> -----Original Message-----
> From: Jan Hubicka <[email protected]>
> Sent: 01 December 2025 11:59
> To: Kugan Vivekanandarajah <[email protected]>
> Cc: [email protected]; Jakub Jelinek <[email protected]>; Richard
> Biener <[email protected]>; Tamar Christina
> <[email protected]>
> Subject: Re: [PATCH 2/4] Implement hierarchical discriminators for AutoFDO
>
> >
> > From 01f9fe56970d66d6931abffb0360088c6fb0107e Mon Sep 17
> 00:00:00 2001
> > From: Kugan Vivekanandarajah <[email protected]>
> > Date: Tue, 11 Nov 2025 18:11:18 -0800
> > Subject: [PATCH 2/4] [AutoFDO] Add hierarchical discriminator for vectorizer
> >
> > Add hierarchical discriminator support for vectorized loop versioning.
> > Assigns copyid discriminators to distinguish vectorized and scalar versions.
> >
> > gcc/ChangeLog:
> >
> > * tree-vect-loop-manip.cc (vect_loop_versioning): Assign copyid
> > discriminators for vectorized and scalar loop versions.
> > * tree-vect-loop.cc (vect_transform_loop): Assign copyid
> > discriminators for vectorized and scalar loop versions.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.dg/hierarchical-discriminator-vect-version.c: New test.
>
> This also looks good to me, though it needs approval from vectorizer
> maintainers.
>
> Honza
> >
> > Signed-off-by: Kugan Vivekanandarajah <[email protected]>
> > ---
> > .../hierarchical-discriminator-vect-version.c | 44 +++++++++++++++++++
> > gcc/tree-vect-loop-manip.cc | 17 +++++++
> > gcc/tree-vect-loop.cc | 12 +++++
> > 3 files changed, 73 insertions(+)
> > create mode 100644 gcc/testsuite/gcc.dg/hierarchical-discriminator-vect-
> version.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/hierarchical-discriminator-vect-version.c
> b/gcc/testsuite/gcc.dg/hierarchical-discriminator-vect-version.c
> > new file mode 100644
> > index 00000000000..36493cc95aa
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/hierarchical-discriminator-vect-version.c
> > @@ -0,0 +1,44 @@
> > +/* Test hierarchical discriminators for vectorized loops.
> > + This test verifies that the vectorizer assigns unique discriminators to
> > + distinguish the main vectorized loop from epilog loops in AutoFDO
> profiles.
> > +
> > + Expected compilation:
> > + - Main loop vectorized with VF=4
> > + - Epilog loop for remainder
> > +
> > + { dg-do compile }
> > + { dg-options "-O3 -g -ftree-vectorize " }
> > + { dg-require-effective-target vect_int } */
> > +
> > +void
> > +test_vectorize (int *restrict a, int *restrict b, int *restrict c, int n)
> > +{
> > + int i;
> > + for (i = 0; i < n; i++)
> > + {
> > + a[i] = b[i] + c[i];
> > + }
> > +}
> > +
> > +/* Hierarchical discriminator format:
> [Base:8][Multiplicity:7][CopyID:11][Unused:6]
> > +
> > + Discriminators encode:
> > + - Base (bits 0-7): Front-end discriminator for same-line statements
> > + - Multiplicity (bits 8-14): Unroll/vectorization factor
> > + - CopyID (bits 15-25): Loop variant identifier
> > +
> > + Expected loop variants:
> > +
> > + 1. Main vectorized loop:
> > + - CopyID=1 (DISCRIMINATOR_LOOP_VERSION_VECTORIZED)
> > + - Multiplicity=4 (vectorization factor)
> > + - Calculation: base=0 | (4 << 8) | (1 << 15) = 33792
> > +
> > + 2. Eplog loop:
> > + - CopyID=7 (DISCRIMINATOR_LOOP_EPILOG)
> > + - Multiplicity=0
> > + - Calculation: base=0 | (2 << 8) | (7 << 15) = 229376. */
> > +
> > +
> > +/* { dg-final { scan-assembler "\\.loc 1 19 10 is_stmt 0 discriminator
> 33792" } } */
> > +/* { dg-final { scan-assembler "\\.loc 1 19 10 is_stmt 0 discriminator
> 229376" } } */
> > diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc
> > index 20141dbc2e5..893838f6d7e 100644
> > --- a/gcc/tree-vect-loop-manip.cc
> > +++ b/gcc/tree-vect-loop-manip.cc
> > @@ -53,6 +53,7 @@ along with GCC; see the file COPYING3. If not see
> > #include "langhooks.h"
> > #include "tree-vector-builder.h"
> > #include "optabs-tree.h"
> > +#include "hierarchical_discriminator.h"
> >
> >
> /******************************************************************
> *******
> > Simple Loop Peeling Utilities
> > @@ -3357,6 +3358,10 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree
> niters, tree nitersm1,
> > gcc_assert (prolog);
> > prolog->force_vectorize = false;
> >
> > + /* Assign hierarchical discriminators to distinguish prolog loop.
