datasketches 0.2.0 → 0.2.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +16 -0
- data/LICENSE +40 -3
- data/NOTICE +1 -1
- data/README.md +7 -7
- data/ext/datasketches/extconf.rb +1 -1
- data/ext/datasketches/theta_wrapper.cpp +20 -4
- data/lib/datasketches/version.rb +1 -1
- data/vendor/datasketches-cpp/CMakeLists.txt +31 -3
- data/vendor/datasketches-cpp/LICENSE +40 -3
- data/vendor/datasketches-cpp/MANIFEST.in +3 -0
- data/vendor/datasketches-cpp/NOTICE +1 -1
- data/vendor/datasketches-cpp/README.md +76 -9
- data/vendor/datasketches-cpp/cmake/DataSketchesConfig.cmake.in +10 -0
- data/vendor/datasketches-cpp/common/CMakeLists.txt +14 -13
- data/vendor/datasketches-cpp/common/include/MurmurHash3.h +11 -7
- data/vendor/datasketches-cpp/common/include/binomial_bounds.hpp +8 -8
- data/vendor/datasketches-cpp/common/include/bounds_binomial_proportions.hpp +12 -15
- data/vendor/datasketches-cpp/common/include/common_defs.hpp +26 -0
- data/vendor/datasketches-cpp/common/include/conditional_forward.hpp +20 -8
- data/vendor/datasketches-cpp/common/include/count_zeros.hpp +2 -2
- data/vendor/datasketches-cpp/common/include/serde.hpp +7 -7
- data/vendor/datasketches-cpp/cpc/CMakeLists.txt +15 -35
- data/vendor/datasketches-cpp/cpc/include/cpc_common.hpp +10 -3
- data/vendor/datasketches-cpp/cpc/include/cpc_compressor.hpp +19 -19
- data/vendor/datasketches-cpp/cpc/include/cpc_compressor_impl.hpp +91 -89
- data/vendor/datasketches-cpp/cpc/include/cpc_sketch.hpp +15 -2
- data/vendor/datasketches-cpp/cpc/include/cpc_sketch_impl.hpp +126 -90
- data/vendor/datasketches-cpp/cpc/include/cpc_union.hpp +1 -1
- data/vendor/datasketches-cpp/cpc/include/cpc_union_impl.hpp +22 -20
- data/vendor/datasketches-cpp/cpc/include/cpc_util.hpp +10 -10
- data/vendor/datasketches-cpp/cpc/include/icon_estimator.hpp +4 -4
- data/vendor/datasketches-cpp/cpc/include/u32_table.hpp +8 -8
- data/vendor/datasketches-cpp/cpc/include/u32_table_impl.hpp +14 -14
- data/vendor/datasketches-cpp/cpc/test/compression_test.cpp +10 -10
- data/vendor/datasketches-cpp/cpc/test/cpc_sketch_allocation_test.cpp +17 -0
- data/vendor/datasketches-cpp/cpc/test/cpc_sketch_test.cpp +25 -0
- data/vendor/datasketches-cpp/cpc/test/cpc_union_test.cpp +1 -1
- data/vendor/datasketches-cpp/fi/CMakeLists.txt +5 -15
- data/vendor/datasketches-cpp/fi/include/frequent_items_sketch_impl.hpp +69 -82
- data/vendor/datasketches-cpp/fi/include/reverse_purge_hash_map_impl.hpp +10 -10
- data/vendor/datasketches-cpp/fi/test/reverse_purge_hash_map_test.cpp +2 -2
- data/vendor/datasketches-cpp/hll/CMakeLists.txt +33 -56
- data/vendor/datasketches-cpp/hll/include/AuxHashMap-internal.hpp +60 -63
- data/vendor/datasketches-cpp/hll/include/AuxHashMap.hpp +19 -19
- data/vendor/datasketches-cpp/hll/include/CompositeInterpolationXTable-internal.hpp +15 -15
- data/vendor/datasketches-cpp/hll/include/CompositeInterpolationXTable.hpp +3 -3
- data/vendor/datasketches-cpp/hll/include/CouponHashSet-internal.hpp +74 -76
- data/vendor/datasketches-cpp/hll/include/CouponHashSet.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/CouponList-internal.hpp +110 -113
- data/vendor/datasketches-cpp/hll/include/CouponList.hpp +13 -13
- data/vendor/datasketches-cpp/hll/include/CubicInterpolation-internal.hpp +2 -4
- data/vendor/datasketches-cpp/hll/include/HarmonicNumbers-internal.hpp +1 -1
- data/vendor/datasketches-cpp/hll/include/Hll4Array-internal.hpp +80 -76
- data/vendor/datasketches-cpp/hll/include/Hll4Array.hpp +9 -9
- data/vendor/datasketches-cpp/hll/include/Hll6Array-internal.hpp +26 -26
- data/vendor/datasketches-cpp/hll/include/Hll6Array.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/Hll8Array-internal.hpp +33 -33
- data/vendor/datasketches-cpp/hll/include/Hll8Array.hpp +6 -6
- data/vendor/datasketches-cpp/hll/include/HllArray-internal.hpp +205 -209
- data/vendor/datasketches-cpp/hll/include/HllArray.hpp +36 -36
- data/vendor/datasketches-cpp/hll/include/HllSketch-internal.hpp +34 -32
- data/vendor/datasketches-cpp/hll/include/HllSketchImpl-internal.hpp +22 -22
- data/vendor/datasketches-cpp/hll/include/HllSketchImpl.hpp +13 -13
- data/vendor/datasketches-cpp/hll/include/HllSketchImplFactory.hpp +15 -15
- data/vendor/datasketches-cpp/hll/include/HllUnion-internal.hpp +61 -61
- data/vendor/datasketches-cpp/hll/include/HllUtil.hpp +120 -127
- data/vendor/datasketches-cpp/hll/include/coupon_iterator-internal.hpp +9 -9
- data/vendor/datasketches-cpp/hll/include/coupon_iterator.hpp +5 -5
- data/vendor/datasketches-cpp/hll/include/hll.hpp +21 -21
- data/vendor/datasketches-cpp/hll/test/AuxHashMapTest.cpp +1 -1
- data/vendor/datasketches-cpp/hll/test/CouponHashSetTest.cpp +34 -34
- data/vendor/datasketches-cpp/hll/test/CouponListTest.cpp +25 -25
- data/vendor/datasketches-cpp/hll/test/CrossCountingTest.cpp +2 -2
- data/vendor/datasketches-cpp/hll/test/HllArrayTest.cpp +35 -35
- data/vendor/datasketches-cpp/hll/test/HllSketchTest.cpp +15 -15
- data/vendor/datasketches-cpp/hll/test/HllUnionTest.cpp +10 -14
- data/vendor/datasketches-cpp/hll/test/IsomorphicTest.cpp +3 -3
- data/vendor/datasketches-cpp/hll/test/ToFromByteArrayTest.cpp +4 -4
- data/vendor/datasketches-cpp/kll/CMakeLists.txt +9 -19
- data/vendor/datasketches-cpp/kll/include/kll_helper.hpp +5 -4
- data/vendor/datasketches-cpp/kll/include/kll_helper_impl.hpp +6 -6
- data/vendor/datasketches-cpp/kll/include/kll_quantile_calculator.hpp +14 -6
- data/vendor/datasketches-cpp/kll/include/kll_quantile_calculator_impl.hpp +39 -24
- data/vendor/datasketches-cpp/kll/include/kll_sketch.hpp +41 -4
- data/vendor/datasketches-cpp/kll/include/kll_sketch_impl.hpp +76 -64
- data/vendor/datasketches-cpp/kll/include/kolmogorov_smirnov.hpp +67 -0
