This is the mail archive of the
glibc-cvs@sourceware.org
mailing list for the glibc project.
GNU C Library master sources branch master updated. glibc-2.26-532-gdd5bc7f
- From: siddhesh at sourceware dot org
- To: glibc-cvs at sourceware dot org
- Date: 5 Oct 2017 16:55:31 -0000
- Subject: GNU C Library master sources branch master updated. glibc-2.26-532-gdd5bc7f
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "GNU C Library master sources".
The branch, master has been updated
via dd5bc7f1b385b29d0f90aefe4d9756b35011709b (commit)
via 5bfb04042dfadc335cde63afe892be140cbe6f31 (commit)
via 36bb8edf5178247933dfeee71812cda93587d32e (commit)
via 9ec87fd2b11ffce5c27c8eacb9deaa20e2373ff5 (commit)
from 1e9522c61c7a544d59db32cb7fbbd42e6793d848 (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
http://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=dd5bc7f1b385b29d0f90aefe4d9756b35011709b
commit dd5bc7f1b385b29d0f90aefe4d9756b35011709b
Author: Siddhesh Poyarekar <siddhesh@sourceware.org>
Date: Thu Oct 5 21:55:31 2017 +0530
aarch64: Optimized implementation of memmove for Qualcomm Falkor
This is an optimized memmove implementation for the Qualcomm Falkor
processor core. Due to the way the falkor memcpy needs to be written,
code cannot be easily shared between memmove and memcpy like in case
of other aarch64 memcpy implementations due to which this routine is
separate. The underlying principle is the same as that of memcpy
where it tries to use registers with the same lower 4 bits for
fetching the same stream, thus optimizing hardware prefetcher
performance.
The memcpy copy loop copies 64 bytes at a time using the same register
pair since that's the way to train the hardware prefetcher on the
falkor core. memmove cannot quite do that since it needs to avoid
overlaps, so it does the next best thing, i.e. has a 32 byte loop with
a 32 byte end (prefetch a loop ahead to account for overlapping
locations) with register pairs that alias so that they hit the same
prefetcher. Due to this difference in loop size, they have to
currently be separate implementations but efforts are on to try and
get memmove to fall back into memcpy whenever it can without simply
duplicating all of the code.
Performance:
The routine fares around 20-25% better than the generic memmove for
most medium to large sizes (i.e. > 128 bytes) for the new walking
memmove benchmark (memmove-walk) with an unexplained regression
between 1K and 2K. The minor regression is something worth looking
into for us, but the remaining gains are significant enough that we
would like this included upstream as we looking into the cause for the
regression. Here is a snippet of the numbers as generated from the
microbenchmark by the compare_strings script. Comparisons are against
__memmove_generic:
Function: memmove
Variant: walk
__memmove_thunderx __memmove_falkor __memmove_generic
========================================================================================================================
<snip>
length=16384: 12508800.00 ( 6.09%) 11486800.00 ( 13.76%) 13319600.00
length=16400: 13614200.00 ( -0.67%) 11585000.00 ( 14.33%) 13523600.00
length=16385: 13448400.00 ( 0.10%) 11732700.00 ( 12.84%) 13461200.00
length=16399: 13594100.00 ( -0.22%) 11859600.00 ( 12.57%) 13564400.00
length=16386: 13211600.00 ( 1.13%) 11503800.00 ( 13.91%) 13362400.00
length=16398: 13218600.00 ( 2.12%) 11573200.00 ( 14.30%) 13504700.00
length=16387: 13510900.00 ( -0.37%) 11744200.00 ( 12.76%) 13461300.00
length=16397: 13603700.00 ( -0.15%) 11878200.00 ( 12.55%) 13583200.00
