On Thu, 27 Jan 2022, Jiufu Guo wrote:
> Hi Richard,
>
> Richard Biener <rguent...@suse.de> writes:
>
> > On Tue, 25 Jan 2022, Richard Biener wrote:
> >
> >> On Tue, 25 Jan 2022, Jiufu Guo wrote:
> >>
> ...
> >> > >
> >> > > The point is that we may not change the iteration number at which
> >> > > overflow occurs since that alters the result of the < compare.
> >> > > Only if we know there is no overflow with the present expression
> >> > > during the loop evaluation we can do the transform and then only
> >> > > if we do not introduce overflow.
> >> > Exactly, this is also what I mean :)
> >> >
> >> > >
> >> > > We are basically doing the transform that fold_comparison
> >> > > in fold-const.cc does:
> >> > >
> >> > > /* Transform comparisons of the form X +- C1 CMP Y +- C2 to
> >> > > X CMP Y +- C2 +- C1 for signed X, Y. This is valid if
> >> > > the resulting offset is smaller in absolute value than the
> >> > > original one and has the same sign. */
> >> > > if (ANY_INTEGRAL_TYPE_P (TREE_TYPE (arg0))
> >> > > && TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (arg0))
> >> > > ...
> >> > >
> >> > This transform seems not the scenario which we are care about in
> >> > number_of_iterations_cond.
> >> > For example, 'X + 1 < Y + 4' ==> 'X < Y + 3' would be correct if
> >> > no overflow happen.
> >> > But for loop, the niter for '{X, 1} < {Y, 4}' would be totally
> >> > different with niter for '{X, 0} < {Y, 3}'.
> >> > for (iv0 = X, iv1 = Y; iv0 < iv1; iv0 += 1, iv1 += 4)
> >> > in this loop, iv1 walks to overflow faster, step is 4.
> >> > vs.
> >> > for (iv0 = X, iv1 = Y; iv0 < iv1; iv1 += 3) (iv1 overflow slow)
> >> > in this loop, iv1 overflows slower, step is 3.
> >>
> >> Huh? But we are _exactly_ doing this, analyzing {X, + 1} < {Y, + 4}
> >> as X < {Y, + 3} (well, OK, we're only trying {X, -3} which now
> >> fails - we should try the other way around as well).
> >>
> >> > Actually, the transformation 'X + 1 < Y + 4' ==> 'X < Y + 3',
> >> > may not always correct. e.g. for below code, X=6, and Y=2147483645
> >> > it may output "b0 + 1 < b1 + 4 is true".
> >>
> >> But Y + 4 overflows with 2147483645 so X + 1 < Y + 4 invokes UB
> >> and we can ignore this situation.
> >>
> ...
> >> > >> > Or it may also ok if we can compute an assumption, under which
> >> > >> > all three ivs are not overflowed/wrapped.
> >> > >> >
> >> > >> >> only guaranteed if the absolute value of C1 - C2 is smaller than
> >> > >> >> that of C1 and if it has the same sign.
> >> > >> > I'm thinking this in another way:
> >> > >> > When trying to do this transform in number_of_iterations_cond,
> >> > >> > GT/GE is inverted to LT/LE, then the compare code would be:
> >> > >> > LT/LE or NE.
> >> > >> >
> >> > >> > For LT/LE, like {X, C1} < {Y, C2}, we can look it as iv0 is
> >> > >> > chasing iv1. We would able to assume X < Y (may_be_zero would
> >> > >> > be set later via number_of_iterations_lt/le).
> >> > >> > 1. If C1 < C2, iv0 can never catch up iv1. For examples:
> >> > >> > {X, 1} < {Y, 2}; {X, -2} < {Y, -1}; {X, -2} < {Y, 1}.
> >> > >> > And there must be at least one overflow/wrap in iv0,iv1, or iv.
> >> > >> > This indicates, if the sign of (C1 - C1) is negative, then the
> >> > >> > transform would be incorrect.
> >> > >> > 2. If C1 > C2, we still need to make sure all the ivs (iv0,
> >> > >> > iv1 and combined iv) are not wrapped.
