Merge "perfetto: fix PERFETTO_ANNOTATE_BENIGN_RACE_SIZED macro" into main
diff --git a/src/trace_processor/db/query_executor_unittest.cc b/src/trace_processor/db/query_executor_unittest.cc
index e5022c0..849c1cb 100644
--- a/src/trace_processor/db/query_executor_unittest.cc
+++ b/src/trace_processor/db/query_executor_unittest.cc
@@ -223,7 +223,7 @@
ASSERT_THAT(rm.GetAllIndices(), ElementsAre(0u, 4u));
}
-TEST(QueryExecutor, ArrangementOverlaySubsetInputRange) {
+TEST(QueryExecutor, ArrangementStorageSubsetInputRange) {
std::unique_ptr<storage::Storage> fake =
storage::FakeStorage::SearchSubset(5u, RowMap::Range(2u, 4u));
@@ -237,7 +237,7 @@
ASSERT_THAT(rm.GetAllIndices(), ElementsAre(2u));
}
-TEST(QueryExecutor, ArrangementOverlaySubsetInputBitvector) {
+TEST(QueryExecutor, ArrangementStorageSubsetInputBitvector) {
std::unique_ptr<storage::Storage> fake =
storage::FakeStorage::SearchSubset(5u, BitVector({0, 0, 1, 1, 0}));
diff --git a/src/trace_processor/db/storage/arrangement_storage.cc b/src/trace_processor/db/storage/arrangement_storage.cc
index ee777f4..b0171db 100644
--- a/src/trace_processor/db/storage/arrangement_storage.cc
+++ b/src/trace_processor/db/storage/arrangement_storage.cc
@@ -60,6 +60,7 @@
}
} else {
BitVector storage_bitvector = std::move(storage_result).TakeIfBitVector();
+ PERFETTO_DCHECK(storage_bitvector.size() == *max_i + 1);
// After benchmarking, it turns out this complexity *is* actually worthwhile
// and has a noticable impact on the performance of this function in real
@@ -67,13 +68,13 @@
// Fast path: we compare as many groups of 64 elements as we can.
// This should be very easy for the compiler to auto-vectorize.
+ const uint32_t* arrangement_idx = arrangement.data() + in.start;
uint32_t fast_path_elements = builder.BitsInCompleteWordsUntilFull();
- uint32_t cur_idx = 0;
for (uint32_t i = 0; i < fast_path_elements; i += BitVector::kBitsInWord) {
uint64_t word = 0;
// This part should be optimised by SIMD and is expected to be fast.
- for (uint32_t k = 0; k < BitVector::kBitsInWord; ++k, ++cur_idx) {
- bool comp_result = storage_bitvector.IsSet((*arrangement_)[cur_idx]);
+ for (uint32_t k = 0; k < BitVector::kBitsInWord; ++k, ++arrangement_idx) {
+ bool comp_result = storage_bitvector.IsSet(*arrangement_idx);
word |= static_cast<uint64_t>(comp_result) << k;
}
builder.AppendWord(word);
@@ -81,8 +82,8 @@
// Slow path: we compare <64 elements and append to fill the Builder.
uint32_t back_elements = builder.BitsUntilFull();
- for (uint32_t i = 0; i < back_elements; ++i, ++cur_idx) {
- builder.Append(storage_bitvector.IsSet((*arrangement_)[cur_idx]));
+ for (uint32_t i = 0; i < back_elements; ++i, ++arrangement_idx) {
+ builder.Append(storage_bitvector.IsSet(*arrangement_idx));
}
}
return RangeOrBitVector(std::move(builder).Build());
