Zeroing local variables

[TimoVJL]

Code Select Expand

AMD Ryzen 5 3400G with Radeon Vega Graphics     (SSE4)

20775   cycles for 100 * ClearLocals (fast+bloated)
21998   cycles for 100 * ClearLocals (slow+compact)
??      cycles for 100 * one nop
74      cycles for 100 * 10 nops
1733    cycles for 100 * 100 nops

18085   cycles for 100 * ClearLocals (fast+bloated)
22850   cycles for 100 * ClearLocals (slow+compact)
??      cycles for 100 * one nop
74      cycles for 100 * 10 nops
1731    cycles for 100 * 100 nops

18147   cycles for 100 * ClearLocals (fast+bloated)
21744   cycles for 100 * ClearLocals (slow+compact)
??      cycles for 100 * one nop
72      cycles for 100 * 10 nops
1736    cycles for 100 * 100 nops

19727   cycles for 100 * ClearLocals (fast+bloated)
20353   cycles for 100 * ClearLocals (slow+compact)
??      cycles for 100 * one nop
79      cycles for 100 * 10 nops
3249    cycles for 100 * 100 nops

51      bytes for ClearLocals (fast+bloated)
39      bytes for ClearLocals (slow+compact)
1       bytes for one nop
10      bytes for 10 nops
100     bytes for 100 nops


[daydreamer]

It's good to know pad nops before innerloop don't take many clock cycles
But what about the 2,3,4,5,6,7,8,9 byte nops? Is there a difference between 9 single byte nops vs one 9byte nop (66h nop [eax*8+ebx+dword address] something)

[jj2007]

Good question, why don't you test it?