aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorAndrew Cagney <cagney@redhat.com>2000-06-23 07:44:00 +0000
committerAndrew Cagney <cagney@redhat.com>2000-06-23 07:44:00 +0000
commitf09ded24229f68245da53149e1ac46a8066057cd (patch)
tree156cba54aedc61f23eaa4883f71b3030e9e06e4b
parent70d6ecf3bf23fe901a2e52e645e200e4f49c3006 (diff)
downloadgdb-f09ded24229f68245da53149e1ac46a8066057cd.zip
gdb-f09ded24229f68245da53149e1ac46a8066057cd.tar.gz
gdb-f09ded24229f68245da53149e1ac46a8066057cd.tar.bz2
When FP registers are full, store FP arguments on stack and not in
integer registers.
-rw-r--r--gdb/ChangeLog6
-rw-r--r--gdb/mips-tdep.c41
2 files changed, 35 insertions, 12 deletions
diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index c000743..6338d11 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,9 @@
+Fri Jun 23 16:20:21 2000 Andrew Cagney <cagney@b1.cygnus.com>
+
+ * mips-tdep.c (fp_register_arg_p): New function.
+ (mips_push_arguments): Use. Do not pass floating point arguments
+ on in an integer register.
+
2000-06-21 Pierre Muller <muller@ics.u-strasbg.fr>
* symfile.c (init_filename_language_table): add ".pas", ".p" and ".pp"
as pascal source file extensions.
diff --git a/gdb/mips-tdep.c b/gdb/mips-tdep.c
index 0922cee..6ba42a2 100644
--- a/gdb/mips-tdep.c
+++ b/gdb/mips-tdep.c
@@ -2062,6 +2062,23 @@ setup_arbitrary_frame (argc, argv)
return create_new_frame (argv[0], argv[1]);
}
+/* According to the current ABI, should the type be passed in a
+ floating-point register (assuming that there is space)? When there
+ is no FPU, FP are not even considered as possibile candidates for
+ FP registers and, consequently this returns false - forces FP
+ arguments into integer registers. */
+
+static int
+fp_register_arg_p (enum type_code typecode, struct type *arg_type)
+{
+ return ((typecode == TYPE_CODE_FLT
+ || (MIPS_EABI
+ && (typecode == TYPE_CODE_STRUCT || typecode == TYPE_CODE_UNION)
+ && TYPE_NFIELDS (arg_type) == 1
+ && TYPE_CODE (TYPE_FIELD_TYPE (arg_type, 0)) == TYPE_CODE_FLT))
+ && MIPS_FPU_TYPE != MIPS_FPU_NONE);
+}
+
CORE_ADDR
mips_push_arguments (nargs, args, sp, struct_return, struct_addr)
int nargs;
@@ -2165,13 +2182,8 @@ mips_push_arguments (nargs, args, sp, struct_return, struct_addr)
/* MIPS_EABI squeeses a struct that contains a single floating
point value into an FP register instead of pusing it onto the
stack. */
- if ((typecode == TYPE_CODE_FLT
- || (MIPS_EABI
- && (typecode == TYPE_CODE_STRUCT || typecode == TYPE_CODE_UNION)
- && TYPE_NFIELDS (arg_type) == 1
- && TYPE_CODE (TYPE_FIELD_TYPE (arg_type, 0)) == TYPE_CODE_FLT))
- && float_argreg <= MIPS_LAST_FP_ARG_REGNUM
- && MIPS_FPU_TYPE != MIPS_FPU_NONE)
+ if (fp_register_arg_p (typecode, arg_type)
+ && float_argreg <= MIPS_LAST_FP_ARG_REGNUM)
{
if (!FP_REGISTER_DOUBLE && len == 8)
{
@@ -2241,14 +2253,17 @@ mips_push_arguments (nargs, args, sp, struct_return, struct_addr)
are treated specially: Irix cc passes them in registers
where gcc sometimes puts them on the stack. For maximum
compatibility, we will put them in both places. */
-
int odd_sized_struct = ((len > MIPS_SAVED_REGSIZE) &&
(len % MIPS_SAVED_REGSIZE != 0));
+ /* Note: Floating-point values that didn't fit into an FP
+ register are only written to memory. */
while (len > 0)
{
int partial_len = len < MIPS_SAVED_REGSIZE ? len : MIPS_SAVED_REGSIZE;
- if (argreg > MIPS_LAST_ARG_REGNUM || odd_sized_struct)
+ if (argreg > MIPS_LAST_ARG_REGNUM
+ || odd_sized_struct
+ || fp_register_arg_p (typecode, arg_type))
{
/* Write this portion of the argument to the stack. */
/* Should shorter than int integer values be
@@ -2291,9 +2306,11 @@ mips_push_arguments (nargs, args, sp, struct_return, struct_addr)
write_memory (addr, val, partial_len);
}
- /* Note!!! This is NOT an else clause.
- Odd sized structs may go thru BOTH paths. */
- if (argreg <= MIPS_LAST_ARG_REGNUM)
+ /* Note!!! This is NOT an else clause. Odd sized
+ structs may go thru BOTH paths. Floating point
+ arguments will not. */
+ if (argreg <= MIPS_LAST_ARG_REGNUM
+ && !fp_register_arg_p (typecode, arg_type))
{
LONGEST regval = extract_unsigned_integer (val, partial_len);