blocks, the mesh program that exercises every block collective of a multi-warp node, as PTX from C
$ sexc add @ajax/mesh-prog-blocks- total 3
- drwxr-xr-x-../
- drwxr-xr-x-Oct 27 03:26v/# 1 version
- drwxr-xr-x-Oct 27 03:26files/# the files of 0.1.0
- latest
- 0.1.0, published Oct 27 03:26 by ajax
- owners
- ajax
- license
- MIT
- dependencies
- @ajax/mesh-rt ^0.1.0, @ajax/mesh-rt-sim-mailbox ^0.1.0, @ajax/mesh-ref-block ^0.1.0, @ajax/mesh-check ^0.1.0
- dependents
- none yet
- keywords
- mesh alphaos gpu ptx program collectives
- homepage
- https://github.com/thomasdavis/alpha-c-mesh/tree/main/programs/blocks
- repository
- https://github.com/thomasdavis/alpha-c-mesh
1# @ajax/mesh-prog-blocks 2 3The block collectives of a multi-warp node, for kcheck: every thread computes every reduction, scan, 4a broadcast and a shared-memory round trip on @ajax/mesh-ref-block's test values (zeros, subnormals, 5large values); op 2 lets warps 1.. leave at once while warp 0 runs a warp sum. `variants/blocks-w4` is 6the same image with 4 warps a node. 7 8- **Ported from:** alpha-rust-mesh `programs/blocks/src/lib.rs`. 9- **Tests:** kcheck's B1, B2 and B3 on the simulator: 32-, 4- and 1-warp nodes, six seeds, every header 10 and per-thread word bitwise equal to @ajax/mesh-ref-block (subnormals included: the simulator's 11 flush-to-zero matches the cluster float mode); the 4-warp f32 sum equals the 1024-leaf tree's; 64 12 early exits, a stop/start, and a task after it.