length=16388: 13461700.00 ( -0.13%) 11558000.00 ( 14.03%) 13444100.00
length=16396: 13517500.00 ( -0.03%) 11561300.00 ( 14.45%) 13513900.00
length=16389: 13534100.00 ( 0.17%) 11756800.00 ( 13.28%) 13556900.00
length=16395: 13585600.00 ( 0.11%) 11791800.00 ( 13.30%) 13601200.00
length=16390: 13480100.00 ( -0.13%) 11685500.00 ( 13.20%) 13462100.00
length=16394: 13529900.00 ( -0.23%) 11549800.00 ( 14.43%) 13498200.00
length=16391: 13595400.00 ( -0.26%) 11768200.00 ( 13.22%) 13560600.00
length=16393: 13567000.00 ( 0.20%) 11779700.00 ( 13.35%) 13594700.00
length=32768: 71308800.00 ( -6.53%) 50220800.00 ( 24.98%) 66939200.00
length=32784: 72100800.00 (-11.55%) 50114100.00 ( 22.47%) 64636300.00
length=32769: 71767000.00 ( -7.10%) 51238400.00 ( 23.54%) 67010000.00
length=32783: 70113700.00 (-40.95%) 51129000.00 ( -2.78%) 49744400.00
length=32770: 71367600.00 ( -6.52%) 50244700.00 ( 25.01%) 67000900.00
length=32782: 64366700.00 ( 4.71%) 50101400.00 ( 25.83%) 67545600.00
length=32771: 71440100.00 ( -6.51%) 51263900.00 ( 23.57%) 67074900.00
length=32781: 66993000.00 ( 0.34%) 51108300.00 ( 23.97%) 67220300.00
length=32772: 71443900.00 (-60.50%) 50062100.00 (-12.47%) 44512600.00
length=32780: 71759100.00 ( -6.58%) 50263200.00 ( 25.35%) 67328600.00
length=32773: 71714900.00 (-33.21%) 51076600.00 ( 5.12%) 53835400.00
length=32779: 71756900.00 ( -6.56%) 51290800.00 ( 23.83%) 67337800.00
length=32774: 59689300.00 (-34.55%) 50068400.00 (-12.86%) 44363300.00
length=32778: 71847500.00 (-18.20%) 50084100.00 ( 17.61%) 60786500.00
length=32775: 71599300.00 ( -6.54%) 51278200.00 ( 23.70%) 67204800.00
length=32777: 71862900.00 (-60.85%) 51094000.00 (-14.36%) 44677900.00
length=65536: 282848000.00 ( -6.60%) 199187000.00 ( 24.93%) 265325000.00
length=65552: 243285000.00 (-41.61%) 198512000.00 (-15.54%) 171805000.00
length=65537: 255415000.00 (-23.47%) 202499000.00 ( 2.11%) 206858000.00
length=65551: 280122000.00 (-62.95%) 203349000.00 (-18.29%) 171911000.00
length=65538: 283676000.00 (-14.46%) 198368000.00 ( 19.96%) 247848000.00
length=65550: 275566000.00 (-51.76%) 198494000.00 ( -9.31%) 181581000.00
length=65539: 283699000.00 ( -6.58%) 203453000.00 ( 23.57%) 266195000.00
length=65549: 286572000.00 ( -6.65%) 202607000.00 ( 24.60%) 268712000.00
length=65540: 283710000.00 ( -6.59%) 199161000.00 ( 25.17%) 266160000.00
length=65548: 237573000.00 ( 11.48%) 198462000.00 ( 26.06%) 268395000.00
length=65541: 284150000.00 ( -6.58%) 203273000.00 ( 23.75%) 266600000.00
length=65547: 286250000.00 ( -6.70%) 202594000.00 ( 24.48%) 268263000.00
length=65542: 284167000.00 ( -6.60%) 199122000.00 ( 25.31%) 266584000.00
length=65546: 285656000.00 ( -6.59%) 198443000.00 ( 25.95%) 268002000.00
length=65543: 284600000.00 ( -6.58%) 203247000.00 ( 23.89%) 267030000.00
length=65545: 285665000.00 ( -6.40%) 202575000.00 ( 24.55%) 268472000.00
<snip>
* sysdeps/aarch64/multiarch/Makefile (sysdep_routines): Add
memmove_falkor.
* sysdeps/aarch64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Likewise.
* sysdeps/aarch64/multiarch/memmove.c: Likewise.
* sysdeps/aarch64/multiarch/memmove_falkor.S: New file.
diff --git a/ChangeLog b/ChangeLog
index 6d4b5f2..74c51ea 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,12 @@
2017-10-05 Siddhesh Poyarekar <siddhesh@sourceware.org>
+ * sysdeps/aarch64/multiarch/Makefile (sysdep_routines): Add
+ memmove_falkor.
+ * sysdeps/aarch64/multiarch/ifunc-impl-list.c
+ (__libc_ifunc_impl_list): Likewise.
+ * sysdeps/aarch64/multiarch/memmove.c: Likewise.
+ * sysdeps/aarch64/multiarch/memmove_falkor.S: New file.
