Re: [Qemu-devel] [PATCH v8 03/11] accel: collecting JIT statistics

["Alex Bennée" <alex.bennee@linaro.org>]

vandersonmr <vandersonmr2@gmail.com> writes:

> If a TB has a TBS (TBStatistics) with the TB_JIT_STATS
> enabled then we collect statistics of its translation
> processes and code translation.
>
> Collecting the number of host instructions seems to be
> not simple as it would imply in having to modify several
> target source files. So, for now, we are only collecting
> the size of the host gen code.
>
> Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
> Signed-off-by: Vanderson M. do Rosario <vandersonmr2@gmail.com>
> ---
>  accel/tcg/translate-all.c | 15 ++++++++++++++-
>  accel/tcg/translator.c    |  4 ++++
>  include/exec/tb-stats.h   | 15 +++++++++++++++
>  tcg/tcg.c                 | 23 +++++++++++++++++++++++
>  tcg/tcg.h                 |  2 ++
>  5 files changed, 58 insertions(+), 1 deletion(-)
>
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index e72aeba682..fb2fe0fa1f 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -1705,6 +1705,7 @@ static TBStatistics *tb_get_stats(tb_page_addr_t phys_pc, target_ulong pc,
>           * then just make the new TB point to the older TBStatistic
>           */
>          g_free(new_stats);
> +        ((TBStatistics *) existing_stats)->tb = current_tb;

This seems out of place and again I can't see what we are doing with
this information yet.

