the kernel's memcpy and memmove is very inefficient. But the glibc version is
quite fast, in some cases it is 10 times faster than the kernel version. So I
introduce some memory copy macros and functions of the glibc to improve the
kernel version's performance.
The strategy of the memory functions is:
1. Copy bytes until the destination pointer is aligned.
2. Copy words in unrolled loops. If the source and destination are not aligned
in the same way, use word memory operations, but shift and merge two read
words before writing.
3. Copy the few remaining bytes.
Signed-off-by: Miao Xie <[email protected]>
---
include/linux/memcopy.h | 226 ++++++++++++++++++++++++++
lib/Makefile | 3 +-
lib/memcopy.c | 403 +++++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 631 insertions(+), 1 deletions(-)
create mode 100644 include/linux/memcopy.h
create mode 100644 lib/memcopy.c
diff --git a/include/linux/memcopy.h b/include/linux/memcopy.h
new file mode 100644
index 0000000..9c65ac8
--- /dev/null
+++ b/include/linux/memcopy.h
@@ -0,0 +1,226 @@
+/*
+ * memcopy.h -- definitions for memory copy functions. Generic C version.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2.1 of the License, or (at your option)
+ * any later version.
+ *
+ * This program 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 General
+ * Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
+ *
+ * The code is derived from the GNU C Library.
+ * Copyright (C) 1991, 1992, 1993, 1997, 2004 Free Software Foundation, Inc.
+ */
+#ifndef _LINUX_MEMCOPY_H_
+#define _LINUX_MEMCOPY_H_
+
+/*
+ * The strategy of the memory functions is:
+ *
+ * 1. Copy bytes until the destination pointer is aligned.
+ *
+ * 2. Copy words in unrolled loops. If the source and destination
+ * are not aligned in the same way, use word memory operations,
+ * but shift and merge two read words before writing.
+ *
+ * 3. Copy the few remaining bytes.
+ *
+ * This is fast on processors that have at least 10 registers for
+ * allocation by GCC, and that can access memory at reg+const in one
+ * instruction.
+ */
+
+#include <linux/types.h>
+#include <linux/compiler.h>
+#include <asm/byteorder.h>
+
+/*
+ * The macros defined in this file are:
+ *
+ * BYTE_COPY_FWD(dst_beg_ptr, src_beg_ptr, nbytes_to_copy)
+ *
+ * BYTE_COPY_BWD(dst_end_ptr, src_end_ptr, nbytes_to_copy)
+ *
+ * WORD_COPY_FWD(dst_beg_ptr, src_beg_ptr, nbytes_remaining, nbytes_to_copy)
+ *
+ * WORD_COPY_BWD(dst_end_ptr, src_end_ptr, nbytes_remaining, nbytes_to_copy)
+ *
+ * MERGE(old_word, sh_1, new_word, sh_2)
+ *
+ * MEM_COPY_FWD(dst_beg_ptr, src_beg_ptr, nbytes_to_copy)
+ *
+ * MEM_COPY_BWD(dst_end_ptr, src_end_ptr, nbytes_to_copy)
+ */
+
+#define OP_T_THRESHOLD 16
+
+/*
+ * Type to use for aligned memory operations.
+ * This should normally be the biggest type supported by a single load
+ * and store.
+ */
+#define op_t unsigned long int
+#define OPSIZ (sizeof(op_t))
+
+/* Type to use for unaligned operations. */
+typedef unsigned char byte;
+
+#ifndef MERGE
+# ifdef __LITTLE_ENDIAN
+# define MERGE(w0, sh_1, w1, sh_2) (((w0) >> (sh_1)) | ((w1) << (sh_2)))
+# elif defined(__BIG_ENDIAN)
+# define MERGE(w0, sh_1, w1, sh_2) (((w0) << (sh_1)) | ((w1) >> (sh_2)))
+# else
+# error "Macro MERGE() hasn't defined!"
+# endif
+#endif
+
+/*
+ * Copy exactly NBYTES bytes from SRC_BP to DST_BP,
+ * without any assumptions about alignment of the pointers.
