Re: [PATCH 1/2] RISC-V: Add memcpy, memmove, memset and strcmp implementations optimized for size.

[email protected] Mon, 20 May 2019 18:29:58 -0500

Newsgroups	gmane.comp.lib.newlib
Message-ID	<[email protected]>

Thanks for you notice, Jim. Currently I cannot use git send-email, so I 
try to provide the patches in attachment.

Jim Wilson писал 2019-05-20 16:09:
> On Fri, May 17, 2019 at 4:53 PM <[email protected]> wrote:
>> This patch adds implementations of memcpy, memmove, memset and strcmp
>> optimized for size. The changes have been tested in
>> riscv/riscv-gnu-toolchain by riscv-dejagnu with
>> riscv-sim.exp/riscv-sim-nano.exp.
> 
> The patch was corrupted by the mailer.   Long lines were line wrapped.
> Lines that should have only a single space are now blank lines.  Line
> that should start with one space now have two.  Using git send-email
> should work better if you can use that.  Otherwise, you probably need
> to include the patches as attachments instead of trying to insert the
> text of the patch into an email.
> 
> Optimitech is working on a contract for SiFive, so having the SiFive
> name in the copyright statements is OK.  Though I don't mind if the
> Optimitech name is in there too.
> 
> I tested an earlier version of this (without the memmove-stub.c fix),
> and it gets a thumbs up from me, though I'm not an official newlib
> maintainer so can't approve it.
> 
> Jim

0001-RISC-V-Add-memcpy-memmove-memset-and-strcmp-implemen.patch (text/x-diff, 9.4 KB)

From 99b42109278140daac310e3064c59ac21db42564 Mon Sep 17 00:00:00 2001
From: Ilia Diachkov <[email protected]>
Date: Tue, 21 May 2019 01:48:46 +0300
Subject: [PATCH 1/2] RISC-V: Add memcpy, memmove, memset and strcmp
 implementations optimized for size.

---
 newlib/libc/machine/riscv/Makefile.am    |  3 ++-
 newlib/libc/machine/riscv/Makefile.in    | 23 +++++++++++++++++-
 newlib/libc/machine/riscv/memcpy-asm.S   | 33 ++++++++++++++++++++++++++
 newlib/libc/machine/riscv/memcpy.c       |  6 +++++
 newlib/libc/machine/riscv/memmove-stub.c | 14 +++++++++++
 newlib/libc/machine/riscv/memmove.S      | 40 ++++++++++++++++++++++++++++++++
 newlib/libc/machine/riscv/memset.S       | 15 ++++++++++++
 newlib/libc/machine/riscv/strcmp.S       | 16 +++++++++++++
 8 files changed, 148 insertions(+), 2 deletions(-)
 create mode 100644 newlib/libc/machine/riscv/memcpy-asm.S
 create mode 100644 newlib/libc/machine/riscv/memmove-stub.c
 create mode 100644 newlib/libc/machine/riscv/memmove.S

diff --git a/newlib/libc/machine/riscv/Makefile.am b/newlib/libc/machine/riscv/Makefile.am
index 676df3e..017b4be 100644
--- a/newlib/libc/machine/riscv/Makefile.am
+++ b/newlib/libc/machine/riscv/Makefile.am
@@ -8,7 +8,8 @@ AM_CCASFLAGS = $(INCLUDES)
 
 noinst_LIBRARIES = lib.a
 
-lib_a_SOURCES = memset.S memcpy.c strlen.c strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
+lib_a_SOURCES = memmove.S memmove-stub.c memset.S memcpy-asm.S memcpy.c strlen.c \
+	strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
 lib_a_CCASFLAGS=$(AM_CCASFLAGS)
 lib_a_CFLAGS=$(AM_CFLAGS)
 
