* [SH, committed] PR 54236 add another addc case
@ 2014-05-21 8:10 Oleg Endo
0 siblings, 0 replies; only message in thread
From: Oleg Endo @ 2014-05-21 8:10 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 519 bytes --]
Hi,
The attached patch adds another addc pattern in order to catch 'reg + T'
opportunities.
Tested on rev 210624 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.
Committed as rev 210682.
Cheers,
Oleg
gcc/ChangeLog:
PR target/54236
* config/sh/sh.md (*addc_r_1): Rename to addc_t_r. Remove empty
constraints.
(*addc_r_t): Add new insn_and_split.
testsuite/ChangeLog:
PR target/54236
* gcc.target/sh/pr54236-3.c: New.
[-- Attachment #2: sh_pr54236_5.patch --]
[-- Type: text/x-patch, Size: 2858 bytes --]
Index: gcc/testsuite/gcc.target/sh/pr54236-3.c
===================================================================
--- gcc/testsuite/gcc.target/sh/pr54236-3.c (revision 0)
+++ gcc/testsuite/gcc.target/sh/pr54236-3.c (revision 0)
@@ -0,0 +1,31 @@
+/* Tests to check the utilization of the addc and subc instructions.
+ If everything works as expected we won't see any movt instructions in
+ these cases. */
+/* { dg-do compile } */
+/* { dg-options "-O1" } */
+/* { dg-skip-if "" { "sh*-*-*" } { "-m5*"} { "" } } */
+/* { dg-final { scan-assembler-times "addc" 1 } } */
+/* { dg-final { scan-assembler-times "subc" 1 } } */
+/* { dg-final { scan-assembler-not "movt" } } */
+
+int
+test_000 (int* x, unsigned int c)
+{
+ /* 1x addc */
+ int s = 0;
+ unsigned int i;
+ for (i = 0; i < c; ++i)
+ s += ! (x[i] & 0x3000);
+ return s;
+}
+
+int
+test_001 (int* x, unsigned int c)
+{
+ /* 1x subc */
+ int s = 0;
+ unsigned int i;
+ for (i = 0; i < c; ++i)
+ s -= ! (x[i] & 0x3000);
+ return s;
+}
Index: gcc/config/sh/sh.md
===================================================================
--- gcc/config/sh/sh.md (revision 205675)
+++ gcc/config/sh/sh.md (working copy)
@@ -1830,6 +1830,8 @@
;; We allow a reg or 0 for one of the operands in order to be able to
;; do 'reg + T' sequences. Reload will load the constant 0 into the reg
;; as needed.
+;; FIXME: The load of constant 0 should be split out before reload, or else
+;; it will be difficult to hoist or combine the constant load.
(define_insn "*addc"
[(set (match_operand:SI 0 "arith_reg_dest" "=r")
(plus:SI (plus:SI (match_operand:SI 1 "arith_reg_operand" "%0")
@@ -1898,10 +1900,10 @@
;; can be scheduled much better since the load of the constant can be
;; done earlier, before any comparison insns that store the result in
;; the T bit.
-(define_insn_and_split "*addc_r_1"
- [(set (match_operand:SI 0 "arith_reg_dest" "")
- (plus:SI (match_operand:SI 1 "t_reg_operand" "")
- (match_operand:SI 2 "arith_reg_operand" "")))
+(define_insn_and_split "*addc_t_r"
+ [(set (match_operand:SI 0 "arith_reg_dest")
+ (plus:SI (match_operand:SI 1 "t_reg_operand")
+ (match_operand:SI 2 "arith_reg_operand")))
(clobber (reg:SI T_REG))]
"TARGET_SH1"
"#"
@@ -1911,6 +1913,19 @@
(match_dup 1)))
(clobber (reg:SI T_REG))])])
+(define_insn_and_split "*addc_r_t"
+ [(set (match_operand:SI 0 "arith_reg_dest")
+ (plus:SI (match_operand:SI 1 "arith_reg_operand")
+ (match_operand:SI 2 "t_reg_operand")))
+ (clobber (reg:SI T_REG))]
+ "TARGET_SH1"
+ "#"
+ "&& 1"
+ [(parallel [(set (match_dup 0)
+ (plus:SI (plus:SI (match_dup 1) (const_int 0))
+ (match_dup 2)))
+ (clobber (reg:SI T_REG))])])
+
;; Use shlr-addc to do 'reg + (reg & 1)'.
(define_insn_and_split "*addc_r_lsb"
[(set (match_operand:SI 0 "arith_reg_dest")
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2014-05-21 8:10 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-05-21 8:10 [SH, committed] PR 54236 add another addc case Oleg Endo
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).