API tools faq
paste
Advertisement
Guest User

[PATCH v2] tile: optimize strnlen using SIMD instructions

a guest
Sep 2nd, 2013
77
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 5.38 KB | None | 0 0
raw download clone embed print report
  1. Using strlen as a model, add length checking to create strnlen.
  2.  
  3. Signed-off-by: Ken Steele <ken@tilera.com>
  4. Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
  5. ---
  6. v2 of the patch adds an "& 31" to the shift count for 32-bit strnlen.
  7. It turns out this doesn't affect the generated code at all, and it's
  8. technically more correct, as pointed out by Max Filippov <jcmvbkbc@gmail.com>.
  9.  
  10. arch/tile/include/asm/string.h | 2 ++
  11. arch/tile/lib/Makefile | 2 +-
  12. arch/tile/lib/strnlen_32.c | 47 +++++++++++++++++++++++++++++++++++++++++
  13. arch/tile/lib/strnlen_64.c | 48 ++++++++++++++++++++++++++++++++++++++++++
  14. 4 files changed, 98 insertions(+), 1 deletion(-)
  15. create mode 100644 arch/tile/lib/strnlen_32.c
  16. create mode 100644 arch/tile/lib/strnlen_64.c
  17.  
  18. diff --git a/arch/tile/include/asm/string.h b/arch/tile/include/asm/string.h
  19. index 7535cf1..92b271b 100644
  20. --- a/arch/tile/include/asm/string.h
  21. +++ b/arch/tile/include/asm/string.h
  22. @@ -21,8 +21,10 @@
  23. #define __HAVE_ARCH_MEMMOVE
  24. #define __HAVE_ARCH_STRCHR
  25. #define __HAVE_ARCH_STRLEN
  26. +#define __HAVE_ARCH_STRNLEN
  27.  
  28. extern __kernel_size_t strlen(const char *);
  29. +extern __kernel_size_t strnlen(const char *, __kernel_size_t);
  30. extern char *strchr(const char *s, int c);
  31. extern void *memchr(const void *s, int c, size_t n);
  32. extern void *memset(void *, int, __kernel_size_t);
  33. diff --git a/arch/tile/lib/Makefile b/arch/tile/lib/Makefile
  34. index 985f598..5d84437 100644
  35. --- a/arch/tile/lib/Makefile
  36. +++ b/arch/tile/lib/Makefile
  37. @@ -4,7 +4,7 @@
  38.  
  39. lib-y = cacheflush.o checksum.o cpumask.o delay.o uaccess.o \
  40. memmove.o memcpy_$(BITS).o memchr_$(BITS).o memset_$(BITS).o \
  41. - strchr_$(BITS).o strlen_$(BITS).o
  42. + strchr_$(BITS).o strlen_$(BITS).o strnlen_$(BITS).o
  43.  
  44. ifeq ($(CONFIG_TILEGX),y)
  45. CFLAGS_REMOVE_memcpy_user_64.o = -fno-omit-frame-pointer
  46. diff --git a/arch/tile/lib/strnlen_32.c b/arch/tile/lib/strnlen_32.c
  47. new file mode 100644
  48. index 0000000..1434141
  49. --- /dev/null
  50. +++ b/arch/tile/lib/strnlen_32.c
  51. @@ -0,0 +1,47 @@
  52. +/*
  53. + * Copyright 2013 Tilera Corporation. All Rights Reserved.
  54. + *
  55. + * This program is free software; you can redistribute it and/or
  56. + * modify it under the terms of the GNU General Public License
  57. + * as published by the Free Software Foundation, version 2.
  58. + *
  59. + * This program is distributed in the hope that it will be useful, but
  60. + * WITHOUT ANY WARRANTY; without even the implied warranty of
  61. + * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  62. + * NON INFRINGEMENT. See the GNU General Public License for
  63. + * more details.