+ */
+#ifndef BYTE_COPY_FWD
+#define BYTE_COPY_FWD(dst_bp, src_bp, nbytes) \
+do { \
+ size_t __nbytes = (nbytes); \
+ while (__nbytes > 0) { \
+ byte __x = ((byte *) src_bp)[0]; \
+ src_bp += 1; \
+ __nbytes -= 1; \
+ ((byte *) dst_bp)[0] = __x; \
+ dst_bp += 1; \
+ } \
+} while (0)
+#endif
+
+/*
+ * Copy exactly NBYTES_TO_COPY bytes from SRC_END_PTR to DST_END_PTR,
+ * beginning at the bytes right before the pointers and continuing towards
+ * smaller addresses. Don't assume anything about alignment of the
+ * pointers.
+ */
+#ifndef BYTE_COPY_BWD
+#define BYTE_COPY_BWD(dst_ep, src_ep, nbytes) \
+do { \
+ size_t __nbytes = (nbytes); \
+ while (__nbytes > 0) { \
+ byte __x; \
+ src_ep -= 1; \
+ __x = ((byte *) src_ep)[0]; \
+ dst_ep -= 1; \
+ __nbytes -= 1; \
+ ((byte *) dst_ep)[0] = __x; \
+ } \
+} while (0)
+#endif
+/*
+ * Copy *up to* NBYTES bytes from SRC_BP to DST_BP, with
+ * the assumption that DST_BP is aligned on an OPSIZ multiple. If
+ * not all bytes could be easily copied, store remaining number of bytes
+ * in NBYTES_LEFT, otherwise store 0.
+ */
+extern void _wordcopy_fwd_aligned(long int, long int, size_t);
+extern void _wordcopy_fwd_dest_aligned(long int, long int, size_t);
+#ifndef WORD_COPY_FWD
+#define WORD_COPY_FWD(dst_bp, src_bp, nbytes_left, nbytes) \
+do { \
+ if (src_bp % OPSIZ == 0) \
+ _wordcopy_fwd_aligned (dst_bp, src_bp, (nbytes) / OPSIZ); \
+ else \
+ _wordcopy_fwd_dest_aligned (dst_bp, src_bp, (nbytes) / OPSIZ);\
+ \
+ src_bp += (nbytes) & -OPSIZ; \
+ dst_bp += (nbytes) & -OPSIZ; \
+ (nbytes_left) = (nbytes) % OPSIZ; \
+} while (0)
+#endif
+
+/*
+ * Copy *up to* NBYTES_TO_COPY bytes from SRC_END_PTR to DST_END_PTR,
+ * beginning at the words (of type op_t) right before the pointers and
+ * continuing towards smaller addresses. May take advantage of that
+ * DST_END_PTR is aligned on an OPSIZ multiple. If not all bytes could be
+ * easily copied, store remaining number of bytes in NBYTES_REMAINING,
+ * otherwise store 0.
+ */
+extern void _wordcopy_bwd_aligned(long int, long int, size_t);
+extern void _wordcopy_bwd_dest_aligned(long int, long int, size_t);
+#ifndef WORD_COPY_BWD
+#define WORD_COPY_BWD(dst_ep, src_ep, nbytes_left, nbytes) \
+do { \
+ if (src_ep % OPSIZ == 0) \
+ _wordcopy_bwd_aligned (dst_ep, src_ep, (nbytes) / OPSIZ); \
+ else \
+ _wordcopy_bwd_dest_aligned (dst_ep, src_ep, (nbytes) / OPSIZ);\
+ \
+ src_ep -= (nbytes) & -OPSIZ; \
+ dst_ep -= (nbytes) & -OPSIZ; \
+ (nbytes_left) = (nbytes) % OPSIZ; \
+} while (0)
+#endif
+
+/* Copy memory from the beginning to the end */
+#ifndef MEM_COPY_FWD
+static __always_inline void mem_copy_fwd(unsigned long dstp,
+ unsigned long srcp,
+ size_t count)
+{
+ /* If there not too few bytes to copy, use word copy. */
+ if (count >= OP_T_THRESHOLD) {
+ /* Copy just a few bytes to make dstp aligned. */
+ count -= (-dstp) % OPSIZ;
+ BYTE_COPY_FWD(dstp, srcp, (-dstp) % OPSIZ);
+
+ /*
+ * Copy from srcp to dstp taking advantage of the known
+ * alignment of dstp. Number if bytes remaining is put in
+ * the third argument.
