Research-Stack/6-Documentation/tiddlywiki-local/wiki/tiddlers/NUVMAP Metadata Blitter.tid
2026-05-11 14:49:17 -05:00

73 lines
1.7 KiB
Text

created: 20260511000000000
modified: 20260511000000000
tags: ResearchStack NUVMAP GPU WebGPU Shader MetadataBlitter SIMD HOLD
title: NUVMAP Metadata Blitter
type: text/vnd.tiddlywiki
! NUVMAP Metadata Blitter
Status: `SHADER_PRIMITIVE_WITH_HOLD_BOUNDARIES`
The NUVMAP metadata blitter is a small WebGPU compute shader that treats the GPU
as a dataport fabric for metadata sorting.
Runner:
```text
5-Applications/scripts/nuvmap_metadata_blitter.wgsl
```
Documentation:
```text
5-Applications/scripts/nuvmap_metadata_blitter.md
```
!! Purpose
```text
metadata shard
-> fixed-size word windows
-> SIMD-wide blitter pass
-> sortable route keys
-> packed metric words
-> radix grouping
-> waveprobe / metaprobe selection
-> NUVMAP receipts
```
The shader does not sort by itself. It emits deterministic keys and metrics that
a later radix/sort pass or CPU verifier can consume.
!! Key Word
```text
bits 31..24 hash8 phase seed
bits 23..17 density7 byte pressure
bits 16..10 delta7 shock / torsion gradient
bits 9..4 zero6 void / low-information throat
bits 3..0 flags4 boundary condition
```
!! Dispatch Shape
```text
@compute @workgroup_size(256, 1, 1)
dispatch = ceil(record_count / 256)
```
Each invocation scans at most `64` words. Larger records should be split into
fixed windows so the kernel remains predictable and shardable.
!! Claim Boundary
This shader emits route keys and metrics only. It does not decompress, classify
semantic meaning, publish data on-chain, claim GPU acceleration, or claim
compression improvement. GPU claims require CPU/GPU parity receipts.
Related:
* [[Hyper-Soliton Radix Bins]]
* [[NUVMAP Chain Protocol Plan]]
* [[Blockchain L3 Self-Scan Scheduler]]
* [[NUVMAP GPU Burst Check Plan]]