+
* benchtests/bench-memmove-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/sysdeps/aarch64/multiarch/Makefile b/sysdeps/aarch64/multiarch/Makefile
index 164ba1a..9aa1e79 100644
--- a/sysdeps/aarch64/multiarch/Makefile
+++ b/sysdeps/aarch64/multiarch/Makefile
@@ -1,3 +1,4 @@
ifeq ($(subdir),string)
-sysdep_routines += memcpy_generic memcpy_thunderx memcpy_falkor
+sysdep_routines += memcpy_generic memcpy_thunderx memcpy_falkor \
+ memmove_falkor
endif
diff --git a/sysdeps/aarch64/multiarch/ifunc-impl-list.c b/sysdeps/aarch64/multiarch/ifunc-impl-list.c
index 8e873b3..2cb74d5 100644
--- a/sysdeps/aarch64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/aarch64/multiarch/ifunc-impl-list.c
@@ -44,6 +44,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_generic))
IFUNC_IMPL (i, name, memmove,
IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_thunderx)
+ IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_falkor)
IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_generic))
return i;
diff --git a/sysdeps/aarch64/multiarch/memmove.c b/sysdeps/aarch64/multiarch/memmove.c
index 34c6b29..016f03e 100644
--- a/sysdeps/aarch64/multiarch/memmove.c
+++ b/sysdeps/aarch64/multiarch/memmove.c
@@ -30,9 +30,14 @@ extern __typeof (__redirect_memmove) __libc_memmove;
extern __typeof (__redirect_memmove) __memmove_generic attribute_hidden;
extern __typeof (__redirect_memmove) __memmove_thunderx attribute_hidden;
+extern __typeof (__redirect_memmove) __memmove_falkor attribute_hidden;
libc_ifunc (__libc_memmove,
- IS_THUNDERX (midr) ? __memmove_thunderx : __memmove_generic);
+ (IS_THUNDERX (midr)
+ ? __memmove_thunderx
+ : (IS_FALKOR (midr)
+ ? __memmove_falkor
+ : __memmove_generic)));
# undef memmove
strong_alias (__libc_memmove, memmove);
diff --git a/sysdeps/aarch64/multiarch/memmove_falkor.S b/sysdeps/aarch64/multiarch/memmove_falkor.S
new file mode 100644
index 0000000..3a4e6a2
--- /dev/null
+++ b/sysdeps/aarch64/multiarch/memmove_falkor.S
@@ -0,0 +1,232 @@
+/* Copyright (C) 2017 Free Software Foundation, Inc.
+
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library. If not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+
+/* Assumptions: ARMv8-a, AArch64, falkor, unaligned accesses. */
+
+#define dstin x0
+#define src x1
+#define count x2
+#define dstlen x3
+#define dst x3
+#define srcend x4
+#define dstend x5
+#define A_l x6
+#define A_lw w6
+#define A_h x7
+#define A_hw w7
+#define B_l x8
+#define B_lw w8
+#define B_h x9
+#define C_l x10
+#define C_h x11
+#define D_l x12
+#define D_h x13
+#define E_l src
+#define E_h count
+#define F_l srcend
+#define F_h dst
+#define tmp1 x14
+
+/* Alias with A_l and A_h to train the prefetcher. */
+#define Q_l x22
+#define Q_h x23
+
+/* RATIONALE:
+
+ The copy has 4 distinct parts:
+ * Small copies of 16 bytes and under
+ * Medium sized copies of 17-96 bytes
+ * Large copies where the source address is higher than the destination
+ (forward copies)
+ * Large copies where the destination address is higher than the source
+ (copy backward, or move).
+
+ We use only two registerpairs x6,x7 and x22,x23 for the copies and copy 32
+ bytes at a time to correctly train the hardware prefetcher for better
+ throughput. */
+ENTRY_ALIGN (__memmove_falkor, 6)
+
+ sub tmp1, dstin, src
+ add srcend, src, count
+ add dstend, dstin, count
+ cmp count, 96
