Hello,
I somehow initially forgot to implement DImode GBR based loads/stores.
Attached patch does that and also fixes a problem with the GBR address
mode optimization.
Tested on rev 192417 with
make -k check RUNTESTFLAGS="--target_board=sh-sim
\{-m2/-ml,-m2/-mb,-m2a/-mb,-m4/-ml,-m4/-mb,-m4a/-ml,-m4a/-mb}"
and no new failures.
OK?
Cheers,
Oleg
gcc/ChangeLog:
PR target/54760
* config/sh/sh.c (sh_find_base_reg_disp): Stop searching insns
when hitting a call insn if GBR is marked as call used.
* config/sh/iterators.md (QIHISIDI): New mode iterator.
* config/sh/predicates.md (gbr_address_mem): New predicate.
* config/sh/sh.md (*movdi_gbr_load, *movdi_gbr_store): New
insn_and_split.
Use QIHISIDI instead of QIHISI in unnamed GBR addressing splits.
testsuite/ChangeLog:
PR target/54760
* gcc.target/sh/pr54760-2.c: Add long long and unsigned long
long test functions.
* gcc.target/sh/pr54760-4.c: New.
Index: gcc/config/sh/sh.c
===================================================================
--- gcc/config/sh/sh.c (revision 192417)
+++ gcc/config/sh/sh.c (working copy)
@@ -13383,6 +13383,10 @@
for (rtx i = prev_nonnote_insn (insn); i != NULL;
i = prev_nonnote_insn (i))
{
+ if (REGNO_REG_SET_P (regs_invalidated_by_call_regset, GBR_REG)
+ && CALL_P (i))
+ break;
+
if (!NONJUMP_INSN_P (i))
continue;
Index: gcc/config/sh/sh.md
===================================================================
--- gcc/config/sh/sh.md (revision 192417)
+++ gcc/config/sh/sh.md (working copy)
@@ -10277,6 +10277,47 @@
"mov.<bwl> %0,@(0,gbr)"
[(set_attr "type" "store")])
+;; DImode memory accesses have to be split in two SImode accesses.
+;; Split them before reload, so that it gets a better chance to figure out
+;; how to deal with the R0 restriction for the individual SImode accesses.
+;; Do not match this insn during or after reload because it can't be split
+;; afterwards.
+(define_insn_and_split "*movdi_gbr_load"
+ [(set (match_operand:DI 0 "register_operand")
+ (match_operand:DI 1 "gbr_address_mem"))]
+ "TARGET_SH1 && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(set (match_dup 3) (match_dup 5))
+ (set (match_dup 4) (match_dup 6))]
+{
+ /* Swap low/high part load order on little endian, so that the result reg
+ of the second load can be used better. */
+ int off = TARGET_LITTLE_ENDIAN ? 1 : 0;
+ operands[3 + off] = gen_lowpart (SImode, operands[0]);
+ operands[5 + off] = gen_lowpart (SImode, operands[1]);
+ operands[4 - off] = gen_highpart (SImode, operands[0]);
+ operands[6 - off] = gen_highpart (SImode, operands[1]);
+})
+
+(define_insn_and_split "*movdi_gbr_store"
+ [(set (match_operand:DI 0 "gbr_address_mem")
+ (match_operand:DI 1 "register_operand"))]
+ "TARGET_SH1 && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(set (match_dup 3) (match_dup 5))
+ (set (match_dup 4) (match_dup 6))]
+{
+ /* Swap low/high part store order on big endian, so that stores of function
+ call results can save a reg copy. */
+ int off = TARGET_LITTLE_ENDIAN ? 0 : 1;
+ operands[3 + off] = gen_lowpart (SImode, operands[0]);
+ operands[5 + off] = gen_lowpart (SImode, operands[1]);
+ operands[4 - off] = gen_highpart (SImode, operands[0]);
+ operands[6 - off] = gen_highpart (SImode, operands[1]);
+})
+
;; Sometimes memory accesses do not get combined with the store_gbr insn,
;; in particular when the displacements are in the range of the regular move
;; insns. Thus, in the first split pass after the combine pass we search
@@ -10287,15 +10328,15 @@
;; other operand) and there's no point of doing it if the GBR is not
;; referenced in a function at all.
