This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [SH] Introduce treg_set_expr
- From: Oleg Endo <oleg dot endo at t-online dot de>
- To: gcc-patches at gcc dot gnu dot org
- Date: Tue, 27 Jan 2015 01:02:51 +0100
- Subject: Re: [SH] Introduce treg_set_expr
- Authentication-results: sourceware.org; auth=none
- References: <20150117 dot 224013 dot 482322450 dot kkojima at rr dot iij4u dot or dot jp> <1421583943 dot 2376 dot 19 dot camel at yam-132-YW-E178-FTW> <1421708391 dot 2376 dot 27 dot camel at yam-132-YW-E178-FTW> <20150120 dot 200532 dot 63808742 dot kkojima at rr dot iij4u dot or dot jp> <1421887903 dot 2376 dot 57 dot camel at yam-132-YW-E178-FTW> <1421966761 dot 2376 dot 78 dot camel at yam-132-YW-E178-FTW>
On Thu, 2015-01-22 at 23:46 +0100, Oleg Endo wrote:
> I will install this the patch from
> https://gcc.gnu.org/ml/gcc-patches/2015-01/msg01743.html
> in 24h if there are no further objections.
>
The above mentioned patch has been committed as r220081. This is a
small follow up patch that fixes two oversights. One is that there's no
point in trying to fit a tst insn constant by right shifting it if it
already fits into K08 (currently done when optimizing for size).
The other oversight was a code path that hasn't been triggered yet in
sh_split_treg_set_expr, into which I ran when trying out more stuff with
the treg_set_expr facility.
Attached patch has been committed as r220144.
Cheers,
Oleg
gcc/ChangeLog:
PR target/49263
* config/sh/sh.c (sh_split_treg_set_expr): Invoke emit_insn before
remove_insn.
* config/sh/sh.md (tstsi_t): Don't try to optimize constant with right
shifts if it already fits into K08.
gcc/testsuite/ChangeLog:
PR target/49263
* gcc.target/sh/pr49263-4.c: New.
Index: gcc/config/sh/sh.c
===================================================================
--- gcc/config/sh/sh.c (revision 220133)
+++ gcc/config/sh/sh.c (working copy)
@@ -14466,6 +14466,8 @@
fprintf (dump_file, "trailing nott insn %d\n", INSN_UID (nott_insn));
}
+ emit_insn (insnlist.first);
+
if (nott_insn != NULL && append_nott)
{
if (dump_file)
@@ -14475,8 +14477,6 @@
append_nott = false;
}
- emit_insn (insnlist.first);
-
if (append_nott)
nott_insn = emit_insn (gen_nott (get_t_reg_rtx ()));
Index: gcc/config/sh/sh.md
===================================================================
--- gcc/config/sh/sh.md (revision 220133)
+++ gcc/config/sh/sh.md (working copy)
@@ -742,9 +742,13 @@
variant instead and load the constant into a reg. For that we'd need
to do some analysis. */
- if ((op1val & 0xFFFF) == 0
- && CONST_OK_FOR_K08 (op1val >> 16) && optimize_size)
+ if (CONST_OK_FOR_K08 (op1val))
{
+ /* Do nothing. */
+ }
+ else if ((op1val & 0xFFFF) == 0
+ && CONST_OK_FOR_K08 (op1val >> 16) && optimize_size)
+ {
/* Use a swap.w insn to do a shift + reg copy (to R0) in one insn. */
op1val = op1val >> 16;
rtx r = gen_reg_rtx (SImode);
Index: gcc/testsuite/gcc.target/sh/pr49263-4.c
===================================================================
--- gcc/testsuite/gcc.target/sh/pr49263-4.c (revision 0)
+++ gcc/testsuite/gcc.target/sh/pr49263-4.c (revision 0)
@@ -0,0 +1,10 @@
+/* Verify that TST #imm, R0 instruction is generated if the constant
+ allows it when compiling for -Os. */
+/* { dg-do compile } */
+/* { dg-options "-Os" } */
+/* { dg-final { scan-assembler-not "and" } } */
+/* { dg-final { scan-assembler-not "extu" } } */
+/* { dg-final { scan-assembler-not "exts" } } */
+/* { dg-final { scan-assembler-not "shlr" } } */
+
+#include "pr49263.c"