+ ccmp tmp1, count, 2, hi
+ b.lo L(move_long)
+
+ cmp count, 16
+ b.ls L(copy16)
+ cmp count, 96
+ b.hi L(copy_long)
+
+ /* Medium copies: 17..96 bytes. */
+ sub tmp1, count, 1
+ ldp A_l, A_h, [src]
+ tbnz tmp1, 6, L(copy96)
+ ldp D_l, D_h, [srcend, -16]
+ tbz tmp1, 5, 1f
+ ldp B_l, B_h, [src, 16]
+ ldp C_l, C_h, [srcend, -32]
+ stp B_l, B_h, [dstin, 16]
+ stp C_l, C_h, [dstend, -32]
+1:
+ stp A_l, A_h, [dstin]
+ stp D_l, D_h, [dstend, -16]
+ ret
+
+ .p2align 4
+ /* Small copies: 0..16 bytes. */
+L(copy16):
+ cmp count, 8
+ b.lo 1f
+ ldr A_l, [src]
+ ldr A_h, [srcend, -8]
+ str A_l, [dstin]
+ str A_h, [dstend, -8]
+ ret
+ .p2align 4
+1:
+ /* 4-7 */
+ tbz count, 2, 1f
+ ldr A_lw, [src]
+ ldr A_hw, [srcend, -4]
+ str A_lw, [dstin]
+ str A_hw, [dstend, -4]
+ ret
+ .p2align 4
+1:
+ /* 2-3 */
+ tbz count, 1, 1f
+ ldrh A_lw, [src]
+ ldrh A_hw, [srcend, -2]
+ strh A_lw, [dstin]
+ strh A_hw, [dstend, -2]
+ ret
+ .p2align 4
+1:
+ /* 0-1 */
+ tbz count, 0, 1f
+ ldrb A_lw, [src]
+ strb A_lw, [dstin]
+1: ret
+
+ .p2align 4
+ /* Copy 64..96 bytes. Copy 64 bytes from the start and
+ 32 bytes from the end. */
+L(copy96):
+ ldp B_l, B_h, [src, 16]
+ ldp C_l, C_h, [src, 32]
+ ldp D_l, D_h, [src, 48]
+ ldp E_l, E_h, [srcend, -32]
+ ldp F_l, F_h, [srcend, -16]
+ stp A_l, A_h, [dstin]
+ stp B_l, B_h, [dstin, 16]
+ stp C_l, C_h, [dstin, 32]
+ stp D_l, D_h, [dstin, 48]
+ stp E_l, E_h, [dstend, -32]
+ stp F_l, F_h, [dstend, -16]
+ ret
+
+ /* Align SRC to 16 byte alignment so that we don't cross cache line
+ boundaries on both loads and stores. There are at least 96 bytes
+ to copy, so copy 16 bytes unaligned and then align. The loop
+ copies 32 bytes per iteration and prefetches one iteration ahead. */
+
+ .p2align 4
+L(copy_long):
+ sub count, count, 64 + 16 /* Test and readjust count. */
+ mov B_l, Q_l
+ mov B_h, Q_h
+ ldp A_l, A_h, [src]
+ and tmp1, src, 15
+ bic src, src, 15
+ sub dst, dstin, tmp1
+ add count, count, tmp1 /* Count is now 16 too large. */
+ ldp Q_l, Q_h, [src, 16]!
+ stp A_l, A_h, [dstin]
+ ldp A_l, A_h, [src, 16]!
+
+L(loop64):
+ subs count, count, 32
+ stp Q_l, Q_h, [dst, 16]
+ ldp Q_l, Q_h, [src, 16]!
+ stp A_l, A_h, [dst, 32]!
+ ldp A_l, A_h, [src, 16]!
+ b.hi L(loop64)
+
+ /* Write the last full set of 32 bytes. The remainder is at most 32
+ bytes, so it is safe to always copy 32 bytes from the end even if
+ there is just 1 byte left. */
+L(last64):
+ ldp C_l, C_h, [srcend, -32]
+ stp Q_l, Q_h, [dst, 16]
+ ldp Q_l, Q_h, [srcend, -16]
+ stp A_l, A_h, [dst, 32]
+ stp C_l, C_h, [dstend, -32]
+ stp Q_l, Q_h, [dstend, -16]
+ mov Q_l, B_l
+ mov Q_h, B_h
+ ret
+
+ .p2align 4
+L(move_long):
+ cbz tmp1, 3f
+
+ mov B_l, Q_l
+ mov B_h, Q_h
+
+ /* Align SRCEND to 16 byte alignment so that we don't cross cache line
+ boundaries on both loads and stores. There are at least 96 bytes
+ to copy, so copy 16 bytes unaligned and then align. The loop
+ copies 32 bytes per iteration and prefetches one iteration ahead. */
+
+ ldp A_l, A_h, [srcend, -16]
+ and tmp1, srcend, 15
+ sub srcend, srcend, tmp1
+ ldp Q_l, Q_h, [srcend, -16]!
+ stp A_l, A_h, [dstend, -16]
+ sub count, count, tmp1
+ ldp A_l, A_h, [srcend, -16]!
+ sub dstend, dstend, tmp1
+ sub count, count, 64
+
+1:
+ subs count, count, 32
+ stp Q_l, Q_h, [dstend, -16]
+ ldp Q_l, Q_h, [srcend, -16]!
+ stp A_l, A_h, [dstend, -32]!
+ ldp A_l, A_h, [srcend, -16]!