> > + Prolog loops are for alignment and are not vectorized. */
> > + assign_discriminators_to_loop (prolog, 0,
> DISCRIMINATOR_LOOP_PROLOG);
> > +
> > first_loop = prolog;
> > reset_original_copy_tables ();
> >
> > @@ -3462,6 +3467,12 @@ vect_do_peeling (loop_vec_info loop_vinfo, tree
> niters, tree nitersm1,
> > epilog->force_vectorize = false;
> > bb_before_epilog = loop_preheader_edge (epilog)->src;
> >
> > + /* Assign hierarchical discriminators to distinguish epilog loop.
> > + Only assign if it's a scalar epilog. If it will be vectorized
> > + (vect_epilogues), discriminators will be assigned. */
> > + if (!vect_epilogues)
> > + assign_discriminators_to_loop (epilog, 0,
> DISCRIMINATOR_LOOP_EPILOG);
> > +
> > /* Scalar version loop may be preferred. In this case, add guard
> > and skip to epilog. Note this only happens when the number of
> > iterations of loop is unknown at compile time, otherwise this
> > @@ -4359,6 +4370,12 @@ vect_loop_versioning (loop_vec_info
> loop_vinfo,
> > gcc_assert (nloop);
> > nloop = get_loop_copy (loop);
> >
> > + /* Assign hierarchical discriminators to distinguish loop versions.
> > + Only assign to the scalar version here; the vectorized version will
> > + get discriminators later during transformation/peeling. */
> > + assign_discriminators_to_loop (nloop, 0,
> > +
I think the patch is OK, but did you mean to add the discriminator only when
we don't re-use the loop version created by if-convert?
i.e. should this be moved outside of the if block?
Thanks,
Tamar
> DISCRIMINATOR_LOOP_VERSION_SCALAR);
> > +
> > /* For cycle vectorization with SLP we rely on the PHI arguments
> > appearing in the same order as the SLP node operands which for the
> > loop PHI nodes means the preheader edge dest index needs to remain
> > diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
> > index 9320bf8e878..4723370d838 100644
> > --- a/gcc/tree-vect-loop.cc
> > +++ b/gcc/tree-vect-loop.cc
> > @@ -59,6 +59,7 @@ along with GCC; see the file COPYING3. If not see
> > #include "case-cfn-macros.h"
> > #include "langhooks.h"
> > #include "opts.h"
> > +#include "hierarchical_discriminator.h"
> >
> > /* Loop Vectorization Pass.
> >
> > @@ -11101,6 +11102,17 @@ vect_transform_loop (loop_vec_info
> loop_vinfo, gimple *loop_vectorized_call)
> > &step_vector, &niters_vector_mult_vf, th,
> > check_profitability, niters_no_overflow,
> > &advance);
> > +
> > + /* Assign hierarchical discriminators to the vectorized loop. */
> > + poly_uint64 vf_val = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
> > + unsigned int vf_int = constant_lower_bound (vf_val);
> > + if (vf_int > DISCR_MULTIPLICITY_MAX)
> > + vf_int = DISCR_MULTIPLICITY_MAX;
> > + unsigned int copyid = LOOP_VINFO_EPILOGUE_P (loop_vinfo)
> > + ? DISCRIMINATOR_LOOP_EPILOG_VECTORIZED
> > + : DISCRIMINATOR_LOOP_VERSION_VECTORIZED;
> > + assign_discriminators_to_loop (loop, vf_int, copyid);
> > +
> > if (LOOP_VINFO_SCALAR_LOOP (loop_vinfo)
> > && LOOP_VINFO_SCALAR_LOOP_SCALING (loop_vinfo).initialized_p ())
> > {
> > --
> > 2.34.1
> >
