bad at that job?
#asahi · 2025-08-26 19:38
chaos_princesstwb: why not? Let e-cores help out, once p-cores are out of work, the workload should migrate
#asahi · 2025-08-26 19:38
opticronI also do a bit on the OSS side of things for community management and general project steering
#asahi · 2025-08-26 19:37
opticronI work at OAR, so I port it to new boards/platforms/write drivers/etc.
#asahi · 2025-08-26 19:37
tpw_ruleswhat do you use rtems for?
#asahi · 2025-08-26 19:37
twbI wonder if that's optimal
#asahi · 2025-08-26 19:36
opticronI think the builder I'm using just uses all cores
#asahi · 2025-08-26 19:35
opticron(for reference, my build test is the rtems-source-builder aarch64 toolchain build)
#asahi · 2025-08-26 19:35
twbWhat's the sensible approach when you have AMP instead of SMP? Do you just hard-code MAKEFLAGS=j4 when you have 4 performance cores and 4 mobility cores?
#asahi · 2025-08-26 19:34
twbI remember compiling busybox with -j (i.e. infinitely parallel) would effectively forkbomb myself
#asahi · 2025-08-26 19:33
twbI guess the other obvious(?) guess is whethere MAKEFLAGS=j$(nproc) is similar
#asahi · 2025-08-26 19:32
twbI'm usually doing mmdebstrap so cross-arch is actually running all the post-install scripts through CPU emulation
#asahi · 2025-08-26 19:31
opticronall unit tests occur in a different step elsewhere
#asahi · 2025-08-26 19:31
opticroncorrect
#asahi · 2025-08-26 19:31
twbI guess unless there are unit tests it's never *executing* the cross-compiled code under qemu-user so it doesn't matter
#asahi · 2025-08-26 19:31
opticronbut yes, the built compiler is always run natively
#asahi · 2025-08-26 19:31
opticrontwb, just newlib with the built cross-compiler
#asahi · 2025-08-26 19:20
chaos_princessExcept for gcc, but you just do --diable-bootstrap or w/e
#asahi · 2025-08-26 19:20
chaos_princessIt shouldnt matter much
#asahi · 2025-08-26 19:17
twbopticron: are you cross-compiling in any/all of those scenarios?
#asahi · 2025-08-26 19:15
chaos_princessAs in, you can't use all of max memory bw with only cpu
#asahi · 2025-08-26 19:14
chaos_princessMemory bandwidth doesnt matter _that_ much
#asahi · 2025-08-26 19:14
chaos_princessSame-ish roughly
#asahi · 2025-08-26 19:12
opticronjust to see if it differs
#asahi · 2025-08-26 19:12
opticronthough I'd be interested in someone running the test for a M1Pro
#asahi · 2025-08-26 19:11
opticronI suspect that the memory bandwidth is critical for that task
#asahi · 2025-08-26 19:11
opticron7 minutes for a full fresh build with hot caches where the workhorse dual xeon at work takes 22 minutes for the same compile and even my gaming desktop takes 13 minutes
#asahi · 2025-08-26 19:10
opticronchaos_princess, I have a MBP M1 Max and it is by far the fastest machine I own for my regular workload of compiling GCC, GDB, support libraries, and newlib for the target platform
#asahi · 2025-08-26 18:56
twb"should we use kelvin" "no what cpu would ever run below 15°C even in anarctica"
#asahi · 2025-08-26 18:56
twbmade me want to go back and see the meeting with the guys designing the sensor spec
#asahi · 2025-08-26 18:55
twbI did like that peltier rig intel trialled for a couple of years, where the sensors reported all cores running at 0°C because they couldn't report negative numbers
#asahi · 2025-08-26 18:53
tpw_rules(hot air rises and heat sinks)
#asahi · 2025-08-26 18:53
tpw_rulesespecially with today's phones that can't lie flat on a table anymore
#asahi · 2025-08-26 18:53
tpw_rulesthere's a great book on heatsinks which somewhat supports that
#asahi · 2025-08-26 18:51
twbI don't have a citation so I'm happy to just retract
#asahi · 2025-08-26 18:50
chaos_princessi would really want to see the measurement setup, not immediately calling bs, but it feels unintuitive
#asahi · 2025-08-26 18:50
twbbut convection
#asahi · 2025-08-26 18:50
chaos_princessair is an insulator?
#asahi · 2025-08-26 18:50
chaos_princesswhat
#asahi · 2025-08-26 18:49
twbbecause if they're lying on the table they can't dump heat into the table as fast as they could into the air
#asahi · 2025-08-26 18:49
twbI read somewhere phones run faster if you hold them upright

plus ›