On Thu, Jun 25, 2026 at 1:38 AM Philipp Tomsich
<[email protected]> wrote:
>
> Andrea,
>
> Agreed, staging this into a series will go a long way towards making
> it more reviewable.
> We'll resubmit a restructured series.
>
> Proposed split:
> 1. The single flat carry form: a top-level '+' chain of the four 32x32
> partial products plus one overflow-compare carry. This is what
> multiply_uint64_hw64_generic lowers to (and has been a separate commit
> in our tree anyway).
> 2.[n]. One decomposition variant per patch: carry-long, two-carry,
> ladder, ladder-long, low-plusl; each with the matching test.
> 3. The 2-arg-PHI carry form (`if (overflow) hi += 1;`). This is
> strictly additive on top of the earlier.
>
> The bulk of the "complexity" is due to the same longhand written
> several different ways across sealcrypto and the Hacker's Delight
> variants.
> The mid-end hands us several canonical forms.
>
> Regarding "worrying about the target": we gate the high-part emit on
> the 2N scalar mode existing because on a target without a native wide
> multiply, we'd turn a working hand-rolled longhand into a
> __multi3-style libgcc call.
> I'd recommend to keep that check.

Why not implement the opposite flow in say widenning_multiply when the
target does not have a widening multiply?
This will allow for when someone say just wrote
((__int128)long_var1)*((__int128)long_var2) in the first place.
and then canonical form during gimple is one thing?

Thanks,
Andrea

>
> Most of the +3300 lines is testsuite; the logic is ~1.2k. The split
> above should make each patch individually digestible in a single
> sitting.
>
> Thanks,
> Philipp
>
>
> On Thu, 25 Jun 2026 at 01:05, Andrea Pinski
> <[email protected]> wrote:
> >
> > On Wed, Jun 24, 2026 at 2:09 PM Philipp Tomsich
> > <[email protected]> wrote:
> > >
> > > Also, this restores feature-parity with LLVM.
> > > The equivalent support was added in late November to LLVM:
> > > https://github.com/llvm/llvm-project/pull/168396
> >
> > I have a big question. Why is it much more complex than LLVM's 
> > implementation?
> > Can we start with a simple implementation and add more things to it?
> > Also each revision seems to get even more complex and adds much more things.
> > So let's start with assuming no basic block changes needed.
> > And then we can add the second case.
> > Also start with not worrying about what the target has, that can be
> > lowered during expand and/or during widen_multiply.
> >
> > This will allow it easier to review this huge patch. Because it seems
> > like each time it is getting bigger and bigger.
> >
> > Thanks,
> > Andrea
> >
> > >
> > >
> > > On Wed, 24 Jun 2026 at 16:14, Konstantinos Eleftheriou
> > > <[email protected]> wrote:
> > > >
> > > >
> > > > This patch series teaches GCC to recognize longhand 64x64->128
> > > > wide-multiplication idioms and replace them with native multiply
> > > > instructions: a widening multiply followed by a right shift for the
> > > > high part, and plain MULT_EXPR for the low part.
> > > >
> > > > Portable C/C++ code that needs a 128-bit product on a 64-bit target
> > > > often resorts to a longhand decomposition: split operands into 32-bit
> > > > halves, compute four partial products, and propagate carries manually.
> > > > This pattern appears in a number of real-world codebases, including
> > > > SPEC2026's 750.sealcrypto_r (seal/util/uintarith.h) and several
> > > > examples from Hacker's Delight.  Targets like AArch64 (mul/umulh) and
> > > > x86-64 can compute the full 128-bit product in one or two instructions,
> > > > but GCC does not currently fold the longhand sequence back to these.
