diff options
Diffstat (limited to 'neozip/arch/x86/chunkset_sse2.c')
| -rw-r--r-- | neozip/arch/x86/chunkset_sse2.c | 50 |
1 files changed, 50 insertions, 0 deletions
diff --git a/neozip/arch/x86/chunkset_sse2.c b/neozip/arch/x86/chunkset_sse2.c new file mode 100644 index 0000000000..633ab6e64f --- /dev/null +++ b/neozip/arch/x86/chunkset_sse2.c @@ -0,0 +1,50 @@ +/* chunkset_sse2.c -- SSE2 inline functions to copy small data chunks. + * For conditions of distribution and use, see copyright notice in zlib.h + */ + +#ifdef X86_SSE2 + +#include "zbuild.h" +#include "zmemory.h" + +#include <immintrin.h> + +typedef __m128i chunk_t; + +#define HAVE_CHUNKMEMSET_2 +#define HAVE_CHUNKMEMSET_4 +#define HAVE_CHUNKMEMSET_8 + +static inline void chunkmemset_2(uint8_t *from, chunk_t *chunk) { + *chunk = _mm_set1_epi16(zng_memread_2(from)); +} + +static inline void chunkmemset_4(uint8_t *from, chunk_t *chunk) { + *chunk = _mm_set1_epi32(zng_memread_4(from)); +} + +static inline void chunkmemset_8(uint8_t *from, chunk_t *chunk) { + *chunk = _mm_set1_epi64x(zng_memread_8(from)); +} + +static inline void loadchunk(uint8_t const *s, chunk_t *chunk) { + *chunk = _mm_loadu_si128((__m128i *)s); +} + +static inline void storechunk(uint8_t *out, chunk_t *chunk) { + _mm_storeu_si128((__m128i *)out, *chunk); +} + +#define CHUNKSIZE chunksize_sse2 +#define CHUNKCOPY chunkcopy_sse2 +#define CHUNKUNROLL chunkunroll_sse2 +#define CHUNKMEMSET chunkmemset_sse2 +#define CHUNKMEMSET_SAFE chunkmemset_safe_sse2 + +#include "chunkset_tpl.h" + +#define INFLATE_FAST inflate_fast_sse2 + +#include "inffast_tpl.h" + +#endif |