> >> > >> > For C2 > 0, {Y,C2} should not cross MAX before {X, C1} catch up.
> >> > >> > the assumption may like : (MAX-Y)/C2 > (Y-X)/(C1-C1)
> >> > >> There is still some cases: iv0 step is too large, then iv0 wraps
> >> > >> first, e.g. {MAX-5, 10} < {MAX-3, 1}. For this, the assumption
> >> > >> would need to and with (MAX-X)/C1 > (Y-X)/(C1-C1).
> >> > >>
> >> > >> > For C1 < 0, {X,C1} should not down cross MIN
> >> > >> > the assumption may like : (X-MIN)/-C1 > (Y-X)/(C1-C1)
> >> > >> Also add the assumption: (Y-MIN)/-C2 > (Y-X)/(C1-C1)
> >> > >>
> >> > >> > For C1 > 0 and C2 < 0, iv0 and iv1 are walking to each other,
> >> > >> > it would be almost safe.
> >> > >> For this case, we may still add the assumption to avoid wraping
> >> > >> at the first iteration.
> >> > >>
> >> > >> BR,
> >> > >> Jiufu
> >> > >>
> >> > >> >
> >> > >> > For NE, it seems more interesting. The transformation depends
> >> > >> > on 3 things: 1. the relation between X and Y; 2 the sign
> >> > >> > of (C1-C2); 3. if iv0 and iv1 can be equal finally.
> >> > >> > The 3rd one may be more special.
> >> > >> > The good news is, number_of_iterations_ne seems able to handle NE.
> >> > >> >
> >> > >> >>
> >> > >> >> With the same reasoning we then know the new IV0 doesn't overflow.
> >> > >> >>
> >> > >> >> So something like the following. IIRC I've proposed sth similar
> >> > >> >> a while back. I'm going to give it some testing, collecting
> >> > >> >> testcases from related PRs.
> >> > >> >>
> >> > >> >> diff --git a/gcc/tree-ssa-loop-niter.cc
> >> > >> >> b/gcc/tree-ssa-loop-niter.cc
> >> > >> >> index b767056aeb0..74fa4f66ee2 100644
> >> > ...
> >> > >> >> + if (TREE_CODE (step) != INTEGER_CST
> >> > >> >> + || !iv0->no_overflow || !iv1->no_overflow)
> >> > >> >> + {
> >> > >> > I was also trying to leverage no_overflow of iv0 and iv1. While it
> >> > >> > seems
> >> > >> > the computation logic of no_overflow is related to the type of IV.
> >> > >> > If the
> >> > >> > type of IV is signed, the C semantics may be used, overflow in
> >> > >> > signed
> >> > >> > IV are treated UB, and then no_overflow would be true.
> >> > >> >
> >> > >> > For unsigned IV, no_overflow would be false, even for the cases
> >> > >> > which
> >> > >> > looks like:
> >> > >> > "{10, 2} < {20, 1}", which would be ok to compute niter.
> >> > >
> >> > > IIRC no_overflow is determined by SCEV which might also use niter
> >> > > analysis. For the case of {10, +2} < {20, +1} there is no need to
> >> > > compute it as {10, +1} < 20 and we hopefully deal with this in
> >> > > other code paths (in fact with base and step all constant we
> >> > > can simply solve the linear equation for 'n' - maybe that's a
> >> > > capability we should add to number_of_iterations_cond).
> >> >
> >> > Thanks for point this out.
> >> > Yes, for const base(s) and step(s), we have other code path
> >> > to deal with (e.g. loop_niter_by_eval).
> >> >
> >> > For {10, +2}, what I really mean is about the no_overflow iv(s)
> >> > on unsigned. Sorry the misleading words.