(define_split
- [(set (match_operand:QIHISI 0 "register_operand")
- (match_operand:QIHISI 1 "memory_operand"))]
+ [(set (match_operand:QIHISIDI 0 "register_operand")
+ (match_operand:QIHISIDI 1 "memory_operand"))]
"TARGET_SH1 && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 0) (match_dup 1))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[1]);
if (gbr_mem != NULL_RTX)
- operands[1] = change_address (operands[1], GET_MODE (operands[1]), gbr_mem);
+ operands[1] = replace_equiv_address (operands[1], gbr_mem);
else
FAIL;
})
@@ -10309,7 +10350,7 @@
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[1]);
if (gbr_mem != NULL_RTX)
- operands[1] = change_address (operands[1], GET_MODE (operands[1]), gbr_mem);
+ operands[1] = replace_equiv_address (operands[1], gbr_mem);
else
FAIL;
})
@@ -10328,23 +10369,22 @@
if (gbr_mem != NULL_RTX)
{
operands[2] = gen_reg_rtx (GET_MODE (operands[1]));
- operands[1] = change_address (operands[1], GET_MODE (operands[1]),
- gbr_mem);
+ operands[1] = replace_equiv_address (operands[1], gbr_mem);
}
else
FAIL;
})
(define_split
- [(set (match_operand:QIHISI 0 "memory_operand")
- (match_operand:QIHISI 1 "register_operand"))]
+ [(set (match_operand:QIHISIDI 0 "memory_operand")
+ (match_operand:QIHISIDI 1 "register_operand"))]
"TARGET_SH1 && !reload_in_progress && !reload_completed
&& df_regs_ever_live_p (GBR_REG)"
[(set (match_dup 0) (match_dup 1))]
{
rtx gbr_mem = sh_find_equiv_gbr_addr (curr_insn, operands[0]);
if (gbr_mem != NULL_RTX)
- operands[0] = change_address (operands[0], GET_MODE (operands[0]), gbr_mem);
+ operands[0] = replace_equiv_address (operands[0], gbr_mem);
else
FAIL;
})
Index: gcc/config/sh/iterators.md
===================================================================
--- gcc/config/sh/iterators.md (revision 192417)
+++ gcc/config/sh/iterators.md (working copy)
@@ -18,6 +18,7 @@
;; along with GCC; see the file COPYING3. If not see
;; <http://www.gnu.org/licenses/>.
+(define_mode_iterator QIHISIDI [QI HI SI DI])
(define_mode_iterator QIHISI [QI HI SI])
(define_mode_iterator QIHI [QI HI])
(define_mode_iterator HISI [HI SI])
Index: gcc/config/sh/predicates.md
===================================================================
--- gcc/config/sh/predicates.md (revision 192417)
+++ gcc/config/sh/predicates.md (working copy)
@@ -1139,3 +1139,20 @@
return INTVAL (op) >= 0 && INTVAL (op) <= max_disp;
})
+
+;; A predicate that determines whether OP is a valid GBR addressing mode
+;; memory reference.
+(define_predicate "gbr_address_mem"
+ (match_code "mem")
+{
+ rtx addr = XEXP (op, 0);
+
+ if (REG_P (addr) && REGNO (addr) == GBR_REG)
+ return true;
+ if (GET_CODE (addr) == PLUS
+ && REG_P (XEXP (addr, 0)) && REGNO (XEXP (addr, 0)) == GBR_REG
+ && gbr_displacement (XEXP (addr, 1), mode))
+ return true;
+
+ return false;
+})
Index: gcc/testsuite/gcc.target/sh/pr54760-2.c
===================================================================
--- gcc/testsuite/gcc.target/sh/pr54760-2.c (revision 192417)
+++ gcc/testsuite/gcc.target/sh/pr54760-2.c (working copy)
@@ -9,107 +9,129 @@
/* ---------------------------------------------------------------------------
Simple GBR load.
