github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-11-18 02:00:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

Enable sheng32 and sheng64 on Arm

Browse Source 
Signed-off-by: Yoan Picchi <yoan.picchi@arm.com>
This commit is contained in:
Yoan Picchi
2024-04-04 09:46:23 +00:00
parent 94ee8b9c1c
commit b839bc3033
6 changed files with 153 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
src/nfa/sheng_impl.h
View File

@@ -96,7 +96,7 @@ char SHENG_IMPL(u8 *state, NfaCallback cb, void *ctxt, const struct sheng *s,
return MO_CONTINUE_MATCHING;
}
#if defined(HAVE_AVX512VBMI)
#if defined(HAVE_AVX512VBMI) || defined(HAVE_SVE)
static really_inline
char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
const struct sheng32 *s,
@@ -114,14 +114,28 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
}
DEBUG_PRINTF("Scanning %lli bytes\n", (s64a)(end - start));
#if defined(HAVE_SVE)
const svbool_t lane_pred_32 = svwhilelt_b8(0, 32);
svuint8_t cur_state = svdup_u8(*state);
svuint8_t tbl_mask = svdup_u8((unsigned char)0x1F);
const m512 *masks = s->succ_masks;
#else
m512 cur_state = set1_64x8(*state);
const m512 *masks = s->succ_masks;
#endif
while (likely(cur_buf != end)) {
const u8 c = *cur_buf;
#if defined(HAVE_SVE)
svuint8_t succ_mask = svld1(lane_pred_32, (const u8*)(masks + c));
cur_state = svtbl(succ_mask, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 tmp = svlastb(lane_pred_32, cur_state);
#else
const m512 succ_mask = masks[c];
cur_state = vpermb512(cur_state, succ_mask);
const u8 tmp = movd512(cur_state);
#endif
DEBUG_PRINTF("c: %02hhx '%c'\n", c, ourisprint(c) ? c : '?');
DEBUG_PRINTF("s: %u (flag: %u)\n", tmp & SHENG32_STATE_MASK,
@@ -153,7 +167,11 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
}
cur_buf++;
}
#if defined(HAVE_SVE)
*state = svlastb(lane_pred_32, cur_state);
#else
*state = movd512(cur_state);
#endif
*scan_end = cur_buf;
return MO_CONTINUE_MATCHING;
}
@@ -175,14 +193,28 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
}
DEBUG_PRINTF("Scanning %lli bytes\n", (s64a)(end - start));
#if defined(HAVE_SVE)
const svbool_t lane_pred_64 = svwhilelt_b8(0, 64);
svuint8_t cur_state = svdup_u8(*state);
svuint8_t tbl_mask = svdup_u8((unsigned char)0x3F);
const m512 *masks = s->succ_masks;
#else
m512 cur_state = set1_64x8(*state);
const m512 *masks = s->succ_masks;
#endif
while (likely(cur_buf != end)) {
const u8 c = *cur_buf;
#if defined(HAVE_SVE)
svuint8_t succ_mask = svld1(lane_pred_64, (const u8*)(masks + c));
cur_state = svtbl(succ_mask, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 tmp = svlastb(lane_pred_64, cur_state);
#else
const m512 succ_mask = masks[c];
cur_state = vpermb512(cur_state, succ_mask);
const u8 tmp = movd512(cur_state);
#endif
DEBUG_PRINTF("c: %02hhx '%c'\n", c, ourisprint(c) ? c : '?');
DEBUG_PRINTF("s: %u (flag: %u)\n", tmp & SHENG64_STATE_MASK,
@@ -214,7 +246,11 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
}
cur_buf++;
}
#if defined(HAVE_SVE)
*state = svlastb(lane_pred_64, cur_state);
#else
*state = movd512(cur_state);
#endif
*scan_end = cur_buf;
return MO_CONTINUE_MATCHING;
}