> >> > For no_overflow, it is set at some places, including
> >> > number_of_iterations_xxx. :), Before number_of_iterations_xxx,
> >> > no_overflow could be calculated in simple_iv_with_niters:
> >> > ```c
> >> > iv->no_overflow = !folded_casts && nowrap_type_p (type);
> >> > ```
> >> > nowrap_type_p checks if overflow is UB on type through macro
> >> > TYPE_OVERFLOW_UNDEFINED. For signed, it is UB; for unsigned
> >> > it is different.
> >>
> >> Yes.
> >>
> >> > For example as below code, no_overflow is set as false for iv0
> >> > and iv1, and then niter was not computed quickly.
> >> > ```c
> >> > unsigned __attribute__ ((noinline))
> >> > foo (unsigned b0, unsigned b1)
> >> > {
> >> > unsigned n = 0;
> >> > for (; b0 < b1; b0 += 2, b1 += 1)
> >> > n++;
> >> > return n;
> >> > }
> >>
> >> There's no difference to before my patch of course. There are
> >> some code paths in number_of_iterations_lt that use assumptions
> >> to prove the combined IV does not wrap, just for this case
> >> we give up too early. I'm currently looking at rectifying this
> >> with small incremental changes.
> >
> > Like the one below which should handle the PR81196 case when
> > integer IVs are used. I suspect we can do something similar
> > for IVs where we do not know the original overflow status
> > (we have to register assumptions for both original IVs _and_
> > the new adjusted one).
> Agree with you, that without accurate overflow status, it may
> not safe to combine the two iv steps directly.
>
> For those cases where integer IVs are used, one kind of case,
> as shown in your patch, is IV0 increasing and IV1 descreasing.
> looks like "{b0, +C1} < {b1, -C2}". For this kind of cases,
> because "abs(step) > abs(iv0->step)", so more assumptions are
> needed. Maybe simple assumption is ok for most cases:
> `(b0 > MIN + C2) && (b1 < MAX - C1 - C2).`
> 'b0 > MIN + C2' would help to make sure IV1 does not overflow,
> and 'b1 < MAX - C1 - C2' guard that IV0 and combined IV do not
> walk cross MAX.
>
> In a previous patch, I tried to add more assumptions. While
> the assumption contains a few expensive expressions(e.g. DIV),
> because it tries to handle more cases accurately even for
> unsigned integer whose 'no_overflow' information is not known.
Yes, I've seen this. Btw, my attempt failed and will regress
the miscompile testcases again - I have not yet analyzed why.
> >
> > And as noted we can try rewriting to the other IV.
> >
> > I do wonder how important these are and what improvements we need
> > to include in backports (I think we do want to fix the original issue
> > on branches).
> I also have the feeling that combining two IVs are rare in cases.
> It may not benefit benchmarks too much. It seems this kind of
> case may be not be used widely and is not on a hot path in spec.
Indeed. So my plan is to have the current state settle a bit
(you've seen I've moved over to the other niter correctness issue)
to see whether there's any further bad fallout also to assess
safeness of backporting. For the next stage1 we can see how to
improve these cases, I'd also like to add some statistics so we
can see how many loops we have where we are not able to compute
a symbolic number of iterations (and why), possibly at loop_init
pass time.
Richard.
> Thanks for your thoughts and sugguestions!
>
> BR,
> Jiufu
> >
> > Richard.
> >
> > From f46855709dd45603d18f2dcd8403f5b060c164f0 Mon Sep 17 00:00:00 2001
> > From: Richard Biener <rguent...@suse.de>
> > Date: Tue, 25 Jan 2022 13:11:57 +0100
> > Subject: [PATCH] tree-optimization/104214 - improve IV analysis with integer
> > IV compares
> > To: gcc-patches@gcc.gnu.org
> >
> > For rewriting BASE0 + STEP0 cmp BASE1 + STEP1 as
> > BASE0 + STEP0 - STEP1 cmp BASE1 for signed integers we can use
> > niter assumptions to ensure that BASE0 + STEP0 - STEP1 does not
> > overflow instead of giving up when we cannot prove this
> > statically.
> >
> > We can use the existing assert_no_overflow_lt for this and make it
> > efficient for LE_EXPR also by rewriting LE_EXPR IV compares to LT_EXPR
> > earlier.
