From: Kyrylo Tkachov <[email protected]>
EXPAND_MEMORY requires a MEM result but does not state whether the access
is a load or a store. Its conservative store semantics avoid unsafe
MEM_NOTRAP claims, but two known load callers lose valid nontrapping
information.
On this patch's parent, a memory-only inline asm input is:
(mem/u:SI (...) [ input+0 ])
Restore its load semantics:
(mem/u/c:SI (...) [ input+0 ])
The partial-load expander has the same problem. It uses EXPAND_WRITE even
though its MEM is the source of a masked or length-limited load. Its source
MEM changes from:
(mem/u:V4SI (...))
to:
(mem/u/c:V4SI (...))
Use EXPAND_MEMORY for partial loads because they require a MEM, then set
MEM_NOTRAP from tree_could_trap_p at both caller sites.
Bootstrapped and tested on aarch64-unknown-linux-gnu. Tested on
x86_64-pc-linux-gnu.
PR middle-end/127185
gcc/ChangeLog:
* cfgexpand.cc (expand_asm_stmt): Restore load MEM_NOTRAP for memory
inputs.
* internal-fn.cc: Include tree-eh.h.
(expand_partial_load_optab_fn): Expand the source as memory and
restore load MEM_NOTRAP.
gcc/testsuite/ChangeLog:
* g++.dg/opt/mem-notrap-load-asm.C: New test.
* gcc.target/i386/mem-notrap-mask-load.c: New test.
Signed-off-by: Kyrylo Tkachov <[email protected]>
---
gcc/cfgexpand.cc | 5 +++-
gcc/internal-fn.cc | 4 +++-
.../g++.dg/opt/mem-notrap-load-asm.C | 14 +++++++++++
.../gcc.target/i386/mem-notrap-mask-load.c | 23 +++++++++++++++++++
4 files changed, 44 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/g++.dg/opt/mem-notrap-load-asm.C
create mode 100644 gcc/testsuite/gcc.target/i386/mem-notrap-mask-load.c
diff --git a/gcc/cfgexpand.cc b/gcc/cfgexpand.cc
index 245ba357b5e..4bd311b11a3 100644
--- a/gcc/cfgexpand.cc
+++ b/gcc/cfgexpand.cc
@@ -3781,7 +3781,10 @@ expand_asm_stmt (gasm *stmt)
if (GET_CODE (op) == CONCAT)
op = force_reg (GET_MODE (op), op);
else if (MEM_P (op))
- op = validize_mem (op);
+ {
+ op = validize_mem (op);
+ MEM_NOTRAP_P (op) = !tree_could_trap_p (val);
+ }
if (asm_operand_ok (op, constraint, NULL) <= 0)
{
diff --git a/gcc/internal-fn.cc b/gcc/internal-fn.cc
index eae0ca9bf51..3fdf89e0e6e 100644
--- a/gcc/internal-fn.cc
+++ b/gcc/internal-fn.cc
@@ -25,6 +25,7 @@ along with GCC; see the file COPYING3. If not see
#include "target.h"
#include "rtl.h"
#include "tree.h"
+#include "tree-eh.h"
#include "gimple.h"
#include "predict.h"
#include "stringpool.h"
@@ -3144,8 +3145,9 @@ expand_partial_load_optab_fn (internal_fn ifn, gcall
*stmt, convert_optab optab)
icode = convert_optab_handler (optab, TYPE_MODE (type),
TYPE_MODE (TREE_TYPE (maskt)));
- mem = expand_expr (rhs, NULL_RTX, VOIDmode, EXPAND_WRITE);
+ mem = expand_expr (rhs, NULL_RTX, VOIDmode, EXPAND_MEMORY);
gcc_assert (MEM_P (mem));
+ MEM_NOTRAP_P (mem) = !tree_could_trap_p (rhs);
/* The built MEM_REF does not accurately reflect that the load
is only partial. Clear it. */
set_mem_expr (mem, NULL_TREE);
diff --git a/gcc/testsuite/g++.dg/opt/mem-notrap-load-asm.C
b/gcc/testsuite/g++.dg/opt/mem-notrap-load-asm.C
new file mode 100644
index 00000000000..797d6793c0d
--- /dev/null
+++ b/gcc/testsuite/g++.dg/opt/mem-notrap-load-asm.C
@@ -0,0 +1,14 @@
+// A memory-only asm input is a load even though expansion must return a MEM.
+
+// { dg-do compile }
+// { dg-options "-O2 -fnon-call-exceptions -fdump-rtl-expand" }
+
+extern const int input;
+
+void
+load ()
+{
+ asm volatile ("" : : "m" (input));
+}
+
+// { dg-final { scan-rtl-dump {\(mem(?:/[a-z])*/c(?:/[a-z])*:[^\n\r]*\[[0-9]+
input\+0} "expand" } }
diff --git a/gcc/testsuite/gcc.target/i386/mem-notrap-mask-load.c
b/gcc/testsuite/gcc.target/i386/mem-notrap-mask-load.c
new file mode 100644
index 00000000000..bc510e6ad4f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/mem-notrap-mask-load.c
@@ -0,0 +1,23 @@
+/* A masked load is a read even though its expander requires a MEM. */
+
+/* { dg-do compile } */
+/* { dg-options "-O -fgimple -mavx2 -mno-avx512f -fnon-call-exceptions
-fdump-tree-optimized -fdump-rtl-expand" } */
+
+typedef int v4si __attribute__ ((vector_size (16)));
+
+extern const v4si values;
+
+v4si __GIMPLE (ssa)
+load (v4si mask)
+{
+ v4si result;
+
+ __BB(2):
+ result_2 = .MASK_LOAD (&values, _Literal (v4si *) 128,
+ mask_1(D),
+ _Literal (v4si) { 0, 0, 0, 0 });
+ return result_2;
+}
+
+/* { dg-final { scan-tree-dump {\.MASK_LOAD} "optimized" } } */
+/* { dg-final { scan-rtl-dump-times {\(mem/u/c:V4SI} 1 "expand" } } */
--
2.50.1 (Apple Git-155)