- data/vendor/datasketches-cpp/kll/include/kolmogorov_smirnov_impl.hpp +78 -0
- data/vendor/datasketches-cpp/kll/test/CMakeLists.txt +1 -0
- data/vendor/datasketches-cpp/kll/test/kll_sketch_test.cpp +133 -46
- data/vendor/datasketches-cpp/kll/test/kolmogorov_smirnov_test.cpp +111 -0
- data/vendor/datasketches-cpp/pyproject.toml +4 -2
- data/vendor/datasketches-cpp/python/CMakeLists.txt +10 -6
- data/vendor/datasketches-cpp/python/README.md +50 -50
- data/vendor/datasketches-cpp/python/pybind11Path.cmd +3 -0
- data/vendor/datasketches-cpp/python/src/cpc_wrapper.cpp +1 -1
- data/vendor/datasketches-cpp/python/src/hll_wrapper.cpp +4 -4
- data/vendor/datasketches-cpp/python/src/kll_wrapper.cpp +1 -1
- data/vendor/datasketches-cpp/python/src/theta_wrapper.cpp +8 -8
- data/vendor/datasketches-cpp/python/src/vector_of_kll.cpp +11 -5
- data/vendor/datasketches-cpp/python/src/vo_wrapper.cpp +2 -2
- data/vendor/datasketches-cpp/python/tests/hll_test.py +1 -1
- data/vendor/datasketches-cpp/python/tests/kll_test.py +2 -2
- data/vendor/datasketches-cpp/python/tests/req_test.py +2 -2
- data/vendor/datasketches-cpp/python/tests/vector_of_kll_test.py +4 -4
- data/vendor/datasketches-cpp/python/tests/vo_test.py +3 -3
- data/vendor/datasketches-cpp/req/CMakeLists.txt +8 -21
- data/vendor/datasketches-cpp/req/include/req_common.hpp +2 -1
- data/vendor/datasketches-cpp/req/include/req_compactor.hpp +4 -4
- data/vendor/datasketches-cpp/req/include/req_compactor_impl.hpp +26 -39
- data/vendor/datasketches-cpp/req/include/req_sketch.hpp +1 -1
- data/vendor/datasketches-cpp/req/include/req_sketch_impl.hpp +13 -11
- data/vendor/datasketches-cpp/req/test/req_sketch_test.cpp +52 -52
- data/vendor/datasketches-cpp/sampling/CMakeLists.txt +5 -9
- data/vendor/datasketches-cpp/sampling/include/var_opt_sketch.hpp +10 -5
- data/vendor/datasketches-cpp/sampling/include/var_opt_sketch_impl.hpp +61 -64
- data/vendor/datasketches-cpp/sampling/include/var_opt_union_impl.hpp +42 -48
- data/vendor/datasketches-cpp/sampling/test/var_opt_sketch_test.cpp +6 -6
- data/vendor/datasketches-cpp/sampling/test/var_opt_union_test.cpp +13 -13
- data/vendor/datasketches-cpp/setup.py +10 -7
- data/vendor/datasketches-cpp/theta/CMakeLists.txt +26 -45
- data/vendor/datasketches-cpp/theta/include/bounds_on_ratios_in_sampled_sets.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/compact_theta_sketch_parser.hpp +67 -0
- data/vendor/datasketches-cpp/theta/include/compact_theta_sketch_parser_impl.hpp +137 -0
- data/vendor/datasketches-cpp/theta/include/theta_constants.hpp +9 -4
- data/vendor/datasketches-cpp/theta/include/theta_helpers.hpp +15 -0
- data/vendor/datasketches-cpp/theta/include/theta_intersection.hpp +9 -4
- data/vendor/datasketches-cpp/theta/include/theta_intersection_base_impl.hpp +6 -6
- data/vendor/datasketches-cpp/theta/include/theta_intersection_impl.hpp +1 -1
- data/vendor/datasketches-cpp/theta/include/theta_jaccard_similarity_base.hpp +18 -14
- data/vendor/datasketches-cpp/theta/include/theta_set_difference_base_impl.hpp +2 -2
- data/vendor/datasketches-cpp/theta/include/theta_sketch.hpp +73 -15
- data/vendor/datasketches-cpp/theta/include/theta_sketch_impl.hpp +247 -103
- data/vendor/datasketches-cpp/theta/include/theta_union.hpp +10 -5
- data/vendor/datasketches-cpp/theta/include/theta_union_base.hpp +3 -1
- data/vendor/datasketches-cpp/theta/include/theta_union_base_impl.hpp +9 -3
- data/vendor/datasketches-cpp/theta/include/theta_union_impl.hpp +8 -5
- data/vendor/datasketches-cpp/theta/include/theta_update_sketch_base.hpp +11 -5
- data/vendor/datasketches-cpp/theta/include/theta_update_sketch_base_impl.hpp +70 -37
- data/vendor/datasketches-cpp/theta/test/CMakeLists.txt +1 -0
- data/vendor/datasketches-cpp/theta/test/theta_a_not_b_test.cpp +23 -1
- data/vendor/datasketches-cpp/theta/test/theta_compact_empty_from_java_v1.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_compact_empty_from_java_v2.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_compact_estimation_from_java_v1.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_compact_estimation_from_java_v2.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_compact_exact_from_java.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_intersection_test.cpp +21 -1
- data/vendor/datasketches-cpp/theta/test/theta_jaccard_similarity_test.cpp +58 -2
- data/vendor/datasketches-cpp/theta/test/theta_setop_test.cpp +445 -0
- data/vendor/datasketches-cpp/theta/test/theta_sketch_test.cpp +437 -1
- data/vendor/datasketches-cpp/theta/test/theta_union_test.cpp +41 -9
- data/vendor/datasketches-cpp/tuple/CMakeLists.txt +18 -33
- data/vendor/datasketches-cpp/tuple/include/array_of_doubles_sketch.hpp +1 -1
- data/vendor/datasketches-cpp/tuple/include/array_of_doubles_sketch_impl.hpp +50 -63
- data/vendor/datasketches-cpp/tuple/include/array_of_doubles_union.hpp +1 -1
- data/vendor/datasketches-cpp/tuple/include/array_of_doubles_union_impl.hpp +3 -3
- data/vendor/datasketches-cpp/tuple/include/tuple_sketch.hpp +13 -9
- data/vendor/datasketches-cpp/tuple/include/tuple_sketch_impl.hpp +84 -78
- data/vendor/datasketches-cpp/tuple/include/tuple_union.hpp +6 -1
- data/vendor/datasketches-cpp/tuple/include/tuple_union_impl.hpp +8 -3
- data/vendor/datasketches-cpp/tuple/test/array_of_doubles_sketch_test.cpp +17 -1
- data/vendor/datasketches-cpp/tuple/test/tuple_a_not_b_test.cpp +17 -17
- data/vendor/datasketches-cpp/tuple/test/tuple_intersection_test.cpp +12 -12
- data/vendor/datasketches-cpp/tuple/test/tuple_jaccard_similarity_test.cpp +5 -5
- data/vendor/datasketches-cpp/tuple/test/tuple_sketch_allocation_test.cpp +1 -1
- data/vendor/datasketches-cpp/tuple/test/tuple_sketch_test.cpp +66 -28