> > > >
> > > > The series is split into two patches:
> > > >
> > > >   1/2  forwprop: Match and fold long-multiply patterns [PR107090]
> > > >
> > > >        Adds match.pd atom patterns plus a forwprop framework that
> > > >        linearises the outer add/ior chain, classifies summands, and
> > > >        matches the multiset against a table of six decomposed
> > > >        variants:
> > > >
> > > >          - carry:       single overflow comparison on the cross-sum
> > > >          - carry-long:  cross-carry with separate high/low accumulation
> > > >          - two-carry:   both cross-carry and low-carry as separate
> > > >                         comparisons
> > > >          - ladder:      sequential accumulation without explicit carry
> > > >                         comparison
> > > >          - ladder-long: ladder with separate high/low accumulation
> > > >          - low-plus:    low part as a direct sum of partial products
> > > >
> > > >        Emits the canonical widening shape
> > > >
> > > >          (N)(((2N) a * (2N) b) >> N)
> > > >
> > > >        for high parts.  pass_optimize_widening_mul later lowers this
> > > >        to WIDEN_MULT_EXPR / MULT_HIGHPART_EXPR on supporting targets.
> > > >        Skipped when the target lacks the 2N scalar mode.
> > > >
> > > >   2/2  match.pd, forwprop: Recognise long-multiply carries written as
> > > >        2-arg PHI
> > > >
> > > >        The 1/2 fold's entry walks a top-level + statement.  Hand-
> > > >        written long-multiply code often writes the carry as a 2-arg
> > > >        PHI (`if (overflow) result += pow2;`) instead, with no
> > > >        top-level + at the result.  2/2 adds cond_carry_add and
> > > >        cond_carry_add_neg match.pd recognizers for that PHI shape
> > > >        and a match_long_mul_phi entry that linearises the no-carry
> > > >        arm, synthesises the carry summand from the PHI bindings,
> > > >        and reuses 1/2's table walk and emit path.
> > > >
> > > > On SPEC2026's 750.sealcrypto_r:
> > > >
> > > >   - AArch64 Neoverse-N1: 25% improvement
> > > >   - x86-64 Zen4:         59% improvement
> > > >
> > > > Compile-time impact: single-threaded recompile of gcc/*.cc with each
> > > > branch's stage3 cc1plus (checking=yes,extra) takes 87:05 on master vs
> > > > 87:55 on this series -- ~1% overhead.
> > > >
> > > > Bootstrapped/regtested on AArch64, x86-64 and PowerPC.
> > > >
> > > > Changes in v6:
> > > > - Reorder so the long-multiply fold (was 2/2) is now 1/2 and a new
> > > >   PHI-form recognition pass is 2/2.  Reverting 2/2 leaves a working
> > > >   long-multiply fold for the flat-shifted-compare carry form.
> > > > - Drop v5's standalone flatten_cond_carry_add driver.  The same
> > > >   cond_carry_add / cond_carry_add_neg match.pd recognizers now feed
> > > >   a match_long_mul_phi entry inside the long-multiply fold, so a
> > > >   PHI-shaped carry folds straight to the wide-multiply output.
> > > > - Factor long_mul_classify_chain, long_mul_classify_match and
> > > >   build_mul_high_seq for sharing between match_long_mul and the new
> > > >   match_long_mul_phi.
> > > > - cond_carry_add_neg uses le / ge instead of gt / lt to encode the
> > > >   carry condition strictly.  v5 inverted the compare via
> > > >   invert_tree_comparison in the flatten driver; v6 synthesises the
> > > >   carry summand directly inside match_long_mul_phi and so requires
> > > >   the recogniser to encode the strict form.
> > > > - Delete forwprop-44/45/46.c; add PHI-form coverage in
> > > >   long-mul-carry.c, long-mul-two-carry.c, long-mul-boundary.c
> > > >   and long-mul-boundary-64.c.
> > > > - Add PHI-form near-miss tests in long-mul-partial.c and
> > > >   operand-swap polarity coverage in long-mul-boundary{,-64}.c.
> > > > - Refresh stale long-mul comment references (check_hilo_and_ops,
> > > >   fold_mul_low_plus) and reword mul_carry_low's :c-on-gt note to
> > > >   the correct LT form (a + b < a).
