64 lines
1.4 KiB
HLSL
64 lines
1.4 KiB
HLSL
struct S_atomic {
|
|
int x;
|
|
uint a;
|
|
uint y;
|
|
};
|
|
|
|
static uint local_invocation_index_1 = 0u;
|
|
groupshared S_atomic wg[10];
|
|
|
|
void compute_main_inner(uint local_invocation_index) {
|
|
uint idx = 0u;
|
|
idx = local_invocation_index;
|
|
[loop] while (true) {
|
|
const uint x_23 = idx;
|
|
if (!((x_23 < 10u))) {
|
|
break;
|
|
}
|
|
const uint x_28 = idx;
|
|
wg[x_28].x = 0;
|
|
uint atomic_result = 0u;
|
|
InterlockedExchange(wg[x_28].a, 0u, atomic_result);
|
|
wg[x_28].y = 0u;
|
|
{
|
|
const uint x_41 = idx;
|
|
idx = (x_41 + 1u);
|
|
}
|
|
}
|
|
GroupMemoryBarrierWithGroupSync();
|
|
uint atomic_result_1 = 0u;
|
|
InterlockedExchange(wg[4].a, 1u, atomic_result_1);
|
|
return;
|
|
}
|
|
|
|
void compute_main_1() {
|
|
const uint x_53 = local_invocation_index_1;
|
|
compute_main_inner(x_53);
|
|
return;
|
|
}
|
|
|
|
struct tint_symbol_1 {
|
|
uint local_invocation_index_1_param : SV_GroupIndex;
|
|
};
|
|
|
|
void compute_main_inner_1(uint local_invocation_index_1_param) {
|
|
{
|
|
[loop] for(uint idx_1 = local_invocation_index_1_param; (idx_1 < 10u); idx_1 = (idx_1 + 1u)) {
|
|
const uint i = idx_1;
|
|
wg[i].x = 0;
|
|
uint atomic_result_2 = 0u;
|
|
InterlockedExchange(wg[i].a, 0u, atomic_result_2);
|
|
wg[i].y = 0u;
|
|
}
|
|
}
|
|
GroupMemoryBarrierWithGroupSync();
|
|
local_invocation_index_1 = local_invocation_index_1_param;
|
|
compute_main_1();
|
|
}
|
|
|
|
[numthreads(1, 1, 1)]
|
|
void compute_main(tint_symbol_1 tint_symbol) {
|
|
compute_main_inner_1(tint_symbol.local_invocation_index_1_param);
|
|
return;
|
|
}
|