+ b.hi 1b
+
+ /* Write the last full set of 32 bytes. The remainder is at most 32
+ bytes, so it is safe to always copy 32 bytes from the start even if
+ there is just 1 byte left. */
+2:
+ ldp C_l, C_h, [src, 16]
+ stp Q_l, Q_h, [dstend, -16]
+ ldp Q_l, Q_h, [src]
+ stp A_l, A_h, [dstend, -32]
+ stp C_l, C_h, [dstin, 16]
+ stp Q_l, Q_h, [dstin]
+ mov Q_l, B_l
+ mov Q_h, B_h
+3: ret
+
+END (__memmove_falkor)
+libc_hidden_builtin_def (__memmove_falkor)
http://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=5bfb04042dfadc335cde63afe892be140cbe6f31
commit 5bfb04042dfadc335cde63afe892be140cbe6f31
Author: Siddhesh Poyarekar <siddhesh@sourceware.org>
Date: Thu Oct 5 21:53:13 2017 +0530
benchtests: Memory walking benchmark for memmove
This benchmark is an attempt to eliminate cache effects from string
benchmarks. The benchmark walks both ways through a large memory area
and copies different sizes of memory and alignments one at a time
instead of looping around in the same memory area. This is a good
metric to have alongside the simple memmove benchmark (which is only
really useful for smaller sizes) especially for larger sizes where the
likelihood of the call being done only once is pretty high.
This benchmark is different from memcpy in that it also tests
overlapping copies.
* benchtests/bench-memmove-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/ChangeLog b/ChangeLog
index 5821c5d..6d4b5f2 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,8 @@
2017-10-05 Siddhesh Poyarekar <siddhesh@sourceware.org>
+ * benchtests/bench-memmove-walk.c: New file.
+ * benchtests/Makefile (string-benchset): Add it.
+
* benchtests/bench-memset-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/benchtests/Makefile b/benchtests/Makefile
index 5a07639..d8681fc 100644
--- a/benchtests/Makefile
+++ b/benchtests/Makefile
@@ -38,7 +38,7 @@ string-benchset := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
strcoll memcpy-large memcpy-random memmove-large memset-large \
- memcpy-walk memset-walk
+ memcpy-walk memset-walk memmove-walk
# Build and run locale-dependent benchmarks only if we're building natively.
ifeq (no,$(cross-compiling))
diff --git a/benchtests/bench-memmove-walk.c b/benchtests/bench-memmove-walk.c
new file mode 100644
index 0000000..54dcd64
--- /dev/null
+++ b/benchtests/bench-memmove-walk.c
@@ -0,0 +1,143 @@
+/* Measure memmove function combined throughput for different alignments.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* This microbenchmark measures the throughput of memmove for various sizes from
+ 1 byte to 32MiB, doubling every iteration and then misaligning by 0-15
+ bytes. The copies are done from source to destination and then back and the
+ source walks forward across the array and the destination walks backward by
+ one byte each, thus measuring misaligned accesses as well. The idea is to
+ avoid caching effects by copying a different string and far enough from each
+ other, walking in different directions so that we can measure prefetcher
+ efficiency (software or hardware) more closely than with a loop copying the
+ same data over and over, which eventually only gives us L1 cache
+ performance. */
+
+#ifndef MEMMOVE_RESULT
+# define MEMMOVE_RESULT(dst, len) dst
+# define START_SIZE 1
+# define MIN_PAGE_SIZE (getpagesize () + 32 * 1024 * 1024)
+# define TEST_MAIN
+# define TEST_NAME "memmove"
+# define TIMEOUT (20 * 60)
+# include "bench-string.h"
+
+IMPL (memmove, 1)
+#endif
+
+#include "json-lib.h"
+
+typedef char *(*proto_t) (char *, const char *, size_t);
+
+static void
+do_one_test (json_ctx_t *json_ctx, impl_t *impl, char *dst, char *src,
+ size_t len)
+{
+ size_t i, iters = MIN_PAGE_SIZE / len;
+ timing_t start, stop, cur;
+
+ char *dst_end = dst + MIN_PAGE_SIZE - len;
+ char *src_end = src + MIN_PAGE_SIZE - len;
+
+ TIMING_NOW (start);
+ /* Copy the entire buffer back and forth, LEN at a time. */
+ for (i = 0; i < iters && dst_end >= dst && src <= src_end; src++, dst_end--)
+ {