> > > >
> > > > Changes in v5:
> > > > - 1/2:
> > > >   - Replace the match.pd simplify on COND_EXPR with cond_carry_add
> > > >     / cond_carry_add_neg match recognizers (cond^), split by gcond
> > > >     polarity, plus a flatten_cond_carry_add driver in
> > > >     tree-ssa-forwprop.cc.  The driver inverts the gcond's
> > > >     comparison for the _neg form.  Modelled on match_saturation_add.
> > > >   - Remove fold_cond_carry_add_profitable_p and the tm_p.h /
> > > >     predict.h includes from gimple-match-head.cc.  The width >
> > > >     MAX_FIXED_MODE_SIZE and width % 2 != 0 guards were
> > > >     prerequisites for the can_mult_highpart_p fallback path, not
> > > >     soundness checks.  type_has_mode_precision_p subsumes them.
> > > >   - Retarget the test scans from phiopt2 to forwprop1.  Add
> > > >     forwprop-46.c covering all four arm/comparison polarities.
> > > >   - forwprop-45.c uses __UINT64_TYPE__ instead of unsigned long
> > > >     and drops the lp64 restriction, covering the type > word_mode
> > > >     regime on 32-bit targets.
> > > > - 2/2:
> > > >   - Lower the high-part as (N)(((2N) op1 * (2N) op2) >> N).
> > > >     pass_optimize_widening_mul rewrites this to WIDEN_MULT_EXPR /
> > > >     MULT_HIGHPART_EXPR on supporting targets.  Removes
> > > >     can_mult_highpart_p queries from forwprop.
> > > >   - Replace the can_mult_highpart_p prefilter in match_long_mul
> > > >     with a targetm.scalar_mode_supported_p check on the 2N mode.
> > > >     Test scans select on int128, mirroring the gate, instead of
> > > >     lp64.
> > > >   - Drop the m_long_mul_fold_p pass parameter and its passes.def
> > > >     arguments.  The long-mul fold runs in every forwprop instance.
> > > >     Test scans retargeted from forwprop2 to forwprop1.
> > > >   - Stop restricting forwprop-44.c to lp64.  With the
> > > >     can_mult_highpart_p gating gone, the fold is target-independent
> > > >     and the test passes on ilp32 targets too.
> > > >
> > > > Changes in v4:
> > > > - 1/2:
> > > >   - Rebuild the guard with per-conjunct reasoning: require both
> > > >     operands to be SSA names (drops degenerate one-side-constant
> > > >     cases that fold trivially elsewhere), require the type to
> > > >     have_mode_precision_p (excludes BITINT_TYPE precision != mode
> > > >     and similar oddities), drop the explicit MAX_FIXED_MODE_SIZE
> > > >     width cap (subsumed by have_mode_precision_p), and gate on the
> > > >     flat optab via can_mult_highpart_p of the 2N mode.
> > > >   - Retain BRANCH_COST >= 2: keep the flatten conditional on a
> > > >     target where the branchless form is generally cheaper.
> > > >   - Rewrite the cover letter to describe the gate as the
> > > >     composition of these conjuncts and to clarify that the
> > > >     transformation now only ever introduces a (mul_hi-like)
> > > >     can_mult_highpart_p shape, not a libgcc multi-precision call.
> > > > - 2/2:
> > > >   - Convert per-variant fold_mul_* functions into a
> > > >     table-driven long_mul fold framework.
> > > >   - Migrate each variant into a row in long_mul_table (six
> > > >     HIGH_PART, six LOW_PART rows) keyed by (kind, extract).
> > > >   - Add cross-summand consistency checks
> > > >     (long_mul_check_consistency, long_mul_check_two_carries,
> > > >     long_mul_check_low_plus_defer) shared across rows.
