Expand description
Workgroup simulation: invocation IDs, shared memory. Workgroup simulation — the parity engine’s model of invocation coordination.
GPU backends must reproduce the exact barrier synchronization, shared-memory layout, and invocation-ID arithmetic that this module defines. The conform gate compares GPU dispatch output against this deterministic CPU simulation; any divergence in control flow uniformity or workgroup memory semantics is a bug.
Structs§
- Buffer
Map - Small-N buffer lookup keyed by interned
Arc<str>names. - Invocation
- One paused or running invocation.
- Invocation
Ids - Identity of one compute invocation.
- Local
Slots - Shared slot layout for all locals in one program.
- Memory
- Shared execution memory for storage and current workgroup buffers.
Enums§
- Frame
- Interpreter continuation stack.
Constants§
- MAX_
WORKGROUP_ BYTES - Maximum per-workgroup shared memory the reference interpreter will allocate.