Improve DeltaTree update routine #5200
Description
Enhancement
On DeltaTree update routine (afterNodeUpdated), the call to getDelta() can be effectively translated by the compiler to SIMD memory scanning.
Since compiler has already decides that afterNodeUpdated is too large to be inlined, we can actually dispatch this function to multiple targets. Indeed, it enjoys different codegen for avx targets:
Experiments have shown that, it can provide up to 12% speed up at IDC environment:
2022-06-23T08:53:15+08:00
Running dbms/src/Storages/DeltaMerge/Benchmark/bench_deltatree
Run on (40 X 2394.97 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x20)
L1 Instruction 32 KiB (x20)
L2 Unified 256 KiB (x20)
L3 Unified 25600 KiB (x2)
Load Average: 8.91, 23.44, 19.41
------------------------------------------------------------------------------------------------
Benchmark Time CPU Iterations
------------------------------------------------------------------------------------------------
BM_deltatree_dispatch<true>/iterations:50/threads:40 1764554 ns 59946595 ns 2000
BM_deltatree_dispatch<false>/iterations:50/threads:40 1553962 ns 55583999 ns 2000