[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #13 from CVS Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:2dc5d6b1e7ec8822f5bd78761962ca2c85d4a2b4 commit r13-4607-g2dc5d6b1e7ec8822f5bd78761962ca2c85d4a2b4 Author: Richard Biener Date: Mon Dec 12 08:13:33 2022 +0100 tree-optimization/89317 - another pattern for >x != p + 4 As seen in the original testcase for PR89317 we are missing comparison simplification patterns for >x != p + 4. Fixed by making an existing one apply. To make the pattern apply during CCP we need to simplify ccp_fold to not use GENERIC folding of conditions but also use GIMPLE folding. PR tree-optimization/89317 * tree-ssa-ccp.cc (ccp_fold): Handle GIMPLE_COND via gimple_fold_stmt_to_constant_1. * match.pd ( != + c): Apply to pointer_plus with non-ADDR_EXPR base as well. * gcc.dg/tree-ssa/pr89317.c: Amend.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #12 from rguenther at suse dot de --- On Sun, 11 Dec 2022, glisse at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 > > --- Comment #11 from Marc Glisse --- > (In reply to Richard Biener from comment #10) > > Should be fixed in GCC 13. > > If I compile the original testcase with -O3, I get for test2: > > _1 = this_6(D) + 16; > _2 = _6(D)->data1; > if (_1 != _2) > > so we should probably also handle comparisons and not just subtractions. For > this particular testcase, the relevant optimizations still happen and RTL > cleans up the comparison, so it is ok, but the pattern appears in other PRs > like PR 106677. I'll see to add those.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #11 from Marc Glisse --- (In reply to Richard Biener from comment #10) > Should be fixed in GCC 13. If I compile the original testcase with -O3, I get for test2: _1 = this_6(D) + 16; _2 = _6(D)->data1; if (_1 != _2) so we should probably also handle comparisons and not just subtractions. For this particular testcase, the relevant optimizations still happen and RTL cleans up the comparison, so it is ok, but the pattern appears in other PRs like PR 106677.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 Richard Biener changed: What|Removed |Added Status|ASSIGNED|RESOLVED Resolution|--- |FIXED Target Milestone|--- |13.0 --- Comment #10 from Richard Biener --- Should be fixed in GCC 13.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #9 from CVS Commits --- The master branch has been updated by Richard Biener : https://gcc.gnu.org/g:d13b86f932ff7b9d8f41483fd869c637b67d4dec commit r13-4600-gd13b86f932ff7b9d8f41483fd869c637b67d4dec Author: Richard Biener Date: Sun Dec 11 14:07:34 2022 +0100 tree-optimization/89317 - missed folding of (p + 4) - >d The PR notices we fail to simplify a_4 = _3(D)->data; b_5 = x_3(D) + 16; _1 = b_5 - a_4; together with the enabler handling ADDR_EXPR leafs in separate stmts in match.pd the suggested patterns work. PR tree-optimization/89317 * match.pd ((p + b) - >c -> b - offsetof(c)): New patterns. * gcc.dg/tree-ssa/pr89317.c: New testcase.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 Richard Biener changed: What|Removed |Added Assignee|pinskia at gcc dot gnu.org |rguenth at gcc dot gnu.org --- Comment #8 from Richard Biener --- I have a patch in testing for the missed optimization on GIMPLE.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317
--- Comment #7 from Andrew Pinski ---
(In reply to Andrew Pinski from comment #6)
> Better one:
> struct b { int data[16]; };
>
> int foo (struct b *x)
> {
> int *a = x->data;
> int *b = ((int*)x) + 4;
> return b - a;
> }
Note we can handle address of a variable case already; that is:
```
struct b { int data[16]; };
int foo (struct b x)
{
int *a = x.data;
int *b = ((int*)) + 4;
return b - a;
}
```
is optimized during ccp1.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317
--- Comment #6 from Andrew Pinski ---
(In reply to Andrew Pinski from comment #5)
> Reduced testcase for that issue:
> struct { int data[16]; } *x;
> int foo (int n1)
> {
> int *a = x->data;
> int *b = ((int*)x) + 4;
> return b - a;
> }
Better one:
struct b { int data[16]; };
int foo (struct b *x)
{
int *a = x->data;
int *b = ((int*)x) + 4;
return b - a;
}
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317
Andrew Pinski changed:
What|Removed |Added
Keywords||TREE
--- Comment #5 from Andrew Pinski ---
Note on the trunk, these two functions now produce the same code. BUT we are
still missing a gimple level optimization.
The RTL level can figure out:
_1 = this_6(D) + 16;
_2 = _6(D)->data1;
_19 = _1 - _2;
__n_20 = _19 /[ex] 2;
if (_19 > 0)
goto ; [89.00%]
else
goto ; [11.00%]
That _19 is 16 and the condition is always true.
Reduced testcase for that issue:
struct { int data[16]; } *x;
int foo (int n1)
{
int *a = x->data;
int *b = ((int*)x) + 4;
return b - a;
}
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 Andrew Pinski changed: What|Removed |Added Severity|normal |enhancement
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317
Andrew Pinski changed:
What|Removed |Added
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
--- Comment #4 from Andrew Pinski ---
Mine.
After my patch for PR 102216, we get:
_1 = _6(D)->data1;
_2 = this_6(D) + 16;
_19 = _2 - _1;
I suspect if we add:
(simplify
(pointer_diff (pointer_plus @0 @1) ADDR_EXPR@2)
(with { poly_int64 diff; }
(if (ptr_difference_const (@0, @2, ))
(plus { build_int_cst_type (type, diff); } (convert @1)
(simplify
(pointer_diff ADDR_EXPR@0 (pointer_plus @1 @2))
(with { poly_int64 diff; }
(if (ptr_difference_const (@0, @1, ))
(minus { build_int_cst_type (type, diff); } (convert @2)
This will work.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 Richard Biener changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2019-02-18 CC||rguenth at gcc dot gnu.org Ever confirmed|0 |1
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #3 from Marc Glisse --- If we avoid changing the type from int16_t to uint16_t, we get 2 calls to memmove. They don't get expanded, probably because we fail to simplify the size to 16 in gimple: _1 = [(void *)this_5(D) + 16B]; _2 = _5(D)->data1; _10 = _1 - _2;
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #2 from Daniel Fruzynski --- Yes, I mean inefficient.
[Bug tree-optimization/89317] Ineffective code from std::copy
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89317 --- Comment #1 from Jonathan Wakely --- You mean inefficient, right? The effects are correct as far as I can see.
