58 lines
1.3 KiB
HLSL
58 lines
1.3 KiB
HLSL
|
struct S_atomic {
|
||
|
int x;
|
||
|
uint a[10];
|
||
|
uint y;
|
||
|
};
|
||
|
|
||
|
static uint local_invocation_index_1 = 0u;
|
||
|
groupshared S_atomic wg;
|
||
|
|
||
|
void compute_main_inner(uint local_invocation_index) {
|
||
|
uint idx = 0u;
|
||
|
wg.x = 0;
|
||
|
wg.y = 0u;
|
||
|
idx = local_invocation_index;
|
||
|
{
|
||
|
[loop] for(; !(!((idx < 10u))); idx = (idx + 1u)) {
|
||
|
uint atomic_result = 0u;
|
||
|
InterlockedExchange(wg.a[idx], 0u, atomic_result);
|
||
|
}
|
||
|
}
|
||
|
GroupMemoryBarrierWithGroupSync();
|
||
|
uint atomic_result_1 = 0u;
|
||
|
InterlockedExchange(wg.a[4], 1u, atomic_result_1);
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
void compute_main_1() {
|
||
|
compute_main_inner(local_invocation_index_1);
|
||
|
return;
|
||
|
}
|
||
|
|
||
|
struct tint_symbol_1 {
|
||
|
uint local_invocation_index_1_param : SV_GroupIndex;
|
||
|
};
|
||
|
|
||
|
void compute_main_inner_1(uint local_invocation_index_1_param) {
|
||
|
{
|
||
|
wg.x = 0;
|
||
|
wg.y = 0u;
|
||
|
}
|
||
|
{
|
||
|
[loop] for(uint idx_1 = local_invocation_index_1_param; (idx_1 < 10u); idx_1 = (idx_1 + 1u)) {
|
||
|
const uint i = idx_1;
|
||
|
uint atomic_result_2 = 0u;
|
||
|
InterlockedExchange(wg.a[i], 0u, atomic_result_2);
|
||
|
}
|
||
|
}
|
||
|
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;
|
||
|
}
|