The decompiler needs to use data size knowledge it has to produce better results, question is, how!

[Wall-AF]

The decompiler has access to the sizes of the different base types but doesn't really seem to make use of this knowledge. A good example is when a type like long naturally required more than a single register to represent it, the decompiler ends up producing code like CONCATXX(*((int)(long *)&LONG_103065ce + X),(int)*(long *)&LONG_103065ce) because, in this case, two stack pushes are required as 2 registers/memory locations are needed/used. This is exacerbated when addition/subtraction is performed as assembly such as this x86 extract ADD word ptr [g_nSumDo10ms_65d2],0x1; ADC word ptr [g_nSumDo10ms_65d2+X],0x0 produces

                        bVar4 = -2 < (uint)g_nSumDo10ms_65d2;
                        (uint)g_nSumDo10ms_65d2 += 1;
                        g_nSumDo10ms_65d2._X_X_ += (uint)bVar4;

making any subsequent arithmetic harder to follow. (In the C above I replaced the hex representation of -2 with that value.)