local pointers with an EXPLICIT lower bound 1

[ya0037]

I have tried to test my code and confronted with st strange:
making local pointers with an EXPLICIT lower bound seems to generate faster code.
Is this right? In case it is, why?

 

[xwb]

Have you looked at the code generation?

 

[ya0037]

Sorry, could you explain what do you mean by code generation and how I can look at that?

 

[xwb]

Depends on the compiler - some compilers will generate an assembler listing if requested to do so. You can then compare the ones with and without the lower bound.

Alternatively, in the debugger, set a breakpoint where the code is accessing an array index and have a look at the disassembly.

At a guess, if you specify a lower bound of 0, it will be faster (since it is easier to compute arrays from a 0 index). If it is any other lower bound, the speed will be the same.

 

[xwb]

Just to clarify - say there are 3 arrays

integer a(-5:5), b(0:10), c(11)

And there are 3 computations

a(2) = 10
b(2) = 10
c(2) = 10

To compute the index

addr(a(2)) = addr(a) + (2 - (-5)) * size(integer)
addr(b(2)) = addr(b) + (2 - (0)) * size(integer) optimized to b + 2 * size(integer)
addr(c(2)) = addr(c) + (2 - (1)) * size(integer)

The computation of b has one instruction less so in theory, it will be faster.