From: Evandro Menezes <e.menezes@samsung.com>
To: James Greenhalgh <james.greenhalgh@arm.com>
Cc: 'gcc-patches' <gcc-patches@gcc.gnu.org>,
'Marcus Shawcroft' <Marcus.Shawcroft@arm.com>,
'Kyrill Tkachov' <kyrylo.tkachov@arm.com>,
Andrew Pinski <pinskia@gmail.com>
Subject: Re: [PATCH 1/4][AArch64] Add scheduling and cost models for Exynos M1
Date: Thu, 05 Nov 2015 17:31:00 -0000 [thread overview]
Message-ID: <563B9275.5080207@samsung.com> (raw)
In-Reply-To: <20151105092243.GA31968@arm.com>
[-- Attachment #1: Type: text/plain, Size: 2567 bytes --]
James,
Since other members of the "tune_params" structure were signed integers,
even though negative numbers would make no sense for most either, I
followed the same pattern.
Regardless, here's a patch with unsigned integers as you requested:
[AArch64] Add extra tuning parameters for target processors
2015-11-05 Evandro Menezes <e.menezes@samsung.com>
gcc/
* config/aarch64/aarch64-protos.h (tune_params): Add new members
"max_case_values" and "cache_line_size".
* config/aarch64/aarch64.c (aarch64_case_values_threshold): New
function.
(aarch64_override_options_internal): Tune heuristics based on new
members in "tune_params".
(TARGET_CASE_VALUES_THRESHOLD): Define macro.
Please, commit if it's alright.
Thank you,
--
Evandro Menezes
On 11/05/2015 03:22 AM, James Greenhalgh wrote:
> On Wed, Nov 04, 2015 at 05:21:03PM -0600, Evandro Menezes wrote:
>> Please, ignore the previous patch. This is the intended patch.
>>
>> Sorry.
>>
>> --
>> Evandro Menezes
>>
>> On 11/04/2015 05:18 PM, Evandro Menezes wrote:
>>> This patch adds extra tuning information about AArch64 targets:
>>>
>>> * Maximum number of case values before resorting to a jump table
>>> The default values assumed independently of the specific backends
>>> may be rather low for modern processors, which sport quite efficient
>>> direct branch prediction, whereas indirect branch prediction is
>>> still typically not so efficient. This value may be specifically
>>> set for a processor or left at zero to use the default values.
>>> * L1 cache line size
>>> The auto-prefetcher uses this information when emitting software
>>> prefetch insns.
>>>
>>> Please, commit if it's alright.
>>>
>>> Thank you,
>>>
> Thanks for the patch,
>
>> diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
>> index 81792bc..ecf4685 100644
>> --- a/gcc/config/aarch64/aarch64-protos.h
>> +++ b/gcc/config/aarch64/aarch64-protos.h
>> @@ -195,6 +195,9 @@ struct tune_params
>> int vec_reassoc_width;
>> int min_div_recip_mul_sf;
>> int min_div_recip_mul_df;
>> + int max_case_values; /* Case values threshold; or 0 for the default. */
> If we're using an int, how about -1 as the sentinel value? (Maybe someone
> really likes jump tables!). Otherwise, make this an unsigned int?
>
>> +
>> + int cache_line_size; /* Cache line size; or 0 for the default. */
> unsigned int?
>
> The patch is otherwise OK, though it needs a ChangeLog.
>
> Thanks,
> James
>
>
[-- Attachment #2: target.patch --]
[-- Type: text/x-patch, Size: 4631 bytes --]
diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
index 81792bc..ecf4685 100644
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -195,6 +195,9 @@ struct tune_params
int vec_reassoc_width;
int min_div_recip_mul_sf;
int min_div_recip_mul_df;
+ unsigned int max_case_values; /* Case values threshold; or 0 for the default. */
+
+ unsigned int cache_line_size; /* Cache line size; or 0 for the default. */
/* An enum specifying how to take into account CPU autoprefetch capabilities
during instruction scheduling:
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index 5c8604f..e7f1c07 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -355,6 +355,8 @@ static const struct tune_params generic_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NONE) /* tune_flags. */
};
@@ -378,6 +380,8 @@ static const struct tune_params cortexa53_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_WEAK, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NONE) /* tune_flags. */
};
@@ -401,6 +405,8 @@ static const struct tune_params cortexa57_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_WEAK, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_RENAME_FMA_REGS) /* tune_flags. */
};
@@ -424,6 +430,8 @@ static const struct tune_params cortexa72_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NONE) /* tune_flags. */
};
@@ -446,6 +454,8 @@ static const struct tune_params thunderx_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NONE) /* tune_flags. */
};
@@ -468,6 +478,8 @@ static const struct tune_params xgene1_tunings =
1, /* vec_reassoc_width. */
2, /* min_div_recip_mul_sf. */
2, /* min_div_recip_mul_df. */
+ 0, /* max_case_values. */
+ 0, /* cache_line_size. */
tune_params::AUTOPREFETCHER_OFF, /* autoprefetcher_model. */
(AARCH64_EXTRA_TUNE_NONE) /* tune_flags. */
};
@@ -3242,6 +3254,20 @@ aarch64_cannot_force_const_mem (machine_mode mode ATTRIBUTE_UNUSED, rtx x)
return aarch64_tls_referenced_p (x);
}
+/* Implement TARGET_CASE_VALUES_THRESHOLD. */
+
+static unsigned int
+aarch64_case_values_threshold (void)
+{
+ /* Use the specified limit for the number of cases before using jump
+ tables at higher optimization levels. */
+ if (optimize > 2
+ && selected_cpu->tune->max_case_values != 0)
+ return selected_cpu->tune->max_case_values;
+ else
+ return default_case_values_threshold ();
+}
+
/* Return true if register REGNO is a valid index register.