- data/vendor/datasketches-cpp/tuple/test/tuple_union_test.cpp +19 -12
- metadata +18 -7
- data/vendor/datasketches-cpp/theta/test/theta_update_empty_from_java.sk +0 -0
- data/vendor/datasketches-cpp/theta/test/theta_update_estimation_from_java.sk +0 -0
@@ -53,9 +53,7 @@ first_interesting_column(0),
|
|
53
53
|
kxp(1 << lg_k),
|
54
54
|
hip_est_accum(0)
|
55
55
|
{
|
56
|
-
|
57
|
-
throw std::invalid_argument("lg_k must be >= " + std::to_string(CPC_MIN_LG_K) + " and <= " + std::to_string(CPC_MAX_LG_K) + ": " + std::to_string(lg_k));
|
58
|
-
}
|
56
|
+
check_lg_k(lg_k);
|
59
57
|
}
|
60
58
|
|
61
59
|
template<typename A>
|
@@ -176,7 +174,7 @@ void cpc_sketch_alloc<A>::update(float value) {
|
|
176
174
|
|
177
175
|
static inline uint32_t row_col_from_two_hashes(uint64_t hash0, uint64_t hash1, uint8_t lg_k) {
|
178
176
|
if (lg_k > 26) throw std::logic_error("lg_k > 26");
|
179
|
-
const
|
177
|
+
const uint32_t k = 1 << lg_k;
|
180
178
|
uint8_t col = count_leading_zeros_in_u64(hash1); // 0 <= col <= 64
|
181
179
|
if (col > 63) col = 63; // clip so that 0 <= col <= 63
|
182
180
|
const uint32_t row = hash0 & (k - 1);
|
@@ -188,7 +186,7 @@ static inline uint32_t row_col_from_two_hashes(uint64_t hash0, uint64_t hash1, u
|
|
188
186
|
}
|
189
187
|
|
190
188
|
template<typename A>
|
191
|
-
void cpc_sketch_alloc<A>::update(const void* value,
|
189
|
+
void cpc_sketch_alloc<A>::update(const void* value, size_t size) {
|
192
190
|
HashState hashes;
|
193
191
|
MurmurHash3_x64_128(value, size, seed, hashes);
|
194
192
|
row_col_update(row_col_from_two_hashes(hashes.h1, hashes.h2, lg_k));
|
@@ -208,7 +206,7 @@ void cpc_sketch_alloc<A>::row_col_update(uint32_t row_col) {
|
|
208
206
|
|
209
207
|
template<typename A>
|
210
208
|
void cpc_sketch_alloc<A>::update_sparse(uint32_t row_col) {
|
211
|
-
const
|
209
|
+
const uint32_t k = 1 << lg_k;
|
212
210
|
const uint64_t c32pre = static_cast<uint64_t>(num_coupons) << 5;
|
213
211
|
if (c32pre >= 3 * k) throw std::logic_error("c32pre >= 3 * k"); // C < 3K/32, in other words flavor == SPARSE
|
214
212
|
bool is_novel = surprising_value_table.maybe_insert(row_col);
|
@@ -224,7 +222,7 @@ void cpc_sketch_alloc<A>::update_sparse(uint32_t row_col) {
|
|
224
222
|
template<typename A>
|
225
223
|
void cpc_sketch_alloc<A>::update_windowed(uint32_t row_col) {
|
226
224
|
if (window_offset > 56) throw std::logic_error("wrong window offset");
|
227
|
-
const
|
225
|
+
const uint32_t k = 1 << lg_k;
|
228
226
|
const uint64_t c32pre = static_cast<uint64_t>(num_coupons) << 5;
|
229
227
|
if (c32pre < 3 * k) throw std::logic_error("c32pre < 3 * k"); // C < 3K/32, in other words flavor >= HYBRID
|
230
228
|
const uint64_t c8pre = static_cast<uint64_t>(num_coupons) << 3;
|
@@ -266,7 +264,7 @@ void cpc_sketch_alloc<A>::update_windowed(uint32_t row_col) {
|
|
266
264
|
// Call this whenever a new coupon has been collected.
|
267
265
|
template<typename A>
|
268
266
|
void cpc_sketch_alloc<A>::update_hip(uint32_t row_col) {
|
269
|
-
const
|
267
|
+
const uint32_t k = 1 << lg_k;
|
270
268
|
const uint8_t col = row_col & 63;
|
271
269
|
const double one_over_p = static_cast<double>(k) / kxp;
|
272
270
|
hip_est_accum += one_over_p;
|
@@ -276,7 +274,7 @@ void cpc_sketch_alloc<A>::update_hip(uint32_t row_col) {
|
|
276
274
|
// In terms of flavor, this promotes SPARSE to HYBRID
|
277
275
|
template<typename A>
|
278
276
|
void cpc_sketch_alloc<A>::promote_sparse_to_windowed() {
|
279
|
-
const
|
277
|
+
const uint32_t k = 1 << lg_k;
|
280
278
|
const uint64_t c32 = static_cast<uint64_t>(num_coupons) << 5;
|
281
279
|
if (!(c32 == 3 * k || (lg_k == 4 && c32 > 3 * k))) throw std::logic_error("wrong c32");
|
282
280
|
|
@@ -285,16 +283,16 @@ void cpc_sketch_alloc<A>::promote_sparse_to_windowed() {
|
|
285
283
|
u32_table<A> new_table(2, 6 + lg_k, sliding_window.get_allocator());
|
286
284
|
|
287
285
|
const uint32_t* old_slots = surprising_value_table.get_slots();
|
288
|
-
const
|
286
|
+
const uint32_t old_num_slots = 1 << surprising_value_table.get_lg_size();
|
289
287
|
|
290
288
|
if (window_offset != 0) throw std::logic_error("window_offset != 0");
|
291
289
|
|
292
|
-
for (
|
290
|
+
for (uint32_t i = 0; i < old_num_slots; i++) {
|
293
291
|
const uint32_t row_col = old_slots[i];
|
294
292
|
if (row_col != UINT32_MAX) {
|
295
293
|
const uint8_t col = row_col & 63;
|
296
294
|
if (col < 8) {
|
297
|
-
const
|
295
|
+
const uint32_t row = row_col >> 6;
|
298
296
|
sliding_window[row] |= 1 << col;
|
299
297
|
} else {
|
300
298
|
// cannot use u32_table::must_insert(), because it doesn't provide for growth
|
@@ -314,7 +312,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
314
312
|
if (new_offset != determine_correct_offset(lg_k, num_coupons)) throw std::logic_error("new_offset is wrong");
|
315
313
|
|
316
314
|
if (sliding_window.size() == 0) throw std::logic_error("no sliding window");
|
317
|
-
const
|
315
|
+
const uint32_t k = 1 << lg_k;
|
318
316
|
|
319
317
|
// Construct the full-sized bit matrix that corresponds to the sketch
|
320
318
|
vector_u64<A> bit_matrix = build_bit_matrix();
|
@@ -328,7 +326,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
328
326
|
const uint64_t mask_for_flipping_early_zone = (static_cast<uint64_t>(1) << new_offset) - 1;
|
329
327
|
uint64_t all_surprises_ored = 0;
|
330
328
|
|
331
|
-
for (
|
329
|
+
for (uint32_t i = 0; i < k; i++) {
|
332
330
|
uint64_t pattern = bit_matrix[i];
|
333
331
|
sliding_window[i] = (pattern >> new_offset) & 0xff;
|
334
332
|
pattern &= mask_for_clearing_window;
|
@@ -357,7 +355,7 @@ void cpc_sketch_alloc<A>::move_window() {
|
|
357
355
|
// so that it will reflect changes that were previously outside the mantissa.
