[Bug tree-optimization/71336] Suboptimal code generated for "(a & 1) ? (CST1 + CST2) : CST1"

[daniel.barboza at oss dot qualcomm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71336

Daniel Henrique Barboza <daniel.barboza at oss dot qualcomm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |daniel.barboza at oss dot qualcomm
                   |                            |.com

--- Comment #13 from Daniel Henrique Barboza <daniel.barboza at oss dot qualcomm.com> ---
A fix proposal was sent to the ML (currently on v2):

https://gcc.gnu.org/pipermail/gcc-patches/2026-March/710154.html

The idea is to generalize the idea Jeff gave in comment #12: for a pattern
like:

int test(int a) {
    return a & 1 ? 7 : 3;
}

We can simplify it to 

3 + (a & 1) * (7 - 3);

This will remove the conditional and targets can optimize the PLUS pattern at
their leisure.  In an armv8-a target, using current trunk, this function will
generate:

test:
.LFB0:
        .cfi_startproc
        tst     x0, 1   // 38   [c=8 l=4]  *anddi3nr_compare0_zextract
        mov     w1, 3   // 41   [c=4 l=4]  *movsi_aarch64/3
        mov     w0, 7   // 42   [c=4 l=4]  *movsi_aarch64/3
        csel    w0, w1, w0, eq  // 17   [c=4 l=4]  *cmovsi_insn/0
        ret             // 47   [c=0 l=4]  *do_return
        .cfi_endproc


With the proposed fix:

.LFB0:
        .cfi_startproc
        ubfiz   w0, w0, 2, 1    // 7    [c=4 l=4]  *andim_ashiftsi_bfiz
        add     w0, w0, 3       // 13   [c=4 l=4]  *addsi3_aarch64/0
        ret             // 21   [c=0 l=4]  *do_return
        .cfi_endproc


For RISC-V the generated .s is:

test:
.LFB0:
        .cfi_startproc
        andi    a0,a0,1 # 6     [c=4 l=4]  *anddi3/1
        slliw   a0,a0,2 # 7     [c=8 l=4]  ashlsi3_extend
        addi    a0,a0,3 # 16    [c=4 l=4]  *adddi3/1
        ret             # 29    [c=0 l=4]  simple_return
        .cfi_endproc



This idea of using a mult to re-create an immediate to eliminate a conditional
is also used in 56110 and 123967.  I'll send patches for those shortly.

design is going to be used in 56110 and 123967 as well.