| cbuffer cbuffer_u : register(b0, space0) { |
| uint4 u[4]; |
| }; |
| static matrix<float16_t, 2, 4> p[4] = (matrix<float16_t, 2, 4>[4])0; |
| |
| matrix<float16_t, 2, 4> u_load_1(uint offset) { |
| const uint scalar_offset = ((offset + 0u)) / 4; |
| uint4 ubo_load_1 = u[scalar_offset / 4]; |
| uint2 ubo_load = ((scalar_offset & 2) ? ubo_load_1.zw : ubo_load_1.xy); |
| vector<float16_t, 2> ubo_load_xz = vector<float16_t, 2>(f16tof32(ubo_load & 0xFFFF)); |
| vector<float16_t, 2> ubo_load_yw = vector<float16_t, 2>(f16tof32(ubo_load >> 16)); |
| const uint scalar_offset_1 = ((offset + 8u)) / 4; |
| uint4 ubo_load_3 = u[scalar_offset_1 / 4]; |
| uint2 ubo_load_2 = ((scalar_offset_1 & 2) ? ubo_load_3.zw : ubo_load_3.xy); |
| vector<float16_t, 2> ubo_load_2_xz = vector<float16_t, 2>(f16tof32(ubo_load_2 & 0xFFFF)); |
| vector<float16_t, 2> ubo_load_2_yw = vector<float16_t, 2>(f16tof32(ubo_load_2 >> 16)); |
| return matrix<float16_t, 2, 4>(vector<float16_t, 4>(ubo_load_xz[0], ubo_load_yw[0], ubo_load_xz[1], ubo_load_yw[1]), vector<float16_t, 4>(ubo_load_2_xz[0], ubo_load_2_yw[0], ubo_load_2_xz[1], ubo_load_2_yw[1])); |
| } |
| |
| typedef matrix<float16_t, 2, 4> u_load_ret[4]; |
| u_load_ret u_load(uint offset) { |
| matrix<float16_t, 2, 4> arr[4] = (matrix<float16_t, 2, 4>[4])0; |
| { |
| for(uint i = 0u; (i < 4u); i = (i + 1u)) { |
| arr[i] = u_load_1((offset + (i * 16u))); |
| } |
| } |
| return arr; |
| } |
| |
| [numthreads(1, 1, 1)] |
| void f() { |
| p = u_load(0u); |
| p[1] = u_load_1(32u); |
| uint2 ubo_load_4 = u[0].zw; |
| vector<float16_t, 2> ubo_load_4_xz = vector<float16_t, 2>(f16tof32(ubo_load_4 & 0xFFFF)); |
| vector<float16_t, 2> ubo_load_4_yw = vector<float16_t, 2>(f16tof32(ubo_load_4 >> 16)); |
| p[1][0] = vector<float16_t, 4>(ubo_load_4_xz[0], ubo_load_4_yw[0], ubo_load_4_xz[1], ubo_load_4_yw[1]).ywxz; |
| p[1][0].x = float16_t(f16tof32(((u[0].z) & 0xFFFF))); |
| return; |
| } |