+ */
+ WORD_COPY_FWD(dstp, srcp, count, count);
+
+ /* Fall out and copy the tail. */
+ }
+
+ /* There are just a few bytes to copy. Use byte memory operations. */
+ BYTE_COPY_FWD(dstp, srcp, count);
+}
+#endif
+
+/* Copy memory from the end to the beginning. */
+#ifndef MEM_COPY_BWD
+static __always_inline void mem_copy_bwd(unsigned long dstp,
+ unsigned long srcp,
+ size_t count)
+{
+ srcp += count;
+ dstp += count;
+
+ /* If there not too few bytes to copy, use word copy. */
+ if (count >= OP_T_THRESHOLD) {
+ /* Copy just a few bytes to make dstp aligned. */
+ count -= dstp % OPSIZ;
+ BYTE_COPY_BWD(dstp, srcp, dstp % OPSIZ);
+
+ /*
+ * Copy from srcp to dstp taking advantage of the known
+ * alignment of dstp. Number if bytes remaining is put in
+ * the third argument.
+ */
+ WORD_COPY_BWD(dstp, srcp, count, count);
+
+ /* Fall out and copy the tail. */
+ }
+
+ /* There are just a few bytes to copy. Use byte memory operations. */
+ BYTE_COPY_BWD (dstp, srcp, count);
+}
+#endif
+
+#endif
diff --git a/lib/Makefile b/lib/Makefile
index e6a3763..10a319e 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -12,7 +12,8 @@ lib-y := ctype.o string.o vsprintf.o cmdline.o \
idr.o int_sqrt.o extable.o prio_tree.o \
sha1.o irq_regs.o reciprocal_div.o argv_split.o \
proportions.o prio_heap.o ratelimit.o show_mem.o \
- is_single_threaded.o plist.o decompress.o flex_array.o
+ is_single_threaded.o plist.o decompress.o flex_array.o \
+ memcopy.o
lib-$(CONFIG_MMU) += ioremap.o
lib-$(CONFIG_SMP) += cpumask.o
diff --git a/lib/memcopy.c b/lib/memcopy.c
new file mode 100644
index 0000000..70fb6b2
--- /dev/null
+++ b/lib/memcopy.c
@@ -0,0 +1,403 @@
+/*
+ * memcopy.c -- subroutines for memory copy functions.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2.1 of the License, or (at your option)
+ * any later version.
+ *
+ * This program 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 General
+ * Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
+ *
+ * The code is derived from the GNU C Library.
+ * Copyright (C) 1991, 1992, 1993, 1997, 2004 Free Software Foundation, Inc.
+ */
+
+/* BE VERY CAREFUL IF YOU CHANGE THIS CODE...! */
+
+#include <linux/memcopy.h>
+
+/*
+ * _wordcopy_fwd_aligned -- Copy block beginning at SRCP to block beginning
+ * at DSTP with LEN `op_t' words (not LEN bytes!).
+ * Both SRCP and DSTP should be aligned for memory operations on `op_t's.
+ */
+void _wordcopy_fwd_aligned (long int dstp, long int srcp, size_t len)
+{
+ op_t a0, a1;
+
+ switch (len % 8) {
+ case 2:
+ a0 = ((op_t *) srcp)[0];
+ srcp -= 6 * OPSIZ;
+ dstp -= 7 * OPSIZ;
+ len += 6;
+ goto do1;
+ case 3:
+ a1 = ((op_t *) srcp)[0];
+ srcp -= 5 * OPSIZ;
+ dstp -= 6 * OPSIZ;
+ len += 5;
+ goto do2;
+ case 4:
+ a0 = ((op_t *) srcp)[0];
+ srcp -= 4 * OPSIZ;
+ dstp -= 5 * OPSIZ;
+ len += 4;
+ goto do3;
+ case 5:
+ a1 = ((op_t *) srcp)[0];
+ srcp -= 3 * OPSIZ;
+ dstp -= 4 * OPSIZ;
+ len += 3;
+ goto do4;
+ case 6:
+ a0 = ((op_t *) srcp)[0];
+ srcp -= 2 * OPSIZ;
+ dstp -= 3 * OPSIZ;
+ len += 2;