STRICT_P is true if REG_OK_STRICT is in effect. */
@@ -7672,6 +7698,13 @@ aarch64_override_options_internal (struct gcc_options *opts)
opts->x_param_values,
global_options_set.x_param_values);
+ /* Set the L1 cache line size. */
+ if (selected_cpu->tune->cache_line_size != 0)
+ maybe_set_param_value (PARAM_L1_CACHE_LINE_SIZE,
+ selected_cpu->tune->cache_line_size,
+ opts->x_param_values,
+ global_options_set.x_param_values);
+
aarch64_override_options_after_change_1 (opts);
}
@@ -13385,6 +13418,7 @@ aarch64_promoted_type (const_tree t)
return float_type_node;
return NULL_TREE;
}
+
#undef TARGET_ADDRESS_COST
#define TARGET_ADDRESS_COST aarch64_address_cost
@@ -13432,6 +13466,9 @@ aarch64_promoted_type (const_tree t)
#undef TARGET_CANNOT_FORCE_CONST_MEM
#define TARGET_CANNOT_FORCE_CONST_MEM aarch64_cannot_force_const_mem
+#undef TARGET_CASE_VALUES_THRESHOLD
+#define TARGET_CASE_VALUES_THRESHOLD aarch64_case_values_threshold
+
#undef TARGET_CONDITIONAL_REGISTER_USAGE
#define TARGET_CONDITIONAL_REGISTER_USAGE aarch64_conditional_register_usage
next prev parent reply other threads:[~2015-11-05 17:31 UTC|newest]
Thread overview: 52+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-10-27 23:38 [AArch64] " Evandro Menezes
2015-10-28 10:40 ` James Greenhalgh
2015-10-28 10:45 ` Andrew Pinski
2015-10-28 21:58 ` Evandro Menezes
2015-10-29 23:02 ` Evandro Menezes
2015-10-28 11:01 ` Kyrill Tkachov
2015-10-29 19:38 ` Evandro Menezes
2015-11-04 23:10 ` [PATCH 0/4][AArch64] " Evandro Menezes
2015-11-04 23:18 ` [PATCH 1/4][AArch64] " Evandro Menezes
2015-11-04 23:21 ` Evandro Menezes
2015-11-05 9:22 ` James Greenhalgh
2015-11-05 17:31 ` Evandro Menezes [this message]
2015-11-12 14:47 ` James Greenhalgh
2015-11-05 20:51 ` [PATCH 2/4][AArch64] Increase the loop peeling limit Evandro Menezes
2015-11-19 22:04 ` Evandro Menezes
2015-11-20 11:53 ` James Greenhalgh
2015-12-03 21:07 ` Evandro Menezes
2015-12-14 11:26 ` James Greenhalgh
2015-12-15 23:34 ` Evandro Menezes
2015-12-16 11:24 ` Richard Earnshaw (lists)
2015-12-16 12:42 ` Richard Biener
2015-12-16 20:11 ` Evandro Menezes
2016-01-08 22:55 ` Evandro Menezes
2016-02-03 19:46 ` Evandro Menezes
2016-03-16 19:48 ` Evandro Menezes
2015-11-05 23:30 ` [PATCH 3/4][AArch64] Add scheduling model for Exynos M1 Evandro Menezes
2015-11-05 23:30 ` Evandro Menezes
2015-11-09 23:06 ` Evandro Menezes
2015-11-10 17:50 ` [PATCH 3a/4][AArch64] Add attribute for compatibility with ARM pipeline models Evandro Menezes
2015-11-10 18:01 ` Ramana Radhakrishnan
2015-11-10 18:03 ` Ramana Radhakrishnan
2015-11-12 14:55 ` James Greenhalgh
2015-11-12 15:39 ` Evandro Menezes
2015-11-12 17:32 ` Evandro Menezes
2015-11-19 22:05 ` Evandro Menezes
2015-11-20 12:27 ` James Greenhalgh
2015-11-20 14:34 ` Kyrill Tkachov
2015-11-20 15:56 ` Evandro Menezes
2015-11-20 15:55 ` Evandro Menezes
2015-11-20 16:16 ` James Greenhalgh
2015-11-10 17:54 ` [PATCH 3b/4][AArch64] Add scheduling model for Exynos M1 Evandro Menezes
2015-11-19 22:06 ` Evandro Menezes
2015-11-20 17:17 ` James Greenhalgh
2015-11-20 22:07 ` Evandro Menezes
2015-12-03 20:58 ` Evandro Menezes
2015-12-04 9:25 ` Kyrill Tkachov
2015-12-07 19:55 ` Evandro Menezes
2015-11-06 0:09 ` [PATCH 4/4][AArch64] Add cost " Evandro Menezes
2015-11-19 22:06 ` Evandro Menezes
2015-11-20 17:19 ` James Greenhalgh
2015-11-24 9:56 ` Kyrill Tkachov
2015-12-03 20:49 ` Evandro Menezes
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=563B9275.5080207@samsung.com \
--to=e.menezes@samsung.com \
--cc=Marcus.Shawcroft@arm.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=james.greenhalgh@arm.com \
--cc=kyrylo.tkachov@arm.com \
--cc=pinskia@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).