Přeskočit obsah

Time and Benchmark

Set of functions to time measurement and Benchmark.

Initialization

The best way to initialize all Ikarus functions is to call MEM_InitAll() in the Init_Global() initialization function.

Warning

If you want to use Ikarus in Gothic 1, it is best to define your own Init_Global() function and call it from every world initialization function.

1
MEM_InitAll();

Implementation

Ikarus.d on GitHub

Time functions

MEM_GetSystemTime

Returns the elapsed time since Gothic started. 

1
func int MEM_GetSystemTime()
Return value

The function returns the elapsed time since the start of Gothic in milliseconds. This value is used for timing measurements, in the BenchmarkMS functions.

MEM_GetPerformanceCounter

Call to the WinAPI QueryPerformanceCounter function. 

1
func int MEM_GetPerformanceCounter()
Return value

The function returns a value representing the number of elapsed ticks since the system was started. This value is used for timing measurements, in the BenchmarkPC functions.

Benchmark functions

Tip

For reliable results, avoid measuring a single run of a function; instead, measure the total duration of multiple runs (e.g., 1000). This is crucial, especially for very fast functions, as a single run can distort the measurement. Use _N benchmark functions to include a parameter specifying the number of runs for function f.

Choose the parameter n to ensure meaningful results. If n executions take less than a millisecond, obtaining a return value in milliseconds has no sense. For very fast functions, the time spent in the benchmark function, not in f, significantly affects the measurement, falsifying the result. Reliable measurements are achievable only for functions with sufficient slowness.

For reference, here is a timing for some operations (in nanoseconds, i.e., billionths of a second): 

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
- Function call (jumping back and forth): 30ns
- Elementary calculation (e.g., i = i + 1): 130ns
- Wld_IsTime: 200ns
- MEM_ReadInt, MEM_WriteInt: 350ns
- Hlp_StrCmp("Hello", "Hello"): 500ns
- MEM_InstToPtr: 1400ns
- (small) Allocate and free memory: 9700ns
- CALL__stdcall (in empty function): 29000ns
- MEM_GetParserSymb: 280000ns

- Iteration of the benchmark function: 300ns

MEM_BenchmarkMS

Benchmark of the execution time for a specified function. (Milliseconds) 

1
func int MEM_BenchmarkMS(var func f)
Parameters
  • var func f
    Function to benchmark

Return value

The function returns the duration of a function execution in milliseconds.

MEM_BenchmarkMMS

Benchmark of the execution time for a specified function. (microseconds) 

1
func int MEM_BenchmarkMMS(var func f)
Parameters
  • var func f
    Function to benchmark

Return value

The function returns the duration of a function execution in microseconds.

MEM_BenchmarkPC

Benchmark of the execution time for a specified function, using the Performancecounter. 

1
func int MEM_BenchmarkMS(var func f)
Parameters
  • var func f
    Function to benchmark

Return value

The function returns the number of Performancecounter ticks the function needs.

MEM_BenchmarkMS_N

MEM_BenchmarkMS, but with the parameter to specify the number of function runs. 

1
func int MEM_BenchmarkMS_N(var func f, var int n)
Parameters
  • var func f
    Function to benchmark
  • var int n
    Number of runs

Return value

The function returns a summed duration of multiple (n) runs of the function in milliseconds.

MEM_BenchmarkMMS_N

MEM_BenchmarkMMS, but with the parameter to specify the number of function runs. 

1
func int MEM_BenchmarkMMS_N(var func f, var int n)
Parameters
  • var func f
    Function to benchmark
  • var int n
    Number of runs

Return value

The function returns a summed duration of multiple (n) runs of the function in microseconds.

MEM_BenchmarkPC_N

MEM_BenchmarkPC, but with the parameter to specify the number of function runs. 

1
func int MEM_BenchmarkPC_N(var func f, var int n)
Parameters
  • var func f
    Function to benchmark
  • var int n
    Number of runs

Return value

The function returns a summed number of Performancecounter ticks needed to execute function multiple (n) times.