|
358
356
|
template<typename A>
|
359
357
|
void cpc_sketch_alloc<A>::refresh_kxp(const uint64_t* bit_matrix) {
|
360
|
-
const
|
358
|
+
const uint32_t k = 1 << lg_k;
|
361
359
|
|
362
360
|
// for improved numerical accuracy, we separately sum the bytes of the U64's
|
363
361
|
double byte_sums[8]; // allocating on the stack
|
@@ -383,7 +381,9 @@ void cpc_sketch_alloc<A>::refresh_kxp(const uint64_t* bit_matrix) {
|
|
383
381
|
|
384
382
|
template<typename A>
|
385
383
|
string<A> cpc_sketch_alloc<A>::to_string() const {
|
386
|
-
|
384
|
+
// Using a temporary stream for implementation here does not comply with AllocatorAwareContainer requirements.
|
385
|
+
// The stream does not support passing an allocator instance, and alternatives are complicated.
|
386
|
+
std::ostringstream os;
|
387
387
|
os << "### CPC sketch summary:" << std::endl;
|
388
388
|
os << " lg_k : " << std::to_string(lg_k) << std::endl;
|
389
389
|
os << " seed hash : " << std::hex << compute_seed_hash(seed) << std::dec << std::endl;
|
@@ -394,14 +394,14 @@ string<A> cpc_sketch_alloc<A>::to_string() const {
|
|
394
394
|
os << " HIP estimate : " << hip_est_accum << std::endl;
|
395
395
|
os << " kxp : " << kxp << std::endl;
|
396
396
|
}
|
397
|
-
os << "
|
397
|
+
os << " interesting col: " << std::to_string(first_interesting_column) << std::endl;
|
398
398
|
os << " table entries : " << surprising_value_table.get_num_items() << std::endl;
|
399
399
|
os << " window : " << (sliding_window.size() == 0 ? "not " : "") << "allocated" << std::endl;
|
400
400
|
if (sliding_window.size() > 0) {
|
401
401
|
os << " window offset : " << std::to_string(window_offset) << std::endl;
|
402
402
|
}
|
403
403
|
os << "### End sketch summary" << std::endl;
|
404
|
-
return os.str();
|
404
|
+
return string<A>(os.str().c_str(), sliding_window.get_allocator());
|
405
405
|
}
|
406
406
|
|
407
407
|
template<typename A>
|
@@ -415,44 +415,44 @@ void cpc_sketch_alloc<A>::serialize(std::ostream& os) const {
|
|
415
415
|
const bool has_table = compressed.table_data.size() > 0;
|
416
416
|
const bool has_window = compressed.window_data.size() > 0;
|
417
417
|
const uint8_t preamble_ints = get_preamble_ints(num_coupons, has_hip, has_table, has_window);
|
418
|
-
|
418
|
+
write(os, preamble_ints);
|
419
419
|
const uint8_t serial_version = SERIAL_VERSION;
|
420
|
-
|
420
|
+
write(os, serial_version);
|
421
421
|
const uint8_t family = FAMILY;
|
422
|
-
|
423
|
-
|
424
|
-
|
422
|
+
write(os, family);
|
423
|
+
write(os, lg_k);
|
424
|
+
write(os, first_interesting_column);
|
425
425
|
const uint8_t flags_byte(
|
426
426
|
(1 << flags::IS_COMPRESSED)
|
427
427
|
| (has_hip ? 1 << flags::HAS_HIP : 0)
|
428
428
|
| (has_table ? 1 << flags::HAS_TABLE : 0)
|
429
429
|
| (has_window ? 1 << flags::HAS_WINDOW : 0)
|
430
430
|
);
|
431
|
-
|
431
|
+
write(os, flags_byte);
|
432
432
|
const uint16_t seed_hash(compute_seed_hash(seed));
|
433
|
-
|
433
|
+
write(os, seed_hash);
|
434
434
|
if (!is_empty()) {
|
435
|
-
|
435
|
+
write(os, num_coupons);
|
436
436
|
if (has_table && has_window) {
|
437
437
|
// if there is no window it is the same as number of coupons
|
438
|
-
|
438
|
+
write(os, compressed.table_num_entries);
|
439
439
|
// HIP values can be in two different places in the sequence of fields
|
440
440
|
// this is the first HIP decision point
|
441
441
|
if (has_hip) write_hip(os);
|
442
442
|
}
|
443
443
|
if (has_table) {
|
444
|
-
|
444
|
+
write(os, compressed.table_data_words);
|
445
445
|
}
|
446
446
|
if (has_window) {
|
447
|
-
|
447
|
+
write(os, compressed.window_data_words);
|
448
448
|
}
|
449
449
|
// this is the second HIP decision point
|
450
450
|
if (has_hip && !(has_table && has_window)) write_hip(os);
|
451
451
|
if (has_window) {
|
452
|
-
|
452
|
+
write(os, compressed.window_data.data(), compressed.window_data_words * sizeof(uint32_t));
|
453
453
|
}
|
454
454
|
if (has_table) {
|
455
|
-
|
455
|
+
write(os, compressed.table_data.data(), compressed.table_data_words * sizeof(uint32_t));
|
456
456
|
}
|
457
457
|
}
|
458
458
|
}
|
@@ -471,36 +471,36 @@ vector_u8<A> cpc_sketch_alloc<A>::serialize(unsigned header_size_bytes) const {
|
|
471
471
|
const size_t size = header_size_bytes + (preamble_ints + compressed.table_data_words + compressed.window_data_words) * sizeof(uint32_t);
|
472
472
|
vector_u8<A> bytes(size, 0, sliding_window.get_allocator());
|
473
473
|
uint8_t* ptr = bytes.data() + header_size_bytes;
|
474
|
-
ptr += copy_to_mem(
|
474
|
+
ptr += copy_to_mem(preamble_ints, ptr);
|
475
475
|
const uint8_t serial_version = SERIAL_VERSION;
|
476
|
-
ptr += copy_to_mem(
|
476
|
+
ptr += copy_to_mem(serial_version, ptr);
|
477
477
|
const uint8_t family = FAMILY;
|
478
|
-
ptr += copy_to_mem(
|
479
|
-
ptr += copy_to_mem(
|
480
|
-
ptr += copy_to_mem(
|
478
|
+
ptr += copy_to_mem(family, ptr);
|
479
|
+
ptr += copy_to_mem(lg_k, ptr);
|
480
|
+
ptr += copy_to_mem(first_interesting_column, ptr);
|
481
481
|
const uint8_t flags_byte(
|
482
482
|
(1 << flags::IS_COMPRESSED)
|
483
483
|
| (has_hip ? 1 << flags::HAS_HIP : 0)
|
484
484
|
| (has_table ? 1 << flags::HAS_TABLE : 0)
|
485
485
|
| (has_window ? 1 << flags::HAS_WINDOW : 0)
|
486
486
|
);
|
487
|
-
ptr += copy_to_mem(
|
487
|
+
ptr += copy_to_mem(flags_byte, ptr);
|
488
488
|
const uint16_t seed_hash = compute_seed_hash(seed);
|
489
|
-
ptr += copy_to_mem(
|
489
|
+
ptr += copy_to_mem(seed_hash, ptr);