+ CALL (impl, dst_end, src, len);
+ CALL (impl, src, dst_end, len);
+ i += 2;
+ }
+ TIMING_NOW (stop);
+
+ TIMING_DIFF (cur, start, stop);
+
+ /* Get time taken per function call. */
+ json_element_double (json_ctx, (double) cur * len / i);
+}
+
+static void
+do_test (json_ctx_t *json_ctx, size_t len, bool overlap)
+{
+ json_element_object_begin (json_ctx);
+ json_attr_uint (json_ctx, "length", (double) len);
+ json_array_begin (json_ctx, "timings");
+
+ if (overlap)
+ buf2 = buf1;
+
+ /* First the non-overlapping moves. */
+ FOR_EACH_IMPL (impl, 0)
+ do_one_test (json_ctx, impl, (char *) buf2, (char *) buf1, len);
+
+ json_array_end (json_ctx);
+ json_element_object_end (json_ctx);
+}
+
+int
+test_main (void)
+{
+ json_ctx_t json_ctx;
+ size_t i;
+
+ test_init ();
+
+ json_init (&json_ctx, 0, stdout);
+
+ json_document_begin (&json_ctx);
+ json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
+
+ json_attr_object_begin (&json_ctx, "functions");
+ json_attr_object_begin (&json_ctx, "memmove");
+ json_attr_string (&json_ctx, "bench-variant", "walk");
+
+ json_array_begin (&json_ctx, "ifuncs");
+ FOR_EACH_IMPL (impl, 0)
+ json_element_string (&json_ctx, impl->name);
+ json_array_end (&json_ctx);
+
+ json_array_begin (&json_ctx, "results");
+ /* Non-overlapping buffers. */
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ {
+ /* Test length alignments from 0-16 bytes. */
+ for (int j = 0; j < 8; j++)
+ {
+ do_test (&json_ctx, i + j, false);
+ do_test (&json_ctx, i + 16 - j, false);
+ }
+ }
+
+ /* Overlapping buffers. */
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ {
+ /* Test length alignments from 0-16 bytes. */
+ for (int j = 0; j < 8; j++)
+ {
+ do_test (&json_ctx, i + j, true);
+ do_test (&json_ctx, i + 16 - j, true);
+ }
+ }
+
+ json_array_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_document_end (&json_ctx);
+
+ return ret;
+}
+
+#include <support/test-driver.c>
http://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=36bb8edf5178247933dfeee71812cda93587d32e
commit 36bb8edf5178247933dfeee71812cda93587d32e
Author: Siddhesh Poyarekar <siddhesh@sourceware.org>
Date: Thu Oct 5 21:51:41 2017 +0530
benchtests: Memory walking benchmark for memset
This benchmark is an attempt to eliminate cache effects from string
benchmarks. The benchmark walks backward through a large memory area
and sets different sizes of memory and alignments one at a time
instead of looping around in the same memory area. This is a good
metric to have alongside the simple memset benchmark (which is only
really useful for smaller sizes) especially for larger sizes where the
likelihood of the call being done only once is pretty high.
* benchtests/bench-memset-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/ChangeLog b/ChangeLog
index a86faeb..5821c5d 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,8 @@
2017-10-05 Siddhesh Poyarekar <siddhesh@sourceware.org>
+ * benchtests/bench-memset-walk.c: New file.
+ * benchtests/Makefile (string-benchset): Add it.
+
* benchtests/bench-memcpy-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/benchtests/Makefile b/benchtests/Makefile
index d086cc6..5a07639 100644
--- a/benchtests/Makefile
+++ b/benchtests/Makefile
@@ -38,7 +38,7 @@ string-benchset := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
strcoll memcpy-large memcpy-random memmove-large memset-large \
- memcpy-walk
+ memcpy-walk memset-walk
# Build and run locale-dependent benchmarks only if we're building natively.
ifeq (no,$(cross-compiling))
diff --git a/benchtests/bench-memset-walk.c b/benchtests/bench-memset-walk.c
new file mode 100644
index 0000000..59d2626
--- /dev/null
+++ b/benchtests/bench-memset-walk.c
@@ -0,0 +1,138 @@
+/* Measure memset function throughput with large data sizes.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define TEST_MAIN
+#ifndef WIDE
+# define TEST_NAME "memset"
+#else
+# define TEST_NAME "wmemset"
+#endif /* WIDE */
+#define START_SIZE (1)
