[Bug tree-optimization/102971] New: GCC cannot understand >>32 pattern

[Bug tree-optimization/102971] New: GCC cannot understand >>32 pattern

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102971

            Bug ID: 102971
           Summary: GCC cannot understand >>32 pattern
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: unlvsur at live dot com
  Target Milestone: ---

https://godbolt.org/z/eeG8fnY1z

I want to seperate a uint_least64_t to 2x uint_least32_t or merge 2x
uint_least32_t to uint_least64_t. GCC just cannot understand the pattern at all
here.

The assembly it generates is very bad for things like this.

[Bug tree-optimization/102971] GCC cannot understand >>32 pattern

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102971

--- Comment #1 from cqwrteur <unlvsur at live dot com> ---
Here is clang
https://godbolt.org/z/7YjW1Ezrx

[Bug tree-optimization/102971] GCC cannot understand >>32 pattern

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102971

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
#include <cstdint>

std::uint64_t umul_naive(std::uint64_t a,std::uint64_t b,std::uint64_t& high)
noexcept
{
        std::uint32_t a0(static_cast<std::uint32_t>(a));
        std::uint32_t a1(static_cast<std::uint32_t>(a>>32));
        std::uint32_t b0(static_cast<std::uint32_t>(b));
        std::uint32_t b1(static_cast<std::uint32_t>(b>>32));
        std::uint64_t c00(static_cast<std::uint64_t>(a0)*b0);
        std::uint64_t c01(static_cast<std::uint64_t>(a0)*b1);
        std::uint64_t c10(static_cast<std::uint64_t>(a1)*b0);
        std::uint64_t c11(static_cast<std::uint64_t>(a1)*b1);

        std::uint64_t d0{static_cast<std::uint32_t>(c00)};
        std::uint64_t c00_high{c00>>32};
        std::uint64_t c01_low{static_cast<std::uint32_t>(c01)};
        std::uint64_t c01_high{c01>>32};
        std::uint64_t c10_low{static_cast<std::uint32_t>(c10)};
        std::uint64_t c10_high{c10>>32};

        std::uint64_t d1{c00_high+c01_low+c10_low};
        std::uint64_t d2{(d1>>32)+c10_high+c01_high};
        std::uint64_t d3{(d2>>32)+c11};
        high=d2|(d3<<32);
        return d0|(d1<<32);
}

[Bug target/102971] GCC cannot understand >>32 pattern

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102971

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
          Component|middle-end                  |target
             Target|i?86                        |i?86-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-10-28

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  With -mno-stv the assembly looks much better so something is wrong
there with costing when deciding to do 64bit math in SSE regs as opposed to
GPR pairs.