Unexpected Higher UNet Compute Buffer Sizes in Winograd vs. im2col Implementations

[rmatif]

I'm developing a Flutter wrapper around the sdcpp implementation and have been testing the performance. While I'm seeing a 1.15-1.2x speedup on Android compared to the original sdcpp repo, I've noticed an unexpected pattern in memory usage where the Winograd implementation shows higher UNet compute buffer sizes compared to im2col, which seems counterintuitive given that im2col typically requires more memory for intermediate matrices. Typically like here

Here are some results I have compiled during my testings :

Stable Diffusion 1.5 (SD1.5)

Resolution	Winograd UNET Buffer Size	im2col UNET Buffer Size	Difference (Winograd - im2col)
512x512	611.79 MB	559.71 MB	+52.08 MB
384x384	244.16 MB	192.08 MB	+52.08 MB
256x256	100.41 MB	49.43 MB	+50.98 MB

Stable Diffusion XL (SDXL)

Resolution	Winograd UNET Buffer Size	im2col UNET Buffer Size	Difference (Winograd - im2col)
1024x1024	864.29 MB	830.19 MB	+34.10 MB
512x512	156.29 MB	131.85 MB	+24.44 MB
384x384	113.32 MB	95.47 MB	+17.85 MB
256x256	96.29 MB	60.31 MB	+35.98 MB

I also tried compiling with OpenCL, as you mention in your report that it is supported on Android. However, I encountered crashes and compilation errors. I would appreciate it if you could confirm its support and include a guide in the README

EDIT :

For anyone wondering how to build CLBlast on Termux (note: it has worse performance than CPU-only execution since only some operations are offloaded to the GPU, so it's generally not worth it):

Install dependencies and libraries :

pkg update
pkg upgrade
pkg install git cmake clblast opencl-headers binutils termux-elf-cleaner ocl-icd clinfo mesa opencl-clhpp opencl-vendor-driver

Note : binutils-is-llvm can do the work just as binutils

build sdcpp

git clone https://github.com/SealAILab/stable-diffusion-cpp.git 
cd stable-diffusion-cpp/
cmake -B build -DGGML_CLBLAST=ON -DCMAKE_EXE_LINKER_FLAGS="-fuse-ld=bfd" -DCMAKE_SHARED_LINKER_FLAGS="-fuse-ld=bfd" 
cmake --build build --config Release -j8

fixing alignment

cp build/bin/sd ../
cd ..
termux-elf-cleaner sd

You should now have a fully working CLBlast binary. I got my Mali GPU working out of the box, but Adreno GPUs might require additional steps. See here: termux/termux-packages#16852