> >
> > 2022-01-25 Richard Biener <rguent...@suse.de>
> >
> > PR tree-optimization/104214
> > * tree-ssa-loop-niter.cc (number_of_iterations_le): Refactor
> > into ...
> > (number_of_iterations_le_to_lt): ... this, just doing
> > the iv->base rewriting and assumption registering.
> > (number_of_iterations_cond): Rewrite LE_EXPR into LT_EXPR
> > earlier. When rewriting BASE0 + STEP0 cmp BASE1 + STEP1
> > as BASE0 + STEP0 - STEP1 cmp BASE1 would fail for LT_EXPR
> > because of possible overflow register assumptions instead.
> >
> > * gcc.dg/vect/pr81196-3.c: New testcase variant.
> > ---
> > gcc/testsuite/gcc.dg/vect/pr81196-3.c | 12 +++++
> > gcc/tree-ssa-loop-niter.cc | 67 +++++++++++++++------------
> > 2 files changed, 49 insertions(+), 30 deletions(-)
> > create mode 100644 gcc/testsuite/gcc.dg/vect/pr81196-3.c
> >
> > diff --git a/gcc/testsuite/gcc.dg/vect/pr81196-3.c
> > b/gcc/testsuite/gcc.dg/vect/pr81196-3.c
> > new file mode 100644
> > index 00000000000..bcdd815dc5d
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/vect/pr81196-3.c
> > @@ -0,0 +1,12 @@
> > +/* { dg-do compile } */
> > +/* { dg-require-effective-target vect_int } */
> > +
> > +void b (int *p, int j, int k)
> > +{
> > + p = (int *)__builtin_assume_aligned(p, __BIGGEST_ALIGNMENT__);
> > + int i = 0;
> > + for(; j < k; ++j, --k)
> > + p[i++] = 1;
> > +}
> > +
> > +/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
> > diff --git a/gcc/tree-ssa-loop-niter.cc b/gcc/tree-ssa-loop-niter.cc
> > index d33095b8e03..b5f3d4b4a8d 100644
> > --- a/gcc/tree-ssa-loop-niter.cc
> > +++ b/gcc/tree-ssa-loop-niter.cc
> > @@ -1721,17 +1721,17 @@ number_of_iterations_lt (class loop *loop, tree
> > type, affine_iv *iv0,
> > return true;
> > }
> >
> > -/* Determines number of iterations of loop whose ending condition
> > - is IV0 <= IV1. TYPE is the type of the iv. The number of
> > - iterations is stored to NITER. EXIT_MUST_BE_TAKEN is true if
> > - we know that this condition must eventually become false (we derived
> > this
> > +/* Rewrite the IV0 <= IV1 condition to IV0 < IV1 by adjusting one of
> > + the IVs bases. TYPE is the type of the iv. Assumptions are
> > + recorded to NITER. EXIT_MUST_BE_TAKEN is true if we know that this
> > + condition must eventually become false (we derived this
> > earlier, and possibly set NITER->assumptions to make sure this
> > - is the case). BNDS bounds the difference IV1->base - IV0->base. */
> > + is the case). */
> >
> > static bool
> > -number_of_iterations_le (class loop *loop, tree type, affine_iv *iv0,
> > - affine_iv *iv1, class tree_niter_desc *niter,
> > - bool exit_must_be_taken, bounds *bnds)
> > +number_of_iterations_le_to_lt (tree type, affine_iv *iv0, affine_iv *iv1,
> > + class tree_niter_desc *niter,
> > + bool exit_must_be_taken)
> > {
> > tree assumption;
> > tree type1 = type;
> > @@ -1777,10 +1777,7 @@ number_of_iterations_le (class loop *loop, tree
> > type, affine_iv *iv0,
> > iv0->base = fold_build2 (MINUS_EXPR, type1,
> > iv0->base, build_int_cst (type1, 1));
> >
> > - bounds_add (bnds, 1, type1);
> > -
> > - return number_of_iterations_lt (loop, type, iv0, iv1, niter,
> > exit_must_be_taken,
> > - bnds);
> > + return true;
> > }
> >
> > /* Dumps description of affine induction variable IV to FILE. */
