[SPARC] Fix PR target/114416
Checks
| Context |
Check |
Description |
| linaro-tcwg-bot/tcwg_gcc_build--master-arm |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_check--master-arm |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 |
success
|
Testing passed
|
| linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 |
success
|
Testing passed
|
Commit Message
For the 20th anniversary of https://gcc.gnu.org/gcc-3.4/sparc-abi.html, a new
calling convention incompatibility with the vendor compiler (and the ABI) has
been discovered in 64-bit mode, affecting small structures containing arrays
of floating-point components. The decision has been made to fix it on Solaris
only at this point.
Bootstrapped/regtested on SPARC/Solaris 11.4 and SPARC64/Linux by Rainer (many
thanks again) and applied on the mainline.
2024-04-25 Eric Botcazou <ebotcazou@adacore.com>
PR target/114416
* config/sparc/sparc.h (SUN_V9_ABI_COMPATIBILITY): New macro.
* config/sparc/sol2.h (SUN_V9_ABI_COMPATIBILITY): Redefine it.
* config/sparc/sparc.cc (fp_type_for_abi): New predicate.
(traverse_record_type): Use it to spot floating-point types.
(compute_fp_layout): Also deal with array types.
2024-04-25 Eric Botcazou <ebotcazou@adacore.com>
* gcc.target/sparc/small-struct-1.c: New test.
Comments
> For the 20th anniversary of https://gcc.gnu.org/gcc-3.4/sparc-abi.html, a
> new calling convention incompatibility with the vendor compiler (and the
> ABI) has been discovered in 64-bit mode, affecting small structures
> containing arrays of floating-point components. The decision has been made
> to fix it on Solaris only at this point.
Documented by the attached patch, validated with W3C's Validator and applied.
@@ -456,3 +456,6 @@ extern const char *host_detect_local_cpu (int argc, const char **argv);
#undef SPARC_LOW_FE_EXCEPT_VALUES
#define SPARC_LOW_FE_EXCEPT_VALUES 1
+
+#undef SUN_V9_ABI_COMPATIBILITY
+#define SUN_V9_ABI_COMPATIBILITY 1
@@ -6782,6 +6782,22 @@ sparc_pass_by_reference (cumulative_args_t, const function_arg_info &arg)
|| GET_MODE_SIZE (mode) > 16);
}
+/* Return true if TYPE is considered as a floating-point type by the ABI. */
+
+static bool
+fp_type_for_abi (const_tree type)
+{
+ /* This is the original GCC implementation. */
+ if (FLOAT_TYPE_P (type) || VECTOR_TYPE_P (type))
+ return true;
+
+ /* This has been introduced in GCC 14 to match the vendor compiler. */
+ if (SUN_V9_ABI_COMPATIBILITY && TREE_CODE (type) == ARRAY_TYPE)
+ return fp_type_for_abi (TREE_TYPE (type));
+
+ return false;
+}
+
/* Traverse the record TYPE recursively and call FUNC on its fields.
NAMED is true if this is for a named parameter. DATA is passed
to FUNC for each field. OFFSET is the starting position and
@@ -6820,8 +6836,7 @@ traverse_record_type (const_tree type, bool named, T *data,
packed);
else
{
- const bool fp_type
- = FLOAT_TYPE_P (field_type) || VECTOR_TYPE_P (field_type);
+ const bool fp_type = fp_type_for_abi (field_type);
Func (field, bitpos, fp_type && named && !packed && TARGET_FPU,
data);
}
@@ -7072,6 +7087,13 @@ compute_fp_layout (const_tree field, int bitpos, assign_data_t *data,
mode = TYPE_MODE (TREE_TYPE (TREE_TYPE (field)));
nregs = 2;
}
+ else if (TREE_CODE (TREE_TYPE (field)) == ARRAY_TYPE)
+ {
+ tree elt_type = strip_array_types (TREE_TYPE (field));
+ mode = TYPE_MODE (elt_type);
+ nregs
+ = int_size_in_bytes (TREE_TYPE (field)) / int_size_in_bytes (elt_type);
+ }
else
nregs = 1;
@@ -1700,3 +1700,6 @@ extern int sparc_indent_opcode;
#define SPARC_LOW_FE_EXCEPT_VALUES 0
#define TARGET_SUPPORTS_WIDE_INT 1
+
+/* Define this to 1 to accept ABI changes to match the vendor compiler. */
+#define SUN_V9_ABI_COMPATIBILITY 0