+ goto do5;
+ case 7:
+ a1 = ((op_t *) srcp)[0];
+ srcp -= 1 * OPSIZ;
+ dstp -= 2 * OPSIZ;
+ len += 1;
+ goto do6;
+ case 0:
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ return;
+ a0 = ((op_t *) srcp)[0];
+ srcp -= 0 * OPSIZ;
+ dstp -= 1 * OPSIZ;
+ goto do7;
+ case 1:
+ a1 = ((op_t *) srcp)[0];
+ srcp -=-1 * OPSIZ;
+ dstp -= 0 * OPSIZ;
+ len -= 1;
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ goto do0;
+ goto do8; /* No-op. */
+ }
+
+ do {
+do8:
+ a0 = ((op_t *) srcp)[0];
+ ((op_t *) dstp)[0] = a1;
+do7:
+ a1 = ((op_t *) srcp)[1];
+ ((op_t *) dstp)[1] = a0;
+do6:
+ a0 = ((op_t *) srcp)[2];
+ ((op_t *) dstp)[2] = a1;
+do5:
+ a1 = ((op_t *) srcp)[3];
+ ((op_t *) dstp)[3] = a0;
+do4:
+ a0 = ((op_t *) srcp)[4];
+ ((op_t *) dstp)[4] = a1;
+do3:
+ a1 = ((op_t *) srcp)[5];
+ ((op_t *) dstp)[5] = a0;
+do2:
+ a0 = ((op_t *) srcp)[6];
+ ((op_t *) dstp)[6] = a1;
+do1:
+ a1 = ((op_t *) srcp)[7];
+ ((op_t *) dstp)[7] = a0;
+
+ srcp += 8 * OPSIZ;
+ dstp += 8 * OPSIZ;
+ len -= 8;
+ } while (len != 0);
+
+ /*
+ * This is the right position for do0. Please don't move it into
+ * the loop.
+ */
+do0:
+ ((op_t *) dstp)[0] = a1;
+}
+
+/*
+ * _wordcopy_fwd_dest_aligned -- Copy block beginning at SRCP to block
+ * beginning at DSTP with LEN `op_t' words (not LEN bytes!). DSTP should
+ * be aligned for memory operations on `op_t's, but SRCP must *not* be aligned.
+ */
+
+void _wordcopy_fwd_dest_aligned (long int dstp, long int srcp, size_t len)
+{
+ op_t a0, a1, a2, a3;
+ int sh_1, sh_2;
+
+ /*
+ * Calculate how to shift a word read at the memory operation aligned
+ * srcp to make it aligned for copy.
+ */
+ sh_1 = 8 * (srcp % OPSIZ);
+ sh_2 = 8 * OPSIZ - sh_1;
+
+ /*
+ * Make SRCP aligned by rounding it down to the beginning of the `op_t'
+ * it points in the middle of.
+ */
+ srcp &= -OPSIZ;
+
+ switch (len % 4) {
+ case 2:
+ a1 = ((op_t *) srcp)[0];
+ a2 = ((op_t *) srcp)[1];
+ srcp -= 1 * OPSIZ;
+ dstp -= 3 * OPSIZ;
+ len += 2;
+ goto do1;
+ case 3:
+ a0 = ((op_t *) srcp)[0];
+ a1 = ((op_t *) srcp)[1];
+ srcp -= 0 * OPSIZ;
+ dstp -= 2 * OPSIZ;
+ len += 1;
+ goto do2;
+ case 0:
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ return;
+ a3 = ((op_t *) srcp)[0];
+ a0 = ((op_t *) srcp)[1];
+ srcp -=-1 * OPSIZ;
+ dstp -= 1 * OPSIZ;
+ len += 0;
+ goto do3;
+ case 1:
+ a2 = ((op_t *) srcp)[0];
+ a3 = ((op_t *) srcp)[1];
+ srcp -=-2 * OPSIZ;
+ dstp -= 0 * OPSIZ;
+ len -= 1;
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ goto do0;
+ goto do4; /* No-op. */
+ }
+
+ do {
+do4:
+ a0 = ((op_t *) srcp)[0];
+ ((op_t *) dstp)[0] = MERGE (a2, sh_1, a3, sh_2);
+do3:
+ a1 = ((op_t *) srcp)[1];
+ ((op_t *) dstp)[1] = MERGE (a3, sh_1, a0, sh_2);
+do2:
+ a2 = ((op_t *) srcp)[2];
+ ((op_t *) dstp)[2] = MERGE (a0, sh_1, a1, sh_2);
+do1:
+ a3 = ((op_t *) srcp)[3];
+ ((op_t *) dstp)[3] = MERGE (a1, sh_1, a2, sh_2);
+
+ srcp += 4 * OPSIZ;
+ dstp += 4 * OPSIZ;
+ len -= 4;
+ } while (len != 0);
+
+ /*
+ * This is the right position for do0. Please don't move it into
+ * the loop.
