* [PATCH][AARCH64] Enable compare branch fusion
@ 2019-12-24 15:59 Wilco Dijkstra
2020-01-16 18:03 ` Wilco Dijkstra
2020-01-17 10:02 ` Richard Sandiford
0 siblings, 2 replies; 4+ messages in thread
From: Wilco Dijkstra @ 2019-12-24 15:59 UTC (permalink / raw)
To: GCC Patches; +Cc: Richard Earnshaw, Kyrylo Tkachov, Richard Sandiford
Enable the most basic form of compare-branch fusion since various CPUs
support it. This has no measurable effect on cores which don't support
branch fusion, but increases fusion opportunities on cores which do.
Bootstrapped on AArch64, OK for commit?
ChangeLog:
2019-12-24 Wilco Dijkstra <wdijkstr@arm.com>
* config/aarch64/aarch64.c (generic_tunings): Add branch fusion.
(neoversen1_tunings): Likewise.
--
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index a3b18b381e1748f8fe5e522bdec4f7c850821fe8..1c32a3543bec4031cc9b641973101829c77296b5 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -726,7 +726,7 @@ static const struct tune_params generic_tunings =
SVE_NOT_IMPLEMENTED, /* sve_width */
4, /* memmov_cost */
2, /* issue_rate */
- (AARCH64_FUSE_AES_AESMC), /* fusible_ops */
+ (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
"16:12", /* function_align. */
"4", /* jump_align. */
"8", /* loop_align. */
@@ -1130,7 +1130,7 @@ static const struct tune_params neoversen1_tunings =
SVE_NOT_IMPLEMENTED, /* sve_width */
4, /* memmov_cost */
3, /* issue_rate */
- AARCH64_FUSE_AES_AESMC, /* fusible_ops */
+ (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
"32:16", /* function_align. */
"32:16", /* jump_align. */
"32:16", /* loop_align. */
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH][AARCH64] Enable compare branch fusion
2019-12-24 15:59 [PATCH][AARCH64] Enable compare branch fusion Wilco Dijkstra
@ 2020-01-16 18:03 ` Wilco Dijkstra
2020-01-17 10:02 ` Richard Sandiford
1 sibling, 0 replies; 4+ messages in thread
From: Wilco Dijkstra @ 2020-01-16 18:03 UTC (permalink / raw)
To: GCC Patches; +Cc: Richard Earnshaw, Kyrylo Tkachov, Richard Sandiford
ping
Enable the most basic form of compare-branch fusion since various CPUs
support it. This has no measurable effect on cores which don't support
branch fusion, but increases fusion opportunities on cores which do.
Bootstrapped on AArch64, OK for commit?
ChangeLog:
2019-12-24 Wilco Dijkstra <wdijkstr@arm.com>
* config/aarch64/aarch64.c (generic_tunings): Add branch fusion.
(neoversen1_tunings): Likewise.
--
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index a3b18b381e1748f8fe5e522bdec4f7c850821fe8..1c32a3543bec4031cc9b641973101829c77296b5 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -726,7 +726,7 @@ static const struct tune_params generic_tunings =
SVE_NOT_IMPLEMENTED, /* sve_width */
4, /* memmov_cost */
2, /* issue_rate */
- (AARCH64_FUSE_AES_AESMC), /* fusible_ops */
+ (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
"16:12", /* function_align. */
"4", /* jump_align. */
"8", /* loop_align. */
@@ -1130,7 +1130,7 @@ static const struct tune_params neoversen1_tunings =
SVE_NOT_IMPLEMENTED, /* sve_width */
4, /* memmov_cost */
3, /* issue_rate */
- AARCH64_FUSE_AES_AESMC, /* fusible_ops */
+ (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
"32:16", /* function_align. */
"32:16", /* jump_align. */
"32:16", /* loop_align. */
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH][AARCH64] Enable compare branch fusion
2019-12-24 15:59 [PATCH][AARCH64] Enable compare branch fusion Wilco Dijkstra
2020-01-16 18:03 ` Wilco Dijkstra
@ 2020-01-17 10:02 ` Richard Sandiford
2020-01-17 15:58 ` Wilco Dijkstra
1 sibling, 1 reply; 4+ messages in thread
From: Richard Sandiford @ 2020-01-17 10:02 UTC (permalink / raw)
To: Wilco Dijkstra; +Cc: GCC Patches, Richard Earnshaw, Kyrylo Tkachov
Wilco Dijkstra <Wilco.Dijkstra@arm.com> writes:
> Enable the most basic form of compare-branch fusion since various CPUs
> support it. This has no measurable effect on cores which don't support
> branch fusion, but increases fusion opportunities on cores which do.
