GNAT (GCC) Profile Guided Compilation

[Keean Schupke <keean.schupke@googlemail.com>]

I am developing a Monte-Carlo simulation engine, and I have implementations in C++ and Ada. All performance figures come from the same reference hardware. Both are built using GCC, and similar optimisation is applied to each (same optimisation level, cpu arch, inlining etc).

My C++ implementation achieves 40,000 simulations per second.

My Ada port initially achieves 32,000 iterations per second.

I refactored the Ada port to use System.Address to access the main data structure (as in the 'C++' version benchmarking showed a clear performance benefit to *ptr_x compared to array[x])

The Ada port using System.Address achieves 40,000 simulations per second - so this is looking good.

However, when I use profile guided compilation, the C++ code performance leaps to 56,000 simulations per second, whereas the Ada only achieves 44,000 simulations per second.

Anyone have any ideas why profile guided compilation performs so poorly with Ada compared to C++? The initial with normal compilation seemed very promising.


Cheers,
Keean.