[Bug c/110119] New: RISC-V: RVV --param=riscv-autovec-preference=fixed-vlmax ICE

[Bug c/110119] New: RISC-V: RVV --param=riscv-autovec-preference=fixed-vlmax ICE

[juzhe.zhong at rivai dot ai]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110119

            Bug ID: 110119
           Summary: RISC-V: RVV
                    --param=riscv-autovec-preference=fixed-vlmax ICE
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: juzhe.zhong at rivai dot ai
  Target Milestone: ---

This following cases:

#include "riscv_vector.h"

typedef int8_t vnx2qi __attribute__ ((vector_size (2)));

__attribute__ ((noipa)) vnx2qi
f_vnx2qi (int8_t a, int8_t b, int8_t *out)
{
  vnx2qi v = {a, b};
  return v;
}

__attribute__ ((noipa)) vnx2qi
f_vnx2qi_2 (vnx2qi a, int8_t *out)
{
  return a;
}

__attribute__ ((noipa)) vint32m1_t
f_vint32m1 (int8_t * a, int8_t *out)
{
  vint32m1_t v = *(vint32m1_t*)a;
  return v;
}

are all causing ICE:

0x1667a6d crash_signal
        ../../../riscv-gcc/gcc/toplev.cc:314
0x10a476c hard_function_value(tree_node const*, tree_node const*, tree_node
const*, int)
        ../../../riscv-gcc/gcc/explow.cc:2258
0x116f092 aggregate_value_p(tree_node const*, tree_node const*)
        ../../../riscv-gcc/gcc/function.cc:2126
0x117938e allocate_struct_function(tree_node*, bool)
        ../../../riscv-gcc/gcc/function.cc:4831
0xd0c7b7 store_parm_decls()
        ../../../riscv-gcc/gcc/c/c-decl.cc:10743
0xd7811e c_parser_declaration_or_fndef
        ../../../riscv-gcc/gcc/c/c-parser.cc:2805
0xd75ea7 c_parser_external_declaration
        ../../../riscv-gcc/gcc/c/c-parser.cc:1925
0xd7592e c_parser_translation_unit
        ../../../riscv-gcc/gcc/c/c-parser.cc:1779
0xdc1d2b c_parse_file()
        ../../../riscv-gcc/gcc/c/c-parser.cc:24657
0xe5f20b c_common_parse_file()
        ../../../riscv-gcc/gcc/c-family/c-opts.cc:1248

The reason of this ICE is because this walkaround code:

  /* TODO: Currently, it will cause an ICE for --param
     riscv-autovec-preference=fixed-vlmax. So, we just return NULL_RTX here
     let GCC generate loads/stores. Ideally, we should either warn the user not
     to use an RVV vector type as function argument or support the calling
     convention directly.  */
  if (riscv_v_ext_mode_p (mode))
    return NULL_RTX;

in riscv_get_arg_info.

From my experience (my local downstream GCC doesn't have such issue), I am
100% sure that such issue will be addressed when we support vector calling
convention in GCC. 

And we already have calling convention in "rvv-next" branch, there is no
difference between "rvv-next" and my internal downstream RVV GCC.

The calling convention implementation of "rvv-next" is totally same as LLVM.
However, the calling convention is not ratified yet.

I don't have time to figure out the approach without supporting calling
convention and I don't think we need this approach.

I believe we will ultimately remove this approach (without supporting calling
convention) since we will have vector calling convention.

Also, this ICE only happens when we specify
--param=riscv-autovec-preference=fixed-vlmax

Base on these circumstances, I suggest we support vector calling convention and
enable it when --param=riscv-autovec-preference=fixed-vlmax is enable.
I think such GCC internal compile option to enable vector calling convention
doesn't break the RISC-V standard.

Any suggestions?

[Bug target/110119] RISC-V: RVV --param=riscv-autovec-preference=fixed-vlmax ICE

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110119

--- Comment #1 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Pan Li <panli@gcc.gnu.org>:

https://gcc.gnu.org/g:0ec3fbb5903ac3ad735b3154e814b46724fe1a27

commit r14-1829-g0ec3fbb5903ac3ad735b3154e814b46724fe1a27
Author: Lehua Ding <lehua.ding@rivai.ai>
Date:   Wed Jun 14 19:56:11 2023 +0800

    RISC-V: Ensure vector args and return use function stack to pass [PR110119]

    The V2 patch address comments from Juzhe, thanks.

    Hi,

    The reason for this bug is that in the case where the vector register is
set
    to a fixed length (with `--param=riscv-autovec-preference=fixed-vlmax`
option),
    TARGET_PASS_BY_REFERENCE thinks that variables of type vint32m1 can be
passed
    through two scalar registers, but when GCC calls FUNCTION_VALUE (call
function
    riscv_get_arg_info inside) it returns NULL_RTX. These two functions are not
    unified. The current treatment is to pass all vector arguments and returns
    through the function stack, and a new calling convention for vector
registers
    will be added in the future.

    https://github.com/riscv-non-isa/riscv-elf-psabi-doc/
   
https://github.com/palmer-dabbelt/riscv-elf-psabi-doc/commit/126fa719972ff998a8a239c47d506c7809aea363

    Best,
    Lehua

    gcc/ChangeLog:
            PR target/110119
            * config/riscv/riscv.cc (riscv_get_arg_info): Return NULL_RTX for
vector mode
            (riscv_pass_by_reference): Return true for vector mode

    gcc/testsuite/ChangeLog:
            PR target/110119
            * gcc.target/riscv/rvv/base/pr110119-1.c: New test.
            * gcc.target/riscv/rvv/base/pr110119-2.c: New test.

[Bug target/110119] RISC-V: RVV --param=riscv-autovec-preference=fixed-vlmax ICE

[juzhe.zhong at rivai dot ai]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110119

JuzheZhong <juzhe.zhong at rivai dot ai> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED

--- Comment #2 from JuzheZhong <juzhe.zhong at rivai dot ai> ---
fix on trunk.