*/
-#define func(name, type, disp)\
- int \
+#define func(name, rettype, type, disp)\
+ rettype \
name ## _tp_load (void) \
{ \
type* tp = (type*)__builtin_thread_pointer (); \
return tp[disp]; \
}
-func (test00, int, 0)
-func (test01, int, 5)
-func (test02, int, 255)
+func (test00, int, int, 0)
+func (test01, int, int, 5)
+func (test02, int, int, 255)
-func (test03, short, 0)
-func (test04, short, 5)
-func (test05, short, 255)
+func (test03, int, short, 0)
+func (test04, int, short, 5)
+func (test05, int, short, 255)
-func (test06, char, 0)
-func (test07, char, 5)
-func (test08, char, 255)
+func (test06, int, char, 0)
+func (test07, int, char, 5)
+func (test08, int, char, 255)
-func (test09, unsigned int, 0)
-func (test10, unsigned int, 5)
-func (test11, unsigned int, 255)
+func (test09, int, unsigned int, 0)
+func (test10, int, unsigned int, 5)
+func (test11, int, unsigned int, 255)
-func (test12, unsigned short, 0)
-func (test13, unsigned short, 5)
-func (test14, unsigned short, 255)
+func (test12, int, unsigned short, 0)
+func (test13, int, unsigned short, 5)
+func (test14, int, unsigned short, 255)
-func (test15, unsigned char, 0)
-func (test16, unsigned char, 5)
-func (test17, unsigned char, 255)
+func (test15, int, unsigned char, 0)
+func (test16, int, unsigned char, 5)
+func (test17, int, unsigned char, 255)
+func (test18, long long, long long, 0)
+func (test19, long long, long long, 5)
+func (test20, long long, long long, 127)
+
+func (test21, long long, unsigned long long, 0)
+func (test22, long long, unsigned long long, 5)
+func (test23, long long, unsigned long long, 127)
+
#undef func
/* ---------------------------------------------------------------------------
Simple GBR store.
*/
-#define func(name, type, disp)\
+#define func(name, argtype, type, disp)\
void \
- name ## _tp_store (int a) \
+ name ## _tp_store (argtype a) \
{ \
type* tp = (type*)__builtin_thread_pointer (); \
tp[disp] = (type)a; \
}
-func (test00, int, 0)
-func (test01, int, 5)
-func (test02, int, 255)
+func (test00, int, int, 0)
+func (test01, int, int, 5)
+func (test02, int, int, 255)
-func (test03, short, 0)
-func (test04, short, 5)
-func (test05, short, 255)
+func (test03, int, short, 0)
+func (test04, int, short, 5)
+func (test05, int, short, 255)
-func (test06, char, 0)
-func (test07, char, 5)
-func (test08, char, 255)
+func (test06, int, char, 0)
+func (test07, int, char, 5)
+func (test08, int, char, 255)
-func (test09, unsigned int, 0)
-func (test10, unsigned int, 5)
-func (test11, unsigned int, 255)
+func (test09, int, unsigned int, 0)
+func (test10, int, unsigned int, 5)
+func (test11, int, unsigned int, 255)
-func (test12, unsigned short, 0)
-func (test13, unsigned short, 5)
-func (test14, unsigned short, 255)
+func (test12, int, unsigned short, 0)
+func (test13, int, unsigned short, 5)
+func (test14, int, unsigned short, 255)
-func (test15, unsigned char, 0)
-func (test16, unsigned char, 5)
-func (test17, unsigned char, 255)
+func (test15, int, unsigned char, 0)
+func (test16, int, unsigned char, 5)
+func (test17, int, unsigned char, 255)
+func (test18, long long, long long, 0)
+func (test19, long long, long long, 5)
+func (test20, long long, long long, 127)
+
+func (test21, long long, unsigned long long, 0)
+func (test22, long long, unsigned long long, 5)
+func (test23, long long, unsigned long long, 127)
+
#undef func
/* ---------------------------------------------------------------------------
Arithmetic on the result of a GBR load.
