diff options
Diffstat (limited to 'strcmp-delete-align-for-loop_aligned.patch')
| -rw-r--r-- | strcmp-delete-align-for-loop_aligned.patch | 32 |
1 files changed, 32 insertions, 0 deletions
diff --git a/strcmp-delete-align-for-loop_aligned.patch b/strcmp-delete-align-for-loop_aligned.patch new file mode 100644 index 0000000..cf5b15a --- /dev/null +++ b/strcmp-delete-align-for-loop_aligned.patch @@ -0,0 +1,32 @@ +From 9bbffed83b93f633b272368fc536a4f24e9942e6 Mon Sep 17 00:00:00 2001 +From: Yang Yanchao <yangyanchao6@huawei.com> +Date: Mon, 21 Feb 2022 14:25:25 +0800 +Subject: [PATCH] strcmp: delete align for loop_aligned + +In Kunpeng-920, the performance of strcmp deteriorates only +when the 16 to 23 characters are different.Or the string is +only 16-23 characters.That shows 2 misses per iteration which +means this is a branch predictor issue indeed. +In the preceding scenario, strcmp performance is 300% worse than expected. + +Fortunately, this problem can be solved by modifying the alignment of the functions. +--- + sysdeps/aarch64/strcmp.S | 2 -- + 1 file changed, 2 deletions(-) + +diff --git a/sysdeps/aarch64/strcmp.S b/sysdeps/aarch64/strcmp.S +index f225d718..7a048b66 100644 +--- a/sysdeps/aarch64/strcmp.S ++++ b/sysdeps/aarch64/strcmp.S +@@ -71,8 +71,6 @@ ENTRY(strcmp) + b.ne L(misaligned8) + cbnz tmp, L(mutual_align) + +- .p2align 4 +- + L(loop_aligned): + ldr data2, [src1, off2] + ldr data1, [src1], 8 +-- +2.33.0 + |