* Re: [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants
@ 2015-05-12 15:47 Uros Bizjak
2015-05-13 8:39 ` Ilya Enkovich
0 siblings, 1 reply; 5+ messages in thread
From: Uros Bizjak @ 2015-05-12 15:47 UTC (permalink / raw)
To: gcc-patches
Cc: Илья
Энкович
Hello!
>> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
>>
>> PR target/65103
>> * config/i386/i386.c (ix86_rtx_costs): We want to propagate
>> link time constants into adress expressions and therefore set
>> their cost to 0.
>>
>> gcc/testsuite/
>>
>> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
>>
>> PR target/65103
>> * gcc.target/i386/pr65103-3.c: New.
>>
>>
>> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
>> index b3971b8..341a157 100644
>> --- a/gcc/config/i386/i386.c
>> +++ b/gcc/config/i386/i386.c
>> @@ -42039,7 +42039,8 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
>> && !(TARGET_64BIT
>> && (GET_CODE (x) == LABEL_REF
>> || (GET_CODE (x) == SYMBOL_REF
>> - && SYMBOL_REF_LOCAL_P (x)))))
>> + && SYMBOL_REF_LOCAL_P (x))))
>> + && (TARGET_64BIT || GET_CODE (x) != CONST))
Please also add a one-line comment for the new condition.
OK with this change.
Uros.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants
2015-05-12 15:47 [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants Uros Bizjak
@ 2015-05-13 8:39 ` Ilya Enkovich
0 siblings, 0 replies; 5+ messages in thread
From: Ilya Enkovich @ 2015-05-13 8:39 UTC (permalink / raw)
To: Uros Bizjak; +Cc: gcc-patches
On 12 May 17:33, Uros Bizjak wrote:
> Hello!
>
> >> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
> >>
> >> PR target/65103
> >> * config/i386/i386.c (ix86_rtx_costs): We want to propagate
> >> link time constants into adress expressions and therefore set
> >> their cost to 0.
> >>
> >> gcc/testsuite/
> >>
> >> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
> >>
> >> PR target/65103
> >> * gcc.target/i386/pr65103-3.c: New.
> >>
> >>
> >> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> >> index b3971b8..341a157 100644
> >> --- a/gcc/config/i386/i386.c
> >> +++ b/gcc/config/i386/i386.c
> >> @@ -42039,7 +42039,8 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
> >> && !(TARGET_64BIT
> >> && (GET_CODE (x) == LABEL_REF
> >> || (GET_CODE (x) == SYMBOL_REF
> >> - && SYMBOL_REF_LOCAL_P (x)))))
> >> + && SYMBOL_REF_LOCAL_P (x))))
> >> + && (TARGET_64BIT || GET_CODE (x) != CONST))
>
> Please also add a one-line comment for the new condition.
>
> OK with this change.
>
> Uros.
Thanks! Here is committed version.
Ilya
--
gcc/
2015-05-13 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* config/i386/i386.c (ix86_rtx_costs): We want to propagate
link time constants into adress expressions and therefore set
their cost to 0.
gcc/testsuite/
2015-05-13 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* gcc.target/i386/pr65103-3.c: New.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index fd52d89..d739f89 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -42005,7 +42005,9 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
&& !(TARGET_64BIT
&& (GET_CODE (x) == LABEL_REF
|| (GET_CODE (x) == SYMBOL_REF
- && SYMBOL_REF_LOCAL_P (x)))))
+ && SYMBOL_REF_LOCAL_P (x))))
+ /* Use 0 cost for CONST to improve its propagation. */
+ && (TARGET_64BIT || GET_CODE (x) != CONST))
*total = 1;
else
*total = 0;
diff --git a/gcc/testsuite/gcc.target/i386/pr65103-3.c b/gcc/testsuite/gcc.target/i386/pr65103-3.c
new file mode 100644
index 0000000..eddf20b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr65103-3.c
@@ -0,0 +1,19 @@
+/* { dg-do compile { target ia32 } } */
+/* { dg-require-effective-target pie } */
+/* { dg-options "-O2 -fPIE" } */
+/* { dg-final { scan-assembler-not "GOTOFF," } } */
+
+static int *data[100];
+
+void test (long a, long b)
+{
+ do
+ {
+ if( a < b )
+ {
+ data[a] = data[b];
+ a++;
+ }
+ }
+ while (a <= b);
+}
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants
2015-03-10 16:14 ` Ilya Enkovich
@ 2015-05-12 14:43 ` Ilya Enkovich
0 siblings, 0 replies; 5+ messages in thread
From: Ilya Enkovich @ 2015-05-12 14:43 UTC (permalink / raw)
To: gcc-patches
Ping
2015-03-10 19:14 GMT+03:00 Ilya Enkovich <enkovich.gnu@gmail.com>:
> On 10 Mar 18:11, Ilya Enkovich wrote:
>> Hi,
>>
>> This patch changes rtx cost for address constants to 0 similar to what we have in address cost hook beacuse we expect them to be propagated into address expression anyway. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk or stage 1?
>>
>> Thanks,
>> Ilya
>
> Updated ChangeLog and test.
>
> Thanks,
> Ilya
> --
> gcc/
>
> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
>
> PR target/65103
> * config/i386/i386.c (ix86_rtx_costs): We want to propagate
> link time constants into adress expressions and therefore set
> their cost to 0.
>
> gcc/testsuite/
>
> 2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
>
> PR target/65103
> * gcc.target/i386/pr65103-3.c: New.