|
490
490
|
if (!is_empty()) {
|
491
|
-
ptr += copy_to_mem(
|
491
|
+
ptr += copy_to_mem(num_coupons, ptr);
|
492
492
|
if (has_table && has_window) {
|
493
493
|
// if there is no window it is the same as number of coupons
|
494
|
-
ptr += copy_to_mem(
|
494
|
+
ptr += copy_to_mem(compressed.table_num_entries, ptr);
|
495
495
|
// HIP values can be in two different places in the sequence of fields
|
496
496
|
// this is the first HIP decision point
|
497
497
|
if (has_hip) ptr += copy_hip_to_mem(ptr);
|
498
498
|
}
|
499
499
|
if (has_table) {
|
500
|
-
ptr += copy_to_mem(
|
500
|
+
ptr += copy_to_mem(compressed.table_data_words, ptr);
|
501
501
|
}
|
502
502
|
if (has_window) {
|
503
|
-
ptr += copy_to_mem(
|
503
|
+
ptr += copy_to_mem(compressed.window_data_words, ptr);
|
504
504
|
}
|
505
505
|
// this is the second HIP decision point
|
506
506
|
if (has_hip && !(has_table && has_window)) ptr += copy_hip_to_mem(ptr);
|
@@ -517,20 +517,13 @@ vector_u8<A> cpc_sketch_alloc<A>::serialize(unsigned header_size_bytes) const {
|
|
517
517
|
|
518
518
|
template<typename A>
|
519
519
|
cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(std::istream& is, uint64_t seed, const A& allocator) {
|
520
|
-
|
521
|
-
|
522
|
-
uint8_t
|
523
|
-
|
524
|
-
uint8_t
|
525
|
-
|
526
|
-
|
527
|
-
is.read((char*)&lg_k, sizeof(lg_k));
|
528
|
-
uint8_t first_interesting_column;
|
529
|
-
is.read((char*)&first_interesting_column, sizeof(first_interesting_column));
|
530
|
-
uint8_t flags_byte;
|
531
|
-
is.read((char*)&flags_byte, sizeof(flags_byte));
|
532
|
-
uint16_t seed_hash;
|
533
|
-
is.read((char*)&seed_hash, sizeof(seed_hash));
|
520
|
+
const auto preamble_ints = read<uint8_t>(is);
|
521
|
+
const auto serial_version = read<uint8_t>(is);
|
522
|
+
const auto family_id = read<uint8_t>(is);
|
523
|
+
const auto lg_k = read<uint8_t>(is);
|
524
|
+
const auto first_interesting_column = read<uint8_t>(is);
|
525
|
+
const auto flags_byte = read<uint8_t>(is);
|
526
|
+
const auto seed_hash = read<uint16_t>(is);
|
534
527
|
const bool has_hip = flags_byte & (1 << flags::HAS_HIP);
|
535
528
|
const bool has_table = flags_byte & (1 << flags::HAS_TABLE);
|
536
529
|
const bool has_window = flags_byte & (1 << flags::HAS_WINDOW);
|
@@ -542,31 +535,31 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(std::istream& is, uint64_t
|
|
542
535
|
double kxp = 0;
|
543
536
|
double hip_est_accum = 0;
|
544
537
|
if (has_table || has_window) {
|
545
|
-
|
538
|
+
num_coupons = read<uint32_t>(is);
|
546
539
|
if (has_table && has_window) {
|
547
|
-
|
540
|
+
compressed.table_num_entries = read<uint32_t>(is);
|
548
541
|
if (has_hip) {
|
549
|
-
|
550
|
-
|
542
|
+
kxp = read<double>(is);
|
543
|
+
hip_est_accum = read<double>(is);
|
551
544
|
}
|
552
545
|
}
|
553
546
|
if (has_table) {
|
554
|
-
|
547
|
+
compressed.table_data_words = read<uint32_t>(is);
|
555
548
|
}
|
556
549
|
if (has_window) {
|
557
|
-
|
550
|
+
compressed.window_data_words = read<uint32_t>(is);
|
558
551
|
}
|
559
552
|
if (has_hip && !(has_table && has_window)) {
|
560
|
-
|
561
|
-
|
553
|
+
kxp = read<double>(is);
|
554
|
+
hip_est_accum = read<double>(is);
|
562
555
|
}
|
563
556
|
if (has_window) {
|
564
557
|
compressed.window_data.resize(compressed.window_data_words);
|
565
|
-
|
558
|
+
read(is, compressed.window_data.data(), compressed.window_data_words * sizeof(uint32_t));
|
566
559
|
}
|
567
560
|
if (has_table) {
|
568
561
|
compressed.table_data.resize(compressed.table_data_words);
|
569
|
-
|
562
|
+
read(is, compressed.table_data.data(), compressed.table_data_words * sizeof(uint32_t));
|
570
563
|
}
|
571
564
|
if (!has_window) compressed.table_num_entries = num_coupons;
|
572
565
|
}
|
@@ -602,19 +595,19 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
602
595
|
const char* ptr = static_cast<const char*>(bytes);
|
603
596
|
const char* base = static_cast<const char*>(bytes);
|
604
597
|
uint8_t preamble_ints;
|
605
|
-
ptr += copy_from_mem(ptr,
|
598
|
+
ptr += copy_from_mem(ptr, preamble_ints);
|
606
599
|
uint8_t serial_version;
|
607
|
-
ptr += copy_from_mem(ptr,
|
600
|
+
ptr += copy_from_mem(ptr, serial_version);
|
608
601
|
uint8_t family_id;
|
609
|
-
ptr += copy_from_mem(ptr,
|
602
|
+
ptr += copy_from_mem(ptr, family_id);
|
610
603
|
uint8_t lg_k;
|
611
|
-
ptr += copy_from_mem(ptr,
|
604
|
+
ptr += copy_from_mem(ptr, lg_k);
|
612
605
|
uint8_t first_interesting_column;
|
613
|
-
ptr += copy_from_mem(ptr,
|
606
|
+
ptr += copy_from_mem(ptr, first_interesting_column);
|
614
607
|
uint8_t flags_byte;
|
615
|
-
ptr += copy_from_mem(ptr,
|
608
|
+
ptr += copy_from_mem(ptr, flags_byte);
|
616
609
|
uint16_t seed_hash;
|
617
|
-
ptr += copy_from_mem(ptr,
|
610
|
+
ptr += copy_from_mem(ptr, seed_hash);
|
618
611
|
const bool has_hip = flags_byte & (1 << flags::HAS_HIP);
|
619
612
|
const bool has_table = flags_byte & (1 << flags::HAS_TABLE);
|
620
613
|
const bool has_window = flags_byte & (1 << flags::HAS_WINDOW);
|
@@ -628,28 +621,28 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
628
621
|
double hip_est_accum = 0;
|
629
622
|
if (has_table || has_window) {
|
630
623
|
check_memory_size(ptr - base + sizeof(num_coupons), size);
|
631
|
-
ptr += copy_from_mem(ptr,
|
624
|
+
ptr += copy_from_mem(ptr, num_coupons);
|
632
625
|
if (has_table && has_window) {
|
633
626
|
check_memory_size(ptr - base + sizeof(compressed.table_num_entries), size);
