goldsteinn
goldsteinn
Hi, Nice stuff! A few questions: 1. Which version of GLIBC where you using? 2. What CPU where these benchmarks taken on?
Instead of generating the `info` hash bits with a mask + shift, we can just use an unsigned shift from the end as that will bring in zeros. With that...
Summary: Change layout: [63:48][47:0] -> [63:16][ 15:0] [extra][ ptr] -> [ ptr][extra] Previous PackedSyncPtr was organized s.t: [63:48][47:0] [extra][ ptr] So access pointer required a 48-bit bitmask: `data_.getData() & (-1ull...
The `popcnt` feature checks weren't actually checking the ISA feature or cpuid. This caused an illegal instruction exception on x86 hardware that doesn't support `popcnt` when compiled with MSVC. Since...
The lock only has two-states, so `atomic_exchange` works and is as or more performant in any situation.
- **[PatternMatching] Add generic API for matching constants using custom conditions** - **[InstCombine] Add example usage for new `Checked` matcher API** The new API is: `m_CheckedInt(Lambda)`/`m_CheckedFp(Lambda)` - Matches non-undef constants...
- **[ValueTracking] Add basic tests tracking `or disjoint` conditions as `add`; NFC** - **[ValueTracking] Tracking `or disjoint` conditions as `add` in Assumption/DomCondition Cache**
Generally would think events in the kernel are just noise in this context, but maybe not.
It would allow anyone w/o having to have the older/newer hardware to compare. It would also make this repo an amazing resource for quickly checking numbers.