+#define MIN_PAGE_SIZE (getpagesize () + 32 * 1024 * 1024)
+#define TIMEOUT (20 * 60)
+#include "bench-string.h"
+
+#ifndef WIDE
+# define MEMSET memset
+# define CHAR char
+# define SIMPLE_MEMSET simple_memset
+# define MEMCMP memcmp
+#else
+# include <wchar.h>
+# define MEMSET wmemset
+# define CHAR wchar_t
+# define SIMPLE_MEMSET simple_wmemset
+# define MEMCMP wmemcmp
+#endif /* WIDE */
+
+#include <assert.h>
+#include "json-lib.h"
+
+
+typedef CHAR *(*proto_t) (CHAR *, int, size_t);
+
+CHAR *
+inhibit_loop_to_libcall
+SIMPLE_MEMSET (CHAR *s, int c, size_t n)
+{
+ CHAR *r = s, *end = s + n;
+ while (r < end)
+ *r++ = c;
+ return s;
+}
+
+IMPL (SIMPLE_MEMSET, 1)
+
+static void
+do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s, CHAR *s_end,
+ int c __attribute ((unused)), size_t n)
+{
+ size_t i, iters = MIN_PAGE_SIZE / n;
+ timing_t start, stop, cur;
+
+ TIMING_NOW (start);
+ for (i = 0; i < iters && s <= s_end; s++, i++)
+ CALL (impl, s, c, n);
+ TIMING_NOW (stop);
+
+ TIMING_DIFF (cur, start, stop);
+
+ /* Get time taken per function call. */
+ json_element_double (json_ctx, (double) cur * n / i);
+}
+
+static void
+do_test (json_ctx_t *json_ctx, int c, size_t len)
+{
+ json_element_object_begin (json_ctx);
+ json_attr_uint (json_ctx, "length", len);
+ json_attr_uint (json_ctx, "char", c);
+ json_array_begin (json_ctx, "timings");
+
+ FOR_EACH_IMPL (impl, 0)
+ {
+ do_one_test (json_ctx, impl, (CHAR *) buf1,
+ (CHAR *) buf1 + MIN_PAGE_SIZE - len, c, len);
+ realloc_bufs ();
+ }
+
+ json_array_end (json_ctx);
+ json_element_object_end (json_ctx);
+}
+
+int
+test_main (void)
+{
+ json_ctx_t json_ctx;
+ size_t i;
+
+ test_init ();
+
+ json_init (&json_ctx, 0, stdout);
+
+ json_document_begin (&json_ctx);
+ json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
+
+ json_attr_object_begin (&json_ctx, "functions");
+ json_attr_object_begin (&json_ctx, TEST_NAME);
+ json_attr_string (&json_ctx, "bench-variant", "walk");
+
+ json_array_begin (&json_ctx, "ifuncs");
+ FOR_EACH_IMPL (impl, 0)
+ json_element_string (&json_ctx, impl->name);
+ json_array_end (&json_ctx);
+
+ json_array_begin (&json_ctx, "results");
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ /* Test length alignments from 0-16 bytes. */
+ for (int j = 0; j < i && j < 16; j++)
+ do_test (&json_ctx, 65, i + j);
+
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ for (int j = 0; j < i && j < 16; j++)
+ do_test (&json_ctx, 0, i + j);
+
+ json_array_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_document_end (&json_ctx);
+
+ return ret;
+}
+
+#include <support/test-driver.c>
http://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=9ec87fd2b11ffce5c27c8eacb9deaa20e2373ff5
commit 9ec87fd2b11ffce5c27c8eacb9deaa20e2373ff5
Author: Siddhesh Poyarekar <siddhesh@sourceware.org>
Date: Thu Oct 5 21:49:37 2017 +0530
benchtests: Memory walking benchmark for memcpy
This benchmark is an attempt to eliminate cache effects from string
benchmarks. The benchmark walks both ways through a large memory area
and copies different sizes of memory and alignments one at a time
instead of looping around in the same memory area. This is a good
metric to have alongside the other memcpy benchmarks, especially for
larger sizes where the likelihood of the call being done only once is
pretty high.
* benchtests/bench-memcpy-walk.c: New file.
* benchtests/Makefile (string-benchset): Add it.
diff --git a/ChangeLog b/ChangeLog
index 7ddff74..a86faeb 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,8 @@
+2017-10-05 Siddhesh Poyarekar <siddhesh@sourceware.org>
+
+ * benchtests/bench-memcpy-walk.c: New file.
+ * benchtests/Makefile (string-benchset): Add it.
+
2017-10-05 Florian Weimer <fweimer@redhat.com>
nscd: Eliminate compilation time dependency in the build output.
diff --git a/benchtests/Makefile b/benchtests/Makefile
index 3acc39c..d086cc6 100644
--- a/benchtests/Makefile
+++ b/benchtests/Makefile
@@ -37,7 +37,8 @@ string-benchset := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
