Microsoft adds Linux Bash Shell and Ubuntu Binaries to Win10

[ewmayer]

Microsoft adds Linux Bash Shell and Ubuntu Binaries to Windows 10 | The Hacker News

So let's say I have some multithreaded code which is currently GCC/Posix-builds-only. Will this allow me to build with the usual Posix pthread/thread-affinity support under Win10? Because while MSYS et al. are fine as far single-threaded builds go, || builds are a huge hurdle there.

[wombatman]

Want me to try compiling something? I use the Bash shell in Windows pretty regularly for sieving and prp testing and am able to compile and use, for instance, multithreaded instances of sr1sieve and sr2sieve.

[ewmayer]

Originally Posted by wombatman

Want me to try compiling something? I use the Bash shell in Windows pretty regularly for sieving and prp testing and am able to compile and use, for instance, multithreaded instances of sr1sieve and sr2sieve.

Sure, how about Mlucas v17? If this functions as hoped, you should be able to simply follow the Linux build instructions. Suggest trying on a single sample source file, though, e.g. (this assumes you're in an obj dir which you created within .../src):

gcc -c -O3 -DUSE_SSE2 -DUSE_THREADS ../util.c

[wombatman]

Looks like it works right out of the box:

Using 1 threads in carry step
100 iterations of M3888517 with FFT length 196608 = 192 K
Res64: 579D593FCE0707B2. AvgMaxErr = 0.240951102. MaxErr = 0.300781250. Program: E17.0
Res mod 2^36     =          67881076658
Res mod 2^35 - 1 =          21674900403
Res mod 2^36 - 1 =          42893438228
Clocks = 00:00:00.510

with -nthread 2:

Using 2 threads in carry step
100 iterations of M3888517 with FFT length 196608 = 192 K
Res64: 579D593FCE0707B2. AvgMaxErr = 0.240951102. MaxErr = 0.300781250. Program: E17.0
Res mod 2^36     =          67881076658
Res mod 2^35 - 1 =          21674900403
Res mod 2^36 - 1 =          42893438228
Clocks = 00:00:00.339

Oh, and grep'ing for "error" in build.log returned nothing.

[ewmayer]

Originally Posted by wombatman

Looks like it works right out of the box:

Sweet! Note you should replace USE_SSE2 with one of

USE_AVX
USE_AVX2
USE_AVX512

depending on what the most-recent one of those instruction sets your hardware supports. Binary won't beat Prime95 (yet), but on at least 2 recent platforms (AMD Ryzen and Skylake Xeon, at least Google Cloud's version thereof) it should be pretty close, though optimal jobs/threads-per-job combos to maximize total throughput on the given CPU will often be different between the 2 programs. (On David Stanfill's Ryzen I get best total throughput for my code by running one 1-threaded job per physical core, after building for AVX2, obviously).

[wombatman]

Duly noted!