73 lines · plain
1 i synthesize instructions events2 y synthesize cycles events3 b synthesize branches events (branch misses for Arm SPE)4 c synthesize branches events (calls only)5 r synthesize branches events (returns only)6 x synthesize transactions events7 w synthesize ptwrite events8 p synthesize power events (incl. PSB events for Intel PT)9 o synthesize other events recorded due to the use10 of aux-output (refer to perf record)11 I synthesize interrupt or similar (asynchronous) events12 (e.g. Intel PT Event Trace)13 e synthesize error events14 d create a debug log15 f synthesize first level cache events16 m synthesize last level cache events17 M synthesize memory events18 t synthesize TLB events19 a synthesize remote access events20 g synthesize a call chain (use with i or x)21 G synthesize a call chain on existing event records22 l synthesize last branch entries (use with i or x)23 L synthesize last branch entries on existing event records24 s skip initial number of events25 q quicker (less detailed) decoding26 A approximate IPC27 Z prefer to ignore timestamps (so-called "timeless" decoding)28 T use the timestamp trace as kernel time29 30 The default is all events i.e. the same as --itrace=iybxwpe,31 except for perf script where it is --itrace=ce32 33 In addition, the period (default 100000, except for perf script where it is 1)34 for instructions events can be specified in units of:35 36 i instructions37 t ticks38 ms milliseconds39 us microseconds40 ns nanoseconds (default)41 42 Also the call chain size (default 16, max. 1024) for instructions or43 transactions events can be specified.44 45 Also the number of last branch entries (default 64, max. 1024) for46 instructions or transactions events can be specified.47 48 Similar to options g and l, size may also be specified for options G and L.49 On x86, note that G and L work poorly when data has been recorded with50 large PEBS. Refer linkperf:perf-intel-pt[1] man page for details.51 52 It is also possible to skip events generated (instructions, branches, transactions,53 ptwrite, power) at the beginning. This is useful to ignore initialization code.54 55 --itrace=i0nss100000056 57 skips the first million instructions.58 59 The 'e' option may be followed by flags which affect what errors will or60 will not be reported. Each flag must be preceded by either '+' or '-'.61 The flags are:62 o overflow63 l trace data lost64 65 If supported, the 'd' option may be followed by flags which affect what66 debug messages will or will not be logged. Each flag must be preceded67 by either '+' or '-'. The flags are:68 a all perf events69 e output only on errors (size configurable - see linkperf:perf-config[1])70 o output to stdout71 72 If supported, the 'q' option may be repeated to increase the effect.73