+ */
+do0:
+ ((op_t *) dstp)[0] = MERGE (a2, sh_1, a3, sh_2);
+}
+
+/*
+ * _wordcopy_bwd_aligned -- Copy block finishing right before
+ * SRCP to block finishing right before DSTP with LEN `op_t' words (not LEN
+ * bytes!). Both SRCP and DSTP should be aligned for memory operations
+ * on `op_t's.
+ */
+void _wordcopy_bwd_aligned (long int dstp, long int srcp, size_t len)
+{
+ op_t a0, a1;
+
+ switch (len % 8) {
+ case 2:
+ srcp -= 2 * OPSIZ;
+ dstp -= 1 * OPSIZ;
+ a0 = ((op_t *) srcp)[1];
+ len += 6;
+ goto do1;
+ case 3:
+ srcp -= 3 * OPSIZ;
+ dstp -= 2 * OPSIZ;
+ a1 = ((op_t *) srcp)[2];
+ len += 5;
+ goto do2;
+ case 4:
+ srcp -= 4 * OPSIZ;
+ dstp -= 3 * OPSIZ;
+ a0 = ((op_t *) srcp)[3];
+ len += 4;
+ goto do3;
+ case 5:
+ srcp -= 5 * OPSIZ;
+ dstp -= 4 * OPSIZ;
+ a1 = ((op_t *) srcp)[4];
+ len += 3;
+ goto do4;
+ case 6:
+ srcp -= 6 * OPSIZ;
+ dstp -= 5 * OPSIZ;
+ a0 = ((op_t *) srcp)[5];
+ len += 2;
+ goto do5;
+ case 7:
+ srcp -= 7 * OPSIZ;
+ dstp -= 6 * OPSIZ;
+ a1 = ((op_t *) srcp)[6];
+ len += 1;
+ goto do6;
+ case 0:
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ return;
+ srcp -= 8 * OPSIZ;
+ dstp -= 7 * OPSIZ;
+ a0 = ((op_t *) srcp)[7];
+ goto do7;
+ case 1:
+ srcp -= 9 * OPSIZ;
+ dstp -= 8 * OPSIZ;
+ a1 = ((op_t *) srcp)[8];
+ len -= 1;
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ goto do0;
+ goto do8; /* No-op. */
+ }
+
+ do {
+do8:
+ a0 = ((op_t *) srcp)[7];
+ ((op_t *) dstp)[7] = a1;
+do7:
+ a1 = ((op_t *) srcp)[6];
+ ((op_t *) dstp)[6] = a0;
+do6:
+ a0 = ((op_t *) srcp)[5];
+ ((op_t *) dstp)[5] = a1;
+do5:
+ a1 = ((op_t *) srcp)[4];
+ ((op_t *) dstp)[4] = a0;
+do4:
+ a0 = ((op_t *) srcp)[3];
+ ((op_t *) dstp)[3] = a1;
+do3:
+ a1 = ((op_t *) srcp)[2];
+ ((op_t *) dstp)[2] = a0;
+do2:
+ a0 = ((op_t *) srcp)[1];
+ ((op_t *) dstp)[1] = a1;
+do1:
+ a1 = ((op_t *) srcp)[0];
+ ((op_t *) dstp)[0] = a0;
+
+ srcp -= 8 * OPSIZ;
+ dstp -= 8 * OPSIZ;
+ len -= 8;
+ } while (len != 0);
+
+ /*
+ * This is the right position for do0. Please don't move it into
+ * the loop.
+ */
+do0:
+ ((op_t *) dstp)[7] = a1;
+}
+
+/*
+ * _wordcopy_bwd_dest_aligned -- Copy block finishing right before SRCP to
+ * block finishing right before DSTP with LEN `op_t' words (not LEN bytes!).
