flac: build with cmake, fixes cross compilation

[NickCao]

Description of changes

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 22.11 Release Notes (or backporting 22.05 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

[NickCao]

Built flac and gst_all_1.gst-plugins-good on x86_64-linux and riscv64-linux (cross), not feasible to run nixpkgs-review though.

[ck3d]

LGTM

[ruuda]

The new binary performs noticeably worse. I suspect it ends up passing different optimization flags to the compiler in the end? Could you look into this?

Benchmark 1: result-bin-before/bin/flac -t album/*.flac
  Time (mean ± σ):      5.037 s ±  0.113 s    [User: 4.834 s, System: 0.120 s]
  Range (min … max):    4.955 s …  5.282 s    10 runs
 
Benchmark 2: result-bin-cmake/bin/flac -t album/*.flac
  Time (mean ± σ):      5.307 s ±  0.040 s    [User: 5.072 s, System: 0.136 s]
  Range (min … max):    5.257 s …  5.381 s    10 runs
 
Summary
  'result-bin-before/bin/flac -t album/*.flac' ran
    1.05 ± 0.03 times faster than 'result-bin-cmake/bin/flac -t album/*.flac'

[ck3d]

Good catch!

According to the project documentation, autotools and cmake should be equivalent:
https://github.com/xiph/flac/blob/master/README.md?plain=1#L87

The autotools used following configuration:

  Configuration summary :

    FLAC version : ............................ 1.4.1

    Host CPU : ................................ x86_64
    Host Vendor : ............................. pc
    Host OS : ................................. linux-gnu


    Compiler is GCC : ......................... yes
    GCC version : ............................. 11.3.0
    Compiler is Clang : ....................... no
    SSE optimizations : ....................... yes
    Neon optimizations : ...................... no
    Asm optimizations : ....................... yes
    Ogg/FLAC support : ........................ yes
    Stack protector  : ........................ yes
    Fuzzing support (Clang only) : ............ no

When building with cmake, no configuration summary is available.
It could be that some optimizations are not activated.

I think this is an upstream issue.

[NickCao]

https://github.com/xiph/flac/blob/2440e19a733f045e50cd7f3dbd706b9ed98fc5ef/configure.ac#L508 sets -O3, while it's missing from the cmake build system. AVX is on by default in cmake, ruling out that possibility.

[ruuda]

Hmm, even with b8a03ce, there is a significant difference ... the binary is about 3% larger, and when I measure a perf stat of flac --test, it executes about 8% more instructions. (Using instructions now instead of wall time because it is more stable, in wall time, the old version is 1.05 ± 0.02 times faster on my machine.)

Is there an easy way to trace the compiler invocations, so we can see which flags get passed?

[ruuda]

I am a fool, I was comparing the current PR to master, which still has flac 1.3.4..., the regression is an upstream one, unrelated to this PR. I now no longer find a difference between the two:

Benchmark 1: result-bin-cmake/bin/flac -t album/*.flac
  Time (mean ± σ):      4.854 s ±  0.140 s    [User: 4.753 s, System: 0.096 s]
  Range (min … max):    4.702 s …  5.111 s    10 runs
 
Benchmark 2: result-bin-before/bin/flac -t album/*.flac
  Time (mean ± σ):      4.912 s ±  0.078 s    [User: 4.815 s, System: 0.091 s]
  Range (min … max):    4.783 s …  4.991 s    10 runs
 
Summary
  'result-bin-cmake/bin/flac -t album/*.flac' ran
    1.01 ± 0.03 times faster than 'result-bin-before/bin/flac -t album/*.flac'

[ck3d]

No problem :)

[Mindavi]

I was just preparing a PR and got a merge conflict, thanks for already fixing this! My solution was worse (using configure script patches that already landed upstream).