[Bug c/113065] New: [ARM][NEON] Compiler crashes when doing shift on data with NEON intrinsic type

[Bug c/113065] New: [ARM][NEON] Compiler crashes when doing shift on data with NEON intrinsic type

[clh960524+gcc at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113065

            Bug ID: 113065
           Summary: [ARM][NEON] Compiler crashes when doing shift on data
                    with NEON intrinsic type
           Product: gcc
           Version: 11.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: clh960524+gcc at gmail dot com
  Target Milestone: ---

Created attachment 56900
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56900&action=edit
preprocessor result

Hi developers,

Currently I am working on adding ARMv8-A 32-bit test environment.
However, compiler outputs an internal error when it tries to do shifting on
NEON intrinsic data type.

This bug is confirmed on arm-linux-gnueabihf-gcc version 10.5.0, 11.4.0, and
12.3.0.

For your ease, here is the code to reproduce the issue (test.c):

#include <stddef.h>

#include <arm_neon.h>
#include <arm_acle.h>

#define HEDLEY_STATIC_CAST(T, expr) ((T) (expr))

#define SIMDE_POLY64_TYPE poly64_t
#define SIMDE_POLY64_C(value) (HEDLEY_STATIC_CAST(poly64_t, value ## ull))

typedef SIMDE_POLY64_TYPE simde_poly64;
typedef simde_poly64 simde_poly64_t;
typedef poly128_t simde_poly128_t;

simde_poly128_t
simde_vaddq_p128(simde_poly128_t a, simde_poly128_t b) {
  simde_poly128_t mask = 0xFFFFFFFFFFFFFFFFull;
  mask = mask << 64; 
  mask = mask | 0xFFFFFFFFFFFFFFFFull;
  return b ^ ((0 ^ a) & mask);
}

static int
simde_assert_equal_p64(simde_poly64 a, simde_poly64 b) {
  uint64_t a_ = HEDLEY_STATIC_CAST(uint64_t, a);
  uint64_t b_ = HEDLEY_STATIC_CAST(uint64_t, b);
  if (a_ != b_) {
        return 1;
  }
  return 0;
}

static int
test_simde_vaddq_p128 () {
  struct {
    simde_poly64_t a[2];
    simde_poly64_t b[2];
    simde_poly64_t r[2];
  } test_vec[] = {
    { {  SIMDE_POLY64_C( 2773134259082670592),  SIMDE_POLY64_C(
9031857671464751104) },
      {  SIMDE_POLY64_C(15408611317451976704),  SIMDE_POLY64_C(
5314471842361763840) },
      {  SIMDE_POLY64_C(17557921622677996032),  SIMDE_POLY64_C(
3789586984176121856) } }, 
    { {  SIMDE_POLY64_C(14363410812027238400),  SIMDE_POLY64_C(
7751777860661664768) },
      {  SIMDE_POLY64_C( 4070846577912177664),  SIMDE_POLY64_C(
8033403141717473280) },
      {  SIMDE_POLY64_C(18386960794487266304),  SIMDE_POLY64_C( 
355706841938420736) } }, 
    { {  SIMDE_POLY64_C(17019201421568192512),  SIMDE_POLY64_C(
9077731670917768192) },
      {  SIMDE_POLY64_C( 1093803490337685888), 
SIMDE_POLY64_C(12690135020377692160) },
      {  SIMDE_POLY64_C(16365414741193798016), 
SIMDE_POLY64_C(14836825620528675840) } }, 
    { {  SIMDE_POLY64_C( 8834005665657798656),  SIMDE_POLY64_C(
7315692870291161088) },
      {  SIMDE_POLY64_C( 3583172236868845568), 
SIMDE_POLY64_C(14913079096527761408) },
      {  SIMDE_POLY64_C( 5413693383775240192), 
SIMDE_POLY64_C(12354334249137418240) } }, 
    { {  SIMDE_POLY64_C(16318651152855345152), 
SIMDE_POLY64_C(15283972988033570816) },
      {  SIMDE_POLY64_C( 4795255037585536000), 
SIMDE_POLY64_C(17815911970323484672) },
      {  SIMDE_POLY64_C(11599959615349332992),  SIMDE_POLY64_C(
2532502929754736640) } }, 
    { {  SIMDE_POLY64_C(16540936757178476544),  SIMDE_POLY64_C(
4596081763017794560) },
      {  SIMDE_POLY64_C(15120265319048798208),  SIMDE_POLY64_C(
3251201193519464448) },
      {  SIMDE_POLY64_C( 3771992723781634048),  SIMDE_POLY64_C(
1357279216665519104) } }, 
    { {  SIMDE_POLY64_C( 7991373722374909952), 
SIMDE_POLY64_C(17418001980645539840) },
      {  SIMDE_POLY64_C(  303154826827825408),  SIMDE_POLY64_C(
9983013542091358208) },
      {  SIMDE_POLY64_C( 7697238601746657536),  SIMDE_POLY64_C(
8877691528758722560) } }, 
    { {  SIMDE_POLY64_C(14668463690243614720),  SIMDE_POLY64_C(
4370163632906065408) },
      {  SIMDE_POLY64_C( 3236423722687647744),  SIMDE_POLY64_C(
6051015495717347328) },
      {  SIMDE_POLY64_C(16679868711160412160),  SIMDE_POLY64_C(
8024418529387197952) } }, 
  };

