On Mon, Nov 3, 2025 at 12:36 PM <[email protected]> wrote: > > From: Pan Li <[email protected]> > > There are 3 kinds of widen_mul during the unsigned SAT_MUL pattern, aka > * widen_mul directly, like _3 w* _4 > * convert and the widen_mul, like (uint64_t)_3 *w (uint64_t)_4 > * convert and then mul, like (uint64_t)_3 * (uint64_t)_4 > > All of them will be referenced during different forms of unsigned > SAT_MUL pattern match, but actually we can wrap them into a helper > which present the "widening_mul" sematics. With this helper, some > unnecessary pattern and duplicated code could be eliminated.
OK. Thanks, Richard. > The below test suites are passed for this patch: > 1. The rv64gcv fully regression tests. > 2. The x86 bootstrap tests. > 3. The x86 fully regression tests. > > gcc/ChangeLog: > > * match.pd: Add usmul_widen_mult helper and referenced by > min based unsigned SAT_MUL pattern. > > Signed-off-by: Pan Li <[email protected]> > --- > gcc/match.pd | 64 +++++++++++++++++++++++----------------------------- > 1 file changed, 28 insertions(+), 36 deletions(-) > > diff --git a/gcc/match.pd b/gcc/match.pd > index 6aaf80eee7d..0ea86d97416 100644 > --- a/gcc/match.pd > +++ b/gcc/match.pd > @@ -3656,34 +3656,39 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT) > /* Saturation mult for unsigned integer. */ > (if (INTEGRAL_TYPE_P (type) && TYPE_UNSIGNED (type)) > (for mult_op (mult widen_mult) > - (match (unsigned_integer_sat_mul @0 @1) > - /* SAT_U_MUL (X, Y) = { > - WT x = (WT)a * (WT)b; > - T max = -1; > - if (x > (WT)(max)) > - return max; > - else > - return (T)x; > + (match (usmul_widen_mult @0 @1) > + (mult_op:c (convert@2 @0) (convert @1)) > + (if (types_match (@0, @1) && TYPE_UNSIGNED (TREE_TYPE (@0))) > + (with > + { > + unsigned prec = TYPE_PRECISION (TREE_TYPE (@0)); > + unsigned cvt2_prec = TYPE_PRECISION (TREE_TYPE (@2)); > + bool widen_cvt_p = cvt2_prec > prec; > } > - while WT is uint128_t, T is uint8_t, uint16_t, uint32_t or uint64_t. > */ > - (convert (min (mult_op:c@3 (convert@4 @0) (convert@5 @1)) INTEGER_CST@2)) > - (if (types_match (type, @0, @1)) > - (with > - { > + (if (widen_cvt_p)))))) > + (match (usmul_widen_mult @0 @1) > + (widen_mult:c @0 @1) > + (if (types_match (@0, @1)))) > + (match (unsigned_integer_sat_mul @0 @1) > + /* SAT_U_MUL (X, Y) = { > + WT x = (WT)a * (WT)b; > + T max = -1; > + if (x > (WT)(max)) > + return max; > + else > + return (T)x; > + } > + while WT is uint128_t, T is uint8_t, uint16_t, uint32_t or uint64_t. */ > + (convert (min (usmul_widen_mult@3 @0 @1) INTEGER_CST@2)) > + (if (types_match (type, @0, @1)) > + (with > + { > unsigned prec = TYPE_PRECISION (type); > unsigned widen_prec = TYPE_PRECISION (TREE_TYPE (@3)); > - unsigned cvt4_prec = TYPE_PRECISION (TREE_TYPE (@4)); > - unsigned cvt5_prec = TYPE_PRECISION (TREE_TYPE (@5)); > - > wide_int max = wi::mask (prec, false, widen_prec); > bool c2_is_max_p = wi::eq_p (wi::to_wide (@2), max); > - > - bool widen_mult_p = mult_op == WIDEN_MULT_EXPR && cvt4_prec == > cvt5_prec > - && widen_prec == cvt5_prec * 2 && widen_prec > prec; > - bool mult_p = mult_op == MULT_EXPR && cvt4_prec == cvt5_prec > - && cvt4_prec == widen_prec && widen_prec > prec; > - } > - (if (c2_is_max_p && (widen_mult_p || mult_p))))))) > + } > + (if (c2_is_max_p))))) > (match (unsigned_integer_sat_mul @0 @1) > /* SAT_U_MUL (X, Y) = { > T result; > @@ -3751,19 +3756,6 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT) > bool c2_is_type_precision_p = tree_to_uhwi (@2) == prec; > } > (if (c2_is_type_precision_p && (mult_p || widen_mult_p))))))) > - (match (unsigned_integer_sat_mul @0 @1) > - (convert (min (widen_mult:c@3 @0 @1) INTEGER_CST@2)) > - (if (types_match (type, @0, @1)) > - (with > - { > - unsigned prec = TYPE_PRECISION (type); > - unsigned widen_prec = TYPE_PRECISION (TREE_TYPE (@3)); > - > - wide_int max = wi::mask (prec, false, widen_prec); > - bool c2_is_max_p = wi::eq_p (wi::to_wide (@2), max); > - bool widen_mult_p = prec * 2 == widen_prec; > - } > - (if (c2_is_max_p && widen_mult_p))))) > (match (unsigned_integer_sat_mul @0 @1) > (convert1? > (bit_ior > -- > 2.43.0 >
