Re: [PATCH v5 9/9] perf, tools, script: Support total cycles count

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Tue, Sep 18, 2018 at 05:32:14AM -0700, Andi Kleen escreveu:
> From: Andi Kleen <ak@linux.intel.com>
> 
> For perf script brstackinsn also print a running cycles count.
> This makes it easier to calculate cycle deltas for code sections
> measured with LBRs.
> 
> % perf record -b -a sleep 1
> % perf script -F +brstackinsn
> ...
>         _dl_sysdep_start+330:
>         00007eff9f20583a        insn: 75 c4                     # PRED 24 cycles [24]
>         00007eff9f205800        insn: 48 83 e8 03
>         00007eff9f205804        insn: 48 83 f8 1e
>         00007eff9f205808        insn: 77 26
>         00007eff9f20580a        insn: 48 63 04 81
>         00007eff9f20580e        insn: 48 01 c8
>         00007eff9f205811        insn: ff e0                     # MISPRED 31 cycles [7] 0.71 IPC
>         00007eff9f2059c0        insn: 44 8b 62 08
>         00007eff9f2059c4        insn: e9 67 fe ff ff            # PRED 55 cycles [24] 0.04 IPC
>         00007eff9f205830        insn: 48 83 c2 10
>         00007eff9f205834        insn: 48 8b 02
>         00007eff9f205837        insn: 48 85 c0
>         00007eff9f20583a        insn: 75 c4                     # PRED 68 cycles [13] 0.23 IPC
> 
> Signed-off-by: Andi Kleen <ak@linux.intel.com>
> ---
>  tools/perf/builtin-script.c | 12 ++++++++----
>  1 file changed, 8 insertions(+), 4 deletions(-)
> 
> diff --git a/tools/perf/builtin-script.c b/tools/perf/builtin-script.c
> index 3b73c3226141..84895c861b84 100644
> --- a/tools/perf/builtin-script.c
> +++ b/tools/perf/builtin-script.c
> @@ -913,7 +913,7 @@ static int grab_bb(u8 *buffer, u64 start, u64 end,
>  
>  static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
>  			    struct perf_insn *x, u8 *inbuf, int len,
> -			    int insn, FILE *fp)
> +			    int insn, FILE *fp, int *total_cycles)
>  {
>  	int printed = fprintf(fp, "\t%016" PRIx64 "\t%-30s\t#%s%s%s%s", ip,
>  			      dump_insn(x, ip, inbuf, len, NULL),
> @@ -922,7 +922,8 @@ static int ip__fprintf_jump(uint64_t ip, struct branch_entry *en,
>  			      en->flags.in_tx ? " INTX" : "",
>  			      en->flags.abort ? " ABORT" : "");
>  	if (en->flags.cycles) {
> -		printed += fprintf(fp, " %d cycles", en->flags.cycles);
> +		*total_cycles += en->flags.cycles;
> +		printed += fprintf(fp, " %d cycles [%d]", *total_cycles, en->flags.cycles);
>  		if (insn)
>  			printed += fprintf(fp, " %.2f IPC", (float)insn / en->flags.cycles);
>  	}
> @@ -979,6 +980,7 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
>  	u8 buffer[MAXBB];
>  	unsigned off;
>  	struct symbol *lastsym = NULL;
> +	int total_cycles = 0;
>  
>  	if (!(br && br->nr))
>  		return 0;
> @@ -999,7 +1001,7 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
>  		printed += ip__fprintf_sym(br->entries[nr - 1].from, thread,
>  					   x.cpumode, x.cpu, &lastsym, attr, fp);
>  		printed += ip__fprintf_jump(br->entries[nr - 1].from, &br->entries[nr - 1],
> -					    &x, buffer, len, 0, fp);
> +					    &x, buffer, len, 0, fp, &total_cycles);
>  	}
>  
>  	/* Print all blocks */
> @@ -1027,7 +1029,9 @@ static int perf_sample__fprintf_brstackinsn(struct perf_sample *sample,
>  
>  			printed += ip__fprintf_sym(ip, thread, x.cpumode, x.cpu, &lastsym, attr, fp);
>  			if (ip == end) {
> -				printed += ip__fprintf_jump(ip, &br->entries[i], &x, buffer + off, len - off, insn, fp);
> +				printed += ip__fprintf_jump(ip, &br->entries[i],
> +						&x, buffer + off, len - off,
> +						insn, fp, &total_cycles);

Please don't reflow these things like that, just align the next line
with the new thing, parses quicker when reviewing.

- Arnaldo



>  				break;
>  			} else {
>  				printed += fprintf(fp, "\t%016" PRIx64 "\t%s\n", ip,
> -- 
> 2.17.1