+ * DSTP should be aligned for memory operations on `op_t', but SRCP must *not*
+ * be aligned.
+ */
+void _wordcopy_bwd_dest_aligned (long int dstp, long int srcp, size_t len)
+{
+ op_t a0, a1, a2, a3;
+ int sh_1, sh_2;
+
+ /*
+ * Calculate how to shift a word read at the memory operation aligned
+ * srcp to make it aligned for copy.
+ */
+
+ sh_1 = 8 * (srcp % OPSIZ);
+ sh_2 = 8 * OPSIZ - sh_1;
+
+ /*
+ * Make srcp aligned by rounding it down to the beginning of the op_t
+ * it points in the middle of.
+ */
+ srcp &= -OPSIZ;
+ srcp += OPSIZ;
+
+ switch (len % 4) {
+ case 2:
+ srcp -= 3 * OPSIZ;
+ dstp -= 1 * OPSIZ;
+ a2 = ((op_t *) srcp)[2];
+ a1 = ((op_t *) srcp)[1];
+ len += 2;
+ goto do1;
+ case 3:
+ srcp -= 4 * OPSIZ;
+ dstp -= 2 * OPSIZ;
+ a3 = ((op_t *) srcp)[3];
+ a2 = ((op_t *) srcp)[2];
+ len += 1;
+ goto do2;
+ case 0:
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ return;
+ srcp -= 5 * OPSIZ;
+ dstp -= 3 * OPSIZ;
+ a0 = ((op_t *) srcp)[4];
+ a3 = ((op_t *) srcp)[3];
+ goto do3;
+ case 1:
+ srcp -= 6 * OPSIZ;
+ dstp -= 4 * OPSIZ;
+ a1 = ((op_t *) srcp)[5];
+ a0 = ((op_t *) srcp)[4];
+ len -= 1;
+ if (OP_T_THRESHOLD <= 3 * OPSIZ && len == 0)
+ goto do0;
+ goto do4; /* No-op. */
+ }
+
+ do {
+do4:
+ a3 = ((op_t *) srcp)[3];
+ ((op_t *) dstp)[3] = MERGE (a0, sh_1, a1, sh_2);
+do3:
+ a2 = ((op_t *) srcp)[2];
+ ((op_t *) dstp)[2] = MERGE (a3, sh_1, a0, sh_2);
+do2:
+ a1 = ((op_t *) srcp)[1];
+ ((op_t *) dstp)[1] = MERGE (a2, sh_1, a3, sh_2);
+do1:
+ a0 = ((op_t *) srcp)[0];
+ ((op_t *) dstp)[0] = MERGE (a1, sh_1, a2, sh_2);
+
+ srcp -= 4 * OPSIZ;
+ dstp -= 4 * OPSIZ;
+ len -= 4;
+ } while (len != 0);
+
+ /*
+ * This is the right position for do0. Please don't move it into
+ * the loop.
+ */
+do0:
+ ((op_t *) dstp)[3] = MERGE (a0, sh_1, a1, sh_2);
+}
+
--
1.7.0.1
On Wed, 2010-09-01 at 18:36 +0800, Miao Xie wrote:
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms of the GNU 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 kernel is GPL v2, see COPYING. GPL v2.1 is not a suitable license.
On Wed, 01 Sep 2010 17:25:36 +0200, Peter Zijlstra wrote:
> On Wed, 2010-09-01 at 18:36 +0800, Miao Xie wrote:
>> + * This program is free software; you can redistribute it and/or modify it
>> + * under the terms of the GNU 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 kernel is GPL v2, see COPYING. GPL v2.1 is not a suitable license.
Ok, I will change it.
Thanks
Miao
Hiya,
On 02/09/10 15:44, Miao Xie wrote:
> Ok, I will change it.
Did you just change the "2.1" to "2" in your patch, or did you find
a specifically version 2 licensed version of that code ?
If so, where was that ?
cheers!
Chris
--
Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
On Thu, 02 Sep 2010 15:55:44 +1000, Chris Samuel wrote:
> Hiya,
>
> On 02/09/10 15:44, Miao Xie wrote:
>
>> Ok, I will change it.
>
> Did you just change the "2.1" to "2" in your patch, or did you find
> a specifically version 2 licensed version of that code ?