|
634
|
-
ptr += copy_from_mem(ptr,
|
627
|
+
ptr += copy_from_mem(ptr, compressed.table_num_entries);
|
635
628
|
if (has_hip) {
|
636
629
|
check_memory_size(ptr - base + sizeof(kxp) + sizeof(hip_est_accum), size);
|
637
|
-
ptr += copy_from_mem(ptr,
|
638
|
-
ptr += copy_from_mem(ptr,
|
630
|
+
ptr += copy_from_mem(ptr, kxp);
|
631
|
+
ptr += copy_from_mem(ptr, hip_est_accum);
|
639
632
|
}
|
640
633
|
}
|
641
634
|
if (has_table) {
|
642
635
|
check_memory_size(ptr - base + sizeof(compressed.table_data_words), size);
|
643
|
-
ptr += copy_from_mem(ptr,
|
636
|
+
ptr += copy_from_mem(ptr, compressed.table_data_words);
|
644
637
|
}
|
645
638
|
if (has_window) {
|
646
639
|
check_memory_size(ptr - base + sizeof(compressed.window_data_words), size);
|
647
|
-
ptr += copy_from_mem(ptr,
|
640
|
+
ptr += copy_from_mem(ptr, compressed.window_data_words);
|
648
641
|
}
|
649
642
|
if (has_hip && !(has_table && has_window)) {
|
650
643
|
check_memory_size(ptr - base + sizeof(kxp) + sizeof(hip_est_accum), size);
|
651
|
-
ptr += copy_from_mem(ptr,
|
652
|
-
ptr += copy_from_mem(ptr,
|
644
|
+
ptr += copy_from_mem(ptr, kxp);
|
645
|
+
ptr += copy_from_mem(ptr, hip_est_accum);
|
653
646
|
}
|
654
647
|
if (has_window) {
|
655
648
|
compressed.window_data.resize(compressed.window_data_words);
|
@@ -688,6 +681,49 @@ cpc_sketch_alloc<A> cpc_sketch_alloc<A>::deserialize(const void* bytes, size_t s
|
|
688
681
|
std::move(uncompressed.window), has_hip, kxp, hip_est_accum, seed);
|
689
682
|
}
|
690
683
|
|
684
|
+
/*
|
685
|
+
* These empirical values for the 99.9th percentile of size in bytes were measured using 100,000
|
686
|
+
* trials. The value for each trial is the maximum of 5*16=80 measurements that were equally
|
687
|
+
* spaced over values of the quantity C/K between 3.0 and 8.0. This table does not include the
|
688
|
+
* worst-case space for the preamble, which is added by the function.
|
689
|
+
*/
|
690
|
+
static const uint8_t CPC_EMPIRICAL_SIZE_MAX_LGK = 19;
|
691
|
+
static const size_t CPC_EMPIRICAL_MAX_SIZE_BYTES[] = {
|
692
|
+
24, // lg_k = 4
|
693
|
+
36, // lg_k = 5
|
694
|
+
56, // lg_k = 6
|
695
|
+
100, // lg_k = 7
|
696
|
+
180, // lg_k = 8
|
697
|
+
344, // lg_k = 9
|
698
|
+
660, // lg_k = 10
|
699
|
+
1292, // lg_k = 11
|
700
|
+
2540, // lg_k = 12
|
701
|
+
5020, // lg_k = 13
|
702
|
+
9968, // lg_k = 14
|
703
|
+
19836, // lg_k = 15
|
704
|
+
39532, // lg_k = 16
|
705
|
+
78880, // lg_k = 17
|
706
|
+
157516, // lg_k = 18
|
707
|
+
314656 // lg_k = 19
|
708
|
+
};
|
709
|
+
static const double CPC_EMPIRICAL_MAX_SIZE_FACTOR = 0.6; // 0.6 = 4.8 / 8.0
|
710
|
+
static const size_t CPC_MAX_PREAMBLE_SIZE_BYTES = 40;
|
711
|
+
|
712
|
+
template<typename A>
|
713
|
+
size_t cpc_sketch_alloc<A>::get_max_serialized_size_bytes(uint8_t lg_k) {
|
714
|
+
check_lg_k(lg_k);
|
715
|
+
if (lg_k <= CPC_EMPIRICAL_SIZE_MAX_LGK) return CPC_EMPIRICAL_MAX_SIZE_BYTES[lg_k - CPC_MIN_LG_K] + CPC_MAX_PREAMBLE_SIZE_BYTES;
|
716
|
+
const uint32_t k = 1 << lg_k;
|
717
|
+
return (int) (CPC_EMPIRICAL_MAX_SIZE_FACTOR * k) + CPC_MAX_PREAMBLE_SIZE_BYTES;
|
718
|
+
}
|
719
|
+
|
720
|
+
template<typename A>
|
721
|
+
void cpc_sketch_alloc<A>::check_lg_k(uint8_t lg_k) {
|
722
|
+
if (lg_k < CPC_MIN_LG_K || lg_k > CPC_MAX_LG_K) {
|
723
|
+
throw std::invalid_argument("lg_k must be >= " + std::to_string(CPC_MIN_LG_K) + " and <= " + std::to_string(CPC_MAX_LG_K) + ": " + std::to_string(lg_k));
|
724
|
+
}
|
725
|
+
}
|
726
|
+
|
691
727
|
template<typename A>
|
692
728
|
uint32_t cpc_sketch_alloc<A>::get_num_coupons() const {
|
693
729
|
return num_coupons;
|
@@ -696,7 +732,7 @@ uint32_t cpc_sketch_alloc<A>::get_num_coupons() const {
|
|
696
732
|
template<typename A>
|
697
733
|
bool cpc_sketch_alloc<A>::validate() const {
|
698
734
|
vector_u64<A> bit_matrix = build_bit_matrix();
|
699
|
-
const uint64_t num_bits_set = count_bits_set_in_matrix(bit_matrix.data(),
|
735
|
+
const uint64_t num_bits_set = count_bits_set_in_matrix(bit_matrix.data(), 1ULL << lg_k);
|
700
736
|
return num_bits_set == num_coupons;
|
701
737
|
}
|
702
738
|
|
@@ -744,7 +780,7 @@ typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor() con
|
|
744
780
|
|
745
781
|
template<typename A>
|
746
782
|
typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor(uint8_t lg_k, uint64_t c) {
|
747
|
-
const
|
783
|
+
const uint32_t k = 1 << lg_k;
|
748
784
|
const uint64_t c2 = c << 1;
|
749
785
|
const uint64_t c8 = c << 3;
|
750
786
|
const uint64_t c32 = c << 5;
|
@@ -757,15 +793,15 @@ typename cpc_sketch_alloc<A>::flavor cpc_sketch_alloc<A>::determine_flavor(uint8
|
|
757
793
|
|
758
794
|
template<typename A>
|
759
795
|
uint8_t cpc_sketch_alloc<A>::determine_correct_offset(uint8_t lg_k, uint64_t c) {
|
760
|
-
const
|
796
|
+
const uint32_t k = 1 << lg_k;
|
761
797
|
const int64_t tmp = static_cast<int64_t>(c << 3) - static_cast<int64_t>(19 * k); // 8C - 19K
|
762
798
|
if (tmp < 0) return 0;
|
763
|
-
return tmp >> (lg_k + 3); // tmp / 8K
|
799
|
+
return static_cast<uint8_t>(tmp >> (lg_k + 3)); // tmp / 8K
|
764
800
|
}
|
765
801
|
|
766
802
|
template<typename A>
|
767
803
|
vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
768
|
-
const
|
804
|
+
const uint32_t k = 1 << lg_k;
|
769
805
|
if (window_offset > 56) throw std::logic_error("offset > 56");
|
770
806
|
|
771
807
|
// Fill the matrix with default rows in which the "early zone" is filled with ones.
