[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 H.J. Lu changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED Target Milestone|--- |10.0 --- Comment #9 from H.J. Lu --- Fixed for GCC 10.
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 Bug 88828 depends on bug 54855, which changed state. Bug 54855 Summary: Unnecessary duplication when performing scalar operation on vector element https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54855 What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 --- Comment #8 from Richard Biener --- Author: rguenth Date: Wed May 15 09:59:37 2019 New Revision: 271204 URL: https://gcc.gnu.org/viewcvs?rev=271204=gcc=rev Log: 2019-05-15 Richard Biener PR tree-optimization/88828 * tree-ssa-forwprop.c (simplify_vector_constructor): Fix bogus check. Modified: trunk/gcc/ChangeLog trunk/gcc/tree-ssa-forwprop.c
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 --- Comment #7 from Andreas Schwab --- ../../gcc/tree-ssa-forwprop.c: In function 'bool simplify_vector_constructor(gimple_stmt_iterator*)': ../../gcc/tree-ssa-forwprop.c:2107:14: error: array subscript 2 is above array bounds of 'tree_node* [2]' [-Werror=array-bounds] 2107 |orig[j] = ref; |~~^ ../../gcc/tree-ssa-forwprop.c:2044:17: note: while referencing 'orig' 2044 | tree op, op2, orig[2], type, elem_type; | ^~~~ ../../gcc/tree-ssa-forwprop.c:2107:14: error: array subscript 2 is above array bounds of 'tree_node* [2]' [-Werror=array-bounds] 2107 |orig[j] = ref; |~~^ ../../gcc/tree-ssa-forwprop.c:2044:17: note: while referencing 'orig' 2044 | tree op, op2, orig[2], type, elem_type; | ^~~~
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 --- Comment #6 from Richard Biener --- Author: rguenth Date: Tue May 14 09:11:15 2019 New Revision: 271153 URL: https://gcc.gnu.org/viewcvs?rev=271153=gcc=rev Log: 2019-05-14 Richard Biener H.J. Lu PR tree-optimization/88828 * tree-ssa-forwprop.c (simplify_vector_constructor): Handle permuting in a single non-constant element not extracted from a vector. * gcc.target/i386/pr88828-1.c: New test. * gcc.target/i386/pr88828-1a.c: Likewise. * gcc.target/i386/pr88828-1b.c: Likewise. * gcc.target/i386/pr88828-1c.c: Likewise. * gcc.target/i386/pr88828-4a.c: Likewise. * gcc.target/i386/pr88828-4b.c: Likewise. * gcc.target/i386/pr88828-5a.c: Likewise. * gcc.target/i386/pr88828-5b.c: Likewise. * gcc.target/i386/pr88828-7.c: Likewise. * gcc.target/i386/pr88828-7a.c: Likewise. * gcc.target/i386/pr88828-7b.c: Likewise. * gcc.target/i386/pr88828-8.c: Likewise. * gcc.target/i386/pr88828-8a.c: Likewise. * gcc.target/i386/pr88828-8b.c: Likewise. * gcc.target/i386/pr88828-9.c: Likewise. * gcc.target/i386/pr88828-9a.c: Likewise. * gcc.target/i386/pr88828-9b.c: Likewise. Added: trunk/gcc/testsuite/gcc.target/i386/pr88828-1.c trunk/gcc/testsuite/gcc.target/i386/pr88828-1a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-1b.c trunk/gcc/testsuite/gcc.target/i386/pr88828-1c.c trunk/gcc/testsuite/gcc.target/i386/pr88828-4a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-4b.c trunk/gcc/testsuite/gcc.target/i386/pr88828-5a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-5b.c trunk/gcc/testsuite/gcc.target/i386/pr88828-7.c trunk/gcc/testsuite/gcc.target/i386/pr88828-7a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-7b.c trunk/gcc/testsuite/gcc.target/i386/pr88828-8.c trunk/gcc/testsuite/gcc.target/i386/pr88828-8a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-8b.c trunk/gcc/testsuite/gcc.target/i386/pr88828-9.c trunk/gcc/testsuite/gcc.target/i386/pr88828-9a.c trunk/gcc/testsuite/gcc.target/i386/pr88828-9b.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-ssa-forwprop.c
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 --- Comment #5 from Richard Biener --- Author: rguenth Date: Mon May 6 12:43:30 2019 New Revision: 270908 URL: https://gcc.gnu.org/viewcvs?rev=270908=gcc=rev Log: 2019-05-06 Richard Biener PR tree-optimization/88828 * tree-ssa-forwprop.c (get_bit_field_ref_def): Split out from... (simplify_vector_constructor): ...here. Handle constants in the constructor. * gcc.target/i386/pr88828-0.c: New testcase. Added: trunk/gcc/testsuite/gcc.target/i386/pr88828-0.c Modified: trunk/gcc/ChangeLog trunk/gcc/testsuite/ChangeLog trunk/gcc/tree-ssa-forwprop.c
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828 Marc Glisse changed: What|Removed |Added See Also||https://gcc.gnu.org/bugzill ||a/show_bug.cgi?id=54855 --- Comment #4 from Marc Glisse --- Comment #3 is similar to PR 54855.
