Request to support aarch64

[trholding]

Make errors out on a aarch64 server

make base.en
#gcc -pthread -O3 -c ggml.c
gcc -pthread -O3 -mcpu=cortex-a72 -mfloat-abi=hard -mfpu=neon-fp-armv8 -mfp16-format=ieee -mno-unaligned-access -c ggml.c
gcc: error: unrecognized command-line option ‘-mfloat-abi=hard’
gcc: error: unrecognized command-line option ‘-mfpu=neon-fp-armv8’
gcc: error: unrecognized command-line option ‘-mfp16-format=ieee’
gcc: error: unrecognized command-line option ‘-mno-unaligned-access’
make: *** [Makefile:7: ggml.o] Error 1

Perhaps this is enough as C flags? : -Ofast -g -mfpu=neon

[WilliamTambellini]

A solution: use/link with
https://github.com/oneapi-src/oneDNN
for onednn to do all the optimizations for the local CPU at runtime whatever the cpu model/brand.

[ggerganov]

@trholding Yes, I think this will work. Obviously, the current Makefile is not very cross-platform - I mainly focus to be able to compile on M1 and Intel processors. With time, it will be improved to support various other platforms.

[trholding]

@ggerganov Works awesome on aarch64!

time make base.en 
bash ./download-ggml-model.sh base.en
Downloading ggml model base.en ...
Model base.en already exists. Skipping download.

===============================================
Running base.en on all samples in ./samples ...
===============================================

----------------------------------------------
[+] Running base.en on samples/jfk.wav ... (run 'ffplay samples/jfk.wav' to listen)
----------------------------------------------

whisper_model_load: loading model from 'models/ggml-base.en.bin'
whisper_model_load: n_vocab       = 51864
whisper_model_load: n_audio_ctx   = 1500
whisper_model_load: n_audio_state = 512
whisper_model_load: n_audio_head  = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx    = 448
whisper_model_load: n_text_state  = 512
whisper_model_load: n_text_head   = 8
whisper_model_load: n_text_layer  = 6
whisper_model_load: n_mels        = 80
whisper_model_load: f16           = 1
whisper_model_load: type          = 2
whisper_model_load: mem_required  = 377.00 MB
whisper_model_load: adding 1607 extra tokens
whisper_model_load: ggml ctx size = 163.43 MB
whisper_model_load: memory size =    22.83 MB 
whisper_model_load: model size  =   140.54 MB

main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, lang = en, task = transcribe, timestamps = 1 ...

[00:00.000 --> 00:11.000]   And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.


whisper_print_timings:     load time =   154.29 ms
whisper_print_timings:      mel time =    82.46 ms
whisper_print_timings:   sample time =     7.87 ms
whisper_print_timings:   encode time =  5479.56 ms / 913.26 ms per layer
whisper_print_timings:   decode time =   553.81 ms / 92.30 ms per layer
whisper_print_timings:    total time =  6278.83 ms


real    0m6.331s
user    0m23.265s
sys     0m0.257s