|
@@ -782,12 +818,12 @@ vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
|
782
818
|
}
|
783
819
|
|
784
820
|
const uint32_t* slots = surprising_value_table.get_slots();
|
785
|
-
const
|
821
|
+
const uint32_t num_slots = 1 << surprising_value_table.get_lg_size();
|
786
822
|
for (size_t i = 0; i < num_slots; i++) {
|
787
823
|
const uint32_t row_col = slots[i];
|
788
824
|
if (row_col != UINT32_MAX) {
|
789
825
|
const uint8_t col = row_col & 63;
|
790
|
-
const
|
826
|
+
const uint32_t row = row_col >> 6;
|
791
827
|
// Flip the specified matrix bit from its default value.
|
792
828
|
// In the "early" zone the bit changes from 1 to 0.
|
793
829
|
// In the "late" zone the bit changes from 0 to 1.
|
@@ -799,8 +835,8 @@ vector_u64<A> cpc_sketch_alloc<A>::build_bit_matrix() const {
|
|
799
835
|
|
800
836
|
template<typename A>
|
801
837
|
void cpc_sketch_alloc<A>::write_hip(std::ostream& os) const {
|
802
|
-
|
803
|
-
|
838
|
+
write(os, kxp);
|
839
|
+
write(os, hip_est_accum);
|
804
840
|
}
|
805
841
|
|
806
842
|
template<typename A>
|
@@ -45,7 +45,7 @@ public:
|
|
45
45
|
* @param lg_k base 2 logarithm of the number of bins in the sketch
|
46
46
|
* @param seed for hash function
|
47
47
|
*/
|
48
|
-
explicit cpc_union_alloc(uint8_t lg_k =
|
48
|
+
explicit cpc_union_alloc(uint8_t lg_k = cpc_constants::DEFAULT_LG_K, uint64_t seed = DEFAULT_SEED, const A& allocator = A());
|
49
49
|
|
50
50
|
cpc_union_alloc(const cpc_union_alloc<A>& other);
|
51
51
|
cpc_union_alloc(cpc_union_alloc<A>&& other) noexcept;
|
@@ -34,7 +34,7 @@ bit_matrix(allocator)
|
|
34
34
|
if (lg_k < CPC_MIN_LG_K || lg_k > CPC_MAX_LG_K) {
|
35
35
|
throw std::invalid_argument("lg_k must be >= " + std::to_string(CPC_MIN_LG_K) + " and <= " + std::to_string(CPC_MAX_LG_K) + ": " + std::to_string(lg_k));
|
36
36
|
}
|
37
|
-
accumulator = new (AllocCpc().allocate(1)) cpc_sketch_alloc<A>(lg_k, seed, allocator);
|
37
|
+
accumulator = new (AllocCpc(allocator).allocate(1)) cpc_sketch_alloc<A>(lg_k, seed, allocator);
|
38
38
|
}
|
39
39
|
|
40
40
|
template<typename A>
|
@@ -45,7 +45,7 @@ accumulator(other.accumulator),
|
|
45
45
|
bit_matrix(other.bit_matrix)
|
46
46
|
{
|
47
47
|
if (accumulator != nullptr) {
|
48
|
-
accumulator = new (AllocCpc().allocate(1)) cpc_sketch_alloc<A>(*other.accumulator);
|
48
|
+
accumulator = new (AllocCpc(accumulator->get_allocator()).allocate(1)) cpc_sketch_alloc<A>(*other.accumulator);
|
49
49
|
}
|
50
50
|
}
|
51
51
|
|
@@ -62,8 +62,9 @@ bit_matrix(std::move(other.bit_matrix))
|
|
62
62
|
template<typename A>
|
63
63
|
cpc_union_alloc<A>::~cpc_union_alloc() {
|
64
64
|
if (accumulator != nullptr) {
|
65
|
+
AllocCpc allocator(accumulator->get_allocator());
|
65
66
|
accumulator->~cpc_sketch_alloc<A>();
|
66
|
-
|
67
|
+
allocator.deallocate(accumulator, 1);
|
67
68
|
}
|
68
69
|
}
|
69
70
|
|
@@ -181,7 +182,7 @@ template<typename A>
|
|
181
182
|
cpc_sketch_alloc<A> cpc_union_alloc<A>::get_result_from_accumulator() const {
|
182
183
|
if (lg_k != accumulator->get_lg_k()) throw std::logic_error("lg_k != accumulator->lg_k");
|
183
184
|
if (accumulator->get_num_coupons() == 0) {
|
184
|
-
return cpc_sketch_alloc<A>(lg_k, seed);
|
185
|
+
return cpc_sketch_alloc<A>(lg_k, seed, accumulator->get_allocator());
|
185
186
|
}
|
186
187
|
if (accumulator->determine_flavor() != cpc_sketch_alloc<A>::flavor::SPARSE) throw std::logic_error("wrong flavor");
|
187
188
|
cpc_sketch_alloc<A> copy(*accumulator);
|
@@ -191,8 +192,8 @@ cpc_sketch_alloc<A> cpc_union_alloc<A>::get_result_from_accumulator() const {
|
|
191
192
|
|
192
193
|
template<typename A>
|
193
194
|
cpc_sketch_alloc<A> cpc_union_alloc<A>::get_result_from_bit_matrix() const {
|
194
|
-
const
|
195
|
-
const
|
195
|
+
const uint32_t k = 1 << lg_k;
|
196
|
+
const uint32_t num_coupons = count_bits_set_in_matrix(bit_matrix.data(), k);
|
196
197
|
|
197
198
|
const auto flavor = cpc_sketch_alloc<A>::determine_flavor(lg_k, num_coupons);
|
198
199
|
if (flavor != cpc_sketch_alloc<A>::flavor::HYBRID && flavor != cpc_sketch_alloc<A>::flavor::PINNED
|
@@ -215,7 +216,7 @@ cpc_sketch_alloc<A> cpc_union_alloc<A>::get_result_from_bit_matrix() const {
|
|
215
216
|
|
216
217
|
// The snowplow effect was caused by processing the rows in order,
|
217
218
|
// but we have fixed it by using a sufficiently large hash table.
