test/tint/extensions/subgroup_matrix/runtime_array_workgroup.wgsl.expected.msl

#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];
};

struct tint_module_vars_struct {
  threadgroup tint_array<uchar, 1024>* v;
};

struct tint_symbol_1 {
  tint_array<uchar, 1024> tint_symbol;
};

void main_inner(uint tint_local_index, tint_module_vars_struct tint_module_vars) {
  {
    uint v_1 = 0u;
    v_1 = tint_local_index;
    while(true) {
      uint const v_2 = v_1;
      if ((v_2 >= 1024u)) {
        break;
      }
      (*tint_module_vars.v)[v_2] = 0u;
      {
        v_1 = (v_2 + 32u);
      }
    }
  }
  (threadgroup_barrier(mem_flags::mem_threadgroup));
  threadgroup tint_array<float, 1>* const view = reinterpret_cast<threadgroup tint_array<float, 1>*>(reinterpret_cast<threadgroup char*>(tint_module_vars.v) + 0u);
  uint const v_3 = as_type<uint>(0);
  uint const v_4 = as_type<uint>(8);
  bool const v_5 = (((v_3 + (v_4 * 7u)) + 8u) <= ((1024u - 0u) / 4u));
  simdgroup_float8x8 v_6 = make_filled_simdgroup_matrix<float, 8, 8>(0.0f);
  (simdgroup_load(v_6, (&(*view)[select(0u, v_3, v_5)]), ulong(select(8u, v_4, v_5)), ulong2(0ul), true));
  simdgroup_float8x8 const m = v_6;
  uint const v_7 = as_type<uint>(0);
  uint const v_8 = as_type<uint>(8);
  bool const v_9 = (((v_7 + (v_8 * 7u)) + 8u) <= ((1024u - 0u) / 4u));
  (simdgroup_store(m, (&(*view)[select(0u, v_7, v_9)]), ulong(select(8u, v_8, v_9)), ulong2(0ul), false));
}

[[max_total_threads_per_threadgroup(32)]]
kernel void v_10(uint tint_local_index [[thread_index_in_threadgroup]], threadgroup tint_symbol_1* v_11 [[threadgroup(0)]]) {
  tint_module_vars_struct const tint_module_vars = tint_module_vars_struct{.v=(&(*v_11).tint_symbol)};
  (main_inner(tint_local_index, tint_module_vars));
}