From: "Alex Bennée" <alex.bennee@linaro.org>
To: qemu-devel@nongnu.org
Cc: Paolo Bonzini <pbonzini@redhat.com>,
vandersonmr <vandersonmr2@gmail.com>,
Richard Henderson <rth@twiddle.net>
Subject: Re: [Qemu-devel] [PATCH v5 04/10] accel: replacing part of CONFIG_PROFILER with TBStats
Date: Thu, 15 Aug 2019 15:54:05 +0100 [thread overview]
Message-ID: <87o90qmqcy.fsf@linaro.org> (raw)
In-Reply-To: <20190815021857.19526-5-vandersonmr2@gmail.com>
vandersonmr <vandersonmr2@gmail.com> writes:
> We add some of the statistics collected in the TCGProfiler
> into the TBStats, having the statistics not only for the whole
> emulation but for each TB. Then, we removed these stats
> from TCGProfiler and reconstruct the information for the
> "info jit" using the sum of all TBStats statistics.
>
> The goal is to have one unique and better way of collecting
> emulation statistics. Moreover, checking dynamiclly if the
> profiling is enabled showed to have an insignificant impact
> on the performance:
> https://wiki.qemu.org/Internships/ProjectIdeas/TCGCodeQuality#Overheads.
>
> Signed-off-by: Vanderson M. do Rosario <vandersonmr2@gmail.com>
> ---
> accel/tcg/tb-stats.c | 95 +++++++++++++++++++++++++++++++++++++++
> accel/tcg/translate-all.c | 8 +---
> include/exec/tb-stats.h | 11 +++++
> tcg/tcg.c | 93 +++++---------------------------------
> tcg/tcg.h | 10 -----
> 5 files changed, 118 insertions(+), 99 deletions(-)
>
> diff --git a/accel/tcg/tb-stats.c b/accel/tcg/tb-stats.c
> index 3489133e9e..9b720d9b86 100644
> --- a/accel/tcg/tb-stats.c
> +++ b/accel/tcg/tb-stats.c
> @@ -1,9 +1,104 @@
> #include "qemu/osdep.h"
>
> #include "disas/disas.h"
> +#include "exec/exec-all.h"
> +#include "tcg.h"
> +
> +#include "qemu/qemu-print.h"
>
> #include "exec/tb-stats.h"
>
> +struct jit_profile_info {
> + uint64_t translations;
> + uint64_t aborted;
> + uint64_t ops;
> + unsigned ops_max;
> + uint64_t del_ops;
> + uint64_t temps;
> + unsigned temps_max;
> + uint64_t host;
> + uint64_t guest;
> + uint64_t search_data;
> +};
> +
> +/* accumulate the statistics from all TBs */
> +static void collect_jit_profile_info(void *p, uint32_t hash, void *userp)
> +{
> + struct jit_profile_info *jpi = userp;
> + TBStatistics *tbs = p;
> +
> + jpi->translations += tbs->translations.total;
> + jpi->ops += tbs->code.num_tcg_ops;
> + if (stat_per_translation(tbs, code.num_tcg_ops) > jpi->ops_max) {
> + jpi->ops_max = stat_per_translation(tbs, code.num_tcg_ops);
> + }
> + jpi->del_ops += tbs->code.deleted_ops;
> + jpi->temps += tbs->code.temps;
> + if (stat_per_translation(tbs, code.temps) > jpi->temps_max) {
> + jpi->temps_max = stat_per_translation(tbs, code.temps);
> + }
> + jpi->host += tbs->code.out_len;
> + jpi->guest += tbs->code.in_len;
> + jpi->search_data += tbs->code.search_out_len;
> +}
> +
> +/* dump JIT statisticis using TCGProfile and TBStats */
> +void dump_jit_profile_info(TCGProfile *s)
> +{
> + if (!tb_stats_collection_enabled()) {
> + return;
> + }
> +
> + struct jit_profile_info *jpi = g_new0(struct jit_profile_info, 1);
> +
> + qht_iter(&tb_ctx.tb_stats, collect_jit_profile_info, jpi);
> +
> + if (jpi->translations) {
> + qemu_printf("translated TBs %" PRId64 "\n", jpi->translations);
> + qemu_printf("avg ops/TB %0.1f max=%d\n",
> + jpi->ops / (double) jpi->translations, jpi->ops_max);
> + qemu_printf("deleted ops/TB %0.2f\n",
> + jpi->del_ops / (double) jpi->translations);
> + qemu_printf("avg temps/TB %0.2f max=%d\n",
> + jpi->temps / (double) jpi->translations, jpi->temps_max);
> + qemu_printf("avg host code/TB %0.1f\n",
> + jpi->host / (double) jpi->translations);