*/
-#define func(name, type, disp, op, opname)\
- int \
- name ## _tp_load_arith_ ##opname (int a) \
+#define func(name, retargtype, type, disp, op, opname)\
+ retargtype \
+ name ## _tp_load_arith_ ##opname (retargtype a) \
{ \
type* tp = (type*)__builtin_thread_pointer (); \
return tp[disp] op a; \
}
#define funcs(op, opname) \
- func (test00, int, 0, op, opname) \
- func (test01, int, 5, op, opname) \
- func (test02, int, 255, op, opname) \
- func (test03, short, 0, op, opname) \
- func (test04, short, 5, op, opname) \
- func (test05, short, 255, op, opname) \
- func (test06, char, 0, op, opname) \
- func (test07, char, 5, op, opname) \
- func (test08, char, 255, op, opname) \
- func (test09, unsigned int, 0, op, opname) \
- func (test10, unsigned int, 5, op, opname) \
- func (test11, unsigned int, 255, op, opname) \
- func (test12, unsigned short, 0, op, opname) \
- func (test13, unsigned short, 5, op, opname) \
- func (test14, unsigned short, 255, op, opname) \
- func (test15, unsigned char, 0, op, opname) \
- func (test16, unsigned char, 5, op, opname) \
- func (test17, unsigned char, 255, op, opname) \
+ func (test00, int, int, 0, op, opname) \
+ func (test01, int, int, 5, op, opname) \
+ func (test02, int, int, 255, op, opname) \
+ func (test03, int, short, 0, op, opname) \
+ func (test04, int, short, 5, op, opname) \
+ func (test05, int, short, 255, op, opname) \
+ func (test06, int, char, 0, op, opname) \
+ func (test07, int, char, 5, op, opname) \
+ func (test08, int, char, 255, op, opname) \
+ func (test09, int, unsigned int, 0, op, opname) \
+ func (test10, int, unsigned int, 5, op, opname) \
+ func (test11, int, unsigned int, 255, op, opname) \
+ func (test12, int, unsigned short, 0, op, opname) \
+ func (test13, int, unsigned short, 5, op, opname) \
+ func (test14, int, unsigned short, 255, op, opname) \
+ func (test15, int, unsigned char, 0, op, opname) \
+ func (test16, int, unsigned char, 5, op, opname) \
+ func (test17, int, unsigned char, 255, op, opname) \
+ func (test18, long long, long long, 0, op, opname) \
+ func (test19, long long, long long, 5, op, opname) \
+ func (test20, long long, long long, 127, op, opname) \
+ func (test21, long long, unsigned long long, 0, op, opname) \
+ func (test22, long long, unsigned long long, 5, op, opname) \
+ func (test23, long long, unsigned long long, 127, op, opname) \
funcs (+, plus)
funcs (-, minus)
@@ -124,8 +146,8 @@
/* ---------------------------------------------------------------------------
Arithmetic of the result of two GBR loads.
*/
-#define func(name, type, disp0, disp1, op, opname)\
- int \
+#define func(name, rettype, type, disp0, disp1, op, opname)\
+ rettype \
name ## _tp_load_load_arith_ ##opname (void) \
{ \
type* tp = (type*)__builtin_thread_pointer (); \
@@ -133,18 +155,22 @@
}
#define funcs(op, opname) \
- func (test00, int, 0, 5, op, opname) \
- func (test02, int, 1, 255, op, opname) \
- func (test03, short, 0, 5, op, opname) \
- func (test05, short, 1, 255, op, opname) \
- func (test06, char, 0, 5, op, opname) \
- func (test08, char, 1, 255, op, opname) \
- func (test09, unsigned int, 0, 5, op, opname) \
- func (test11, unsigned int, 1, 255, op, opname) \
- func (test12, unsigned short, 0, 5, op, opname) \
- func (test14, unsigned short, 1, 255, op, opname) \
- func (test15, unsigned char, 0, 5, op, opname) \
- func (test17, unsigned char, 1, 255, op, opname) \
+ func (test00, int, int, 0, 5, op, opname) \
+ func (test02, int, int, 1, 255, op, opname) \
+ func (test03, int, short, 0, 5, op, opname) \
+ func (test05, int, short, 1, 255, op, opname) \
+ func (test06, int, char, 0, 5, op, opname) \
+ func (test08, int, char, 1, 255, op, opname) \
+ func (test09, int, unsigned int, 0, 5, op, opname) \
+ func (test11, int, unsigned int, 1, 255, op, opname) \
+ func (test12, int, unsigned short, 0, 5, op, opname) \
+ func (test14, int, unsigned short, 1, 255, op, opname) \
+ func (test15, int, unsigned char, 0, 5, op, opname) \
+ func (test17, int, unsigned char, 1, 255, op, opname) \
