From: Eric Botcazou <[email protected]>
This occurs when the construct is totally scalarized by the SRA pass.
gcc/ada/ChangeLog:
* gcc-interface/utils.cc (maybe_pad_type): Propagate the flag
TYPE_REVERSE_STORAGE_ORDER from the inner type to the record type
if the former is also an aggregate type.
Tested on x86_64-pc-linux-gnu, committed on master.
---
gcc/ada/gcc-interface/utils.cc | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/gcc/ada/gcc-interface/utils.cc b/gcc/ada/gcc-interface/utils.cc
index f4fc0cec828..f74b6361b40 100644
--- a/gcc/ada/gcc-interface/utils.cc
+++ b/gcc/ada/gcc-interface/utils.cc
@@ -1763,6 +1763,16 @@ maybe_pad_type (tree type, tree size, unsigned int align,
size_binop (EXACT_DIV_EXPR, TYPE_SIZE (record),
bitsize_unit_node));
+ /* Propagate the storage order from an aggregate type to the padded type.
+ Ideally this should not be necessary, because the flag on the padded
+ type has no semantic effects given that its only field is not scalar.
+ However, when the aggregate type contains a single scalar field, SRA
+ may totally scalarize the padded type without taking into account the
+ intermediate aggregate type and, therefore, may test the flag on the
+ former instead of the latter to generate a reverse memory access. */
+ if (AGGREGATE_TYPE_P (type))
+ TYPE_REVERSE_STORAGE_ORDER (record) = TYPE_REVERSE_STORAGE_ORDER (type);
+
/* If we are changing the alignment and the input type is a record with
BLKmode and a small constant size, try to make a form that has an
integral mode. This might allow the padding record to also have an
--
2.53.0