strcat strchr strchrnul strcmp strcpy strcspn strlen \
strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \
- strcoll memcpy-large memcpy-random memmove-large memset-large
+ strcoll memcpy-large memcpy-random memmove-large memset-large \
+ memcpy-walk
# Build and run locale-dependent benchmarks only if we're building natively.
ifeq (no,$(cross-compiling))
diff --git a/benchtests/bench-memcpy-walk.c b/benchtests/bench-memcpy-walk.c
new file mode 100644
index 0000000..69d467d
--- /dev/null
+++ b/benchtests/bench-memcpy-walk.c
@@ -0,0 +1,127 @@
+/* Measure memcpy function combined throughput for different alignments.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* This microbenchmark measures the throughput of memcpy for various sizes from
+ 1 byte to 32MiB, doubling every iteration and then misaligning by 0-15
+ bytes. The copies are done from source to destination and then back and the
+ source walks forward across the array and the destination walks backward by
+ one byte each, thus measuring misaligned accesses as well. The idea is to
+ avoid caching effects by copying a different string and far enough from each
+ other, walking in different directions so that we can measure prefetcher
+ efficiency (software or hardware) more closely than with a loop copying the
+ same data over and over, which eventually only gives us L1 cache
+ performance. */
+
+#ifndef MEMCPY_RESULT
+# define MEMCPY_RESULT(dst, len) dst
+# define START_SIZE 1
+# define MIN_PAGE_SIZE (getpagesize () + 32 * 1024 * 1024)
+# define TEST_MAIN
+# define TEST_NAME "memcpy"
+# define TIMEOUT (20 * 60)
+# include "bench-string.h"
+
+IMPL (memcpy, 1)
+#endif
+
+#include "json-lib.h"
+
+typedef char *(*proto_t) (char *, const char *, size_t);
+
+static void
+do_one_test (json_ctx_t *json_ctx, impl_t *impl, char *dst, char *src,
+ size_t len)
+{
+ size_t i, iters = MIN_PAGE_SIZE / len;
+ timing_t start, stop, cur;
+
+ char *dst_end = dst + MIN_PAGE_SIZE - len;
+ char *src_end = src + MIN_PAGE_SIZE - len;
+
+ TIMING_NOW (start);
+ /* Copy the entire buffer back and forth, LEN at a time. */
+ for (i = 0; i < iters && dst_end >= dst && src <= src_end; src++, dst_end--)
+ {
+ CALL (impl, dst_end, src, len);
+ CALL (impl, src, dst_end, len);
+ i += 2;
+ }
+ TIMING_NOW (stop);
+
+ TIMING_DIFF (cur, start, stop);
+
+ /* Get time taken per function call. */
+ json_element_double (json_ctx, (double) cur * len / i);
+}
+
+static void
+do_test (json_ctx_t *json_ctx, size_t len)
+{
+ json_element_object_begin (json_ctx);
+ json_attr_uint (json_ctx, "length", (double) len);
+ json_array_begin (json_ctx, "timings");
+
+ FOR_EACH_IMPL (impl, 0)
+ do_one_test (json_ctx, impl, (char *) buf2, (char *) buf1, len);
+
+ json_array_end (json_ctx);
+ json_element_object_end (json_ctx);
+}
+
+int
+test_main (void)
+{
+ json_ctx_t json_ctx;
+ size_t i;
+
+ test_init ();
+
+ json_init (&json_ctx, 0, stdout);
+
+ json_document_begin (&json_ctx);
+ json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
+
+ json_attr_object_begin (&json_ctx, "functions");
+ json_attr_object_begin (&json_ctx, "memcpy");
+ json_attr_string (&json_ctx, "bench-variant", "walk");
+
+ json_array_begin (&json_ctx, "ifuncs");
+ FOR_EACH_IMPL (impl, 0)
+ json_element_string (&json_ctx, impl->name);
+ json_array_end (&json_ctx);
+
+ json_array_begin (&json_ctx, "results");
+ for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1)
+ {
+ /* Test length alignments from 0-16 bytes. */
+ for (int j = 0; j < 8; j++)
+ {
+ do_test (&json_ctx, i + j);
+ do_test (&json_ctx, i + 16 - j);
+ }
+ }
+
+ json_array_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_attr_object_end (&json_ctx);
+ json_document_end (&json_ctx);
+
+ return ret;
+}
+
+#include <support/test-driver.c>
-----------------------------------------------------------------------
Summary of changes:
ChangeLog | 18 ++
benchtests/Makefile | 3 +-
benchtests/bench-memcpy-walk.c | 127 +++++++++++++++
benchtests/bench-memmove-walk.c | 143 ++++++++++++++++
benchtests/bench-memset-walk.c | 138 ++++++++++++++++
sysdeps/aarch64/multiarch/Makefile | 3 +-
sysdeps/aarch64/multiarch/ifunc-impl-list.c | 1 +
sysdeps/aarch64/multiarch/memmove.c | 7 +-
sysdeps/aarch64/multiarch/memmove_falkor.S | 232 +++++++++++++++++++++++++++
9 files changed, 669 insertions(+), 3 deletions(-)
create mode 100644 benchtests/bench-memcpy-walk.c
create mode 100644 benchtests/bench-memmove-walk.c
create mode 100644 benchtests/bench-memset-walk.c
create mode 100644 sysdeps/aarch64/multiarch/memmove_falkor.S
hooks/post-receive
--
GNU C Library master sources