On Thu, 7 May 2020, Richard Sandiford wrote:
> Richard Biener <rguent...@suse.de> writes:
> > This implements patterns combining vector element insertion of
> > vector element extraction to a VEC_PERM_EXPR of both vectors
> > when supported. Plus it adds the more generic identity transform
> > of inserting a piece of itself at the same position.
> >
> > Richard - is there anything I can do to make this SVE aware?
> > I'd need to construct an identity permute and "insert" into
> > that permute that element from the other (or same) vector.
> > I suppose for most element positions that won't work but
> > at least inserting at [0] should? I'm mostly struggling
> > on how to use vec_perm_builder here when nelts is not constant,
> > since it's derived from vec<> can I simply start with
> > a single pattern with 1 stride and then insert by using []?
>
> I guess for SVE we still want to know that the range is safe
> for all VL, so after dropping the is_constant check, we'd
> want something like:
>
> {
> poly_uint64 nelts = TYPE_VECTOR_SUBPARTS (type);
> unsigned int min_nelts = constant_lower_bound (nelts);
> }
> (if (...
> && at + elemsz <= min_nelts)
>
> In theory (hah) it should then just be a case of changing the
> vec_perm_builder constructor to:
>
> vec_perm_builder sel (nelts, min_nelts, 3);
>
> and then iterating over min_nelts * 3 instead of nelts here:
>
> > + for (unsigned i = 0; i < nelts; ++i)
> > + sel.quick_push (i / elemsz == at
> > + ? nelts + elem * elemsz + i % elemsz : i);
>
> So as far as the encoding goes, the first min_nelts elements are arbitrary
> values, and the following two min_nelts sequences form individual linear
> series.
OK - not sure why we need exactly three nelts per pattern here.
It also looks like all the constant_multiple_p () checks constrain
things quite a bit. Oh, and does a BIT_FIELD_REF with poly-int position
extract multiple elements in the end?! For the case we are extracting
a sub-vector and thus elemsz != 1 we constrain it so that this
sub-vector is not of variable size (err, not "independently" so,
whatever that means..)?
My brain hurts... how do you write a GIMPLE testcase for aarch64
SVE covering such cases?
> This ought to be work for both SVE and non-SVE, although obviously
> there's a bit of wasted work for non-SVE.
>
> (And thanks for asking :-))
So like this, it seems to still work on the x86 testcases?
Thanks,
Richard.
This implements patterns combining vector element insertion of
vector element extraction to a VEC_PERM_EXPR of both vectors
when supported. Plus it adds the more generic identity transform
of inserting a piece of itself at the same position.
2020-05-07 Richard Biener <rguent...@suse.de>
PR tree-optimization/94864
PR tree-optimization/94865
* match.pd ((bit_insert @0 (BIT_FIELD_REF @0 ...) ...) -> @0):
New simplification.
((bit_insert @0 (BIT_FIELD_REF @1 ...) ...) -> (vec_perm @0 @1 ..):
Likewise.
* gcc.dg/tree-ssa/forwprop-39.c: New testcase.
* gcc.dg/tree-ssa/forwprop-40.c: Likewise.
