Hi Jeff, Hi Alex,
I recently encountered a problem with the REG_SAVE_BYTES macro in the
mn10300 backend. When compiling some code in -fPID mode it was
producing a different result to the mask computed by
mn10300_get_live_callee_saved_regs. The problem turned out to be that
the macro was not checking the call_really_saved_regs array. Rather
than update the macro, it seemed to me that it would be better to just
make this computation in one place - the
mn10300_get_live_callee_saved_regs function - and avoid problems like
this in the future.
So attached is a patch that does this. Tested with no regressions on
an mn10300-elf and an am33_2.0-linux-gnu toolchain. OK to apply ?
Cheers
Nick
gcc/ChangeLog
2012-06-29 Nick Clifton <ni...@redhat.com>
* config/mn10300/mn10300.c (REG_SAVE_BYTES): Delete.
(mn10300_get_live_callee_saved_regs): If requested return a count
of the number of bytes in the mask.
(mn10300_expand_prologue): Add argument to invocation of
mn10300_get_live_callee_regs.
(mn10300_expand_epilogue): Compute reg_save_bytes by calling
mn10300_get_live_callee_saved_regs.
(mn10300_initial_offset): Likewise.
* config/mn10300/mn10300-protos.h (mn10300_get_live_callee_saved_regs):
Update prototype.
* config/mn10300/mn10300.md (return_ret): Add argument to
invocation of mn10300_get_live_callee_saved_regs.
Index: gcc/config/mn10300/mn10300.md
===================================================================
--- gcc/config/mn10300/mn10300.md (revision 189067)
+++ gcc/config/mn10300/mn10300.md (working copy)
@@ -2048,7 +2048,7 @@
{
/* The RETF insn is up to 3 cycles faster than RET. */
fputs ((mn10300_can_use_retf_insn () ? "\tretf " : "\tret "), asm_out_file);
- mn10300_print_reg_list (asm_out_file, mn10300_get_live_callee_saved_regs ());
+ mn10300_print_reg_list (asm_out_file, mn10300_get_live_callee_saved_regs
(NULL));
fprintf (asm_out_file, ",%d\n", (int) INTVAL (operands[0]));
return "";
})
Index: gcc/config/mn10300/mn10300-protos.h
===================================================================
--- gcc/config/mn10300/mn10300-protos.h (revision 189067)
+++ gcc/config/mn10300/mn10300-protos.h (working copy)
@@ -25,7 +25,7 @@
extern rtx mn10300_legitimize_reload_address (rtx, enum machine_mode,
int, int, int);
extern bool mn10300_function_value_regno_p (const unsigned int);
-extern int mn10300_get_live_callee_saved_regs (void);
+extern unsigned int mn10300_get_live_callee_saved_regs (unsigned int *);
extern bool mn10300_hard_regno_mode_ok (unsigned int, enum machine_mode);
extern bool mn10300_modes_tieable (enum machine_mode, enum machine_mode);
extern const char *mn10300_output_add (rtx[3], bool);
Index: gcc/config/mn10300/mn10300.c
===================================================================
--- gcc/config/mn10300/mn10300.c (revision 189067)
+++ gcc/config/mn10300/mn10300.c (working copy)
@@ -56,18 +56,6 @@
/* Selected processor type for tuning. */
enum processor_type mn10300_tune_cpu = PROCESSOR_DEFAULT;
-/* The size of the callee register save area. Right now we save everything
- on entry since it costs us nothing in code size. It does cost us from a
- speed standpoint, so we want to optimize this sooner or later. */
-#define REG_SAVE_BYTES (4 * df_regs_ever_live_p (2) \
- + 4 * df_regs_ever_live_p (3) \
- + 4 * df_regs_ever_live_p (6) \
- + 4 * df_regs_ever_live_p (7) \
- + 16 * (df_regs_ever_live_p (14) \
- || df_regs_ever_live_p (15) \
- || df_regs_ever_live_p (16) \
- || df_regs_ever_live_p (17)))
-
#define CC_FLAG_Z 1
#define CC_FLAG_N 2
#define CC_FLAG_C 4
@@ -634,21 +622,36 @@
/* Returns the set of live, callee-saved registers as a bitmask. The
callee-saved extended registers cannot be stored individually, so
- all of them will be included in the mask if any one of them is used. */
+ Also returns the number of bytes in the registers in the mask if
+ BYTES_SAVED is not NULL. */
-int
-mn10300_get_live_callee_saved_regs (void)
+unsigned int
+mn10300_get_live_callee_saved_regs (unsigned int * bytes_saved)
{
int mask;
int i;
+ unsigned int count;
- mask = 0;
+ count = mask = 0;
for (i = 0; i <= LAST_EXTENDED_REGNUM; i++)
if (df_regs_ever_live_p (i) && ! call_really_used_regs[i])
- mask |= (1 << i);
+ {
+ mask |= (1 << i);
+ ++ count;
+ }
+
if ((mask & 0x3c000) != 0)
- mask |= 0x3c000;
+ {
+ for (i = 0x04000; i < 0x40000; i <<= 1)
+ if ((mask & i) == 0)
+ ++ count;
+
+ mask |= 0x3c000;
+ }
+ if (bytes_saved)
+ * bytes_saved = count * UNITS_PER_WORD;
+
return mask;
}
@@ -741,7 +744,7 @@
HOST_WIDE_INT size = mn10300_frame_size ();
/* If we use any of the callee-saved registers, save them now. */
- mn10300_gen_multiple_store (mn10300_get_live_callee_saved_regs ());
+ mn10300_gen_multiple_store (mn10300_get_live_callee_saved_regs (NULL));
if (TARGET_AM33_2 && fp_regs_to_save ())
{
@@ -998,8 +1001,10 @@
mn10300_expand_epilogue (void)
{
HOST_WIDE_INT size = mn10300_frame_size ();
- int reg_save_bytes = REG_SAVE_BYTES;
-
+ unsigned int reg_save_bytes;
+
+ mn10300_get_live_callee_saved_regs (& reg_save_bytes);
+
if (TARGET_AM33_2 && fp_regs_to_save ())
{
int num_regs_to_save = fp_regs_to_save (), i;
@@ -1219,7 +1224,7 @@
if (mn10300_can_use_rets_insn ())
emit_jump_insn (ret_rtx);
else
- emit_jump_insn (gen_return_ret (GEN_INT (size + REG_SAVE_BYTES)));
+ emit_jump_insn (gen_return_ret (GEN_INT (size + reg_save_bytes)));
}
/* Recognize the PARALLEL rtx generated by mn10300_gen_multiple_store().
@@ -1434,7 +1439,10 @@
is the size of the callee register save area. */
if (from == ARG_POINTER_REGNUM)
{
- diff += REG_SAVE_BYTES;
+ unsigned int reg_save_bytes;
+
+ mn10300_get_live_callee_saved_regs (& reg_save_bytes);
+ diff += reg_save_bytes;
diff += 4 * fp_regs_to_save ();
}
