Re: [RFC PATCH v3 0/4] Renesas RX FDPIC support

On 7/20/2026 7:36 AM, Yoshinori Sato wrote:
> Hello,
>
> This is v3 of the Renesas RX FDPIC support patch.
>
> This update cleans up and refactors the PIC/FDPIC operand legitimization
> and PC-relative symbol addressing based on feedback and updates.
>
> Summary of changes:
> 1. PIC / FDPIC & PC-Relative Refactoring
>     - Unified PIC operand legitimization helper functions by replacing
>       rx_mov_pic_operands and rx_maybe_pidify_operand with
>       rx_maybe_legitimize_pi_operand and rx_expand_pcrel[cite: 1].
>     - Improved address legitimacy checks (rx_is_legitimate_address) and local
>       read-only symbol handling (rx_is_local_readonly_data_p) for PC-relative
>       addressing[cite: 1].
>     - Updated PIC offset table register tracking in prologue/epilogue and call
>       patterns (clobbering/using PIC_REG where appropriate, and defining
>       EPILOGUE_USES for FDPIC)[cite: 1].
>     - Standardized assembly output printing for GOT/GOTOFF/PLT unspecs using
>       a new helper function (rx_print_operand_with_attrib)[cite: 1].
>
> 2. General Cleanups & Fixes
>     - Cleaned up predicate definitions in predicates.md and header prototypes[cite: 1].
>     - Improved GNU coding style compliance and comment formatting[cite: 1].
>
> Changes since v2:
> - Refactored PIC/FDPIC helper functions and PC-relative address loading.
> - Cleaned up assembly output for GOT/GOTOFF/PLT symbol attributes.
> - (LRA support and DImode arithmetic expansion were split out into separate patches).
>
> Testing:
> - Confirmed no unexpected failures (no regressions) in `make check`.
> - Verified that BusyBox built with this compiler runs correctly on a real
>    Renesas RX62N MCU board.
>
> Specification reference:
> https://gitlab.com/yoshinori.sato/gcc/-/wikis/RX-FDPIC
>
> Thanks,
> Yoshinori Sato
>
> Yoshinori Sato (4):
>    RX: FDPIC support - machine description
>    RX: FDPIC support - helper functions
>    RX: FDPIC support - headers
>    RX: shared libgcc support
So I applied this series in my tester and fired up a test run just to 
see how it'd respond.  This series triggers 631 regression failures.  So 
there's clearly still work to do.

As a sample:


rx-sim: gcc: gcc.c-torture/compile/981006-1.c   -O0  (test for excess 
errors)
rx-sim: gcc: gcc.c-torture/compile/pr100241-1.c   -O0  (test for excess 
errors)
rx-sim: gcc: gcc.c-torture/compile/pr65153.c   -O1  (test for excess errors)
rx-sim: gcc: gcc.c-torture/compile/pr69102.c   -O0  (test for excess errors)
[ ... ]

Odds are there's just a few issues causing the vast majority of the 
failures.  Regardless I would suggest you run the testsuite on your 
submissions to verify they don't cause regressions.

jeff