Re: [PATCH] tree-optimization/108888 - call if-conversion
On Wed, 5 Apr 2023, Andre Vieira (lists) wrote:
> Hi,
>
> The original patch to fix this PR broke the if-conversion of calls into
> IFN_MASK_CALL. This patch restores that original behaviour and makes sure the
> tests added earlier specifically test inbranch SIMD clones.
OOps.
> Bootstrapped and regression tested on aarch64-none-linux-gnu and
> x86_64-pc-linux-gnu.
>
> Is this OK for trunk?
OK.
Thanks,
Richard.
> gcc/ChangeLog:
>
> PR tree-optimization/10
> * tree-if-conv.cc (predicate_statements): Fix gimple call check.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/vect/vect-simd-clone-16.c: Make simd clone inbranch only.
> * gcc.dg/vect/vect-simd-clone-17.c: Likewise.
> * gcc.dg/vect/vect-simd-clone-18.c: Likewise.
>
> On 23/02/2023 10:10, Richard Biener via Gcc-patches wrote:
> > The following makes sure to only predicate calls necessary.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
> >
> > PR tree-optimization/10
> > * tree-if-conv.cc (if_convertible_stmt_p): Set PLF_2 on
> > calls to predicate.
> > (predicate_statements): Only predicate calls with PLF_2.
> >
> > * g++.dg/torture/pr10.C: New testcase.
> > ---
> > gcc/testsuite/g++.dg/torture/pr10.C | 18 ++
> > gcc/tree-if-conv.cc | 17 ++---
> > 2 files changed, 28 insertions(+), 7 deletions(-)
> > create mode 100644 gcc/testsuite/g++.dg/torture/pr10.C
> >
> > diff --git a/gcc/testsuite/g++.dg/torture/pr10.C
> > b/gcc/testsuite/g++.dg/torture/pr10.C
> > new file mode 100644
> > index 000..29a22e21102
> > --- /dev/null
> > +++ b/gcc/testsuite/g++.dg/torture/pr10.C
> > @@ -0,0 +1,18 @@
> > +// { dg-do compile }
> > +
> > +int scaleValueSaturate_scalefactor, scaleValueSaturate___trans_tmp_2,
> > +scaleValuesSaturate_i;
> > +int scaleValueSaturate(int value) {
> > + int result = __builtin_clz(value);
> > + if (value)
> > +if (-result <= scaleValueSaturate_scalefactor)
> > + return 0;
> > + return scaleValueSaturate___trans_tmp_2;
> > +}
> > +short scaleValuesSaturate_dst;
> > +short *scaleValuesSaturate_src;
> > +void scaleValuesSaturate() {
> > + for (; scaleValuesSaturate_i; scaleValuesSaturate_i++)
> > +scaleValuesSaturate_dst =
> > +scaleValueSaturate(scaleValuesSaturate_src[scaleValuesSaturate_i]);
> > +}
> > diff --git a/gcc/tree-if-conv.cc b/gcc/tree-if-conv.cc
> > index a7a8406374d..0e384e36394 100644
> > --- a/gcc/tree-if-conv.cc
> > +++ b/gcc/tree-if-conv.cc
> > @@ -1099,6 +1099,7 @@ if_convertible_stmt_p (gimple *stmt,
> > vec refs)
> >n = n->simdclone->next_clone)
> > if (n->simdclone->inbranch)
> > {
> > + gimple_set_plf (stmt, GF_PLF_2, true);
> > need_to_predicate = true;
> > return true;
> > }
> > @@ -2541,7 +2542,8 @@ predicate_statements (loop_p loop)
> > release_defs (stmt);
> > continue;
> > }
> > - else if (gimple_plf (stmt, GF_PLF_2))
> > + else if (gimple_plf (stmt, GF_PLF_2)
> > + && is_gimple_assign (stmt))
> >{
> > tree lhs = gimple_assign_lhs (stmt);
> > tree mask;
> > @@ -2625,13 +2627,14 @@ predicate_statements (loop_p loop)
> > gimple_assign_set_rhs1 (stmt, ifc_temp_var (type, rhs, ));
> > update_stmt (stmt);
> > }
> > -
> > - /* Convert functions that have a SIMD clone to IFN_MASK_CALL. This
> > -will cause the vectorizer to match the "in branch" clone
> > variants,
> > -and serves to build the mask vector in a natural way. */
> > - gcall *call = dyn_cast (gsi_stmt (gsi));
> > - if (call && !gimple_call_internal_p (call))
> > + else if (gimple_plf (stmt, GF_PLF_2)
> > + && is_gimple_call (stmt))
> > {
> > + /* Convert functions that have a SIMD clone to IFN_MASK_CALL.
