test/tint/builtins/gen/literal/subgroupMatrixMultiplyAccumulate/8b907c.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 {
  device tint_array<half, 1024>* prevent_dce;
};

simdgroup_half8x8 subgroupMatrixMultiplyAccumulate_8b907c() {
  simdgroup_half8x8 const v = simdgroup_half8x8();
  simdgroup_half8x8 const v_1 = simdgroup_half8x8();
  simdgroup_half8x8 const v_2 = simdgroup_half8x8();
  simdgroup_half8x8 v_3 = make_filled_simdgroup_matrix<half, 8, 8>(0.0h);
  simdgroup_multiply_accumulate(v_3, v, v_1, v_2);
  simdgroup_half8x8 res = v_3;
  return res;
}

kernel void compute_main(device tint_array<half, 1024>* prevent_dce [[buffer(0)]]) {
  tint_module_vars_struct const tint_module_vars = tint_module_vars_struct{.prevent_dce=prevent_dce};
  simdgroup_store(subgroupMatrixMultiplyAccumulate_8b907c(), (&(*tint_module_vars.prevent_dce)[0u]), ulong(64u), ulong2(0ul), false);
}