> + qemu_printf("avg search data/TB %0.1f\n",
> + jpi->search_data / (double) jpi->translations);
> +
> + if (s) {
> + int64_t tot = s->interm_time + s->code_time;
> + qemu_printf("JIT cycles %" PRId64 " (%0.3f s at 2.4 GHz)\n",
> + tot, tot / 2.4e9);
> + qemu_printf("cycles/op %0.1f\n",
> + jpi->ops ? (double)tot / jpi->ops : 0);
> + qemu_printf("cycles/in byte %0.1f\n",
> + jpi->guest ? (double)tot / jpi->guest : 0);
> + qemu_printf("cycles/out byte %0.1f\n",
> + jpi->host ? (double)tot / jpi->host : 0);
> + qemu_printf("cycles/search byte %0.1f\n",
> + jpi->search_data ? (double)tot / jpi->search_data : 0);
> + if (tot == 0) {
> + tot = 1;
> + }
> + qemu_printf(" gen_interm time %0.1f%%\n",
> + (double)s->interm_time / tot * 100.0);
> + qemu_printf(" gen_code time %0.1f%%\n",
> + (double)s->code_time / tot * 100.0);
> + qemu_printf("optim./code time %0.1f%%\n",
> + (double)s->opt_time / (s->code_time ? s->code_time : 1)
> + * 100.0);
> + qemu_printf("liveness/code time %0.1f%%\n",
> + (double)s->la_time / (s->code_time ? s->code_time : 1) * 100.0);
> + qemu_printf("cpu_restore count %" PRId64 "\n",
> + s->restore_count);
> + qemu_printf(" avg cycles %0.1f\n",
> + s->restore_count ? (double)s->restore_time / s->restore_count : 0);
> + }
> + }
I think the g_free(jpi) should be moved from the later patches to here.
Otherwise:
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
--
Alex Bennée
next prev parent reply other threads:[~2019-08-15 14:55 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-08-15 2:18 [Qemu-devel] [PATCH v5 00/10] Measure Tiny Code Generation Quality vandersonmr
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 01/10] accel: introducing TBStatistics structure vandersonmr
2019-08-15 13:13 ` Alex Bennée
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 02/10] accel: collecting TB execution count vandersonmr
2019-08-15 13:38 ` Alex Bennée
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 03/10] accel: collecting JIT statistics vandersonmr
2019-08-15 14:29 ` Alex Bennée
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 04/10] accel: replacing part of CONFIG_PROFILER with TBStats vandersonmr
2019-08-15 14:54 ` Alex Bennée [this message]
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 05/10] accel: adding TB_JIT_TIME and full replacing CONFIG_PROFILER vandersonmr
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 06/10] log: adding -d tb_stats to control tbstats vandersonmr
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 07/10] monitor: adding tb_stats hmp command vandersonmr
2019-08-15 8:53 ` Dr. David Alan Gilbert
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 08/10] Adding info [tbs|tb|coverset] commands to HMP. These commands allow the exploration of TBs generated by the TCG. Understand which one hotter, with more guest/host instructions... and examine their guest, host and IR code vandersonmr
2019-08-15 8:59 ` Dr. David Alan Gilbert
2019-08-21 14:16 ` Vanderson Martins do Rosario
2019-08-21 14:29 ` Dr. David Alan Gilbert
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 09/10] monitor: adding new info cfg command vandersonmr
2019-08-15 9:14 ` Dr. David Alan Gilbert
2019-08-15 2:18 ` [Qemu-devel] [PATCH v5 10/10] linux-user: dumping hot TBs at the end of the execution vandersonmr
2019-08-15 14:26 ` Aleksandar Markovic
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=87o90qmqcy.fsf@linaro.org \
--to=alex.bennee@linaro.org \
--cc=pbonzini@redhat.com \
--cc=qemu-devel@nongnu.org \
--cc=rth@twiddle.net \
--cc=vandersonmr2@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).