diff --git a/newlib/libc/machine/riscv/Makefile.in b/newlib/libc/machine/riscv/Makefile.in
index fd6d2ef..304dd35 100644
--- a/newlib/libc/machine/riscv/Makefile.in
+++ b/newlib/libc/machine/riscv/Makefile.in
@@ -70,6 +70,8 @@ ARFLAGS = cru
 lib_a_AR = $(AR) $(ARFLAGS)
 lib_a_LIBADD =
 am_lib_a_OBJECTS = lib_a-memset.$(OBJEXT) lib_a-memcpy.$(OBJEXT) \
+	lib_a-memcpy-asm.$(OBJEXT) \
+	lib_a-memmove.$(OBJEXT) lib_a-memmove-stub.$(OBJEXT) \
 	lib_a-strlen.$(OBJEXT) lib_a-strcpy.$(OBJEXT) \
 	lib_a-strcmp.$(OBJEXT) lib_a-setjmp.$(OBJEXT) \
 	lib_a-ieeefp.$(OBJEXT) lib_a-ffs.$(OBJEXT)
@@ -198,7 +200,8 @@ AUTOMAKE_OPTIONS = cygnus
 INCLUDES = $(NEWLIB_CFLAGS) $(CROSS_CFLAGS) $(TARGET_CFLAGS)
 AM_CCASFLAGS = $(INCLUDES)
 noinst_LIBRARIES = lib.a
-lib_a_SOURCES = memset.S memcpy.c strlen.c strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
+lib_a_SOURCES = memcpy-asm.S memmove.S memmove-stub.c memset.S memcpy.c strlen.c \
+	strcpy.c strcmp.S setjmp.S ieeefp.c ffs.c
 lib_a_CCASFLAGS = $(AM_CCASFLAGS)
 lib_a_CFLAGS = $(AM_CFLAGS)
 ACLOCAL_AMFLAGS = -I ../../.. -I ../../../..
@@ -261,6 +264,18 @@ distclean-compile:
 .S.obj:
 	$(CPPASCOMPILE) -c -o $@ `$(CYGPATH_W) '$<'`
 
+lib_a-memmove.o: memmove.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memmove.o `test -f 'memmove.S' || echo '$(srcdir)/'`memmove.S
+
+lib_a-memmove.obj: memmove.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memmove.obj `if test -f 'memmove.S'; then $(CYGPATH_W) 'memmove.S'; else $(CYGPATH_W) '$(srcdir)/memmove.S'; fi`
+
+lib_a-memcpy-asm.o: memcpy-asm.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memcpy-asm.o `test -f 'memcpy-asm.S' || echo '$(srcdir)/'`memcpy-asm.S
+
+lib_a-memcpy-asm.obj: memcpy-asm.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memcpy-asm.obj `if test -f 'memcpy-asm.S'; then $(CYGPATH_W) 'memcpy-asm.S'; else $(CYGPATH_W) '$(srcdir)/memcpy-asm.S'; fi`
+
 lib_a-memset.o: memset.S
 	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memset.o `test -f 'memset.S' || echo '$(srcdir)/'`memset.S
 
@@ -285,6 +300,12 @@ lib_a-setjmp.obj: setjmp.S
 .c.obj:
 	$(COMPILE) -c `$(CYGPATH_W) '$<'`
 
+lib_a-memmove-stub.o: memmove-stub.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memmove-stub.o `test -f 'memmove-stub.c' || echo '$(srcdir)/'`memmove-stub.c
+
+lib_a-memmove-stub.obj: memmove-stub.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memmove-stub.obj `if test -f 'memmove-stub.c'; then $(CYGPATH_W) 'memmove-stub.c'; else $(CYGPATH_W) '$(srcdir)/memmove-stub.c'; fi`
+
 lib_a-memcpy.o: memcpy.c
 	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memcpy.o `test -f 'memcpy.c' || echo '$(srcdir)/'`memcpy.c
 
diff --git a/newlib/libc/machine/riscv/memcpy-asm.S b/newlib/libc/machine/riscv/memcpy-asm.S
new file mode 100644
index 0000000..33d47a4
--- /dev/null
+++ b/newlib/libc/machine/riscv/memcpy-asm.S
@@ -0,0 +1,33 @@
+/* Copyright (c) 2019  SiFive Inc. All rights reserved.
+
+   This copyrighted material is made available to anyone wishing to use,
+   modify, copy, or redistribute it subject to the terms and conditions
+   of the FreeBSD License.   This program is distributed in the hope that
+   it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
+   including the implied warranties of MERCHANTABILITY or FITNESS FOR
+   A PARTICULAR PURPOSE.  A copy of this license is available at
+   http://www.opensource.org/licenses.
+*/
+
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__) 
+.text
+.global memcpy
+.type	memcpy, @function
+memcpy:
+  mv t1, a0
+  beqz a2, 2f
+
+1:
+  lb t2, 0(a1)
+  sb t2, 0(t1)
+  add   a2, a2, -1
+  add   t1, t1, 1
+  add   a1, a1, 1
+  bnez a2, 1b
+
+2: 
+  ret
+
+  .size	memcpy, .-memcpy
+#endif
+
diff --git a/newlib/libc/machine/riscv/memcpy.c b/newlib/libc/machine/riscv/memcpy.c
index c717f9f..8099b14 100644
--- a/newlib/libc/machine/riscv/memcpy.c
+++ b/newlib/libc/machine/riscv/memcpy.c
@@ -9,6 +9,10 @@
    http://www.opensource.org/licenses.
 */
 
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
+//memcpy defined in memcpy-asm.S
+#else
+
 #include <string.h>
 #include <stdint.h>
 
@@ -79,3 +83,5 @@ small:
     goto small;
   return aa;
 }
+#endif
+
diff --git a/newlib/libc/machine/riscv/memmove-stub.c b/newlib/libc/machine/riscv/memmove-stub.c
new file mode 100644
index 0000000..d882e46
--- /dev/null
+++ b/newlib/libc/machine/riscv/memmove-stub.c
@@ -0,0 +1,14 @@
+/* Copyright (c) 2019  SiFive Inc. All rights reserved.
+
+   This copyrighted material is made available to anyone wishing to use,
+   modify, copy, or redistribute it subject to the terms and conditions
+   of the FreeBSD License.   This program is distributed in the hope that
+   it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
+   including the implied warranties of MERCHANTABILITY or FITNESS FOR
+   A PARTICULAR PURPOSE.  A copy of this license is available at
+   http://www.opensource.org/licenses.
+*/
+
+#if !defined(PREFER_SIZE_OVER_SPEED) && !defined(__OPTIMIZE_SIZE__)
+#include "../../string/memmove.c"
+#endif
diff --git a/newlib/libc/machine/riscv/memmove.S b/newlib/libc/machine/riscv/memmove.S
new file mode 100644
index 0000000..98289a4
--- /dev/null
+++ b/newlib/libc/machine/riscv/memmove.S
@@ -0,0 +1,40 @@
+/* Copyright (c) 2019  SiFive Inc. All rights reserved.
+
+   This copyrighted material is made available to anyone wishing to use,
+   modify, copy, or redistribute it subject to the terms and conditions
+   of the FreeBSD License.   This program is distributed in the hope that
+   it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
+   including the implied warranties of MERCHANTABILITY or FITNESS FOR
+   A PARTICULAR PURPOSE.  A copy of this license is available at
+   http://www.opensource.org/licenses.
+*/
+
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__) 
+.text
+.global memmove
+.type	memmove, @function
+memmove:
+  beqz a2, 2f
+
+  mv t1, a0
+  li a3, 1
+  bgtu  a1, a0, 1f
+
+  li a3, -1
+  addi  a4, a2 , -1
+  add t1, t1, a4
+  add a1, a1, a4
+
+1:
+  lb t2, 0(a1)
+  sb t2, 0(t1)
+  add   a2, a2, -1
+  add   t1, t1, a3
+  add   a1, a1, a3
+  bnez a2, 1b
+
+2: 
+  ret
+
+  .size	memmove, .-memmove
+#endif
diff --git a/newlib/libc/machine/riscv/memset.S b/newlib/libc/machine/riscv/memset.S
index 337ed53..d28a219 100644
--- a/newlib/libc/machine/riscv/memset.S
+++ b/newlib/libc/machine/riscv/memset.S
@@ -13,6 +13,20 @@
 .global memset
 .type	memset, @function
 memset:
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__) 
+  mv t1, a0
+  beqz a2, 2f
+
+1:
+  sb a1, 0(t1)
+  add   a2, a2, -1
+  add   t1, t1, 1
+  bnez a2, 1b
+
+2: 
+  ret
+
+#else
   li t1, 15
   move a4, a0
   bleu a2, t1, .Ltiny
@@ -95,4 +109,5 @@ memset:
   add a2, a2, a5
   bleu a2, t1, .Ltiny
   j .Laligned
+#endif
   .size	memset, .-memset
diff --git a/newlib/libc/machine/riscv/strcmp.S b/newlib/libc/machine/riscv/strcmp.S
index 71c0853..1a9303e 100644
--- a/newlib/libc/machine/riscv/strcmp.S
+++ b/newlib/libc/machine/riscv/strcmp.S
@@ -19,6 +19,21 @@
 .globl strcmp
 .type  strcmp, @function
 strcmp:
+#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__) 
+1:
+  lbu   a2, 0(a0)
+  lbu   a3, 0(a1)
+  add   a0, a0, 1
+  add   a1, a1, 1
+  bne   a2, a3, 2f
+  bnez  a2, 1b
+
+2:
+  sub   a0, a2, a3
+  ret
+
+.size	strcmp, .-strcmp
+#else
   or    a4, a0, a1
   li    t2, -1
   and   a4, a4, SZREG-1
@@ -146,3 +161,4 @@ strcmp:
 mask:
 .dword 0x7f7f7f7f7f7f7f7f
 #endif
+#endif 
-- 
1.8.3.1

0002-RISC-V-Add-_LITE_EXIT-in-crt0.S.patch (text/x-diff, 1.4 KB)

From 192405b7511c54de76d0054c4adb9a2aa90eac3b Mon Sep 17 00:00:00 2001
From: Ilia Diachkov <[email protected]>
Date: Tue, 21 May 2019 01:51:16 +0300
Subject: [PATCH 2/2] RISC-V: Add _LITE_EXIT in crt0.S.

---
 libgloss/riscv/crt0.S | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/libgloss/riscv/crt0.S b/libgloss/riscv/crt0.S
index 3d2a12d..beea921 100644
--- a/libgloss/riscv/crt0.S
+++ b/libgloss/riscv/crt0.S
@@ -9,6 +9,8 @@
    http://www.opensource.org/licenses.
 */
 
+#include "newlib.h"
+
 #=========================================================================
 # crt0.S : Entry point for RISC-V user programs
 #=========================================================================
@@ -30,9 +32,20 @@ _start:
   sub     a2, a2, a0
   li      a1, 0
   call    memset
+#ifdef _LITE_EXIT
+  # Make reference to atexit weak to avoid unconditionally pulling in
+  # support code.  Refer to comments in __atexit.c for more details.
+  .weak   atexit
+  la      a0, atexit
+  beqz    a0, .Lweak_atexit
+  .weak   __libc_fini_array
+#endif
 
   la      a0, __libc_fini_array   # Register global termination functions
   call    atexit                  #  to be called upon exit
+#ifdef _LITE_EXIT
+.Lweak_atexit:
+#endif
   call    __libc_init_array       # Run global initialization functions
 
   lw      a0, 0(sp)                  # a0 = argc
-- 
1.8.3.1