Any reason why examples/ex_blend_bench is 2 times slower on Windows?

[smbape]

Hello I built and ran ex_blend_bench on Windows en Ubuntu WSL.
The Windows version is 2 times slower than the Linux version and I have no idea why.
Can anyone help me?

Windows

Build

mkdir test
cd test
git clone --depth 1 --branch 5.2.11.0 https://github.com/liballeg/allegro5.git allegro5-src
cmake -S allegro5-src -B allegro5-build -DCMAKE_BUILD_TYPE=Release -DWANT_EXAMPLES=ON
cmake --build allegro5-build --config Release
cmake --install allegro5-build --config Release --prefix allegro5-install

set CWD=%CD%
set "PATH=%CWD%\allegro5-install\bin;%PATH%"
cd /d %CWD%\allegro5-src\examples
"%CWD%\allegro5-build\examples\Release\ex_blend_bench.exe"

Results

Benchmark: Plain blit
Please wait...
Time = 5.064 s, 5263 steps
Plain blit: 1039.3 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 5.007 s, 5319 steps
Scaled blit: 1062.31 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 5.158 s, 5494 steps
Rotated blit: 1065.14 FPS
Done

Linux

Build

mkdir test
cd test
git clone --depth 1 --branch 5.2.11.0 https://github.com/liballeg/allegro5.git allegro5-src
cmake -S allegro5-src -B allegro5-build -DCMAKE_BUILD_TYPE=Release -DWANT_EXAMPLES=ON
cmake --build allegro5-build

CWD="$PWD"
cd "$CWD/allegro5-src/examples"
"$CWD/allegro5-build/examples/ex_blend_bench"

Result

Benchmark: Plain blit
Please wait...
Time = 5.10034 s, 11773 steps
Plain blit: 2308.28 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 5.01109 s, 11425 steps
Scaled blit: 2279.94 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 5.1503 s, 12201 steps
Rotated blit: 2368.99 FPS
Done

[SiegeLord]

Linux uses OpenGL by default, and Windows uses Direct3D by default. Perhaps your drivers handle them differently? One way you can switch to OpenGL on windows is to add a file called allegro5.cfg next to the executable with these contents:

[graphics]
driver=opengl

I didn't try Linux, but I did try D3D vs OpenGL and they're roughly comparable on my system (I also checked 5.2.10 vs 5.2.11 in case there was a regression, it doesn't appear to be the case):

5.2.10

D3D

Benchmark: Plain blit
Please wait...
Time = 4.868 s, 21739 steps
Plain blit: 4465.69 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.904 s, 21739 steps
Scaled blit: 4432.91 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.963 s, 22727 steps
Rotated blit: 4579.29 FPS
Done

OGL

Benchmark: Plain blit
Please wait...
Time = 4.963 s, 21739 steps
Plain blit: 4380.21 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.991 s, 21739 steps
Scaled blit: 4355.64 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.843 s, 21739 steps
Rotated blit: 4488.75 FPS
Done

5.2.11

D3D

Benchmark: Plain blit
Please wait...
Time = 4.865 s, 21739 steps
Plain blit: 4468.45 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.893 s, 21739 steps
Scaled blit: 4442.88 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 5.196 s, 23809 steps
Rotated blit: 4582.18 FPS
Done

OGL

Benchmark: Plain blit
Please wait...
Time = 4.866 s, 21739 steps
Plain blit: 4467.53 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.701 s, 20833 steps
Scaled blit: 4431.61 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.756 s, 21739 steps
Rotated blit: 4570.86 FPS
Done

[smbape]

I doubt that it is a driver issue.
The ogl config did not change much.
I have tested it on four computers and the results are the same, being Windows executable is 2 times slower than the Linux WSL executable.
I also ruled out Windows Defender by disabling it during testing.
I can also rule out the CPU/GPU vendor.
Laptop 1 : AMD Ryzen CPU, iGPU + NVidia GPU
Destop 1: AMD Ryzen CPU, NVidia GPU
Laptop 2: Intel Core i7 CPU, iGPU
Laptop 3: AMD Ryzen CPU, iGPU

The performance gain on Linux seems to me too big to just be an OS cause.
It is as if some compiler optimization are missing on Windows.

[SiegeLord]

That is some extensive testing! I am convinced. Alright, could you give me more details about how you compile things? Are you compiling from source? Using Nuget packages?

EDIT: Nevermind, I forgot you gave me the commands earlier. Let me look into things a bit more then.

FWIW, I did test on Linux, and I did not notice an effect as great as the one you saw:

Benchmark: Plain blit
Please wait...
Time = 4.11754 s, 19671 steps
Plain blit: 4777.37 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.94988 s, 23656 steps
Scaled blit: 4779.11 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.93371 s, 24532 steps
Rotated blit: 4972.33 FPS
Done

[SiegeLord]

Would you mind repeating at least one of your tests and comparing 5.2.10 vs 5.2.11 (you'll need to patch the minimum cmake version for 5.2.10, but it should compile fine otherwise).

[smbape]

With version 5.2.10.1, the results are similar.
Patching is not necessary, cmake already has an option for that.

One difference with your benchmark is that you have a much better hardware (twice faster).
The compiler may have more available options.
Some of my computers are quite old (more that 10 years old).
If you can test with a more general public computer, you may have de the same results.

