MIQT Compile benchmarks game

[mappu]

The first time you use MIQT, it must be compiled. The first build can be slow, but after this, any go build takes only a few seconds.

To assist #8 it would be good to collect numbers from different people about how fast MIQT's first compilation is. Please include CPU, Go version, and MIQT Git tag.

🖥️ Example about how to do a clean build and collect all information

# Find CPU
$ lscpu | grep -Fi 'model name'
Model name:                           Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz

# Go version
$ go version
go version go1.19.8 linux/amd64

# Git checkout a well-known state
$ git checkout v0.6.0
$ git show --oneline -s
99a29d0c (HEAD, tag: v0.6.0) doc/CHANGELOG: summary for v0.6.0

# Clear GOCACHE for a clean build
$ go clean -cache
$ du -hs "$(go env GOCACHE)"

# Time build (e.g. Qt5 library package)
$ cd qt
$ /usr/bin/time go build -ldflags '-s -w'
532.26user 45.64system 8:50.40elapsed 108%CPU (0avgtext+0avgdata 4878200maxresident)k
713624inputs+4829920outputs (913major+19294096minor)pagefaults 0swaps

[mappu]

My results:

CPU	OS	Go	C/C++	MIQT	Package	Duration	Notes
AMD Ryzen 5 7600 6-Core	Debian 12	1.19.8	GCC 12.2	v0.6.0	qt5	5m 09s
Intel(R) Core(TM) i7-7700K	Debian 12	1.19.8	GCC 12.2	v0.6.0	qt5	8m 50s
Neoverse-N1 4-Core	Debian 12	1.19.8	GCC 12.2	v0.6.0	qt5	13m 53s
Neoverse-N1 4-Core	Debian 12	1.19.8	GCC 12.2 -O0	v0.6.0	qt5	11m 14s	A
Neoverse-N1 4-Core	Debian 12	1.23.3	GCC 12.2	v0.6.0	qt5	14m 03s	B
Neoverse-N1 4-Core	Debian 12	1.24-devel	GCC 12.2	v0.6.0	qt5	13m 54s	C
Neoverse-N1 4-Core	Debian 12	CL 581336	GCC 12.2	v0.6.0	qt5	14m 01s	D
Neoverse-N1 4-Core	Debian 12	CL 579815	GCC 12.2	v0.6.0	qt5	7m 50s	E
Github Actions run 2024-11	Debian 12 (Docker)	1.19.8	GCC 12.2	v0.6.0	qt5	9m 22s
Github Actions run 2024-11	Debian 12 (Docker)	1.19.8	GCC 12.2	v0.6.0	qt6	11m 39s
Github Actions run 2024-11	Fedora 40 (Docker)	1.22	GCC	v0.6.0	qt6	11m 55s
Github Actions run 2024-12	Debian 12 (Docker)	1.19.8	GCC 12.2	v0.7.0+	qt5	13m 24s
Github Actions run 2024-12	Fedora 41 (Docker)	1.23	GCC	v0.7.0+	qt6	17m 6s
AMD Ryzen 5 7600 2-Core	Win 11 24H2 (VM)	1.22.5	GCC 14.2	v0.7.0+	qt6	46m 42s	F
AMD Ryzen 5 7600 2-Core	Win 11 24H2 (VM)	1.23.1	Clang 18.1	v0.7.0+	qt6	32m 34s	G
AMD Ryzen 5 7600 6-Core	Debian 12	1.19.8	GCC 12.2	v0.7.0+	qt6	8m 49s
AMD Ryzen 5 7600 6-Core	Debian 12	1.19.8	GCC 12.2 -O0	v0.7.0+	qt6	6m 41s
AMD Ryzen 5 7600 6-Core	Debian 12	1.19.8	Clang 14	v0.7.0+	qt6	6m 14s
AMD Ryzen 5 7600 6-Core	Debian 12	1.19.8	Clang 14 -O0	v0.7.0+	qt6	5m 56s
AMD Ryzen 5 7600 2-Core	Debian 12	1.19.8	Clang 14 -O0	v0.7.0+	qt6	6m 02s	H
AMD Ryzen 5 7600 4-Core	FreeBSD 14.2 (VM)	1.21	Clang	v0.7.1	qt6	8m 08s

• A. 1.19.8 linux/arm64 with CGO_CXXFLAGS='-O0' CGO_CFLAGS='-O0'. This reduced the build duration from 833s to 674s (1.24x faster)
• B. Go1.23 seems to have about the same performance as Go1.19.
• C. go version devel go1.24-e67c0f0 Fri Nov 8 21:22:55 2024 +0000 linux/arm64 (tip) (Today's gotip)
• D. go version devel go1.24-d555358bf8 Tue Sep 3 15:39:35 2024 +0000 linux/arm64 (CL 581336)
• E. go version devel go1.23-b0a2790fb08 Wed Apr 17 18:48:29 2024 +0100 linux/arm64 (CL 579815). This was the best result on this CPU.
• F. MSYS2 UCRT64, GCC 14.1, Go from MSYS2, Windows Defender enabled but with C:\msys2\ excluded. It is likely that low VM RAM caused swap thrashing
• G. MSYS2 CLANG64, Clang 18.1, Go from go.dev, Windows Defender enabled but with C:\msys2\ excluded. It is likely that low VM RAM caused swap thrashing
• H. Core limited with taskset -c 0,5

---

Summary:

• Go versions between 1.19 - 1.24 makes no difference in speed, only CL 579815 helps
• Core count (2 vs 6) makes no difference in speed
• Linux builds much faster than Windows
• Qt5 seems to build faster than Qt6
• Clang seems to build faster than GCC
• Building with -O0 is faster than building with Go's default -O2

[infradragon]

13th Gen Intel(R) Core(TM) i5-1340P
Arch Linux, CachyOS v3 repositories and kernel
go 1.24.4
miqt v0.11.0 8962d67 (qt6)
go build -ldflags '-s -w' 1014.23s user 41.90s system 107% cpu 16:20.21 total
seems to be only using a single core