On 2025/07/01 7:40, H.J. Lu wrote:
On Mon, Jun 23, 2025 at 5:41 AM Takayuki 'January June' Suwa
<jjsuwa_sys3...@yahoo.co.jp> wrote:

On 2025/06/23 6:20, H.J. Lu wrote:
On Sun, Jun 22, 2025 at 9:54 PM Max Filippov <jcmvb...@gmail.com> wrote:

On Sun, Jun 22, 2025 at 5:49 AM Takayuki 'January June' Suwa
<jjsuwa_sys3...@yahoo.co.jp> wrote:

On 2025/06/22 6:41, Max Filippov wrote:
On Sat, Jun 21, 2025 at 2:12 PM Takayuki 'January June' Suwa
<jjsuwa_sys3...@yahoo.co.jp> wrote:

That hook has since been deprecated
(commit a670ebde3995481225ec62b29686ec07a21e5c10) and has led to incorrect
results on Xtensa:

        /* example */
        #define <stdint.h>
        uint32_t __attribute__((noinline)) test0(uint32_t a, uint16_t b) {
          return a + b;
        }
        uint32_t __attribute__((noinline)) test1(uint32_t a, uint32_t b) {
          return test0(a, b);
        }

        ;; before (-mabi=call0)
        test0:
           add.n   a2, a3, a2
           ret.n
        test1:
           sext    a3, a3, 15      ;; NG, do not sign-extend
           j.l     test0, a9

        ;; after (-mabi=call0)
        test0:
           extui   a3, a3, 0, 16   ;; OK
           add.n   a2, a3, a2
           ret.n
        test1:
           j.l     test0, a9

With this patch, the result is consistent with other targets such as
AArch64.

gcc/ChangeLog:

           * config/xtensa/xtensa.cc
           (TARGET_PROMOTE_PROTOTYPES, TARGET_PROMOTE_FUNCTION_MODE):
           Remove.
---
     gcc/config/xtensa/xtensa.cc | 5 -----
     1 file changed, 5 deletions(-)

This results in a bunch of ICEs with the following backtrace:

How about leaving TARGET_PROMOTE_FUNCTION_MODE and just removing
TARGET_PROMOTE_PROTOTYPES?

commit a670ebde3995481225ec62b29686ec07a21e5c10
Author: H.J. Lu <hjl.to...@gmail.com>
Date:   Thu Nov 21 07:54:35 2024 +0800

      Drop targetm.promote_prototypes from C, C++ and Ada frontends

dropped targetm.calls.promote_prototypes in FEs.  But

commit 78db4753c9646a372512e6a951fced12f74de0bc
Author: H.J. Lu <hjl.to...@gmail.com>
Date:   Thu Nov 21 08:11:06 2024 +0800

      Honor TARGET_PROMOTE_PROTOTYPES during RTL expand

used targetm.calls.promote_prototypes.   TARGET_PROMOTE_PROTOTYPES
shouldn't be removed.

The sign extension of signed char function arguments is missing with
this change,
breaking the ABI, e.g

void s8(signed char c);
void cs8(signed char *c)
{
      s8(*c);
}

is translated to

cs8:
         entry   sp, 32
         l8ui    a10, a2, 0
         call8   s8
         retw.n

--
Thanks.
-- Max

Thank you for your attention to this issue. I will also withdraw my patch.


xtensa ABI requires sign extension of signed 8/16-bit arguments to 32
bits and zero extension of unsigned 8/16-bit arguments to 32 bits.
TARGET_PROMOTE_PROTOTYPES is an optimization, not an ABI requirement.
xtensa should remove TARGET_PROMOTE_PROTOTYPES and define
a proper TARGET_PROMOTE_FUNCTION_MODE, instead  Please try this:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120888#c4


For the first example I provided, this seems to be fine.
I'll defer to Max for more comprehensive regression testing.

=====
jjsuwa_sys3...@yahoo.co.jp

Reply via email to