Function getL2CacheSize

src/backend/opencl/platform.hpp:103–111

OCL only reports on L1 cache, so we have to estimate the L2 Cache size. From studying many GPU cards, it is noticed that their is a direct correlation between Cache line and L2 Cache size: - 16KB L2 Cache for each bit in Cache line. Example: RTX3070 (4096KB of L2 Cache, 256Bit of Cache line) --> 256*16KB = 4096KB - This is also valid for all AMD GPU's - Exceptions GTX10XX series have 8KB per bit o

//          * GTX10XX series have 8KB per bit of cache line
//          * iGPU (64bit cacheline) have 5KB per bit of cache line
inline size_t getL2CacheSize(const cl::Device& device) {
    const unsigned cacheLine{getMemoryBusWidth(device)};
    return cacheLine * 1024ULL *
           (cacheLine == 64 ? 5
            : device.getInfo<CL_DEVICE_NAME>().find("GTX 10") ==
                    std::string::npos
                ? 16
                : 8);
}

inline unsigned getComputeUnits(const cl::Device& device) {
    return device.getInfo<CL_DEVICE_MAX_COMPUTE_UNITS>();