> > > >   - Drop emission to a libgcc multi-precision call from RTL
> > > >     expansion; defer to pass_optimize_widening_mul / RTL
> > > >     expansion to pick native umul_highpart, a widening multiply, or
> > > >     a synthesised sequence.  Emission is gated on
> > > >     can_mult_highpart_p.
> > > >   - Structural redesign: per-variant fold_mul_* functions
> > > >     consolidated into a single linearise + classify + table-lookup
> > > >     framework (long_mul_table, match_long_mul,
> > > >     long_mul_classify_summand, long_mul_check_consistency).  Each
> > > >     variant is now a row in long_mul_table; consistency checks are
> > > >     shared across rows.
> > > >   - Fast-fail prefilters in match_long_mul: LHS-type prefilter at
> > > >     entry (no legitimate long-mul leaf has a signed / pointer /
> > > >     float / odd-width type) and a can_mult_highpart_p probe before
> > > >     the row loop to skip HIGH_PART rows on unsupported targets.
> > > >   - Bound long_mul_linearize_chain mid-walk by LONG_MUL_MAX_SUMMANDS
> > > >     so an overlong addition / BIT_IOR chain bails immediately rather
> > > >     than after a full traversal.
> > > >   - Emit a dump-file hint pointing at the shared inner addition when
> > > >     long-mul folding rejects a chain because of a multi-used
> > > >     intermediate (caching the partial sum into a single-use SSA
> > > >     name normally enables the fold).
> > > >
> > > > Changes in v3:
> > > > - Moved carry-diamond flattening from forwprop to match.pd,
> > > > replacing ~460 lines of C++ with a 17-line match.pd pattern.
> > > > - Two-carry test scans forwprop3 (the first forwprop after phiopt2,
> > > > since early phiopt restricts which tree codes are allowed).
> > > > - Set location for new sequences.
> > > > - Updated mul_carry_low pattern.
> > > > - Added the `mul_low_plus` pattern.
> > > > - Fixed formatting issues.
> > > >
> > > > Changes in v2:
> > > > - Fixed the testcases by separating the high part's fold count for
> > > > 32-bit and 64-bit targets.
> > > >
> > > > Konstantinos Eleftheriou (2):
> > > >   forwprop: Match and fold long-multiply patterns [PR107090]
> > > >   match.pd, forwprop: Recognise long-multiply carries written as 2-arg
> > > >     PHI
> > > >
> > > >  gcc/match.pd                                  |  177 +++
> > > >  .../gcc.dg/tree-ssa/long-mul-boundary-64.c    |  398 ++++++
> > > >  .../gcc.dg/tree-ssa/long-mul-boundary.c       |  394 ++++++
> > > >  .../gcc.dg/tree-ssa/long-mul-carry.c          |  373 ++++++
> > > >  .../gcc.dg/tree-ssa/long-mul-ladder.c         |  329 +++++
> > > >  .../gcc.dg/tree-ssa/long-mul-low-plus.c       |   54 +
> > > >  .../gcc.dg/tree-ssa/long-mul-partial.c        |  193 +++
> > > >  .../gcc.dg/tree-ssa/long-mul-two-carry.c      |  142 +++
> > > >  gcc/testsuite/gcc.target/aarch64/long_mul.c   |  100 ++
> > > >  gcc/testsuite/gcc.target/i386/long_mul.c      |  100 ++
> > > >  gcc/tree-ssa-forwprop.cc                      | 1066 ++++++++++++++++-
> > > >  11 files changed, 3318 insertions(+), 8 deletions(-)
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-boundary-64.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-boundary.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-carry.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-ladder.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-low-plus.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-partial.c
> > > >  create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/long-mul-two-carry.c
> > > >  create mode 100644 gcc/testsuite/gcc.target/aarch64/long_mul.c
> > > >  create mode 100644 gcc/testsuite/gcc.target/i386/long_mul.c
> > > >
> > > > --
> > > > 2.52.0

Reply via email to