diff options
| author | Mehmet Samet Duman <yongdohyun@projecttick.org> | 2026-04-02 19:56:09 +0300 |
|---|---|---|
| committer | Mehmet Samet Duman <yongdohyun@projecttick.org> | 2026-04-02 19:56:09 +0300 |
| commit | 7fb132859fda54aa96bc9dd46d302b343eeb5a02 (patch) | |
| tree | b43ae77d7451fb470a260c03349a1caf2846c5e5 /neozip/arch/arm/slide_hash_armv6.c | |
| parent | b1e34e861b5d732afe828d58aad2c638135061fd (diff) | |
| parent | c2712b8a345191f6ed79558c089777df94590087 (diff) | |
| download | Project-Tick-7fb132859fda54aa96bc9dd46d302b343eeb5a02.tar.gz Project-Tick-7fb132859fda54aa96bc9dd46d302b343eeb5a02.zip | |
Add 'neozip/' from commit 'c2712b8a345191f6ed79558c089777df94590087'
git-subtree-dir: neozip
git-subtree-mainline: b1e34e861b5d732afe828d58aad2c638135061fd
git-subtree-split: c2712b8a345191f6ed79558c089777df94590087
Diffstat (limited to 'neozip/arch/arm/slide_hash_armv6.c')
| -rw-r--r-- | neozip/arch/arm/slide_hash_armv6.c | 49 |
1 files changed, 49 insertions, 0 deletions
diff --git a/neozip/arch/arm/slide_hash_armv6.c b/neozip/arch/arm/slide_hash_armv6.c new file mode 100644 index 0000000000..b241e6c5e6 --- /dev/null +++ b/neozip/arch/arm/slide_hash_armv6.c @@ -0,0 +1,49 @@ +/* slide_hash_armv6.c -- Optimized hash table shifting for ARMv6 with support for SIMD instructions + * Copyright (C) 2023 Cameron Cawley + * For conditions of distribution and use, see copyright notice in zlib.h + */ + +#ifdef ARM_SIMD + +#include "zbuild.h" +#include "acle_intrins.h" +#include "deflate.h" + +/* SIMD version of hash_chain rebase */ +static inline void slide_hash_chain(Pos *table, uint32_t entries, uint16_t wsize) { + Z_REGISTER uint16x2_t v; + uint16x2_t p0, p1, p2, p3; + Z_REGISTER size_t n; + + size_t size = entries*sizeof(table[0]); + Assert((size % (sizeof(uint16x2_t) * 4) == 0), "hash table size err"); + + Assert(sizeof(Pos) == 2, "Wrong Pos size"); + v = wsize | (wsize << 16); + + n = size / (sizeof(uint16x2_t) * 4); + do { + p0 = *((const uint16x2_t *)(table)); + p1 = *((const uint16x2_t *)(table+2)); + p2 = *((const uint16x2_t *)(table+4)); + p3 = *((const uint16x2_t *)(table+6)); + p0 = __uqsub16(p0, v); + p1 = __uqsub16(p1, v); + p2 = __uqsub16(p2, v); + p3 = __uqsub16(p3, v); + *((uint16x2_t *)(table)) = p0; + *((uint16x2_t *)(table+2)) = p1; + *((uint16x2_t *)(table+4)) = p2; + *((uint16x2_t *)(table+6)) = p3; + table += 8; + } while (--n); +} + +Z_INTERNAL void slide_hash_armv6(deflate_state *s) { + Assert(s->w_size <= UINT16_MAX, "w_size should fit in uint16_t"); + uint16_t wsize = (uint16_t)s->w_size; + + slide_hash_chain(s->head, HASH_SIZE, wsize); + slide_hash_chain(s->prev, wsize, wsize); +} +#endif |
