[Bug c/112713] New: RISC-V: Trunk GCC regression on VSETVL PASS comparing with GCC-13

[Bug c/112713] New: RISC-V: Trunk GCC regression on VSETVL PASS comparing with GCC-13

[juzhe.zhong at rivai dot ai]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112713

            Bug ID: 112713
           Summary: RISC-V: Trunk GCC regression on VSETVL PASS comparing
                    with GCC-13
           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: ---

https://godbolt.org/z/qePhcxd5z

#include "riscv_vector.h"

size_t
foo (char const *buf, size_t len)
{
        size_t sum = 0;
        size_t vl = __riscv_vsetvlmax_e8m8();
        size_t step = vl * 4;
        const char *it = buf, *end = buf + len;
        for(; it + step <= end; ) {
                it += vl;
                vint8m8_t v3 = __riscv_vle8_v_i8m8((void*)it, vl); it += vl;
                vbool1_t m3 = __riscv_vmsgt_vx_i8m8_b1(v3, -65, vl);
                sum += __riscv_vcpop_m_b1(m3, vl);
        }
        return sum;
}

GCC-14 ASM:

foo:
        vsetvli a4,zero,e8,m8,ta,ma
        slli    a5,a4,2
        add     a1,a0,a1
        add     a5,a0,a5
        bltu    a1,a5,.L4
        slli    a6,a4,1
        add     a5,a0,a4
        add     t1,a6,a4
        li      a0,0
        li      a7,-65
        vsetvli zero,zero,e8,m8,ta,ma   ----------> redundant
.L3:
        vle8.v  v8,0(a5)
        add     a5,a5,a6
        add     a3,t1,a5
        vmsgt.vx        v8,v8,a7
        vcpop.m a2,v8
        add     a0,a0,a2
        bgeu    a1,a3,.L3
        ret
.L4:
        li      a0,0
        ret

GCC-13 ASM:

foo:
        vsetvli a4,zero,e8,m8,ta,ma
        slli    a5,a4,2
        add     a1,a0,a1
        add     a5,a0,a5
        bltu    a1,a5,.L4
        slli    a6,a4,1
        add     a5,a0,a4
        add     t1,a6,a4
        li      a0,0
        li      a7,-65
.L3:
        vle8.v  v8,0(a5)
        add     a5,a5,a6
        add     a3,t1,a5
        vmsgt.vx        v24,v8,a7
        vcpop.m a2,v24
        add     a0,a0,a2
        bgeu    a1,a3,.L3
        ret
.L4:
        li      a0,0
        ret

Trunk GCC-14 has a regression that cause a redundant vsetvl.

Such regression need to be fixed and I will take a look at it.

[Bug target/112713] RISC-V: Trunk GCC regression on VSETVL PASS comparing with GCC-13

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112713

--- Comment #1 from GCC 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:9c16ca93641ad460a576a9ed7daf2aadf596193c

commit r14-5897-g9c16ca93641ad460a576a9ed7daf2aadf596193c
Author: Juzhe-Zhong <juzhe.zhong@rivai.ai>
Date:   Mon Nov 27 21:24:12 2023 +0800

    RISC-V: Fix VSETVL PASS regression

    This patch is regression fix patch, not an optimization patch.
    Since trunk GCC generates redundant vsetvl than GCC-13.

    This is the case:

    bb 2:
      def a2 (vsetvl a2, zero)
    bb 3:
      use a2
    bb 4:
      use a2 (vle)

    before this patch:

    bb 2:
    vsetvl a2 zero
    bb 3:
    vsetvl zero, zero ----> should be eliminated.
    bb 4:
    vle.v

    The root cause is we didn't set bb 3 as transparent since the incorrect
codes.
    bb 3 didn't modify "a2" just use it, the VSETVL status from bb 2 can be
available to bb 3 and bb 4:

    bb 2 -> bb 3 -> bb4.

    Another regression fix is anticipation calculation:

    bb 4:
    use a5 (sub)
    use a5 (vle)

    The vle VSETVL status should be considered as anticipated as long as both
sub and vle a5 def are coming from same def.

    Tested on zvl128b no regression.

    I am going to test on zvl256/zvl512/zvl1024

            PR target/112713

    gcc/ChangeLog:

            * config/riscv/riscv-vsetvl.cc
(pre_vsetvl::compute_lcm_local_properties): Fix regression.

    gcc/testsuite/ChangeLog:

            * gcc.target/riscv/rvv/vsetvl/pr112713-1.c: New test.
            * gcc.target/riscv/rvv/vsetvl/pr112713-2.c: New test.

[Bug target/112713] RISC-V: Trunk GCC regression on VSETVL PASS comparing with GCC-13

[juzhe.zhong at rivai dot ai]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112713

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> ---
Fixed