From fd52bc6dc4bfb844995cc63d98682970de1c9fed Mon Sep 17 00:00:00 2001 From: Ulrich Drepper Date: Fri, 28 Oct 2011 18:18:04 -0400 Subject: Clean up x86-64 strcasestr Actually describe in the C code what is going on. --- sysdeps/x86_64/multiarch/strstr.c | 25 ++++++++++++------------- 1 file changed, 12 insertions(+), 13 deletions(-) (limited to 'sysdeps/x86_64') diff --git a/sysdeps/x86_64/multiarch/strstr.c b/sysdeps/x86_64/multiarch/strstr.c index b408b752fa..6e93744140 100644 --- a/sysdeps/x86_64/multiarch/strstr.c +++ b/sysdeps/x86_64/multiarch/strstr.c @@ -1,5 +1,5 @@ /* strstr with SSE4.2 intrinsics - Copyright (C) 2009, 2010 Free Software Foundation, Inc. + Copyright (C) 2009, 2010, 2011 Free Software Foundation, Inc. Contributed by Intel Corporation. This file is part of the GNU C Library. @@ -106,24 +106,22 @@ __m128i_strloadu (const unsigned char * p) #if defined USE_AS_STRCASESTR && !defined STRCASESTR_NONASCII /* Similar to __m128i_strloadu. Convert to lower case for POSIX/C - locale. */ + locale and other which have single-byte letters only in the ASCII + range. */ static inline __m128i -__m128i_strloadu_tolower (const unsigned char *p, __m128i rangeuc, - __m128i u2ldelta) +__m128i_strloadu_tolower (const unsigned char *p, __m128i uclow, + __m128i uchigh, __m128i lcqword) { __m128i frag = __m128i_strloadu (p); -#define UCLOW 0x4040404040404040ULL -#define UCHIGH 0x5b5b5b5b5b5b5b5bULL -#define LCQWORD 0x2020202020202020ULL /* Compare if 'Z' > bytes. Inverted way to get a mask for byte <= 'Z'. */ - __m128i r2 = _mm_cmpgt_epi8 (_mm_set1_epi64x (UCHIGH), frag); + __m128i r2 = _mm_cmpgt_epi8 (uchigh, frag); /* Compare if bytes are > 'A' - 1. */ - __m128i r1 = _mm_cmpgt_epi8 (frag, _mm_set1_epi64x (UCLOW)); + __m128i r1 = _mm_cmpgt_epi8 (frag, uclow); /* Mask byte == ff if byte(r2) <= 'Z' and byte(r1) > 'A' - 1. */ __m128i mask = _mm_and_si128 (r2, r1); /* Apply lowercase bit 6 mask for above mask bytes == ff. */ - return _mm_or_si128 (frag, _mm_and_si128 (mask, _mm_set1_epi64x (LCQWORD))); + return _mm_or_si128 (frag, _mm_and_si128 (mask, lcqword)); } #endif @@ -190,9 +188,10 @@ STRSTR_SSE42 (const unsigned char *s1, const unsigned char *s2) != 0, 0)) return __strcasestr_sse42_nonascii (s1, s2); - const __m128i rangeuc = _mm_set_epi64x (0x0, 0x5a41); - const __m128i u2ldelta = _mm_set1_epi64x (0xe0e0e0e0e0e0e0e0); -# define strloadu(p) __m128i_strloadu_tolower (p, rangeuc, u2ldelta) + const __m128i uclow = _mm_set1_epi8 (0x40); + const __m128i uchigh = _mm_set1_epi8 (0x5b); + const __m128i lcqword = _mm_set1_epi8 (0x20); +# define strloadu(p) __m128i_strloadu_tolower (p, uclow, uchigh, lcqword) # else # define strloadu __m128i_strloadu_tolower # endif -- cgit v1.2.3