On Mon, Mar 9, 2026 at 3:15 PM Daniel Henrique Barboza
<[email protected]> wrote:
>
>
>
> On 3/9/2026 10:31 AM, Richard Biener wrote:
> > On Mon, Mar 9, 2026 at 2:09 PM Daniel Henrique Barboza
> > <[email protected]> wrote:
> >>
> >> From: Daniel Barboza <[email protected]>
> >>
> >> Identify cases where a zero_one comparison is used to conditional
> >> constant assignment and turn that into an unconditional PLUS. For the
> >> code in PR71336:
> >>
> >> int test(int a) {
> >> return a & 1 ? 7 : 3;
> >> }
> >>
> >> We'll turn that into "(a&1) * (7 - 3) + 3", which yields the same
> >> results but without the conditional, promoving more optimization
> >> opportunities. In an armv8-a target the original code generates:
> >>
> >> 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
> >>
> >> With this transformation:
> >>
> >> 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
> >>
> >> Similar gains are noticeable in RISC-V and x86.
> >>
> >> For completeness sake we're also adding the variant "zero_one == 0".
> >>
> >> Bootstrapped and regression tested in x86 and aarch64.
> >
> > This is a very bad transform on targets that cannot do (wide) integer
> > multiplication, like AVR. I would suggest to gate this on types
> > <= word_mode? optab availability isn't going to work since to workaround
> > similar issues AVR for example implements some of these but with
> > explicit libgcc dispatch in the insn patterns.
>
> The original idea behind this transform (and similar ones like 56110 and
> 123967,
> which I'm planning to send in the next few days) was to handle only pow2
> values
> and recover the immediate via lshift. I found that a bit restrictive and
> decided
> to move to 'mult'.
>
> I guess we could use the lshift for all pow2 values and then, for non-pow2
> vals
> that would require a mult, check if type <= word_mode.
I'm not sure if a shift will be much better here. There are targets where
a conditional select is always better than using arithmetic and it is difficult
to recover that at RTL expansion time. So maybe this kind of instruction
selection should happen later, in a more (cost) controlled manner?
I'm aware there are other places / passes in GCC that fall into this
trap for AVR.
It would be nice to develop a workable gating thats applicable to all of those
places. I understand we do not want to delay if-conversion for all targets.
Richard.
>
>
> Thanks,
> Daniel
>
>
>
> >
> > Richard.
> >
> >> PR tree-optimization/71336
> >>
> >> gcc/ChangeLog:
> >>
> >> * match.pd(`zero_one EQ|NE 0 ? CST1:CST2`): New pattern.
> >>
> >> gcc/testsuite/ChangeLog:
> >>
> >> * gcc.dg/tree-ssa/pr71336-2.c: New test.
> >> * gcc.dg/tree-ssa/pr71336.c: New test.
> >> ---
> >> gcc/match.pd | 36 ++++++++++++++
> >> gcc/testsuite/gcc.dg/tree-ssa/pr71336-2.c | 59 +++++++++++++++++++++++
> >> gcc/testsuite/gcc.dg/tree-ssa/pr71336.c | 20 ++++++++
> >> 3 files changed, 115 insertions(+)
> >> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71336-2.c
> >> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71336.c
> >>
> >> diff --git a/gcc/match.pd b/gcc/match.pd
> >> index 7f16fd4e081..d041276c595 100644
> >> --- a/gcc/match.pd
> >> +++ b/gcc/match.pd
> >> @@ -5195,6 +5195,42 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> >> && expr_no_side_effects_p (@2))
> >> (op (mult (convert:type @0) @2) @1))))
> >>
> >> +/* PR71336:
> >> + zero_one != 0 ? CST1 : CST2 -> ((typeof (CST2))zero_one * diff) + CST2,
> >> + where CST1 > CST2 and diff = CST1 - CST2.
