github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-11-19 10:34:25 +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

61
src/nfa/sheng_impl4.h
View File

@@ -283,7 +283,7 @@ 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,
@@ -320,8 +320,15 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
return MO_CONTINUE_MATCHING;
}
#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(end - cur_buf >= 4)) {
const u8 *b1 = cur_buf;
@@ -333,6 +340,23 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
const u8 c3 = *b3;
const u8 c4 = *b4;
#if defined(HAVE_SVE)
svuint8_t succ_mask1 = svld1(lane_pred_32, (const u8*)(masks+c1));
cur_state = svtbl(succ_mask1, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a1 = svlastb(lane_pred_32, cur_state);
svuint8_t succ_mask2 = svld1(lane_pred_32, (const u8*)(masks+c2));
cur_state = svtbl(succ_mask2, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a2 = svlastb(lane_pred_32, cur_state);
svuint8_t succ_mask3 = svld1(lane_pred_32, (const u8*)(masks+c3));
cur_state = svtbl(succ_mask3, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a3 = svlastb(lane_pred_32, cur_state);
svuint8_t succ_mask4 = svld1(lane_pred_32, (const u8*)(masks+c4));
cur_state = svtbl(succ_mask4, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a4 = svlastb(lane_pred_32, cur_state);
#else
const m512 succ_mask1 = masks[c1];
cur_state = vpermb512(cur_state, succ_mask1);
const u8 a1 = movd512(cur_state);
@@ -348,6 +372,7 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
const m512 succ_mask4 = masks[c4];
cur_state = vpermb512(cur_state, succ_mask4);
const u8 a4 = movd512(cur_state);
#endif
DEBUG_PRINTF("c: %02hhx '%c'\n", c1, ourisprint(c1) ? c1 : '?');
DEBUG_PRINTF("s: %u (flag: %u)\n", a1 & SHENG32_STATE_MASK,
@@ -517,7 +542,11 @@ char SHENG32_IMPL(u8 *state, NfaCallback cb, void *ctxt,
};
cur_buf += 4;
}
#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;
}
@@ -541,9 +570,15 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
*scan_end = end;
return MO_CONTINUE_MATCHING;
}
#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(end - cur_buf >= 4)) {
const u8 *b1 = cur_buf;
@@ -555,6 +590,23 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
const u8 c3 = *b3;
const u8 c4 = *b4;
#if defined(HAVE_SVE)
svuint8_t succ_mask1 = svld1(lane_pred_64, (const u8*)(masks+c1));
cur_state = svtbl(succ_mask1, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a1 = svlastb(lane_pred_64, cur_state);
svuint8_t succ_mask2 = svld1(lane_pred_64, (const u8*)(masks+c2));
cur_state = svtbl(succ_mask2, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a2 = svlastb(lane_pred_64, cur_state);
svuint8_t succ_mask3 = svld1(lane_pred_64, (const u8*)(masks+c3));
cur_state = svtbl(succ_mask3, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a3 = svlastb(lane_pred_64, cur_state);
svuint8_t succ_mask4 = svld1(lane_pred_64, (const u8*)(masks+c4));
cur_state = svtbl(succ_mask4, svand_x(svptrue_b8(), tbl_mask, cur_state));
const u8 a4 = svlastb(lane_pred_64, cur_state);
#else
const m512 succ_mask1 = masks[c1];
cur_state = vpermb512(cur_state, succ_mask1);
const u8 a1 = movd512(cur_state);
@@ -570,6 +622,7 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
const m512 succ_mask4 = masks[c4];
cur_state = vpermb512(cur_state, succ_mask4);
const u8 a4 = movd512(cur_state);
#endif
DEBUG_PRINTF("c: %02hhx '%c'\n", c1, ourisprint(c1) ? c1 : '?');
DEBUG_PRINTF("s: %u (flag: %u)\n", a1 & SHENG64_STATE_MASK,
@@ -703,7 +756,11 @@ char SHENG64_IMPL(u8 *state, NfaCallback cb, void *ctxt,
}
cur_buf += 4;
}
#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;
}