> > +This will cause the vectorizer to match the "in branch"
> > +clone variants, and serves to build the mask vector
> > +in a natural way. */
> > + gcall *call = dyn_cast (gsi_stmt (gsi));
> > tree orig_fn = gimple_call_fn (call);
> > int orig_nargs = gimple_call_num_args (call);
> > auto_vec args;
>
--
Richard Biener
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
Re: [PATCH] tree-optimization/108888 - call if-conversion
Hi,
The original patch to fix this PR broke the if-conversion of calls into
IFN_MASK_CALL. This patch restores that original behaviour and makes
sure the tests added earlier specifically test inbranch SIMD clones.
Bootstrapped and regression tested on aarch64-none-linux-gnu and
x86_64-pc-linux-gnu.
Is this OK for trunk?
gcc/ChangeLog:
PR tree-optimization/10
* tree-if-conv.cc (predicate_statements): Fix gimple call check.
gcc/testsuite/ChangeLog:
* gcc.dg/vect/vect-simd-clone-16.c: Make simd clone inbranch only.
* gcc.dg/vect/vect-simd-clone-17.c: Likewise.
* gcc.dg/vect/vect-simd-clone-18.c: Likewise.
On 23/02/2023 10:10, Richard Biener via Gcc-patches wrote:
The following makes sure to only predicate calls necessary.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
PR tree-optimization/10
* tree-if-conv.cc (if_convertible_stmt_p): Set PLF_2 on
calls to predicate.
(predicate_statements): Only predicate calls with PLF_2.
* g++.dg/torture/pr10.C: New testcase.
---
gcc/testsuite/g++.dg/torture/pr10.C | 18 ++
gcc/tree-if-conv.cc | 17 ++---
2 files changed, 28 insertions(+), 7 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/torture/pr10.C
diff --git a/gcc/testsuite/g++.dg/torture/pr10.C
b/gcc/testsuite/g++.dg/torture/pr10.C
new file mode 100644
index 000..29a22e21102
--- /dev/null
+++ b/gcc/testsuite/g++.dg/torture/pr10.C
@@ -0,0 +1,18 @@
+// { dg-do compile }
+
+int scaleValueSaturate_scalefactor, scaleValueSaturate___trans_tmp_2,
+scaleValuesSaturate_i;
+int scaleValueSaturate(int value) {
+ int result = __builtin_clz(value);
+ if (value)
+if (-result <= scaleValueSaturate_scalefactor)
+ return 0;
+ return scaleValueSaturate___trans_tmp_2;
+}
+short scaleValuesSaturate_dst;
+short *scaleValuesSaturate_src;
+void scaleValuesSaturate() {
+ for (; scaleValuesSaturate_i; scaleValuesSaturate_i++)
+scaleValuesSaturate_dst =
+scaleValueSaturate(scaleValuesSaturate_src[scaleValuesSaturate_i]);
+}
diff --git a/gcc/tree-if-conv.cc b/gcc/tree-if-conv.cc
index a7a8406374d..0e384e36394 100644
--- a/gcc/tree-if-conv.cc
+++ b/gcc/tree-if-conv.cc
@@ -1099,6 +1099,7 @@ if_convertible_stmt_p (gimple *stmt,
vec refs)
n = n->simdclone->next_clone)
if (n->simdclone->inbranch)
{
+ gimple_set_plf (stmt, GF_PLF_2, true);
need_to_predicate = true;
return true;
}
@@ -2541,7 +2542,8 @@ predicate_statements (loop_p loop)
release_defs (stmt);
continue;
}
- else if (gimple_plf (stmt, GF_PLF_2))
+ else if (gimple_plf (stmt, GF_PLF_2)
+ && is_gimple_assign (stmt))
{
tree lhs = gimple_assign_lhs (stmt);
tree mask;
@@ -2625,13 +2627,14 @@ predicate_statements (loop_p loop)
gimple_assign_set_rhs1 (stmt, ifc_temp_var (type, rhs, ));
update_stmt (stmt);
}
-
- /* Convert functions that have a SIMD clone to IFN_MASK_CALL. This
-will cause the vectorizer to match the "in branch" clone variants,
-and serves to build the mask vector in a natural way. */
- gcall *call = dyn_cast (gsi_stmt (gsi));
- if (call && !gimple_call_internal_p (call))
+ else if (gimple_plf (stmt, GF_PLF_2)
+ && is_gimple_call (stmt))
{
+ /* Convert functions that have a SIMD clone to IFN_MASK_CALL.
