blob: 4b1fb11376ba3d74dc21b42e2eb789acbe1aca9f [file] [log] [blame]
struct Inner {
scalar_i32 : i32,
@align(16) @size(16)
scalar_f32 : f32,
};
struct S {
scalar_f32 : f32,
scalar_i32 : i32,
scalar_u32 : u32,
vec2_f32 : vec2<f32>,
vec2_i32 : vec2<i32>,
vec2_u32 : vec2<u32>,
vec3_f32 : vec3<f32>,
vec3_i32 : vec3<i32>,
vec3_u32 : vec3<u32>,
vec4_f32 : vec4<f32>,
vec4_i32 : vec4<i32>,
vec4_u32 : vec4<u32>,
mat2x2_f32 : mat2x2<f32>,
mat2x3_f32 : mat2x3<f32>,
mat2x4_f32 : mat2x4<f32>,
mat3x2_f32 : mat3x2<f32>,
mat3x3_f32 : mat3x3<f32>,
mat3x4_f32 : mat3x4<f32>,
mat4x2_f32 : mat4x2<f32>,
mat4x3_f32 : mat4x3<f32>,
mat4x4_f32 : mat4x4<f32>,
@align(16) arr2_vec3_f32 : array<vec3<f32>, 2>,
@align(16) struct_inner : Inner,
@align(16) array_struct_inner : array<Inner, 4>,
};
@binding(0) @group(0) var<uniform> ub : S;
@group(0) @binding(1) var<storage, read_write> s: i32;
@compute @workgroup_size(1)
fn main() {
let scalar_f32 = ub.scalar_f32;
let scalar_i32 = ub.scalar_i32;
let scalar_u32 = ub.scalar_u32;
let vec2_f32 = ub.vec2_f32;
let vec2_i32 = ub.vec2_i32;
let vec2_u32 = ub.vec2_u32;
let vec3_f32 = ub.vec3_f32;
let vec3_i32 = ub.vec3_i32;
let vec3_u32 = ub.vec3_u32;
let vec4_f32 = ub.vec4_f32;
let vec4_i32 = ub.vec4_i32;
let vec4_u32 = ub.vec4_u32;
let mat2x2_f32 = ub.mat2x2_f32;
let mat2x3_f32 = ub.mat2x3_f32;
let mat2x4_f32 = ub.mat2x4_f32;
let mat3x2_f32 = ub.mat3x2_f32;
let mat3x3_f32 = ub.mat3x3_f32;
let mat3x4_f32 = ub.mat3x4_f32;
let mat4x2_f32 = ub.mat4x2_f32;
let mat4x3_f32 = ub.mat4x3_f32;
let mat4x4_f32 = ub.mat4x4_f32;
let arr2_vec3_f32 = ub.arr2_vec3_f32;
let struct_inner = ub.struct_inner;
let array_struct_inner = ub.array_struct_inner;
s = i32(scalar_f32) + scalar_i32 + i32(scalar_u32) +
i32(vec2_f32.x) + vec2_i32.x + i32(vec2_u32.x) +
i32(vec3_f32.y) + vec3_i32.y + i32(vec3_u32.y) +
i32(vec4_f32.z) + vec4_i32.z + i32(vec4_u32.z) +
i32(mat2x2_f32[0].x) + i32(mat2x3_f32[0].x) +
i32(mat2x4_f32[0].x) + i32(mat3x2_f32[0].x) +
i32(mat3x3_f32[0].x) + i32(mat3x4_f32[0].x) +
i32(mat4x2_f32[0].x) + i32(mat4x3_f32[0].x) +
i32(mat4x4_f32[0].x) + i32(arr2_vec3_f32[0].x) +
struct_inner.scalar_i32 + array_struct_inner[0].scalar_i32;
}