[Bug middle-end/115916] New: [15 Regression] wrong code on highway-1.2.0

[Bug middle-end/115916] New: [15 Regression] wrong code on highway-1.2.0

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

            Bug ID: 115916
           Summary: [15 Regression] wrong code on highway-1.2.0
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: slyfox at gcc dot gnu.org
  Target Milestone: ---

Seemingly a recent regression.

On gcc-master r15-2026-g44c9403ed1833a highway-1.2.0 testsuite fails as:

The following tests FAILED:
        1190 -
HwyShuffle4TestGroup/HwyShuffle4Test.TestAllPer4LaneBlockShuffle/EMU128  #
GetParam() = 2305843009213693952 (Subprocess aborted)

Here is the minimized version extracted out of original example:

// $ cat shuffle4_test.cc
#include <stddef.h>
#include <stdint.h>

struct ve {
    ve() = default;
    ve(const ve&) = default;
    ve& operator=(const ve&) = default;

    // note that the code usually uses the first half of this array
    uint8_t raw[16] = {};
};

static ve First8_(void) {
    ve m;
    __builtin_memset(m.raw, 0xff, 8);
    return m;
}

static ve And_(ve a, ve b) {
    ve au;
    __builtin_memcpy(au.raw, a.raw, 16);
    for (size_t i = 0; i < 8; ++i) {
        au.raw[i] &= b.raw[i];
    }
    return au;
}

__attribute__((noipa, optimize(0)))
static void vec_assert(ve a) {
    if (a.raw[6] != 0x06 && a.raw[6] != 0x07)
        __builtin_trap();
}

static ve Reverse4_(ve v) {
    ve ret;
    for (size_t i = 0; i < 8; i += 4) {
        ret.raw[i + 0] = v.raw[i + 3];
        ret.raw[i + 1] = v.raw[i + 2];
        ret.raw[i + 2] = v.raw[i + 1];
        ret.raw[i + 3] = v.raw[i + 0];
    }
    return ret;
}

static ve DupEven_(ve v) {
    for (size_t i = 0; i < 8; i += 2) {
        v.raw[i + 1] = v.raw[i];
    }
    return v;
}

template <bool b>
ve Per4LaneBlockShuffle_(ve v) {
    if (b) {
        return Reverse4_(v);
    } else {
        return DupEven_(v);
    }
}

template <bool b>
static inline __attribute__((always_inline)) void
DoTestPer4LaneBlkShuffle(const ve v) {
    ve actual = Per4LaneBlockShuffle_<b>(v);
    const auto valid_lanes_mask = First8_();
    ve actual_masked = And_(valid_lanes_mask, actual);
    vec_assert(actual_masked);
}

static void DoTestPer4LaneBlkShuffles(const ve v) {
    alignas(128) uint8_t src_lanes[8];
    __builtin_memcpy(src_lanes, v.raw, 8);
    // need both, hm
    DoTestPer4LaneBlkShuffle<true >(v);
    DoTestPer4LaneBlkShuffle<false>(v);
}

__attribute__((noipa, optimize(0)))
static void bug(void) {
   uint8_t iv[8] = {1,2,3,4,5,6,7,8};
   ve v;
   __builtin_memcpy(v.raw, iv, 8);
   DoTestPer4LaneBlkShuffles(v);
}

int main(void) {
    bug();
}

Bad:

$ gcc/xg++ -Bgcc shuffle4_test.cc -O3 -o bug && ./bug
Illegal instruction (core dumped)

Ok:

$ gcc/xg++ -Bgcc shuffle4_test.cc -O2 -o bug && ./bug

$ gcc/xg++ -Bgcc -v
Reading specs from gcc/specs
COLLECT_GCC=gcc/xg++
COLLECT_LTO_WRAPPER=gcc/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /home/slyfox/dev/git/gcc/configure --disable-multilib
--disable-bootstrap --disable-lto --disable-libsanitizer
--disable-libstdcxx-pch --enable-languages=c,c++ --disable-libgomp
--disable-libquadmath --disable-libvtv CFLAGS='-O1 -g0' CXXFLAGS='-O1 -g0'
LDFLAGS='-O1 -g0'
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 15.0.0 20240714 (experimental) (GCC)

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Sergei Trofimovich <slyfox at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu.org
            Summary|[15 Regression] wrong code  |[15 Regression] wrong code
                   |on highway-1.2.0            |on highway-1.2.0 since
                   |                            |r15-2011-ga6f551d079de1d

--- Comment #1 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
Bisect says r15-2011-ga6f551d079de1d introduced it:

commit a6f551d079de1d151b272bcdd3d42316857c9d4e
Author: Jeff Law <jlaw@ventanamicro.com>
Date:   Fri Jul 12 13:11:33 2024 -0600

    [PR rtl-optimization/115876] Fix one of two ubsan reported issues in new
ext-dce.cc code

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

--- Comment #2 from Andreas Schwab <schwab@linux-m68k.org> ---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115876#c7

It broke the ASHIFT case.

-        return mask &gt;&gt; INTVAL (XEXP (x, 1));
+        return (HOST_WIDE_INT)mask &gt;&gt; INTVAL (XEXP (x, 1));</pre>

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Andreas Schwab <schwab@linux-m68k.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zsojka at seznam dot cz

--- Comment #3 from Andreas Schwab <schwab@linux-m68k.org> ---
*** Bug 115877 has been marked as a duplicate of this bug. ***

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Andreas Schwab <schwab@linux-m68k.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sjames at gcc dot gnu.org

--- Comment #4 from Andreas Schwab <schwab@linux-m68k.org> ---
*** Bug 115912 has been marked as a duplicate of this bug. ***

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
   Target Milestone|---                         |15.0

[Bug middle-end/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-07-15

[Bug rtl-optimization/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jeff Law <law@gcc.gnu.org>:

https://gcc.gnu.org/g:94b21f13763638f64e83e7f9959c7f1523b9eaed

commit r15-2048-g94b21f13763638f64e83e7f9959c7f1523b9eaed
Author: Jeff Law <jlaw@ventanamicro.com>
Date:   Mon Jul 15 16:57:44 2024 -0600

    Fix sign/carry bit handling in ext-dce.

    My change to fix a ubsan issue broke handling propagation of the carry/sign
bit
    down through a right shift.  Thanks to Andreas for the analysis and
proposed
    fix and Sergei for the testcase.

            PR rtl-optimization/115876
            PR rtl-optimization/115916
    gcc/
            * ext-dce.cc (carry_backpropagate): Make return type unsigned as
well.
            Cast to signed for right shift to preserve sign bit.

    gcc/testsuite/

            * g++.dg/torture/pr115916.C: New test.

            Co-author: Andreas Schwab <schwab@linux-m68k.org>
            Co-author: Sergei Trofimovich <slyfox at gentoo dot org>

[Bug rtl-optimization/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

Jeffrey A. Law <law at gcc dot gnu.org> changed:

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

--- Comment #6 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Fixed on the trunk.

[Bug rtl-optimization/115916] [15 Regression] wrong code on highway-1.2.0 since r15-2011-ga6f551d079de1d

[slyfox at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115916

--- Comment #7 from Sergei Trofimovich <slyfox at gcc dot gnu.org> ---
The change also fixed highway-1.2.0 testsuite for me. Thank you!