>
> If so, where was that ?
I just change the "2.1" to "2" in your patch, because the orignal code
is LGPL v2.1, LGPL v2.1 permits us to apply the terms of the ordinary
GNU General Public License instead of it.
This is the clause of LGPL v2.1:
3. You may opt to apply the terms of the ordinary GNU General Public
License instead of this License to a given copy of the Library. To do
this, you must alter all the notices that refer to this License, so
that they refer to the ordinary GNU General Public License, version 2,
instead of to this License. (If a newer version than version 2 of the
ordinary GNU General Public License has appeared, then you can specify
that version instead if you wish.) Do not make any other change in these
notices.
Once this change is made in a given copy, it is irreversible for that copy,
so the ordinary GNU General Public License applies to all subsequent copies
and derivative works made from that copy.
This option is useful when you wish to copy part of the code of the Library
into a program that is not a library.
The following is the URL of LGPL v2.1:
http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html
Thanks
Miao
On 02/09/10 16:50, Miao Xie wrote:
> I just change the "2.1" to "2" in your patch, because the
> orignal code is LGPL v2.1, LGPL v2.1 permits us to apply
> the terms of the ordinary GNU General Public License instead
> of it.
Ahhh excellent, I hadn't realised that was possible; well spotted!
cheers,
Chris
--
Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
On Thu, 2010-09-02 at 13:44 +0800, Miao Xie wrote:
> On Wed, 01 Sep 2010 17:25:36 +0200, Peter Zijlstra wrote:
> > On Wed, 2010-09-01 at 18:36 +0800, Miao Xie wrote:
> >> + * This program is free software; you can redistribute it and/or modify it
> >> + * under the terms of the GNU 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 kernel is GPL v2, see COPYING. GPL v2.1 is not a suitable license.
>
> Ok, I will change it.
Are you allowed to change it? from what I understand the FSF owns the
copyright of that code, that least, that's what the preamble I cut away
implied.
On Thu, 2010-09-02 at 09:55 +0200, Peter Zijlstra wrote:
> On Thu, 2010-09-02 at 13:44 +0800, Miao Xie wrote:
> > On Wed, 01 Sep 2010 17:25:36 +0200, Peter Zijlstra wrote:
> > > On Wed, 2010-09-01 at 18:36 +0800, Miao Xie wrote:
> > >> + * This program is free software; you can redistribute it and/or modify it
> > >> + * under the terms of the GNU 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 kernel is GPL v2, see COPYING. GPL v2.1 is not a suitable license.
> >
> > Ok, I will change it.
>
> Are you allowed to change it? from what I understand the FSF owns the
> copyright of that code, that least, that's what the preamble I cut away
> implied.
n/m, I should probably have read more inbox before replying.
On 2 September 2010 14:07, Chris Samuel <[email protected]> wrote:
> On 02/09/10 16:50, Miao Xie wrote:
>
>> I just change the "2.1" to "2" in your patch, because the
>> orignal code is LGPL v2.1, LGPL v2.1 permits us to apply
>> the terms of the ordinary GNU General Public License instead
>> of it.
> Ahhh excellent, I hadn't realised that was possible; well spotted!
Umm, isn't the only one that can do that the copyright holder?
Disclaimer: I am not a laywer
On 02/09/10 18:16, [email protected] wrote:
> Umm, isn't the only one that can do that the copyright holder?
The copyright holder can use whatever license they wish; the LGPL
tells the licensee what rights *they* have, which includes distributing
the software under the (more strict) GPLv2.
cheers,
Chris (not a lawyer either, thankfully!)
--
Chris Samuel : http://www.csamuel.org/ : Melbourne, VIC
On 2 September 2010 15:24, Chris Samuel <[email protected]> wrote:
> On 02/09/10 18:16, [email protected] wrote:
>
>> Umm, isn't the only one that can do that the copyright holder?
>
> The copyright holder can use whatever license they wish; the LGPL
> tells the licensee what rights *they* have, which includes distributing
> the software under the (more strict) GPLv2.
>
Wow, there is even a FAQ.
http://www.gnu.org/licenses/gpl-faq.html#compat-matrix-footnote-7
in the matrix notes 1 and 7 apply, which means that anyone, not just
the license holder may change to GPL v2 from LGPV 2.1
Cheers
Chris