> >> +
> >> + Includes the "zero_one == 0 ? (...)" variant too. */
> >> +(for cmp (ne eq)
> >> + (simplify
> >> + (cond (cmp zero_one_valued_p@0 integer_zerop) INTEGER_CST@1
> >> INTEGER_CST@2)
> >> + (with {
> >> + unsigned HOST_WIDE_INT diff = 0;
> >> +
> >> + if (tree_int_cst_sgn (@1) > 0 && tree_int_cst_sgn (@2) > 0
> >> + && tree_fits_uhwi_p (@1) && tree_fits_uhwi_p (@2))
> >> + {
> >> + if (cmp == NE_EXPR
> >> + && wi::gtu_p (wi::to_wide (@1), wi::to_wide (@2)))
> >> + diff = tree_to_uhwi (@1) - tree_to_uhwi (@2);
> >> +
> >> + if (cmp == EQ_EXPR
> >> + && wi::gtu_p (wi::to_wide (@2), wi::to_wide (@1)))
> >> + diff = tree_to_uhwi (@2) - tree_to_uhwi (@1);
> >> + }
> >> + }
> >> + (if (cmp == NE_EXPR
> >> + && INTEGRAL_TYPE_P (type)
> >> + && INTEGRAL_TYPE_P (TREE_TYPE (@0))
> >> + && diff > 0)
> >> + (plus (mult (convert:type @0) { build_int_cst (type, diff); })
> >> + @2)
> >> + (if (cmp == EQ_EXPR
> >> + && INTEGRAL_TYPE_P (type)
> >> + && INTEGRAL_TYPE_P (TREE_TYPE (@0))
> >> + && diff > 0)
> >> + (plus (mult (convert:type @0) { build_int_cst (type, diff); })
> >> + @1))))))
> >> +
> >> /* ?: Value replacement. */
> >> /* a == 0 ? b : b + a -> b + a */
> >> (for op (plus bit_ior bit_xor)
> >> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71336-2.c
> >> b/gcc/testsuite/gcc.dg/tree-ssa/pr71336-2.c
> >> new file mode 100644
> >> index 00000000000..da44489d3e4
> >> --- /dev/null
> >> +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71336-2.c
> >> @@ -0,0 +1,59 @@
> >> +/* { dg-do run } */
> >> +/* { dg-options "-O1" } */
> >> +
> >> +/* Macro adapted from builtin-object-size-common.h */
> >> +#define FAIL() \
> >> + do { \
> >> + __builtin_printf ("Failure at line: %d\n", __LINE__); \
> >> + abort(); \
> >> + } while (0)
> >> +
> >> +void abort(void);
> >> +
> >> +int test (int a) {
> >> + return a & 1 ? 7 : 3;
> >> +}
> >> +
> >> +int test2 (int a) {
> >> + return a & 1 ? 3 : 7;
> >> +}
> >> +
> >> +int test3 (int a) {
> >> + return (a & 1) == 0 ? 3 : 7;
> >> +}
> >> +
> >> +int test4 (int a) {
> >> + return (a & 1) == 0 ? 7 : 3;
> >> +}
> >> +
> >> +int main (void) {
> >> + if (test (0) != 3)
> >> + FAIL ();
> >> + if (test (1) != 7)
> >> + FAIL ();
> >> + if (test (3) != 7)
> >> + FAIL ();
> >> +
> >> + if (test2 (0) != 7)
> >> + FAIL ();
> >> + if (test2 (1) != 3)
> >> + FAIL ();
> >> + if (test2 (3) != 3)
> >> + FAIL ();
> >> +
> >> + if (test3 (0) != 3)
> >> + FAIL ();
> >> + if (test3 (1) != 7)
> >> + FAIL ();
> >> + if (test3 (2) != 3)
> >> + FAIL ();
> >> +
> >> + if (test4 (0) != 7)
> >> + FAIL ();
> >> + if (test4 (1) != 3)
> >> + FAIL ();
> >> + if (test4 (2) != 7)
> >> + FAIL ();
> >> +
> >> + return 0;
> >> +}
> >> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71336.c
> >> b/gcc/testsuite/gcc.dg/tree-ssa/pr71336.c
> >> new file mode 100644
> >> index 00000000000..fb643bf1eb3
> >> --- /dev/null
> >> +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71336.c
> >> @@ -0,0 +1,20 @@
> >> +/* { dg-additional-options -O1 } */
> >> +/* { dg-additional-options -fdump-tree-gimple } */
> >> +
> >> +int test (int a) {
> >> + return a & 1 ? 7 : 3;
> >> +}
> >> +
> >> +int test2 (int a) {
> >> + return (a & 1) == 0 ? 3 : 7;
> >> +}
> >> +
> >> +int test3 (int a) {
> >> + return a & 1 ? 17 : 3;
> >> +}
> >> +
> >> +int test4 (int a) {
> >> + return (a & 1) == 0 ? 3 : 17;
> >> +}
> >> +
> >> +/* { dg-final { scan-tree-dump-times " goto " 0 gimple } } */
> >> --
> >> 2.43.0
> >>
>
