sin-cos-pair

Status: shipped (Phase 2) — see CHANGELOG.

What it detects

Separate calls to sin(x) and cos(x) within the same function body where both calls share the same argument expression x. The rule matches any two calls — in any order, any number of statements apart — that operate on the same syntactic argument (same identifier, same literal, or structurally identical sub-expression). It does not fire when only one of the two is present, when the arguments differ, or when the results of both calls are already combined via a sincos intrinsic call.

Why it matters on a GPU

On AMD RDNA/RDNA 2/RDNA 3, NVIDIA Turing/Ada Lovelace, and Intel Xe-HPG, sin and cos are transcendental instructions that run on the special-function unit (TALU / transcendental ALU) at one-quarter peak VALU throughput. Two separate calls — sin(x) and cos(x) — occupy two distinct quarter-rate issue slots: on RDNA 3 that is v_sin_f32 followed by v_cos_f32, each at 1/4 rate, for a combined cost of roughly 8 full-rate ALU-equivalent cycles.

The sincos(x, s, c) intrinsic instructs the GPU driver and compiler that both results are needed for the same input. On many architectures the driver can lower this to a single TALU operation that produces both sin and cos simultaneously, or can at minimum pipeline the two transcendental operations from a single angle-reduction step. The angle-reduction (computing x mod 2π and mapping to the unit circle) is the most expensive part of the transcendental evaluation; when computed for sin alone, the same range-reduced argument is discarded and recomputed from scratch for cos. sincos shares that range-reduction across both outputs.

In vertex shaders implementing rotation matrices — a rotation by angle θ requires both sin(θ) and cos(θ) in the matrix entries — and in particle systems applying angular velocity, this pattern is common and hot. The fixture at lines 55-59 shows the exact form: separate sin(angle) and cos(angle) calls used to fill a float2. Replacing with sincos(angle, s, c) is a single source change that halves the transcendental cost for that invocation on hardware that fuses the pair.

Examples

Bad

hlsl

// tests/fixtures/phase2/math.hlsl, lines 55-59 — HIT(sin-cos-pair)
float2 sin_cos_of_same(float angle) {
    float s = sin(angle);   // quarter-rate v_sin_f32; range-reduces angle
    float c = cos(angle);   // quarter-rate v_cos_f32; range-reduces angle again
    return float2(s, c);
}

Good

hlsl

// After machine-applicable fix:
float2 sin_cos_of_same(float angle) {
    float s, c;
    sincos(angle, s, c);   // single range-reduction, potentially one TALU op
    return float2(s, c);
}

Options

none

Fix availability

machine-applicable — Introducing sincos(x, s, c) and replacing the original sin(x) and cos(x) calls with the output variables is a pure textual transformation with no observable semantic change. The output variables are introduced at declaration scope immediately preceding the sincos call. shader-clippy fix applies it automatically.

sin-cos-pair ​

What it detects ​

Why it matters on a GPU ​

Examples ​

Bad ​

Good ​

Options ​

Fix availability ​

See also ​