From 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sat, 16 Apr 2005 15:20:36 -0700 Subject: Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip! --- include/asm-arm/unaligned.h | 191 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 191 insertions(+) create mode 100644 include/asm-arm/unaligned.h (limited to 'include/asm-arm/unaligned.h') diff --git a/include/asm-arm/unaligned.h b/include/asm-arm/unaligned.h new file mode 100644 index 000000000000..1b39c2f322c9 --- /dev/null +++ b/include/asm-arm/unaligned.h @@ -0,0 +1,191 @@ +#ifndef __ASM_ARM_UNALIGNED_H +#define __ASM_ARM_UNALIGNED_H + +#include + +extern int __bug_unaligned_x(void *ptr); + +/* + * What is the most efficient way of loading/storing an unaligned value? + * + * That is the subject of this file. Efficiency here is defined as + * minimum code size with minimum register usage for the common cases. + * It is currently not believed that long longs are common, so we + * trade efficiency for the chars, shorts and longs against the long + * longs. + * + * Current stats with gcc 2.7.2.2 for these functions: + * + * ptrsize get: code regs put: code regs + * 1 1 1 1 2 + * 2 3 2 3 2 + * 4 7 3 7 3 + * 8 20 6 16 6 + * + * gcc 2.95.1 seems to code differently: + * + * ptrsize get: code regs put: code regs + * 1 1 1 1 2 + * 2 3 2 3 2 + * 4 7 4 7 4 + * 8 19 8 15 6 + * + * which may or may not be more efficient (depending upon whether + * you can afford the extra registers). Hopefully the gcc 2.95 + * is inteligent enough to decide if it is better to use the + * extra register, but evidence so far seems to suggest otherwise. + * + * Unfortunately, gcc is not able to optimise the high word + * out of long long >> 32, or the low word from long long << 32 + */ + +#define __get_unaligned_2_le(__p) \ + (__p[0] | __p[1] << 8) + +#define __get_unaligned_2_be(__p) \ + (__p[0] << 8 | __p[1]) + +#define __get_unaligned_4_le(__p) \ + (__p[0] | __p[1] << 8 | __p[2] << 16 | __p[3] << 24) + +#define __get_unaligned_4_be(__p) \ + (__p[0] << 24 | __p[1] << 16 | __p[2] << 8 | __p[3]) + +#define __get_unaligned_le(ptr) \ + ({ \ + __typeof__(*(ptr)) __v; \ + __u8 *__p = (__u8 *)(ptr); \ + switch (sizeof(*(ptr))) { \ + case 1: __v = *(ptr); break; \ + case 2: __v = __get_unaligned_2_le(__p); break; \ + case 4: __v = __get_unaligned_4_le(__p); break; \ + case 8: { \ + unsigned int __v1, __v2; \ + __v2 = __get_unaligned_4_le((__p+4)); \ + __v1 = __get_unaligned_4_le(__p); \ + __v = ((unsigned long long)__v2 << 32 | __v1); \ + } \ + break; \ + default: __v = __bug_unaligned_x(__p); break; \ + } \ + __v; \ + }) + +#define __get_unaligned_be(ptr) \ + ({ \ + __typeof__(*(ptr)) __v; \ + __u8 *__p = (__u8 *)(ptr); \ + switch (sizeof(*(ptr))) { \ + case 1: __v = *(ptr); break; \ + case 2: __v = __get_unaligned_2_be(__p); break; \ + case 4: __v = __get_unaligned_4_be(__p); break; \ + case 8: { \ + unsigned int __v1, __v2; \ + __v2 = __get_unaligned_4_be(__p); \ + __v1 = __get_unaligned_4_be((__p+4)); \ + __v = ((unsigned long long)__v2 << 32 | __v1); \ + } \ + break; \ + default: __v = __bug_unaligned_x(__p); break; \ + } \ + __v; \ + }) + + +static inline void __put_unaligned_2_le(__u32 __v, register __u8 *__p) +{ + *__p++ = __v; + *__p++ = __v >> 8; +} + +static inline void __put_unaligned_2_be(__u32 __v, register __u8 *__p) +{ + *__p++ = __v >> 8; + *__p++ = __v; +} + +static inline void __put_unaligned_4_le(__u32 __v, register __u8 *__p) +{ + __put_unaligned_2_le(__v >> 16, __p + 2); + __put_unaligned_2_le(__v, __p); +} + +static inline void __put_unaligned_4_be(__u32 __v, register __u8 *__p) +{ + __put_unaligned_2_be(__v >> 16, __p); + __put_unaligned_2_be(__v, __p + 2); +} + +static inline void __put_unaligned_8_le(const unsigned long long __v, register __u8 *__p) +{ + /* + * tradeoff: 8 bytes of stack for all unaligned puts (2 + * instructions), or an extra register in the long long + * case - go for the extra register. + */ + __put_unaligned_4_le(__v >> 32, __p+4); + __put_unaligned_4_le(__v, __p); +} + +static inline void __put_unaligned_8_be(const unsigned long long __v, register __u8 *__p) +{ + /* + * tradeoff: 8 bytes of stack for all unaligned puts (2 + * instructions), or an extra register in the long long + * case - go for the extra register. + */ + __put_unaligned_4_be(__v >> 32, __p); + __put_unaligned_4_be(__v, __p+4); +} + +/* + * Try to store an unaligned value as efficiently as possible. + */ +#define __put_unaligned_le(val,ptr) \ + ({ \ + switch (sizeof(*(ptr))) { \ + case 1: \ + *(ptr) = (val); \ + break; \ + case 2: __put_unaligned_2_le((val),(__u8 *)(ptr)); \ + break; \ + case 4: __put_unaligned_4_le((val),(__u8 *)(ptr)); \ + break; \ + case 8: __put_unaligned_8_le((val),(__u8 *)(ptr)); \ + break; \ + default: __bug_unaligned_x(ptr); \ + break; \ + } \ + (void) 0; \ + }) + +#define __put_unaligned_be(val,ptr) \ + ({ \ + switch (sizeof(*(ptr))) { \ + case 1: \ + *(ptr) = (val); \ + break; \ + case 2: __put_unaligned_2_be((val),(__u8 *)(ptr)); \ + break; \ + case 4: __put_unaligned_4_be((val),(__u8 *)(ptr)); \ + break; \ + case 8: __put_unaligned_8_be((val),(__u8 *)(ptr)); \ + break; \ + default: __bug_unaligned_x(ptr); \ + break; \ + } \ + (void) 0; \ + }) + +/* + * Select endianness + */ +#ifndef __ARMEB__ +#define get_unaligned __get_unaligned_le +#define put_unaligned __put_unaligned_le +#else +#define get_unaligned __get_unaligned_be +#define put_unaligned __put_unaligned_be +#endif + +#endif -- cgit v1.2.1