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828
H.J. Lu changed:
What|Removed |Added
CC||crazylht at gmail dot com,
||xuepeng.guo at intel dot com
--- Comment #3 from H.J. Lu ---
Another testcase:
[hjl@gnu-cfl-1 pr88828]$ cat y.i
typedef double __v2df __attribute__ ((__vector_size__ (16)));
typedef double __m128d __attribute__ ((__vector_size__ (16), __may_alias__));
__m128d
_mm_add_sd (__m128d x, __m128d y)
{
__m128d z = __extension__ (__m128d)(__v2df)
{ (((__v2df) x)[0] + ((__v2df) y)[0]), ((__v2df) x)[1] };
return z;
}
[hjl@gnu-cfl-1 pr88828]$ gcc -S -O2 y.i
[hjl@gnu-cfl-1 pr88828]$ cat y.s
.file "y.i"
.text
.p2align 4,,15
.globl _mm_add_sd
.type _mm_add_sd, @function
_mm_add_sd:
.LFB0:
.cfi_startproc
movapd %xmm0, %xmm2
addsd %xmm1, %xmm2
movsd %xmm2, %xmm0
ret
.cfi_endproc
.LFE0:
.size _mm_add_sd, .-_mm_add_sd
.ident "GCC: (GNU) 8.2.1 20190109 (Red Hat 8.2.1-7)"
.section.note.GNU-stack,"",@progbits
[hjl@gnu-cfl-1 pr88828]$
I am expecting
addsd %xmm1, %xmm0
retq
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828
Richard Biener changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed||2019-01-14
CC||rguenth at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #2 from Richard Biener ---
I think there's related bugs. foo1 is optimized OK:
y_4 = BIT_INSERT_EXPR ;
return y_4;
while foo is expanded from
[local count: 1073741824]:
_1 = BIT_FIELD_REF ;
_2 = BIT_FIELD_REF ;
_3 = BIT_FIELD_REF ;
y_6 = {f_5(D), _1, _2, _3};
return y_6;
tree forwprop contains code pattern-matching on vector CONSTRUCTORs,
it could be extended to handle this case I think. IIRC it can detect
arbitrary two-vector permutes already, for the above we could go
through an intermediate
_1 = {f_5(D), f_5(D), ... };
y_6 = VEC_PERM <_1, x_7(D), { }>;
and recognize permutes that only replace a single vector element.
So I think we should optimize
__v4sf
foo (__v4sf x, float f)
{
__v4sf y = __extension__ (__v4sf)
{ f, x[2], x[1], x[3] };
return y;
}
as well, first permuting x and then inserting f (at any position).
[Bug tree-optimization/88828] Inefficient update of the first element of vector registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88828
Andrew Pinski changed:
What|Removed |Added
Keywords||missed-optimization
Target||x86_64
Component|target |tree-optimization
--- Comment #1 from Andrew Pinski ---
I think there are two issues here (maybe only one since I have not tested one
of them).
The first is not recognizing:
typedef float __v4sf __attribute__ ((__vector_size__ (16)));
__v4sf
foo (__v4sf x, float f)
{
__v4sf y = __extension__ (__v4sf)
{ f, x[1], x[2], x[3] };
return y;
}
is the same as:
__v4sf
foo1 (__v4sf x, float f)
{
__v4sf y = x;
y[0] = f;
return y;
}
This is a generic tree optimization issue.
The second is if foo1 is not optimized to what you want it to be. That would
be a target issue.