If you're able to say for the record which cores you tested, then that'd
be good.
> Bootstrapped on AArch64, OK for commit?
>
> ChangeLog:
> 2019-12-24 Wilco Dijkstra <wdijkstr@arm.com>
>
> * config/aarch64/aarch64.c (generic_tunings): Add branch fusion.
> (neoversen1_tunings): Likewise.
OK, thanks. I agree there doesn't seem to be an obvious reason why this
would pessimise any cores significantly. And it looked from a quick
check like all AArch64 cores give these compares the lowest in-use
latency (as expected).
We can revisit this if anyone finds any counterexamples.
Richard
>
> --
> diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
> index a3b18b381e1748f8fe5e522bdec4f7c850821fe8..1c32a3543bec4031cc9b641973101829c77296b5 100644
> --- a/gcc/config/aarch64/aarch64.c
> +++ b/gcc/config/aarch64/aarch64.c
> @@ -726,7 +726,7 @@ static const struct tune_params generic_tunings =
> SVE_NOT_IMPLEMENTED, /* sve_width */
> 4, /* memmov_cost */
> 2, /* issue_rate */
> - (AARCH64_FUSE_AES_AESMC), /* fusible_ops */
> + (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
> "16:12",/* function_align. */
> "4",/* jump_align. */
> "8",/* loop_align. */
> @@ -1130,7 +1130,7 @@ static const struct tune_params neoversen1_tunings =
> SVE_NOT_IMPLEMENTED, /* sve_width */
> 4, /* memmov_cost */
> 3, /* issue_rate */
> - AARCH64_FUSE_AES_AESMC, /* fusible_ops */
> + (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
> "32:16",/* function_align. */
> "32:16",/* jump_align. */
> "32:16",/* loop_align. */
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH][AARCH64] Enable compare branch fusion
2020-01-17 10:02 ` Richard Sandiford
@ 2020-01-17 15:58 ` Wilco Dijkstra
0 siblings, 0 replies; 4+ messages in thread
From: Wilco Dijkstra @ 2020-01-17 15:58 UTC (permalink / raw)
To: Richard Sandiford; +Cc: GCC Patches, Richard Earnshaw, Kyrylo Tkachov
Hi Richard,
> If you're able to say for the record which cores you tested, then that'd
> be good.
I've mostly checked it on Cortex-A57 - if there is any affect, it would be on
older cores.
> OK, thanks. I agree there doesn't seem to be an obvious reason why this
> would pessimise any cores significantly. And it looked from a quick
> check like all AArch64 cores give these compares the lowest in-use
> latency (as expected).
Indeed.
> We can revisit this if anyone finds any counterexamples.
Yes - it's unlikely there are any though!
Cheers,
Wilco
>
> --
> diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
> index a3b18b381e1748f8fe5e522bdec4f7c850821fe8..1c32a3543bec4031cc9b641973101829c77296b5 100644
> --- a/gcc/config/aarch64/aarch64.c
> +++ b/gcc/config/aarch64/aarch64.c
> @@ -726,7 +726,7 @@ static const struct tune_params generic_tunings =
> SVE_NOT_IMPLEMENTED, /* sve_width */
> 4, /* memmov_cost */
> 2, /* issue_rate */
> - (AARCH64_FUSE_AES_AESMC), /* fusible_ops */
> + (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
> "16:12",/* function_align. */
> "4",/* jump_align. */
> "8",/* loop_align. */
> @@ -1130,7 +1130,7 @@ static const struct tune_params neoversen1_tunings =
> SVE_NOT_IMPLEMENTED, /* sve_width */
> 4, /* memmov_cost */
> 3, /* issue_rate */
> - AARCH64_FUSE_AES_AESMC, /* fusible_ops */
> + (AARCH64_FUSE_AES_AESMC | AARCH64_FUSE_CMP_BRANCH), /* fusible_ops */
> "32:16",/* function_align. */
> "32:16",/* jump_align. */
> "32:16",/* loop_align. */
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2020-01-17 14:38 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-12-24 15:59 [PATCH][AARCH64] Enable compare branch fusion Wilco Dijkstra
2020-01-16 18:03 ` Wilco Dijkstra
2020-01-17 10:02 ` Richard Sandiford
2020-01-17 15:58 ` Wilco Dijkstra
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).