+ func (test18, long long, long long, 0, 5, op, opname) \
+ func (test19, long long, long long, 1, 127, op, opname) \
+ func (test20, long long, unsigned long long, 0, 5, op, opname) \
+ func (test21, long long, unsigned long long, 1, 127, op, opname) \
funcs (+, plus)
funcs (-, minus)
@@ -180,6 +206,10 @@
func (test14, unsigned short, 1, 255)
func (test15, unsigned char, 0, 5)
func (test17, unsigned char, 1, 255)
+func (test18, long long, 0, 5)
+func (test19, long long, 1, 127)
+func (test20, unsigned long long, 0, 5)
+func (test21, unsigned long long, 1, 127)
#undef func
@@ -187,33 +217,39 @@
GBR load, arithmetic, GBR store
*/
-#define func(name, type, disp, op, opname)\
+#define func(name, argtype, type, disp, op, opname)\
void \
- name ## _tp_load_arith_store_ ##opname (int a) \
+ name ## _tp_load_arith_store_ ##opname (argtype a) \
{ \
type* tp = (type*)__builtin_thread_pointer (); \
tp[disp] op a; \
}
#define funcs(op, opname) \
- func (test00, int, 0, op, opname) \
- func (test01, int, 5, op, opname) \
- func (test02, int, 255, op, opname) \
- func (test03, short, 0, op, opname) \
- func (test04, short, 5, op, opname) \
- func (test05, short, 255, op, opname) \
- func (test06, char, 0, op, opname) \
- func (test07, char, 5, op, opname) \
- func (test08, char, 255, op, opname) \
- func (test09, unsigned int, 0, op, opname) \
- func (test10, unsigned int, 5, op, opname) \
- func (test11, unsigned int, 255, op, opname) \
- func (test12, unsigned short, 0, op, opname) \
- func (test13, unsigned short, 5, op, opname) \
- func (test14, unsigned short, 255, op, opname) \
- func (test15, unsigned char, 0, op, opname) \
- func (test16, unsigned char, 5, op, opname) \
- func (test17, unsigned char, 255, op, opname) \
+ func (test00, int, int, 0, op, opname) \
+ func (test01, int, int, 5, op, opname) \
+ func (test02, int, int, 255, op, opname) \
+ func (test03, int, short, 0, op, opname) \
+ func (test04, int, short, 5, op, opname) \
+ func (test05, int, short, 255, op, opname) \
+ func (test06, int, char, 0, op, opname) \
+ func (test07, int, char, 5, op, opname) \
+ func (test08, int, char, 255, op, opname) \
+ func (test09, int, unsigned int, 0, op, opname) \
+ func (test10, int, unsigned int, 5, op, opname) \
+ func (test11, int, unsigned int, 255, op, opname) \
+ func (test12, int, unsigned short, 0, op, opname) \
+ func (test13, int, unsigned short, 5, op, opname) \
+ func (test14, int, unsigned short, 255, op, opname) \
+ func (test15, int, unsigned char, 0, op, opname) \
+ func (test16, int, unsigned char, 5, op, opname) \
+ func (test17, int, unsigned char, 255, op, opname) \
+ func (test18, long long, long long, 0, op, opname) \
+ func (test19, long long, long long, 5, op, opname) \
+ func (test20, long long, long long, 127, op, opname) \
+ func (test21, long long, unsigned long long, 0, op, opname) \
+ func (test22, long long, unsigned long long, 5, op, opname) \
+ func (test23, long long, unsigned long long, 127, op, opname) \
funcs (+=, plus)
funcs (-=, minus)
Index: gcc/testsuite/gcc.target/sh/pr54760-4.c
===================================================================
--- gcc/testsuite/gcc.target/sh/pr54760-4.c (revision 0)
+++ gcc/testsuite/gcc.target/sh/pr54760-4.c (revision 0)
@@ -0,0 +1,19 @@
+/* Check that the GBR address optimization does not combine a gbr store
+ and its use when a function call is inbetween, when GBR is a call used
+ register, i.e. it is invalidated by function calls. */
+/* { dg-do compile { target "sh*-*-*" } } */
+/* { dg-options "-O1 -fcall-used-gbr" } */
+/* { dg-skip-if "" { "sh*-*-*" } { "-m5*"} { "" } } */
+/* { dg-final { scan-assembler "stc\tgbr" } } */
+
+extern int test00 (void);
+int
+test01 (int x)
+{
+ /* We must see a stc gbr,rn before the function call, because
+ a function call could modify the gbr. In this case the user requests
+ the old gbr value, before the function call. */
+ int* p = (int*)__builtin_thread_pointer ();
+ p[5] = test00 ();
+ return 0;
+}
