rlx-vulkan 0.2.10

Native Vulkan compute backend for RLX (raw `ash` + embedded SPIR-V compute kernels)
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

#version 450
// Element-wise binary op over the shared f32 arena with per-operand
// trailing-broadcast modulus. op: 0=add 1=sub 2=mul 3=div 4=max 5=min 6=pow.
layout(local_size_x = 256) in;

layout(std430, binding = 0) buffer Arena { float data[]; };

layout(push_constant) uniform PC {
    uint n;       // output element count
    uint a_off;   // f32-element offset of lhs
    uint b_off;   // f32-element offset of rhs
    uint c_off;   // f32-element offset of output
    uint a_mod;   // lhs element count for broadcast (0 = no broadcast)
    uint b_mod;   // rhs element count for broadcast (0 = no broadcast)
    uint op;      // selector
} pc;

void main() {
    uint i = gl_GlobalInvocationID.x;
    if (i >= pc.n) { return; }
    uint ai = (pc.a_mod == 0u) ? i : (i % pc.a_mod);
    uint bi = (pc.b_mod == 0u) ? i : (i % pc.b_mod);
    float a = data[pc.a_off + ai];
    float b = data[pc.b_off + bi];
    float c = 0.0;
    switch (pc.op) {
        case 0u: c = a + b; break;
        case 1u: c = a - b; break;
        case 2u: c = a * b; break;
        case 3u: c = a / b; break;
        case 4u: c = max(a, b); break;
        case 5u: c = min(a, b); break;
        case 6u: c = pow(a, b); break;
        default: c = 0.0; break;
    }
    data[pc.c_off + i] = c;
}