LTTng CLUST: A System-Wide Unified CPU and GPU Tracing Tool for OpenCL Applications

Table 3

Asynchronous OpenCL API function overhead benchmark.

Buffer; Size; (Byte)	Base; ave.; (ns/call)	Base; Std. dev.; (ns/call)	Preload; ave.; (ns/call)	Preload; Std. dev.; (ns/call)	Trace; ave.; (ns/call)	Trace; Std. dev.; (ns/call)	Preload; overhead; (ns/call)	Trace; overhead; (ns/call)
4	149.51	1.47	164.7	1.1	7000.6	261.8	15.2	6851.1
40	158.99	0.92	168.7	1.3	7026.8	289.5	9.7	6867.8
400	156.15	1.50	174.7	1.3	7269.3	240.6	18.5	7113.2
	188.44	1.14	226.7	1.3	7043.6	244.2	38.3	6855.2
	1499.76	5.47	1503.3	5.6	8393.0	227.2	3.6	6893.3
	17805.67	134.31	17862.1	16.1	25404.7	276.3	56.4	7599.0

Sample size = 100; loop size = 1000.