Threads

What may be shared

Object	May be shared between threads?
Definitions	Yes. Definitions may be made, removed and read in any thread: each is made under a lock, and a compilation reads them as they were when it started, so programs may be compiled in several threads at once. The programs kept are kept for every thread, under the same lock.
program	Yes. A compiled program never changes. The vector program-spirv returns is the program's own, the same at each call: read it, and copy it to change it.
vertex-input, uniform, resource, program-binding, slot-layout	Yes. They never change.
The buffer pack-uniforms or pack-value writes	No: it is the caller's, and belongs to one thread at a time.
Targets, such as :vulkan's	Yes. A target never changes.
The SPIR-V grammar's tables, the IR's operation table	Yes. They never change once loaded.
An IR module and everything in it: its types, values, functions and blocks	No: one thread at a time.
An IR builder	No: one thread at a time.
A SPIR-V module of clsl/spirv	No: one thread at a time.
The generator's grammar model	Yes. It never changes once read.

Modules may be translated into SPIR-V in several threads at once, each its own.

Global state

The definitions are the one global state clsl keeps: a registry of what define-function and its kin define, which exists once per image by nature, as the image's functions do, with the programs compile-program compiled from them, a cache keyed by the definitions each was compiled from. Nothing in clsl needs setting up again when a saved image starts.

A uniform of define-uniform is a special variable. pack-uniforms writes it as the thread that packs sees it, so a thread may let-bind it to pack its own value, as for a camera of its own.

Every frame

What a frame loop calls allocates nothing:

• definitions-generation and program-stale-p, which say whether a program needs compiling again;
• pack-uniforms given a :buffer, and pack-value, given doubles and float pairs as double-floats, vectors and matrices of them as vectors of double-floats, and complex numbers of doubles as (complex double-float) values;
• dfloat-parts of a single-float, or of a double-float of a magnitude from 2^-70 to the floats' largest.

Compiling allocates, as it makes a program: compile a program once, and again when program-stale-p says a definition it uses has changed, and read what a pipeline needs of it when the pipeline is made.