  for (size_t i = 0 ; i < (sizeof(test_vec) / sizeof(test_vec[0])) ; i++) {
    simde_poly128_t a = test_vec[i].a[0];
    simde_poly128_t b = test_vec[i].b[0];
    simde_poly128_t r, mask;
    simde_poly64_t top_r, bottom_r;
    a = a << 64;
    a = a | test_vec[i].a[1];
    b =  b << 64;
    b =  b | test_vec[i].b[1];
    r = simde_vaddq_p128(a, b);

    mask = HEDLEY_STATIC_CAST(simde_poly128_t, 0xFFFFFFFFFFFFFFFFull);
    top_r = HEDLEY_STATIC_CAST(simde_poly64_t, (r >> 64) & mask);
    bottom_r = HEDLEY_STATIC_CAST(simde_poly64_t, (r << 64) >> 64);

    simde_assert_equal_p64(top_r, test_vec[i].r[0]);
    simde_assert_equal_p64(bottom_r, test_vec[i].r[1]);
  }

  return 0;
}

int main() {
  test_simde_vaddq_p128();
  return 0;
}

Here's the command for compiling:

arm-linux-gnueabihf-gcc -v -save-temps -mcpu=cortex-a32
-mfpu=crypto-neon-fp-armv8 -Wextra -Werror -Wall -fno-strict-aliasing -fwrapv
-fno-aggressive-loop-optimizations test.c

[Bug c/113065] [ARM][NEON] Compiler crashes when doing shift on data with NEON intrinsic type

[clh960524+gcc at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113065

--- Comment #1 from 陳履軒 <clh960524+gcc at gmail dot com> ---
Here's the result after compiling:

$ arm-linux-gnueabihf-gcc -mcpu=cortex-a32 -mfpu=crypto-neon-fp-armv8 -Wextra
-Werror -Wall -fno-strict-aliasing -fwrapv -fno-aggressive-loop-optimizations
test.c
during RTL pass: expand
test.c: In function ‘simde_vaddq_p128’:
test.c:18:8: internal compiler error: in expand_shift_1, at expmed.c:2668
   18 |   mask = mask << 64;
      |   ~~~~~^~~~~~~~~~~~
0x7f9d29bf4d8f __libc_start_call_main
        ../sysdeps/nptl/libc_start_call_main.h:58
0x7f9d29bf4e3f __libc_start_main_impl
        ../csu/libc-start.c:392
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <file:///usr/share/doc/gcc-11/README.Bugs> for instructions.

[Bug target/113065] [ARM][NEON] Compiler crashes when doing shift on data with NEON intrinsic type

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113065

Alex Coplan <acoplan at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2023-12-18
           Keywords|                            |ice-on-valid-code
                 CC|                            |acoplan at gcc dot gnu.org

--- Comment #2 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Confirmed, here is a reduced testcase:

$ cat t.c
typedef __builtin_neon_poly128 poly128_t;
poly128_t test_simde_vaddq_p128(poly128_t a) { return a << 4; }
$ ./xgcc -B . -c t.c -S -o /dev/null -march=armv8-a+simd -mfloat-abi=hard
during RTL pass: expand
t.c: In function ‘test_simde_vaddq_p128’:
t.c:2:57: internal compiler error: in expand_shift_1, at expmed.cc:2686
    2 | poly128_t test_simde_vaddq_p128(poly128_t a) { return a << 4; }
      |                                                       ~~^~~~
0x225b687 expand_shift_1
        /home/alecop01/toolchain/src/gcc/gcc/expmed.cc:2686
0x225b7a0 expand_variable_shift(tree_code, machine_mode, rtx_def*, tree_node*,
rtx_def*, int)
        /home/alecop01/toolchain/src/gcc/gcc/expmed.cc:2727
0x229007d expand_expr_real_2(separate_ops*, rtx_def*, machine_mode,
expand_modifier)
        /home/alecop01/toolchain/src/gcc/gcc/expr.cc:10256
0x207a203 expand_gimple_stmt_1
        /home/alecop01/toolchain/src/gcc/gcc/cfgexpand.cc:3983
0x207a44c expand_gimple_stmt
        /home/alecop01/toolchain/src/gcc/gcc/cfgexpand.cc:4044
0x20824cc expand_gimple_basic_block
        /home/alecop01/toolchain/src/gcc/gcc/cfgexpand.cc:6100
0x208487d execute
        /home/alecop01/toolchain/src/gcc/gcc/cfgexpand.cc:6835
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.