>
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index b3971b8..341a157 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -42039,7 +42039,8 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
> && !(TARGET_64BIT
> && (GET_CODE (x) == LABEL_REF
> || (GET_CODE (x) == SYMBOL_REF
> - && SYMBOL_REF_LOCAL_P (x)))))
> + && SYMBOL_REF_LOCAL_P (x))))
> + && (TARGET_64BIT || GET_CODE (x) != CONST))
> *total = 1;
> else
> *total = 0;
> diff --git a/gcc/testsuite/gcc.target/i386/pr65103-3.c b/gcc/testsuite/gcc.target/i386/pr65103-3.c
> new file mode 100644
> index 0000000..eddf20b
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr65103-3.c
> @@ -0,0 +1,19 @@
> +/* { dg-do compile { target ia32 } } */
> +/* { dg-require-effective-target pie } */
> +/* { dg-options "-O2 -fPIE" } */
> +/* { dg-final { scan-assembler-not "GOTOFF," } } */
> +
> +static int *data[100];
> +
> +void test (long a, long b)
> +{
> + do
> + {
> + if( a < b )
> + {
> + data[a] = data[b];
> + a++;
> + }
> + }
> + while (a <= b);
> +}
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants
2015-03-10 15:11 Ilya Enkovich
@ 2015-03-10 16:14 ` Ilya Enkovich
2015-05-12 14:43 ` Ilya Enkovich
0 siblings, 1 reply; 5+ messages in thread
From: Ilya Enkovich @ 2015-03-10 16:14 UTC (permalink / raw)
To: gcc-patches
On 10 Mar 18:11, Ilya Enkovich wrote:
> Hi,
>
> This patch changes rtx cost for address constants to 0 similar to what we have in address cost hook beacuse we expect them to be propagated into address expression anyway. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk or stage 1?
>
> Thanks,
> Ilya
Updated ChangeLog and test.
Thanks,
Ilya
--
gcc/
2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* config/i386/i386.c (ix86_rtx_costs): We want to propagate
link time constants into adress expressions and therefore set
their cost to 0.
gcc/testsuite/
2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* gcc.target/i386/pr65103-3.c: New.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index b3971b8..341a157 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -42039,7 +42039,8 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
&& !(TARGET_64BIT
&& (GET_CODE (x) == LABEL_REF
|| (GET_CODE (x) == SYMBOL_REF
- && SYMBOL_REF_LOCAL_P (x)))))
+ && SYMBOL_REF_LOCAL_P (x))))
+ && (TARGET_64BIT || GET_CODE (x) != CONST))
*total = 1;
else
*total = 0;
diff --git a/gcc/testsuite/gcc.target/i386/pr65103-3.c b/gcc/testsuite/gcc.target/i386/pr65103-3.c
new file mode 100644
index 0000000..eddf20b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr65103-3.c
@@ -0,0 +1,19 @@
+/* { dg-do compile { target ia32 } } */
+/* { dg-require-effective-target pie } */
+/* { dg-options "-O2 -fPIE" } */
+/* { dg-final { scan-assembler-not "GOTOFF," } } */
+
+static int *data[100];
+
+void test (long a, long b)
+{
+ do
+ {
+ if( a < b )
+ {
+ data[a] = data[b];
+ a++;
+ }
+ }
+ while (a <= b);
+}
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants
@ 2015-03-10 15:11 Ilya Enkovich
2015-03-10 16:14 ` Ilya Enkovich
0 siblings, 1 reply; 5+ messages in thread
From: Ilya Enkovich @ 2015-03-10 15:11 UTC (permalink / raw)
To: gcc-patches
Hi,
This patch changes rtx cost for address constants to 0 similar to what we have in address cost hook beacuse we expect them to be propagated into address expression anyway. Bootstrapped and tested on x86_64-unknown-linux-gnu. OK for trunk or stage 1?
Thanks,
Ilya
--
gcc/
2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* gcc/config/i386/i386.c (ix86_rtx_costs): We want to propagate
link time constants into adress expressions and therefore set
their cost to 0.
gcc/testsuite/
2015-03-10 Ilya Enkovich <ilya.enkovich@intel.com>
PR target/65103
* gcc.target/i386/pr65103-3.c: New.
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index b3971b8..341a157 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -42039,7 +42039,8 @@ ix86_rtx_costs (rtx x, int code_i, int outer_code_i, int opno, int *total,
&& !(TARGET_64BIT
&& (GET_CODE (x) == LABEL_REF
|| (GET_CODE (x) == SYMBOL_REF
- && SYMBOL_REF_LOCAL_P (x)))))
+ && SYMBOL_REF_LOCAL_P (x))))
+ && (TARGET_64BIT || GET_CODE (x) != CONST))
*total = 1;
else
*total = 0;
diff --git a/gcc/testsuite/gcc.target/i386/pr65103-3.c b/gcc/testsuite/gcc.target/i386/pr65103-3.c
new file mode 100644
index 0000000..1481bf9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/pr65103-3.c
@@ -0,0 +1,18 @@
+/* { dg-do compile { target ia32 } } */
+/* { dg-options "-m32 -O2 -fPIE" } */
+/* { dg-final { scan-assembler-not "GOTOFF," } } */
+
+static int *data[100];
+
+void test (long a, long b)
+{
+ do
+ {
+ if( a < b )
+ {
+ data[a] = data[b];
+ a++;
+ }
+ }
+ while (a <= b);
+}
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2015-05-13 8:23 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-05-12 15:47 [PATCH, PR target/65103, 3/3] Change rtx cost for i386 address constants Uros Bizjak
2015-05-13 8:39 ` Ilya Enkovich
-- strict thread matches above, loose matches on Subject: below --
2015-03-10 15:11 Ilya Enkovich
2015-03-10 16:14 ` Ilya Enkovich
2015-05-12 14:43 ` Ilya Enkovich
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).