> > @@ -1862,6 +1859,17 @@ number_of_iterations_cond (class loop *loop,
> > code = swap_tree_comparison (code);
> > }
> >
> > + /* If the loop exits immediately, there is nothing to do. */
> > + tree tem = fold_binary (code, boolean_type_node, iv0->base, iv1->base);
> > + if (tem && integer_zerop (tem))
> > + {
> > + if (!every_iteration)
> > + return false;
> > + niter->niter = build_int_cst (unsigned_type_for (type), 0);
> > + niter->max = 0;
> > + return true;
> > + }
> > +
> > if (POINTER_TYPE_P (type))
> > {
> > /* Comparison of pointers is undefined unless both iv0 and iv1 point
> > @@ -1884,6 +1892,15 @@ number_of_iterations_cond (class loop *loop,
> > exit_must_be_taken = true;
> > }
> >
> > + /* Turn LE_EXPR to LT_EXPR, registering required assumptions. */
> > + if (code == LE_EXPR)
> > + {
> > + if (!number_of_iterations_le_to_lt (type, iv0, iv1, niter,
> > + exit_must_be_taken))
> > + return false;
> > + code = LT_EXPR;
> > + }
> > +
> > /* We can handle cases which neither of the sides of the comparison is
> > invariant:
> >
> > @@ -1928,15 +1945,21 @@ number_of_iterations_cond (class loop *loop,
> > pointer compares, we also know the resulting IV does not
> > overflow. */
> > ;
> > - else if (code != NE_EXPR)
> > - return false;
> If two lines are removed, some cases in original PR may not work.
> And as you also mentioned above, we have to add more assumptions, then.
> I saw you are try to use number_of_iterations_le_to_lt and
> assert_no_overflow_lt. I still think it would be ok to use
> negative step on combined IV to indicate the IV combining
> is invalid.
>
> Thanks a gain for your comments!
>
> BR,
> Jiufu
>
> > else
> > + /* For LT_EXPR we register the assumptions necessary for
> > + the adjusted IV0 to not overflow. */
> > iv0->no_overflow = false;
> > }
> >
> > iv0->step = step;
> > iv1->step = build_int_cst (step_type, 0);
> > iv1->no_overflow = true;
> > + if (code == LT_EXPR && !iv0->no_overflow)
> > + {
> > + if (!assert_no_overflow_lt (type, iv0, iv1, niter, step))
> > + return false;
> > + /* We will now have iv0->no_overflow == true again. */
> > + }
> > }
> >
> > /* If the result of the comparison is a constant, the loop is weird.
> > More
> > @@ -1945,17 +1968,6 @@ number_of_iterations_cond (class loop *loop,
> > if (integer_zerop (iv0->step) && integer_zerop (iv1->step))
> > return false;
> >
> > - /* If the loop exits immediately, there is nothing to do. */
> > - tree tem = fold_binary (code, boolean_type_node, iv0->base, iv1->base);
> > - if (tem && integer_zerop (tem))
> > - {
> > - if (!every_iteration)
> > - return false;
> > - niter->niter = build_int_cst (unsigned_type_for (type), 0);
> > - niter->max = 0;
> > - return true;
> > - }
> > -
> > /* OK, now we know we have a senseful loop. Handle several cases,
> > depending
> > on what comparison operator is used. */
> > bound_difference (loop, iv1->base, iv0->base, &bnds);
> > @@ -1994,11 +2006,6 @@ number_of_iterations_cond (class loop *loop,
> > exit_must_be_taken, &bnds);
> > break;
> >
> > - case LE_EXPR:
> > - ret = number_of_iterations_le (loop, type, iv0, iv1, niter,
> > - exit_must_be_taken, &bnds);
> > - break;
> > -
> > default:
> > gcc_unreachable ();
> > }
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Ivo Totev; HRB 36809 (AG Nuernberg)