|
218
|
-
for (
|
219
|
+
for (uint32_t i = 0; i < k; i++) {
|
219
220
|
uint64_t pattern = bit_matrix[i];
|
220
221
|
sliding_window[i] = (pattern >> offset) & 0xff;
|
221
222
|
pattern &= mask_for_clearing_window;
|
@@ -242,25 +243,26 @@ cpc_sketch_alloc<A> cpc_union_alloc<A>::get_result_from_bit_matrix() const {
|
|
242
243
|
template<typename A>
|
243
244
|
void cpc_union_alloc<A>::switch_to_bit_matrix() {
|
244
245
|
bit_matrix = accumulator->build_bit_matrix();
|
246
|
+
AllocCpc allocator(accumulator->get_allocator());
|
245
247
|
accumulator->~cpc_sketch_alloc<A>();
|
246
|
-
|
248
|
+
allocator.deallocate(accumulator, 1);
|
247
249
|
accumulator = nullptr;
|
248
250
|
}
|
249
251
|
|
250
252
|
template<typename A>
|
251
253
|
void cpc_union_alloc<A>::walk_table_updating_sketch(const u32_table<A>& table) {
|
252
254
|
const uint32_t* slots = table.get_slots();
|
253
|
-
const
|
255
|
+
const uint32_t num_slots = 1 << table.get_lg_size();
|
254
256
|
const uint64_t dst_mask = (((1 << accumulator->get_lg_k()) - 1) << 6) | 63; // downsamples when dst lgK < src LgK
|
255
257
|
|
256
258
|
// Using a golden ratio stride fixes the snowplow effect.
|
257
259
|
const double golden = 0.6180339887498949025;
|
258
|
-
|
260
|
+
uint32_t stride = static_cast<uint32_t>(golden * static_cast<double>(num_slots));
|
259
261
|
if (stride < 2) throw std::logic_error("stride < 2");
|
260
262
|
if (stride == ((stride >> 1) << 1)) stride += 1; // force the stride to be odd
|
261
263
|
if (stride < 3 || stride >= num_slots) throw std::out_of_range("stride out of range");
|
262
264
|
|
263
|
-
for (
|
265
|
+
for (uint32_t i = 0, j = 0; i < num_slots; i++, j += stride) {
|
264
266
|
j &= num_slots - 1;
|
265
267
|
const uint32_t row_col = slots[j];
|
266
268
|
if (row_col != UINT32_MAX) {
|
@@ -272,13 +274,13 @@ void cpc_union_alloc<A>::walk_table_updating_sketch(const u32_table<A>& table) {
|
|
272
274
|
template<typename A>
|
273
275
|
void cpc_union_alloc<A>::or_table_into_matrix(const u32_table<A>& table) {
|
274
276
|
const uint32_t* slots = table.get_slots();
|
275
|
-
const
|
277
|
+
const uint32_t num_slots = 1 << table.get_lg_size();
|
276
278
|
const uint64_t dest_mask = (1 << lg_k) - 1; // downsamples when dst lgK < sr LgK
|
277
|
-
for (
|
279
|
+
for (uint32_t i = 0; i < num_slots; i++) {
|
278
280
|
const uint32_t row_col = slots[i];
|
279
281
|
if (row_col != UINT32_MAX) {
|
280
282
|
const uint8_t col = row_col & 63;
|
281
|
-
const
|
283
|
+
const uint32_t row = row_col >> 6;
|
282
284
|
bit_matrix[row & dest_mask] |= static_cast<uint64_t>(1) << col; // set the bit
|
283
285
|
}
|
284
286
|
}
|
@@ -288,8 +290,8 @@ template<typename A>
|
|
288
290
|
void cpc_union_alloc<A>::or_window_into_matrix(const vector_u8<A>& sliding_window, uint8_t offset, uint8_t src_lg_k) {
|
289
291
|
if (lg_k > src_lg_k) throw std::logic_error("dst LgK > src LgK");
|
290
292
|
const uint64_t dst_mask = (1 << lg_k) - 1; // downsamples when dst lgK < src LgK
|
291
|
-
const
|
292
|
-
for (
|
293
|
+
const uint32_t src_k = 1 << src_lg_k;
|
294
|
+
for (uint32_t src_row = 0; src_row < src_k; src_row++) {
|
293
295
|
bit_matrix[src_row & dst_mask] |= static_cast<uint64_t>(sliding_window[src_row]) << offset;
|
294
296
|
}
|
295
297
|
}
|
@@ -298,8 +300,8 @@ template<typename A>
|
|
298
300
|
void cpc_union_alloc<A>::or_matrix_into_matrix(const vector_u64<A>& src_matrix, uint8_t src_lg_k) {
|
299
301
|
if (lg_k > src_lg_k) throw std::logic_error("dst LgK > src LgK");
|
300
302
|
const uint64_t dst_mask = (1 << lg_k) - 1; // downsamples when dst lgK < src LgK
|
301
|
-
const
|
302
|
-
for (
|
303
|
+
const uint32_t src_k = 1 << src_lg_k;
|
304
|
+
for (uint32_t src_row = 0; src_row < src_k; src_row++) {
|
303
305
|
bit_matrix[src_row & dst_mask] |= src_matrix[src_row];
|
304
306
|
}
|
305
307
|
}
|
@@ -313,7 +315,7 @@ void cpc_union_alloc<A>::reduce_k(uint8_t new_lg_k) {
|
|
313
315
|
if (accumulator != nullptr) throw std::logic_error("accumulator is not null");
|
314
316
|
vector_u64<A> old_matrix = std::move(bit_matrix);
|
315
317
|
const uint8_t old_lg_k = lg_k;
|
316
|
-
const
|
318
|
+
const uint32_t new_k = 1 << new_lg_k;
|
317
319
|
bit_matrix = vector_u64<A>(new_k, 0, old_matrix.get_allocator());
|
318
320
|
lg_k = new_lg_k;
|
319
321
|
or_matrix_into_matrix(old_matrix, old_lg_k);
|
@@ -324,7 +326,7 @@ void cpc_union_alloc<A>::reduce_k(uint8_t new_lg_k) {
|
|
324
326
|
if (bit_matrix.size() > 0) throw std::logic_error("bit_matrix is not expected");
|
325
327
|
if (!accumulator->is_empty()) {
|
326
328
|
cpc_sketch_alloc<A> old_accumulator(*accumulator);
|
327
|
-
*accumulator = cpc_sketch_alloc<A>(new_lg_k, seed);
|
329
|
+
*accumulator = cpc_sketch_alloc<A>(new_lg_k, seed, old_accumulator.get_allocator());
|
328
330
|
walk_table_updating_sketch(old_accumulator.surprising_value_table);
|
329
331
|
}
|
330
332
|
lg_k = new_lg_k;
|