| #include <metal_stdlib> |
| |
| using namespace metal; |
| |
| template<typename T, size_t N> |
| struct tint_array { |
| const constant T& operator[](size_t i) const constant { return elements[i]; } |
| device T& operator[](size_t i) device { return elements[i]; } |
| const device T& operator[](size_t i) const device { return elements[i]; } |
| thread T& operator[](size_t i) thread { return elements[i]; } |
| const thread T& operator[](size_t i) const thread { return elements[i]; } |
| threadgroup T& operator[](size_t i) threadgroup { return elements[i]; } |
| const threadgroup T& operator[](size_t i) const threadgroup { return elements[i]; } |
| T elements[N]; |
| }; |
| |
| #define TINT_ISOLATE_UB(VOLATILE_NAME) \ |
| {volatile bool VOLATILE_NAME = false; if (VOLATILE_NAME) break;} |
| |
| void tint_zero_workgroup_memory(uint local_idx, threadgroup tint_array<atomic_uint, 4>* const tint_symbol) { |
| for(uint idx = local_idx; (idx < 4u); idx = (idx + 1u)) { |
| TINT_ISOLATE_UB(tint_volatile_false); |
| uint const i = idx; |
| atomic_store_explicit(&((*(tint_symbol))[i]), 0u, memory_order_relaxed); |
| } |
| threadgroup_barrier(mem_flags::mem_threadgroup); |
| } |
| |
| void compute_main_inner(uint local_invocation_index, threadgroup tint_array<atomic_uint, 4>* const tint_symbol_1) { |
| tint_zero_workgroup_memory(local_invocation_index, tint_symbol_1); |
| atomic_store_explicit(&((*(tint_symbol_1))[1]), 1u, memory_order_relaxed); |
| } |
| |
| kernel void compute_main(uint local_invocation_index [[thread_index_in_threadgroup]]) { |
| threadgroup tint_array<atomic_uint, 4> tint_symbol_2; |
| compute_main_inner(local_invocation_index, &(tint_symbol_2)); |
| return; |
| } |
| |