+This will cause the vectorizer to match the "in branch"
+clone variants, and serves to build the mask vector
+in a natural way. */
+ gcall *call = dyn_cast (gsi_stmt (gsi));
tree orig_fn = gimple_call_fn (call);
int orig_nargs = gimple_call_num_args (call);
auto_vec args;diff --git a/gcc/testsuite/gcc.dg/vect/vect-simd-clone-16.c
b/gcc/testsuite/gcc.dg/vect/vect-simd-clone-16.c
index
3ff1cfee05951609d8ca93291d5d7c47cb07ec0d..125ff4f6c8d7df5e289187e523d32e0d12db9769
100644
--- a/gcc/testsuite/gcc.dg/vect/vect-simd-clone-16.c
+++ b/gcc/testsuite/gcc.dg/vect/vect-simd-clone-16.c
@@ -9,7 +9,7 @@
#endif
/* A simple function that will be cloned. */
-#pragma omp declare simd
+#pragma omp declare simd inbranch
TYPE __attribute__((noinline))
foo (TYPE a)
{
diff --git a/gcc/testsuite/gcc.dg/vect/vect-simd-clone-17.c
b/gcc/testsuite/gcc.dg/vect/vect-simd-clone-17.c
index
803e0f25d45c1069633486c7b7d805638db83482..3430d6f5aa4f3ae3ed8bdfda80ef99d5517f15c6
100644
--- a/gcc/testsuite/gcc.dg/vect/vect-simd-clone-17.c
+++ b/gcc/testsuite/gcc.dg/vect/vect-simd-clone-17.c
@@ -9,7 +9,7 @@
#endif
/* A simple function that will be cloned. */
-#pragma omp declare simd
[PATCH] tree-optimization/108888 - call if-conversion
The following makes sure to only predicate calls necessary.
Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.
PR tree-optimization/10
* tree-if-conv.cc (if_convertible_stmt_p): Set PLF_2 on
calls to predicate.
(predicate_statements): Only predicate calls with PLF_2.
* g++.dg/torture/pr10.C: New testcase.
---
gcc/testsuite/g++.dg/torture/pr10.C | 18 ++
gcc/tree-if-conv.cc | 17 ++---
2 files changed, 28 insertions(+), 7 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/torture/pr10.C
diff --git a/gcc/testsuite/g++.dg/torture/pr10.C
b/gcc/testsuite/g++.dg/torture/pr10.C
new file mode 100644
index 000..29a22e21102
--- /dev/null
+++ b/gcc/testsuite/g++.dg/torture/pr10.C
@@ -0,0 +1,18 @@
+// { dg-do compile }
+
+int scaleValueSaturate_scalefactor, scaleValueSaturate___trans_tmp_2,
+scaleValuesSaturate_i;
+int scaleValueSaturate(int value) {
+ int result = __builtin_clz(value);
+ if (value)
+if (-result <= scaleValueSaturate_scalefactor)
+ return 0;
+ return scaleValueSaturate___trans_tmp_2;
+}
+short scaleValuesSaturate_dst;
+short *scaleValuesSaturate_src;
+void scaleValuesSaturate() {
+ for (; scaleValuesSaturate_i; scaleValuesSaturate_i++)
+scaleValuesSaturate_dst =
+scaleValueSaturate(scaleValuesSaturate_src[scaleValuesSaturate_i]);
+}
diff --git a/gcc/tree-if-conv.cc b/gcc/tree-if-conv.cc
index a7a8406374d..0e384e36394 100644
--- a/gcc/tree-if-conv.cc
+++ b/gcc/tree-if-conv.cc
@@ -1099,6 +1099,7 @@ if_convertible_stmt_p (gimple *stmt,
vec refs)
n = n->simdclone->next_clone)
if (n->simdclone->inbranch)
{
+ gimple_set_plf (stmt, GF_PLF_2, true);
need_to_predicate = true;
return true;
}
@@ -2541,7 +2542,8 @@ predicate_statements (loop_p loop)
release_defs (stmt);
continue;
}
- else if (gimple_plf (stmt, GF_PLF_2))
+ else if (gimple_plf (stmt, GF_PLF_2)
+ && is_gimple_assign (stmt))
{
tree lhs = gimple_assign_lhs (stmt);
tree mask;
@@ -2625,13 +2627,14 @@ predicate_statements (loop_p loop)
gimple_assign_set_rhs1 (stmt, ifc_temp_var (type, rhs, ));
update_stmt (stmt);
}
-
- /* Convert functions that have a SIMD clone to IFN_MASK_CALL. This
-will cause the vectorizer to match the "in branch" clone variants,
-and serves to build the mask vector in a natural way. */
- gcall *call = dyn_cast (gsi_stmt (gsi));
- if (call && !gimple_call_internal_p (call))
+ else if (gimple_plf (stmt, GF_PLF_2)
+ && is_gimple_call (stmt))
{
+ /* Convert functions that have a SIMD clone to IFN_MASK_CALL.
+This will cause the vectorizer to match the "in branch"
+clone variants, and serves to build the mask vector
+in a natural way. */
+ gcall *call = dyn_cast (gsi_stmt (gsi));
tree orig_fn = gimple_call_fn (call);
int orig_nargs = gimple_call_num_args (call);
auto_vec args;
--
2.35.3