Build

Windows

mkdir test
cd test
git clone --depth 1 --branch 5.2.10.1 https://github.com/liballeg/allegro5.git allegro5-src
cmake -S allegro5-src -B allegro5-build -DCMAKE_BUILD_TYPE=Release -DWANT_EXAMPLES=ON -DCMAKE_POLICY_VERSION_MINIMUM=3.5
cmake --build allegro5-build --config Release
cmake --install allegro5-build --config Release --prefix allegro5-install

set CWD=%CD%
set "PATH=%CWD%\allegro5-install\bin;%PATH%"
cd /d %CWD%\allegro5-src\examples
"%CWD%\allegro5-build\examples\Release\ex_blend_bench.exe"

Linux

mkdir test
cd test
git clone --depth 1 --branch 5.2.10.1 https://github.com/liballeg/allegro5.git allegro5-src
cmake -S allegro5-src -B allegro5-build -DCMAKE_BUILD_TYPE=Release -DWANT_EXAMPLES=ON -DCMAKE_POLICY_VERSION_MINIMUM=3.5
cmake --build allegro5-build

CWD="$PWD"
cd "$CWD/allegro5-src/examples"
"$CWD/allegro5-build/examples/ex_blend_bench"

Benchmarks

Laptop 1

CPU                         : AMD Ryzen 5 Mobile 4600H
Display adapter 0           : NVIDIA GeForce GTX 1650
Display adapter 1 (primary) : AMD Radeon(TM) Graphics
Results:
    Windows 11: Visual Studio 17 2022
        Benchmark: Plain blit
        Please wait...
        Time = 5.155 s, 5050 steps
        Plain blit: 979.631 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.12 s, 5102 steps
        Scaled blit: 996.484 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 5.241 s, 5319 steps
        Rotated blit: 1014.88 FPS
        Done
    WSL Ubuntu 24.04: gcc version 13.3.0
        Benchmark: Plain blit
        Please wait...
        Time = 4.7807 s, 11062 steps
        Plain blit: 2313.89 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.08241 s, 11486 steps
        Scaled blit: 2259.95 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 4.94737 s, 11781 steps
        Rotated blit: 2381.27 FPS
        Done

Desktop 1

CPU                         : AMD Ryzen 7 1700X
Display adapter 0           : NVIDIA GeForce GTX 970
Display adapter 1 (primary) : Microsoft Remote Display Adapter
Results:
    Windows 10: Visual Studio 17 2022
        Benchmark: Plain blit
        Please wait...
        Time = 4.968 s, 4132 steps
        Plain blit: 831.723 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.024 s, 4201 steps
        Scaled blit: 836.186 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 5.009 s, 4310 steps
        Rotated blit: 860.451 FPS
        Done
    WSL Ubuntu 24.04: gcc version 13.3.0
        Benchmark: Plain blit
        Please wait...
        Time = 4.93884 s, 7959 steps
        Plain blit: 1611.51 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.08161 s, 8281 steps
        Scaled blit: 1629.6 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 5.07969 s, 8545 steps
        Rotated blit: 1682.19 FPS
        Done

Laptop 2

CPU                         : Intel Core i7 4710HQ
Display adapter 0           : Intel(R) HD Graphics 4600
Display adapter 1 (primary) : Microsoft Remote Display Adapter
Results:
    Windows 10: Visual Studio 17 2022
        Benchmark: Plain blit
        Please wait...
        Time = 5.578 s, 2439 steps
        Plain blit: 437.253 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 4.736 s, 2450 steps
        Scaled blit: 517.314 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 4.682 s, 2415 steps
        Rotated blit: 515.805 FPS
        Done
    WSL Ubuntu 24.04: gcc version 13.3.0
        Benchmark: Plain blit
        Please wait...
        Time = 4.90284 s, 4537 steps
        Plain blit: 925.382 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 4.92471 s, 4587 steps
        Scaled blit: 931.425 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 4.93392 s, 4741 steps
        Rotated blit: 960.898 FPS
        Done

Laptop 3

CPU                         : AMD Ryzen 7 PRO 8840U
Display adapter 0 (primary) : Radeon 780M Graphics 
Results:
    Windows 11: Visual Studio 17 2022
        Benchmark: Plain blit
        Please wait...
        Time = 4.562 s, 4000 steps
        Plain blit: 876.808 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.058 s, 4504 steps
        Scaled blit: 890.471 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 5.096 s, 4672 steps
        Rotated blit: 916.797 FPS
        Done
    WSL Ubuntu 24.04: gcc version 13.3.0
        Benchmark: Plain blit
        Please wait...
        Time = 4.94147 s, 9839 steps
        Plain blit: 1991.11 FPS
        Done
        Benchmark: Scaled blit
        Please wait...
        Time = 5.14813 s, 9830 steps
        Scaled blit: 1909.43 FPS
        Done
        Benchmark: Rotated blit
        Please wait...
        Time = 4.30106 s, 8682 steps
        Rotated blit: 2018.57 FPS
        Done

[SiegeLord]

Thanks for doing more testing. I tried this on a weaker machine and using WSL, and I do sort of see a difference (on my more powerful machine the difference remained small):

WSL

Benchmark: Plain blit
Please wait...
Time = 3.99729 s, 6788 steps
Plain blit: 1698.15 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.96245 s, 8409 steps
Scaled blit: 1694.53 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.94029 s, 8650 steps
Rotated blit: 1750.91 FPS
Done

Win

Benchmark: Plain blit
Please wait...
Time = 4.504 s, 6024 steps
Plain blit: 1337.48 FPS
Done
Benchmark: Scaled blit
Please wait...
Time = 4.931 s, 6578 steps
Scaled blit: 1334.01 FPS
Done
Benchmark: Rotated blit
Please wait...
Time = 4.96 s, 6849 steps
Rotated blit: 1380.85 FPS
Done

I can probably work with this. I still don't have an explanation of why this happens though, it'll need some careful profiling.