MCPcopy Create free account
hub / github.com/cortex-js/compute-engine / formatGPUNumber

Function formatGPUNumber

src/compute-engine/compilation/gpu-target.ts:3265–3280  ·  view source on GitHub ↗
(n: number)

Source from the content-addressed store, hash-verified

3263 );
3264 }
3265 // And the mirror obligation, for a scalar the checks above left standing (a
3266 // builtin whose permission table admits a scalar in a slot the language
3267 // nonetheless declares `vecN`). Defensive: with today's tables every mixed
3268 // scalar/`vecN` call already declines above, and the ALL-SCALAR calls — the
3269 // ones this rule exists for — returned before reaching here.
3270 declineVectorOnly();
3271}
3272
3273/**
3274 * Fold a variadic application of a 2-argument shader builtin (`min`/`max`)
3275 * into a left-nested tree of 2-argument calls: `max(max(a, b), c)`. GLSL and
3276 * WGSL do not accept a 3+-argument `min`/`max`, so emitting `max(a, b, c)`
3277 * would be invalid shader source.
3278 *
3279 * Each GENERATED call is validated as it is produced, against the ORIGINAL
3280 * operand shapes (`gpuOperandShape`) rather than against the emitted source:
3281 * once the fold has nested the calls, the CE operand positions no longer line
3282 * up with the emitted argument positions, and the generic gate's source-level
3283 * reconstruction (`gpuSourceIsVector`) cannot see a vector in a BARE

Callers 6

intervalGLSLPreambleFunction · 0.90
createTargetMethod · 0.90
compileGPUSumProductFunction · 0.85
gpu-target.tsFile · 0.85
createTargetFunction · 0.85

Calls 2

toStringMethod · 0.65
isFiniteMethod · 0.45

Tested by

no test coverage detected