>          return existing_stats;
>      } else {
>          return new_stats;
> @@ -1785,13 +1786,18 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
>       */
>      if (tb_stats_collection_enabled()) {
>          tb->tb_stats = tb_get_stats(phys_pc, pc, cs_base, flags, tb);
> +        uint32_t flag = get_default_tbstats_flag();
>
>          if (qemu_log_in_addr_range(tb->pc)) {
> -            uint32_t flag = get_default_tbstats_flag();
>              if (flag & TB_EXEC_STATS) {
>                  tb->tb_stats->stats_enabled |= TB_EXEC_STATS;
>              }
>          }
> +
> +        if (flag & TB_JIT_STATS) {
> +            tb->tb_stats->stats_enabled |= TB_JIT_STATS;
> +            atomic_inc(&tb->tb_stats->translations.total);
> +        }
>      } else {
>          tb->tb_stats = NULL;
>      }
> @@ -1869,6 +1875,10 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
>      atomic_set(&prof->search_out_len, prof->search_out_len + search_size);
>  #endif
>
> +    if (tb_stats_enabled(tb, TB_JIT_STATS)) {
> +        atomic_add(&tb->tb_stats->code.out_len, gen_code_size);
> +    }
> +
>  #ifdef DEBUG_DISAS
>      if (qemu_loglevel_mask(CPU_LOG_TB_OUT_ASM) &&
>          qemu_log_in_addr_range(tb->pc)) {
> @@ -1926,6 +1936,9 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
>      phys_page2 = -1;
>      if ((pc & TARGET_PAGE_MASK) != virt_page2) {
>          phys_page2 = get_page_addr_code(env, virt_page2);
> +        if (tb_stats_enabled(tb, TB_JIT_STATS)) {
> +            atomic_inc(&tb->tb_stats->translations.spanning);
> +        }
>      }
>      /*
>       * No explicit memory barrier is required -- tb_link_page() makes the
> diff --git a/accel/tcg/translator.c b/accel/tcg/translator.c
> index ec6bd829a0..9b2e248b09 100644
> --- a/accel/tcg/translator.c
> +++ b/accel/tcg/translator.c
> @@ -116,6 +116,10 @@ void translator_loop(const TranslatorOps *ops, DisasContextBase *db,
>      db->tb->size = db->pc_next - db->pc_first;
>      db->tb->icount = db->num_insns;
>
> +    if (tb_stats_enabled(tb, TB_JIT_STATS)) {
> +        atomic_add(&db->tb->tb_stats->code.num_guest_inst, db->num_insns);
> +    }
> +
>  #ifdef DEBUG_DISAS
>      if (qemu_loglevel_mask(CPU_LOG_TB_IN_ASM)
>          && qemu_log_in_addr_range(db->pc_first)) {
> diff --git a/include/exec/tb-stats.h b/include/exec/tb-stats.h
> index c4a8715400..b68edd5d24 100644
> --- a/include/exec/tb-stats.h
> +++ b/include/exec/tb-stats.h
> @@ -58,6 +58,20 @@ struct TBStatistics {
>          unsigned long atomic;
>      } executions;
>
> +    struct {
> +        unsigned num_guest_inst;
> +        unsigned num_tcg_ops;
> +        unsigned num_tcg_ops_opt;
> +        unsigned spills;
> +        unsigned out_len;
> +    } code;
> +
> +    struct {
> +        unsigned long total;
> +        unsigned long uncached;
> +        unsigned long spanning;
> +    } translations;
> +
>      /* current TB linked to this TBStatistics */
>      TranslationBlock *tb;
>  };
> @@ -71,6 +85,7 @@ enum TBStatsStatus { TB_STATS_RUNNING, TB_STATS_PAUSED, TB_STATS_STOPPED };
>
>  #define TB_NOTHING    (1 << 0)
>  #define TB_EXEC_STATS (1 << 1)
> +#define TB_JIT_STATS  (1 << 2)
>
>  extern int tcg_collect_tb_stats;
>  extern uint32_t default_tbstats_flag;
> diff --git a/tcg/tcg.c b/tcg/tcg.c
> index 0458eaec57..ae3e7a2217 100644
> --- a/tcg/tcg.c
> +++ b/tcg/tcg.c
> @@ -3125,6 +3125,11 @@ static void temp_sync(TCGContext *s, TCGTemp *ts, TCGRegSet allocated_regs,
>          case TEMP_VAL_REG:
>              tcg_out_st(s, ts->type, ts->reg,
>                         ts->mem_base->reg, ts->mem_offset);
> +
> +            /* Count number of spills */
> +            if (tb_stats_enabled(s->current_tb, TB_JIT_STATS)) {
> +                atomic_inc(&s->current_tb->tb_stats->code.spills);
> +            }
>              break;
>
>          case TEMP_VAL_MEM:
> @@ -3996,6 +4001,8 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
>      int i, num_insns;
>      TCGOp *op;
>
> +    s->current_tb = tb;
> +
>  #ifdef CONFIG_PROFILER
>      {
>          int n = 0;
> @@ -4027,6 +4034,14 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
>      }
>  #endif
>
> +    if (tb_stats_enabled(tb, TB_JIT_STATS)) {
> +        int n = 0;
> +        QTAILQ_FOREACH(op, &s->ops, link) {
> +            n++;
> +        }
> +        atomic_add(&tb->tb_stats->code.num_tcg_ops, n);
> +    }
> +
>  #ifdef CONFIG_DEBUG_TCG
>      /* Ensure all labels referenced have been emitted.  */
>      {
> @@ -4093,6 +4108,14 @@ int tcg_gen_code(TCGContext *s, TranslationBlock *tb)
>      }
>  #endif
>
> +    if (tb_stats_enabled(tb, TB_JIT_STATS)) {
> +        int n = 0;
> +        QTAILQ_FOREACH(op, &s->ops, link) {
> +            n++;
> +        }
> +        atomic_add(&tb->tb_stats->code.num_tcg_ops_opt, n);
> +    }
> +
>      tcg_reg_alloc_start(s);
>
>      s->code_buf = tb->tc.ptr;
> diff --git a/tcg/tcg.h b/tcg/tcg.h
> index 529acb2ed8..b4601162f8 100644
> --- a/tcg/tcg.h
> +++ b/tcg/tcg.h
> @@ -740,6 +740,8 @@ struct TCGContext {
>
>      uint16_t gen_insn_end_off[TCG_MAX_INSNS];
>      target_ulong gen_insn_data[TCG_MAX_INSNS][TARGET_INSN_START_WORDS];
> +
> +    TranslationBlock *current_tb;

As we are only using this to get to tb_stats why not skip the middle man
and go directly to:

       TBStatistics *current_stats;

?

You already have a tbs_stats_enabled() helper 

>  };
>
>  extern TCGContext tcg_init_ctx;


--
Alex Bennée