Segfault with ifort

[ivan-pi]

With the ifort compiler I get a crash after a while.

Build information

$ FC=ifort cmake .. -DFASTGPT_BLAS=Fortran -DCMAKE_BUILD_TYPE=Debug
-- The C compiler identification is GNU 9.4.0
-- The CXX compiler identification is GNU 9.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- The Fortran compiler identification is Intel 20.2.5.20211109
-- Check for working Fortran compiler: /opt/intel/oneapi/compiler/2022.0.2/linux/bin/intel64/ifort
-- Check for working Fortran compiler: /opt/intel/oneapi/compiler/2022.0.2/linux/bin/intel64/ifort  -- works
-- Detecting Fortran compiler ABI info
-- Detecting Fortran compiler ABI info - done
-- Checking whether /opt/intel/oneapi/compiler/2022.0.2/linux/bin/intel64/ifort supports Fortran 90
-- Checking whether /opt/intel/oneapi/compiler/2022.0.2/linux/bin/intel64/ifort supports Fortran 90 -- yes


Configuration results
---------------------
Fortran compiler: /opt/intel/oneapi/compiler/2022.0.2/linux/bin/intel64/ifort
Build type: Debug
Fortran compiler flags: -warn all -check all,noarg_temp_created -traceback -O1 -g
Installation prefix: /usr/local
FASTGPT_BLAS: Fortran
-- Configuring done
-- Generating done
-- Build files have been written to: /home/ivan/lrz/fastGPT/build

Output:

(fastgpt) ivan@maxwell:~/lrz/fastGPT$ gprofng collect app ./build/gpt2 
Creating experiment directory test.10.er (Process ID: 260289) ...
Loading the model...
    done.
Model parameters:
n_vocab = 50257
n_ctx   =  1024
n_embd  =   768
n_layer =    12
n_head  =    12
 
Input parameters:
n_seq                =   6
n_tokens_to_generate =  20
 
Input tokens:
   464   995   286  9439 14448   284
Decoded input as text:
The world of tomorrow belongs to
Running model...
           1         262
           2         661
           3         286
           4        9439
           5          13
           6         198
           7         198
           8         464
           9         995
          10         286
          11        9439
          12       14448
          13         284
          14         262
          15         661
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source             
gpt2               000000000042F1FA  Unknown               Unknown  Unknown
libpthread-2.31.s  00007F2B36905420  Unknown               Unknown  Unknown
gpt2               00000000004268D3  linalg_mp_matmul_          18  linalg_f.f90
gpt2               000000000041EB03  gpt2_mod_mp_gpt2_         165  gpt2.f90
gpt2               00000000004208D6  gpt2_mod_mp_gener         194  gpt2.f90
gpt2               000000000040A56F  MAIN__                     85  main.f90
gpt2               0000000000403862  Unknown               Unknown  Unknown
libc-2.31.so       00007F2B3671D083  __libc_start_main     Unknown  Unknown
gpt2               000000000040376E  Unknown               Unknown  Unknown

I'm guessing this is related to the stack vs heap array issue with the Intel compiler and matmul (https://fortran-lang.discourse.group/t/testing-the-performance-of-matmul-under-default-compiler-settings/4096/27).