Access Violation with ifx /O3 /Qipo (single-file MRE attached)

[Nishimura]

Hello,

I have a reproducible runtime crash in Intel Fortran (ifx) when compiling with optimization and IPO enabled.

Environment:
- Compiler: Intel Fortran Compiler for Intel(R) 64, Version 2025.2.0 Build 20250605
- OS: Windows 11 Pro
- Build environment: cmd.exe

Reproduction:
1) Save attached file as `scat-bug.f90` (single-file MRE).
2) Compile and link:
ifx /fpp /O3 /Qipo /traceback scat-bug.f90
3) Run:
scat-bug.exe

Observed:
- Program crashes with "forrtl: severe (157): Program Exception - access violation".
- run.log (tail) shows:
ALLOCATED(NSTIC)= T
LBOUND= 1 1
UBOUND= 40000 20
i,n= 34 1
forrtl: severe (157): Program Exception - access violation

Additional notes:
- The crash occurs only with `/O3 /Qipo`. Disabling IPO (`/Qipo-`) or compiling with `/Od` avoids the crash.
- The crash occurs at assignment `NSTIC(I,N)=J` while `allocated(NSTIC)` is true.
- Optimization reports (attached) contain `INLINE: MDLPDB_V2` and multiple `DELETE: llvm.memset.*` / `DELETE: llvm.memcpy.*`.
This suggests the compiler removed initialization/copy sequences after inlining/IPO.

Attachments: scat-bug.zip (scat-bug.f90, compile_opt.txt, link_opt.txt, run.log, ifx_version.txt)

Please let me know if you need further details.

Thanks,
T. Nishimura

[Nishimura]

Update (follow-up):

Following up on my previous post — I confirmed that the error still occurs on the latest Intel compiler (Intel Fortran Compiler for Intel(R) 64, Version 2025.3.0).

I reduced the test case further; the shorter example below still reproduces the crash when compiled with:

ifx /O3 /Qipo /traceback scat-bug-v2.f90

I hope this helps the ifx team when investigating and improving future compiler releases.

Thanks,

T. Nishimura

[jonejames235]

Thanks for sharing the detailed report. Since the crash happens only with /O3 /Qipo, it looks like an IPO optimization issue. As a quick workaround, try compiling with /Qipo- or /O2 until Intel releases a patch. You can also report this to Intel via Online Service Center with your attached MRE — it will help them reproduce and fix the bug.

[Nishimura]

Thanks for your suggestion.
Unfortunately, I don’t have access to the Intel Online Service Center since I’m using the free (non-priority) edition of the compiler.
I’ll keep the report here so that Intel engineers can review it — hopefully they can reproduce it and file it internally.
Thanks again for confirming it’s likely IPO-related!

T. Nishimura

[andrew_4619]

I had a play with this code and made additional tests and diagnostics. I agree it is an optimisation bug  but the behaviour is quite baffling!

IPO is the trigger at /03

[Nishimura]

Thanks for taking the time to test it and confirm. Much appreciated!