The following patch avoids infinite recursion during generic folding.
The (cmp (bswap @0) INTEGER_CST@1) simplification relies on
(bswap @1) actually being simplified, if it is not simplified, we just
move the bswap from one operand to the other and if @0 is also INTEGER_CST,
we apply the same rule next.
The reason why bswap @1 isn't folded to INTEGER_CST is that the INTEGER_CST
has TREE_OVERFLOW set on it and fold-const-call.cc predicate punts in
such cases:
static inline bool
integer_cst_p (tree t)
{
return TREE_CODE (t) == INTEGER_CST && !TREE_OVERFLOW (t);
}
The patch uses ! modifier to ensure the bswap is simplified and
extends support to GENERIC by means of requiring !EXPR_P which
is not perfect but a conservative approximation.
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
I didn't bother to un-#ifdef GIMPLE the few use-cases of ! we have
in match.pd since those are not regressions. Let's remember
that for stage1. I also agree this is the safes approach at this
stage.
OK?
Thanks,
Richard.
2022-02-22 Richard Biener <[email protected]>
PR tree-optimization/104644
* doc/match-and-simplify.texi: Amend ! documentation.
* genmatch.cc (expr::gen_transform): Code-generate ! support
for GENERIC.
(parser::parse_expr): Allow ! for GENERIC.
* match.pd (cmp (bswap @0) INTEGER_CST@1): Use ! modifier on
bswap.
* gcc.dg/pr104644.c: New test.
Co-Authored-by: Jakub Jelinek <[email protected]>
---
gcc/doc/match-and-simplify.texi | 6 ++++--
gcc/genmatch.cc | 20 +++++++++-----------
gcc/match.pd | 2 +-
gcc/testsuite/gcc.dg/pr104644.c | 9 +++++++++
4 files changed, 23 insertions(+), 14 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/pr104644.c
diff --git a/gcc/doc/match-and-simplify.texi b/gcc/doc/match-and-simplify.texi
index 63a73ae047c..055a5308e7d 100644
--- a/gcc/doc/match-and-simplify.texi
+++ b/gcc/doc/match-and-simplify.texi
@@ -374,8 +374,10 @@ for example
which moves the outer @code{plus} operation to the inner arms
of the @code{vec_cond} expression but only if the actual plus
-operations both simplify. Note this is currently only supported
-for code generation targeting @code{GIMPLE}.
+operations both simplify. Note that on @code{GENERIC} a simple
+operand means that the result satisfies @code{!EXPR_P} which
+can be limiting if the operation itself simplifies but the
+remaining operand is an (unrelated) expression.
As intermediate conversions are often optional there is a way to
avoid the need to repeat patterns both with and without such
diff --git a/gcc/genmatch.cc b/gcc/genmatch.cc
index 97f6f00fa68..2eda7300821 100644
--- a/gcc/genmatch.cc
+++ b/gcc/genmatch.cc
@@ -2553,19 +2553,20 @@ expr::gen_transform (FILE *f, int indent, const char
*dest, bool gimple,
fprintf_indent (f, indent, "_r%d = fold_build%d_loc (loc, %s, %s",
depth, ops.length(), opr_name, type);
else
- {
- fprintf_indent (f, indent, "{\n");
- fprintf_indent (f, indent, " _r%d = maybe_build_call_expr_loc (loc, "
- "%s, %s, %d", depth, opr_name, type, ops.length());
- }
+ fprintf_indent (f, indent, "_r%d = maybe_build_call_expr_loc (loc, "
+ "%s, %s, %d", depth, opr_name, type, ops.length());
for (unsigned i = 0; i < ops.length (); ++i)
fprintf (f, ", _o%d[%u]", depth, i);
fprintf (f, ");\n");
if (opr->kind != id_base::CODE)
{
- fprintf_indent (f, indent, " if (!_r%d)\n", depth);
- fprintf_indent (f, indent, " goto %s;\n", fail_label);
- fprintf_indent (f, indent, "}\n");
+ fprintf_indent (f, indent, "if (!_r%d)\n", depth);
+ fprintf_indent (f, indent, " goto %s;\n", fail_label);
+ }
+ if (force_leaf)
+ {
+ fprintf_indent (f, indent, "if (EXPR_P (_r%d))\n", depth);
+ fprintf_indent (f, indent, " goto %s;\n", fail_label);
}
if (*opr == CONVERT_EXPR)
{
@@ -4297,9 +4298,6 @@ parser::parse_expr ()
&& token->type == CPP_NOT
&& !(token->flags & PREV_WHITE))
{
- if (!gimple)
- fatal_at (token, "forcing simplification to a leaf is not supported "
- "for GENERIC");
eat_token (CPP_NOT);
e->force_leaf = true;
}
diff --git a/gcc/match.pd b/gcc/match.pd
index cad61848daa..cf78a11ddd4 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3962,7 +3962,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(simplify
(cmp (bswap @0) INTEGER_CST@1)
(with { tree ctype = TREE_TYPE (@1); }
- (cmp (convert:ctype @0) (bswap @1)))))
+ (cmp (convert:ctype @0) (bswap! @1)))))
/* (bswap(x) >> C1) & C2 can sometimes be simplified to (x >> C3) & C2. */
(simplify
(bit_and (convert1? (rshift@0 (convert2? (bswap@4 @1)) INTEGER_CST@2))
diff --git a/gcc/testsuite/gcc.dg/pr104644.c b/gcc/testsuite/gcc.dg/pr104644.c
new file mode 100644
index 00000000000..70bf3a49647
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr104644.c
@@ -0,0 +1,9 @@
+/* PR tree-optimization/104644 */
+/* { dg-do compile } */
+/* { dg-options "-Wno-overflow" } */
+
+int
+foo (void)
+{
+ return __builtin_bswap16 (1.31072e+5f) != (signed char) 1.31072e+5f;
+}
--
2.34.1
