vyre 0.4.0

GPU compute intermediate representation with a standard operation library
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52

use crate::ir::{Expr, Program};
use crate::ops::primitive;
use crate::ops::{AlgebraicLaw, OpSpec, U32_OUTPUTS, U32_U32_INPUTS};

// Saturating addition of two u32 values.



pub const LAWS: &[AlgebraicLaw] = &[
    AlgebraicLaw::Commutative,
    AlgebraicLaw::Bounded {
        lo: 0,
        hi: u32::MAX,
    },
];

/// Saturating u32 addition operation.
#[derive(Debug, Clone, Copy, Default)]
pub struct AddSat;

impl AddSat {
    /// Declarative operation specification.
    pub const SPEC: OpSpec = OpSpec::composition_inlinable(
        "primitive.math.add_sat",
        U32_U32_INPUTS,
        U32_OUTPUTS,
        LAWS,
        Self::program,
    );

    /// Build the canonical IR program.
    #[must_use]
    pub fn program() -> Program {
        primitive::binary_u32_program(|a, b| {
            let sum = Expr::add(a.clone(), b);
            Expr::select(Expr::lt(sum.clone(), a), Expr::u32(u32::MAX), sum)
        })
    }
}

// Backend-specific lowering tripwires for saturating addition.

// WGSL lowering marker for `primitive.math.add_sat`.
//
// Not a stub: this is a zero-overhead Category A marker. `AddSat::program`
// builds concrete IR through `core/src/ops/primitive/math/add_sat/kernel.rs`; `core/src/lower/wgsl/expr.rs` emits WGSL.
// `core/tests/conformance.rs::conformance_all_primitives` verifies
// lowered GPU bytes are bit-exact against the conform CPU reference.
//
// ```wgsl
// _vyre_store_out(idx, select(sum, 4294967295u, (sum < a)));
// ```