https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118068
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jason at gcc dot gnu.org
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Marek/Jason, what do you think about doing cp_walk_tree with
cp_fold_immediate_r callback from cp_fold_function and cp_fully_fold_init?
Though, tried
--- gcc/cp/cp-gimplify.cc.jj 2025-02-01 00:47:07.881079292 +0100
+++ gcc/cp/cp-gimplify.cc 2025-03-06 15:31:11.713951494 +0100
@@ -1204,7 +1204,8 @@ cp_build_init_expr_for_ctor (tree call,
return init;
}
-/* A subroutine of cp_fold_r to handle immediate functions. */
+/* A walk_tree callback for cp_fold_function and cp_fully_fold_init to handle
+ immediate functions. */
static tree
cp_fold_immediate_r (tree *stmt_p, int *walk_subtrees, void *data_)
@@ -1370,45 +1371,8 @@ cp_fold_r (tree *stmt_p, int *walk_subtr
tree stmt = *stmt_p;
enum tree_code code = TREE_CODE (stmt);
- if (cxx_dialect >= cxx20)
- {
- /* Unfortunately we must handle code like
- false ? bar () : 42
- where we have to check bar too. The cp_fold call below could
- fold the ?: into a constant before we've checked it. */
- if (code == COND_EXPR)
- {
- auto then_fn = cp_fold_r, else_fn = cp_fold_r;
- /* See if we can figure out if either of the branches is dead. If it
- is, we don't need to do everything that cp_fold_r does. */
- cp_walk_tree (&TREE_OPERAND (stmt, 0), cp_fold_r, data, nullptr);
- if (integer_zerop (TREE_OPERAND (stmt, 0)))
- then_fn = cp_fold_immediate_r;
- else if (integer_nonzerop (TREE_OPERAND (stmt, 0)))
- else_fn = cp_fold_immediate_r;
-
- if (TREE_OPERAND (stmt, 1))
- cp_walk_tree (&TREE_OPERAND (stmt, 1), then_fn, data,
- nullptr);
- if (TREE_OPERAND (stmt, 2))
- cp_walk_tree (&TREE_OPERAND (stmt, 2), else_fn, data,
- nullptr);
- *walk_subtrees = 0;
- /* Don't return yet, still need the cp_fold below. */
- }
- else
- cp_fold_immediate_r (stmt_p, walk_subtrees, data);
- }
-
*stmt_p = stmt = cp_fold (*stmt_p, data->flags);
- /* For certain trees, like +foo(), the cp_fold above will remove the +,
- and the subsequent tree walk would go straight down to the CALL_EXPR's
- operands, meaning that cp_fold_immediate_r would never see the
- CALL_EXPR. Ew :(. */
- if (TREE_CODE (stmt) == CALL_EXPR && code != CALL_EXPR)
- cp_fold_immediate_r (stmt_p, walk_subtrees, data);
-
if (data->pset.add (stmt))
{
/* Don't walk subtrees of stmts we've already walked once, otherwise
@@ -1537,6 +1501,13 @@ cp_fold_function (tree fndecl)
been constant-evaluated already if possible, so we can safely
pass ff_mce_false. */
cp_fold_data data (ff_genericize | ff_mce_false);
+ /* Do cp_fold_immediate_r in separate whole IL walk instead of during
+ cp_fold_r, as otherwise expressions using results of immediate functions
+ might not be folded as cp_fold is called on those before cp_fold_r is
+ called on their argument. And calling cp_fold_immediate_r during
+ cp_fold can mean evaluation of the immediate functions many times. */
+ if (cxx_dialect >= cxx20)
+ cp_walk_tree (&DECL_SAVED_TREE (fndecl), cp_fold_immediate_r, &data,
NULL);
cp_walk_tree (&DECL_SAVED_TREE (fndecl), cp_fold_r, &data, NULL);
/* This is merely an optimization: if FNDECL has no i-e expressions,
@@ -2910,6 +2881,8 @@ cp_fully_fold_init (tree x)
return x;
x = cp_fully_fold (x, mce_false);
cp_fold_data data (ff_mce_false);
+ if (cxx_dialect >= cxx20)
+ cp_walk_tree (&x, cp_fold_immediate_r, &data, NULL);
cp_walk_tree (&x, cp_fold_r, &data, NULL);
return x;
}
--- gcc/testsuite/g++.target/i386/pr118068.C.jj 2025-03-06 15:25:59.849944651
+0100
+++ gcc/testsuite/g++.target/i386/pr118068.C 2025-03-06 15:25:51.152064565
+0100
@@ -0,0 +1,17 @@
+// PR target/118068
+// { dg-do compile { target c++20 } }
+// { dg-options "-O0 -mavx" }
+
+typedef float V __attribute__((vector_size (32)));
+
+consteval unsigned char
+foo (int x)
+{
+ return x;
+}
+
+V
+bar (V x, V y)
+{
+ return __builtin_ia32_blendps256 (x, y, (int) foo (0x23));
+}
and so far saw
FAIL: g++.dg/cpp23/consteval-if1.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if1.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if1.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if10.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if10.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if10.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if11.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if11.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if11.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if13.C -std=c++23 (test for excess errors)
FAIL: g++.dg/cpp23/consteval-if13.C -std=c++26 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop15.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop15.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop15.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop8.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop17.C -std=gnu++20 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop8.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop1.C -std=c++20 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop17.C -std=gnu++23 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop8.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop17.C -std=gnu++26 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop1.C -std=c++23 (test for excess errors)
FAIL: g++.dg/cpp2a/consteval-prop1.C -std=c++26 (test for excess errors)
regressions.
