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/loongarch/slide_hash_lasx.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/loongarch/slide_hash_lasx.c')
| -rw-r--r-- | neozip/arch/loongarch/slide_hash_lasx.c | 49 |
1 files changed, 49 insertions, 0 deletions
diff --git a/neozip/arch/loongarch/slide_hash_lasx.c b/neozip/arch/loongarch/slide_hash_lasx.c new file mode 100644 index 0000000000..f464779090 --- /dev/null +++ b/neozip/arch/loongarch/slide_hash_lasx.c @@ -0,0 +1,49 @@ +/* + * LASX optimized hash slide, based on Intel AVX2 implementation + * + * Copyright (C) 2017 Intel Corporation + * Copyright (C) 2025 Vladislav Shchapov <vladislav@shchapov.ru> + * Authors: + * Arjan van de Ven <arjan@linux.intel.com> + * Jim Kukunas <james.t.kukunas@linux.intel.com> + * Mika T. Lindqvist <postmaster@raasu.org> + * + * For conditions of distribution and use, see copyright notice in zlib.h + */ + +#ifdef LOONGARCH_LASX + +#include "zbuild.h" +#include "deflate.h" + +#include <lasxintrin.h> + +static inline void slide_hash_chain(Pos *table, uint32_t entries, const __m256i wsize) { + table += entries; + table -= 32; + + do { + __m256i value1, value2, result1, result2; + + value1 = __lasx_xvld(table, 0); + value2 = __lasx_xvld(table, 32); + result1 = __lasx_xvssub_hu(value1, wsize); + result2 = __lasx_xvssub_hu(value2, wsize); + __lasx_xvst(result1, table, 0); + __lasx_xvst(result2, table, 32); + + table -= 32; + entries -= 32; + } while (entries > 0); +} + +Z_INTERNAL void slide_hash_lasx(deflate_state *s) { + Assert(s->w_size <= UINT16_MAX, "w_size should fit in uint16_t"); + uint16_t wsize = (uint16_t)s->w_size; + const __m256i ymm_wsize = __lasx_xvreplgr2vr_h((short)wsize); + + slide_hash_chain(s->head, HASH_SIZE, ymm_wsize); + slide_hash_chain(s->prev, wsize, ymm_wsize); +} + +#endif |