  64. + */
  65. +
  66. +#include <linux/types.h>
  67. +#include <linux/string.h>
  68. +#include <linux/module.h>
  69. +
  70. +size_t strnlen(const char *s, size_t count)
  71. +{
  72. + /* Get an aligned pointer. */
  73. + const uintptr_t s_int = (uintptr_t) s;
  74. + const uint32_t *p = (const uint32_t *)(s_int & -4);
  75. + size_t bytes_read = sizeof(*p) - (s_int & (sizeof(*p) - 1));
  76. + size_t len;
  77. + uint32_t v, bits;
  78. +
  79. + /* Avoid page fault risk by not reading any bytes when count is 0. */
  80. + if (count == 0)
  81. + return 0;
  82. +
  83. + /* Read first word, but force bytes before the string to be nonzero. */
  84. + v = *p | ((1 << ((s_int << 3) & 31)) - 1);
  85. +
  86. + while ((bits = __insn_seqb(v, 0)) == 0) {
  87. + if (bytes_read >= count) {
  88. + /* Read COUNT bytes and didn't find the terminator. */
  89. + return count;
  90. + }
  91. + v = *++p;
  92. + bytes_read += sizeof(v);
  93. + }
  94. +
  95. + len = ((const char *) p) + (__insn_ctz(bits) >> 3) - s;
  96. + return (len < count ? len : count);
  97. +}
  98. +EXPORT_SYMBOL(strnlen);
  99. diff --git a/arch/tile/lib/strnlen_64.c b/arch/tile/lib/strnlen_64.c
  100. new file mode 100644
  101. index 0000000..2e8de6a
  102. --- /dev/null
  103. +++ b/arch/tile/lib/strnlen_64.c
  104. @@ -0,0 +1,48 @@
  105. +/*
  106. + * Copyright 2013 Tilera Corporation. All Rights Reserved.
  107. + *
  108. + * This program is free software; you can redistribute it and/or
  109. + * modify it under the terms of the GNU General Public License
  110. + * as published by the Free Software Foundation, version 2.
  111. + *
  112. + * This program is distributed in the hope that it will be useful, but
  113. + * WITHOUT ANY WARRANTY; without even the implied warranty of
  114. + * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  115. + * NON INFRINGEMENT. See the GNU General Public License for
  116. + * more details.
  117. + */
  118. +
  119. +#include <linux/types.h>
  120. +#include <linux/string.h>
  121. +#include <linux/module.h>
  122. +#include "string-endian.h"
  123. +
  124. +size_t strnlen(const char *s, size_t count)
  125. +{
  126. + /* Get an aligned pointer. */
  127. + const uintptr_t s_int = (uintptr_t) s;
  128. + const uint64_t *p = (const uint64_t *)(s_int & -8);
  129. + size_t bytes_read = sizeof(*p) - (s_int & (sizeof(*p) - 1));
  130. + size_t len;
  131. + uint64_t v, bits;
  132. +
  133. + /* Avoid page fault risk by not reading any bytes when count is 0. */
  134. + if (count == 0)
  135. + return 0;
  136. +
  137. + /* Read and MASK the first word. */
  138. + v = *p | MASK(s_int);
  139. +
  140. + while ((bits = __insn_v1cmpeqi(v, 0)) == 0) {
  141. + if (bytes_read >= count) {
  142. + /* Read COUNT bytes and didn't find the terminator. */
  143. + return count;
  144. + }
  145. + v = *++p;
  146. + bytes_read += sizeof(v);
  147. + }
  148. +
  149. + len = ((const char *) p) + (CFZ(bits) >> 3) - s;
  150. + return (len < count ? len : count);
  151. +}
  152. +EXPORT_SYMBOL(strnlen);
  153. --
  154. 1.8.3.1
  155.  
  156. --
  157. To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
  158. the body of a message to majordomo@vger.kernel.org
  159. More majordomo info at http://vger.kernel.org/majordomo-info.html
  160. Please read the FAQ at http://www.tux.org/lkml/
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!