On Mon, Jan 26, 2026 at 8:35 AM Wilco Dijkstra <[email protected]> wrote:
>
>
> Symbols with a large offset may be emitted into a literal pool.
> aarch64_select_rtx_section() may then select .text or .rodata even if
> the symbol has a dynamic relocation. Checking for CONST_INT or
> CONST_DOUBLE avoids this. Ensure aarch64_cannot_force_const_mem()
> returns true for symbols and labels so that they cannot be placed in
> literal pools. Only the large code model emits symbols in the literal
> pool (no support for PIC/PIE).
>
> Passes regress, OK for commit?
Ok (though question on the testcase).
>
> gcc:
> PR target/123791
> * config/aarch64/aarch64.cc (aarch64_cannot_force_const_mem):
> Always return true for symbols and labels except for large model.
> (aarch64_select_rtx_section): Force .rodata for constants only.
>
> gcc/testsuite:
> PR target/123791
> * gcc.target/aarch64/pr123791.c: New test.
>
> ---
>
> diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> index
> 765c76165443e1829b0bf59e2956b1c5da476d04..b6ab1449ee4d73e9376043361d2738463e366f29
> 100644
> --- a/gcc/config/aarch64/aarch64.cc
> +++ b/gcc/config/aarch64/aarch64.cc
> @@ -11145,19 +11145,12 @@ aarch64_cannot_force_const_mem (machine_mode mode
> ATTRIBUTE_UNUSED, rtx x)
> || aarch64_sme_vq_unspec_p (x, &factor))
> return true;
>
> + /* Only allow symbols in literal pools with the large model (non-PIC). */
> poly_int64 offset;
> rtx base = strip_offset_and_salt (x, &offset);
> - if (SYMBOL_REF_P (base) || LABEL_REF_P (base))
> - {
> - /* We checked for POLY_INT_CST offsets above. */
> - if (aarch64_classify_symbol (base, offset.to_constant ())
> - != SYMBOL_FORCE_TO_MEM)
> - return true;
> - else
> - /* Avoid generating a 64-bit relocation in ILP32; leave
> - to aarch64_expand_mov_immediate to handle it properly. */
> - return mode != ptr_mode;
> - }
> + if ((SYMBOL_REF_P (base) || LABEL_REF_P (base))
> + && aarch64_cmodel != AARCH64_CMODEL_LARGE)
> + return true;
>
> return aarch64_tls_referenced_p (x);
> }
> @@ -14312,11 +14305,14 @@ aarch64_select_rtx_section (machine_mode mode,
> rtx x,
> unsigned HOST_WIDE_INT align)
> {
> + /* Forcing special symbols into the .text section is not correct
> (PR123791).
> + This is only an issue with the large code model. */
> if (aarch64_can_use_per_function_literal_pools_p ())
> return function_section (current_function_decl);
>
> /* When using anchors for constants use the readonly section. */
> - if (known_le (GET_MODE_SIZE (mode), 8))
> + if ((CONST_INT_P (x) || CONST_DOUBLE_P (x))
> + && known_le (GET_MODE_SIZE (mode), 8))
> return readonly_data_section;
>
> return default_elf_select_rtx_section (mode, x, align);
> diff --git a/gcc/testsuite/gcc.target/aarch64/pr123791.c
> b/gcc/testsuite/gcc.target/aarch64/pr123791.c
> new file mode 100644
> index
> 0000000000000000000000000000000000000000..afdf37e8aac1a6451bb1375a32ab68f81b32b36c
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/pr123791.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mpc-relative-literal-loads" } */
> +
> +char *
> +foo ()
> +{
> + return (char *) (__UINTPTR_TYPE__) foo + 7483647;
> +}
> +
> +/* { dg-final { scan-assembler-not "\\.(word|xword)\tfoo" { xfail
> aarch64_large } } } */
This is a real xfail or is it supposed to match for large code model?
If it is a real xfail, please make sure there is a bug report filed
for it (if there is one, please add the bug number to the testcase).
Thanks,
Andrew
>
