Re: [PATCH v2 3/4] perf tools: Fix SMT fallback with large core counts

[John Garry]

On 24/11/2021 00:12, Ian Rogers wrote:
> strtoull can only read a 64-bit bitmap. On an AMD EPYC core_cpus may look
> like:
> 00000000,00000000,00000000,00000001,00000000,00000000,00000000,00000001
> and so the sibling wasn't spotted. Fix by writing a simple hweight string
> parser.
>
> Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
> ---
>   tools/perf/util/smt.c | 68 ++++++++++++++++++++++++++++++++++++-------
>   1 file changed, 58 insertions(+), 10 deletions(-)
>
> diff --git a/tools/perf/util/smt.c b/tools/perf/util/smt.c
> index 34f1b1b1176c..2636be65305a 100644
> --- a/tools/perf/util/smt.c
> +++ b/tools/perf/util/smt.c
> @@ -5,6 +5,56 @@
>   #include "api/fs/fs.h"
>   #include "smt.h"
>   
> +/**
> + * hweight_str - Returns the number of bits set in str. Stops at first non-hex
> + *	       or ',' character.

nit: the wording here could be improved, I think. You really mean we 
iter until we reach a character which is neither a hex nor a ',', while 
I read this as iter until either a non-hex or a ','.

> + */
> +static int hweight_str(char *str)
> +{
> +	int result = 0;
> +
> +	while (*str) {

Can we use __buitin_popcount() to try to simplify?

> +		switch (*str++) {
> +		case '0':
> +		case ',':
> +			break;
> +		case '1':
> +		case '2':
> +		case '4':
> +		case '8':
> +			result++;
> +			break;
> +		case '3':
> +		case '5':
> +		case '6':
> +		case '9':
> +		case 'a':
> +		case 'A':
> +		case 'c':
> +		case 'C':
> +			result += 2;
> +			break;
> +		case '7':
> +		case 'b':
> +		case 'B':
> +		case 'd':
> +		case 'D':
> +		case 'e':
> +		case 'E':
> +			result += 3;
> +			break;
> +		case 'f':
> +		case 'F':
> +			result += 4;
> +			break;
> +		default:
> +			goto done;
> +		}
> +	}
> +done: