| #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 SB_RW { |
| /* 0x0000 */ tint_array<half, 64> arg_0; |
| }; |
| |
| struct tint_module_vars_struct { |
| device SB_RW* sb_rw; |
| }; |
| |
| void subgroupMatrixStore_0522d1(tint_module_vars_struct tint_module_vars) { |
| uint arg_1 = 1u; |
| simdgroup_half8x8 arg_2 = simdgroup_half8x8(); |
| uint arg_4 = 1u; |
| simdgroup_store(arg_2, (&(*tint_module_vars.sb_rw).arg_0[arg_1]), ulong(arg_4), ulong2(0ul), true); |
| } |
| |
| kernel void compute_main(device SB_RW* sb_rw [[buffer(0)]]) { |
| tint_module_vars_struct const tint_module_vars = tint_module_vars_struct{.sb_rw=sb_rw}; |
| subgroupMatrixStore_0522d1(tint_module_vars); |
| } |