DE wrote:
looks smart and tricky but i don't see any advantages over computed indeces but only wasted spaces and time (for allocating additional storage for pointers) at the same time dereferencing a pointer is like computing an index - no computational advantage agai Back to this issue here is the generated code using
1/ NRC array of dimension 4 mov 0x804c18c,%eax mov %eax,-0x20(%ebp) mov -0x20(%ebp),%eax 1> xor %ebx,%ebx mov (%eax,%edi,1),%eax mov %eax,-0x1c(%ebp) mov -0x1c(%ebp),%eax 2> xor %ecx,%ecx mov (%eax,%ebx,1),%esi mov (%esi,%ecx,1),%eax 3> add $0x4,%ecx mov 0x804c140,%edx cmp $0x8,%ecx mov %edx,(%eax) mov 0x804c140,%edx mov %edx,0x4(%eax) jne 3> add $0x4,%ebx cmp $0x8,%ebx jne 2> add $0x4,%edi cmp $0x8,%edi jne 1> 2/ Same array allocated as a float* with index computation mov %eax,0x804c198 mov %eax,-0x14(%ebp) add $0x40,%eax mov %eax,-0x24(%ebp) mov -0x14(%ebp),%eax 1> xor %esi,%esi mov -0x14(%ebp),%edi add $0x4,%eax mov %eax,-0x18(%ebp) mov -0x18(%ebp),%ecx 2> mov %edi,%ebx xor %edx,%edx mov 0x804c144,%eax 3> add $0x1,%edx mov %eax,(%ebx) mov 0x804c144,%eax add $0x8,%ebx mov %eax,(%ecx) add $0x8,%ecx cmp $0x2,%edx jne 3> add$0x1,%esi add$0x10,%edi addl$0x10,-0x18(%ebp) cmp$0x2,%esi jne 2> addl $0x20,-0x14(%ebp) mov -0x24(%ebp),%eax cmp %eax,-0x14(%ebp) jne 1> Lo and behold ;) Doing the computation of index jump int he allcoation helps tremendously. -- ___________________________________________ Joel Falcou - Assistant Professor PARALL Team - LRI - Universite Paris Sud XI Tel : (+33)1 69 15 66 35