weaselab/conflict-set
2
0
Fork 0
Code Issues 4 Packages Releases 13 Activity

Use signed compare
Some checks failed
Tests / Clang total: 3296, failed: 38, passed: 3258
Details
Tests / 64 bit versions total: 3296, passed: 3296
Details
Tests / Debug total: 3294, failed: 76, passed: 3218
Details
Tests / SIMD fallback total: 3296, passed: 3296
Details
Tests / Release [gcc] total: 3296, failed: 38, passed: 3258
Details
Tests / Release [gcc,aarch64] total: 2458, passed: 2458
Details
Tests / Coverage total: 2476, failed: 76, passed: 2400
Details
weaselab/conflict-set/pipeline/head There was a failure building this commit
Details

Browse Source 
Unsigned compare is incorrect here
This commit is contained in:
Andrew Noyes
2024-09-13 22:55:22 -07:00
parent 0814822d82
commit 52201fa4c7
1 changed files with 10 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
ConflictSet.cpp
View File

@@ -2925,11 +2925,18 @@ horizontalMaxUpTo16(InternalVersionT *vs, InternalVersionT z, int len) {
uint32_t zero; uint32_t zero;
memcpy(&zero, &z, sizeof(zero)); memcpy(&zero, &z, sizeof(zero));
auto zeroVec = _mm512_set1_epi32(zero); auto zeroVec = _mm512_set1_epi32(zero);
return InternalVersionT( auto actual = InternalVersionT(
zero + zero +
_mm512_reduce_max_epu32(_mm512_sub_epi32( _mm512_reduce_max_epi32(_mm512_sub_epi32(
_mm512_mask_loadu_epi32(zeroVec, _mm512_int2mask((1 << len) - 1), vs), _mm512_mask_loadu_epi32(zeroVec, _mm512_int2mask((1 << len) - 1), vs),
zeroVec))); zeroVec)));
// Hope it gets vectorized
InternalVersionT max = vs[0];
for (int i = 1; i < len; ++i) {
max = std::max(vs[i], max);
}
assert(actual == max);
return max;
#endif #endif
} }
__attribute__((target("default"))) __attribute__((target("default")))
@@ -2959,7 +2966,7 @@ horizontalMax16(InternalVersionT *vs, InternalVersionT z) {
uint32_t zero; uint32_t zero;
memcpy(&zero, &z, sizeof(zero)); memcpy(&zero, &z, sizeof(zero));
auto zeroVec = _mm512_set1_epi32(zero); auto zeroVec = _mm512_set1_epi32(zero);
return InternalVersionT(zero + _mm512_reduce_max_epu32(_mm512_sub_epi32( return InternalVersionT(zero + _mm512_reduce_max_epi32(_mm512_sub_epi32(
_mm512_loadu_epi32(vs), zeroVec))); _mm512_loadu_epi32(vs), zeroVec)));
#endif #endif
} }