---
gcc/match.pd | 41 +++++++++++++++++++++++++++++
gcc/testsuite/gcc.dg/tree-ssa/forwprop-39.c | 30 +++++++++++++++++++++
gcc/testsuite/gcc.dg/tree-ssa/forwprop-40.c | 18 +++++++++++++
3 files changed, 89 insertions(+)
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/forwprop-39.c
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/forwprop-40.c
diff --git a/gcc/match.pd b/gcc/match.pd
index 9259dd4ddaa..28d3bc459a7 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -5838,6 +5838,47 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
wi::to_wide (@ipos) + isize))
(BIT_FIELD_REF @0 @rsize @rpos)))))
+/* Optimize a bit-insertion of a bit-extraction from the same object,
+ position and size. */
+(simplify
+ (bit_insert @0 (BIT_FIELD_REF@1 @0 @size @pos) @pos)
+ @0)
+
+/* Optimize an element insertion into a vector that is extracted from
+ another vector to a permutation of both vectors. */
+(simplify
+ (bit_insert @0 (BIT_FIELD_REF@1 @2 @size @rpos) @ipos)
+ (if (VECTOR_TYPE_P (type)
+ && types_match (type, TREE_TYPE (@2))
+ && single_use (@1))
+ (with
+ {
+ poly_uint64 nelts = TYPE_VECTOR_SUBPARTS (type);
+ unsigned int min_nelts = constant_lower_bound (nelts);
+ unsigned int elem_idx, at, n_elem;
+ unsigned HOST_WIDE_INT elem_sz
+ = tree_to_uhwi (TYPE_SIZE (TREE_TYPE (type)));
+ }
+ (if (constant_multiple_p (tree_to_poly_uint64 (@rpos),
+ tree_to_poly_uint64 (@size), &elem_idx)
+ && constant_multiple_p (tree_to_poly_uint64 (@ipos),
+ tree_to_poly_uint64 (@size), &at)
+ && constant_multiple_p (tree_to_poly_uint64 (@size), elem_sz, &n_elem)
+ && at + n_elem <= min_nelts)
+ (with
+ {
+ vec_perm_builder sel (nelts, min_nelts, 3);
+ for (unsigned i = 0; i < 3 * min_nelts; ++i)
+ sel.quick_push (i / n_elem == at
+ ? nelts + elem_idx * n_elem + i % n_elem : i);
+ vec_perm_indices indices (sel, @0 == @2 ? 1 : 2, nelts);
+ }
+ (if (can_vec_perm_const_p (TYPE_MODE (type), indices))
+ (vec_perm @0 @2 { vec_perm_indices_to_tree
+ (build_vector_type
+ (build_nonstandard_integer_type
+ (elem_sz, 1), nelts), indices); })))))))
+
(if (canonicalize_math_after_vectorization_p ())
(for fmas (FMA)
(simplify
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/forwprop-39.c
b/gcc/testsuite/gcc.dg/tree-ssa/forwprop-39.c
new file mode 100644
index 00000000000..d1aefa9ee60
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/forwprop-39.c
@@ -0,0 +1,30 @@
+/* { dg-do compile } */
+/* { dg-options "-O -fdump-tree-forwprop1 -fdump-tree-cddce1 -Wno-psabi -w" }
*/
+
+typedef double v2df __attribute__((vector_size(16)));
+
+/* PR94865 */
+v2df move_sd1(v2df a, v2df b)
+{
+ v2df result = a;
+ result[1] = b[1];
+ return result;
+}
+
+/* PR94864 */
+v2df move_sd2(v2df a, v2df b)
+{
+ v2df result = a;
+ result[0] = b[1];
+ return result;
+}
+
+v2df move_nnop(v2df a)
+{
+ v2df result = a;
+ result[1] = a[1];
+ return result;
+}
+
+/* { dg-final { scan-tree-dump-times "VEC_PERM_EXPR" 2 "forwprop1" } } */
+/* { dg-final { scan-tree-dump-times "BIT_FIELD_REF" 0 "cddce1" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/forwprop-40.c
b/gcc/testsuite/gcc.dg/tree-ssa/forwprop-40.c
new file mode 100644
index 00000000000..94329437f0d
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/forwprop-40.c
@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+/* { dg-options "-O -fgimple -fdump-tree-forwprop1 -w -Wno-psabi" } */
+
+typedef int v4si __attribute__((vector_size(16)));
+typedef int v2si __attribute__((vector_size(8)));
+
+v4si __GIMPLE(ssa) bar (v4si a)
+{
+ v2si el;
+ v4si res;
+
+ __BB(2):
+ el_2 = __BIT_FIELD_REF <v2si> (a_1(D), 64u, 64u);
+ res_3 = __BIT_INSERT (a_1(D), el_2, 0u);
+ return res_3;
+}
+
+/* { dg-final { scan-tree-dump "VEC_PERM" "forwprop1" } } */
--
2.13.7
