diff mbox

tcg-runtime: increase hit rate of lookup_tb_ptr

Message ID 20170614202733.GA8420@flamenco
State New
Headers show

Commit Message

Emilio Cota June 14, 2017, 8:27 p.m. UTC 
On Wed, Jun 14, 2017 at 12:48:17 -0700, Richard Henderson wrote:
> We can call tb_htable_lookup even when the tb_jmp_cache
> is completely empty.  Therefore, un-nest most of the code
> dependent on tb != NULL from the read from the cache.
> 
> Signed-off-by: Richard Henderson <rth@twiddle.net>

I just wrote this alternative patch, which does the same thing
as yours. I also measured what the effect of this change
has on the hit rate of lookup_tb_ptr. Feel free to reuse parts
of the patch and/or the commit message!

Thanks,

		E.

--- 8< ---

Strangely, we do not look up the tb in the global hash table
when we get NULL from tb_jmp_cache.

Fix it, which improves the hit rate of lookup_tb_ptr; for instance,
when booting and immediately shutting down debian-arm, the hit
rate improves from
	93.150742% (before this patch)
to
	99.451323 % (after).

While at it, use a variable for the tb_jmp_cache hash and get rid
of the goto's.

Suggested-by: Richard Henderson <rth@twiddle.net>
Suggested-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Emilio G. Cota <cota@braap.org>
---
 tcg-runtime.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)

Comments

Richard Henderson June 15, 2017, 1:21 a.m. UTC | #1 
On 06/14/2017 01:27 PM, Emilio G. Cota wrote:
> On Wed, Jun 14, 2017 at 12:48:17 -0700, Richard Henderson wrote:
>> We can call tb_htable_lookup even when the tb_jmp_cache
>> is completely empty.  Therefore, un-nest most of the code
>> dependent on tb != NULL from the read from the cache.
>>
>> Signed-off-by: Richard Henderson <rth@twiddle.net>
> 
> I just wrote this alternative patch, which does the same thing
> as yours. I also measured what the effect of this change
> has on the hit rate of lookup_tb_ptr. Feel free to reuse parts
> of the patch and/or the commit message!

Thanks.  I'll adjust the commit.


r~
diff mbox

Patch

diff --git a/tcg-runtime.c b/tcg-runtime.c
index 7fa90ce..09324b9 100644
--- a/tcg-runtime.c
+++ b/tcg-runtime.c
@@ -149,23 +149,19 @@  void *HELPER(lookup_tb_ptr)(CPUArchState *env, target_ulong addr)
     CPUState *cpu = ENV_GET_CPU(env);
     TranslationBlock *tb;
     target_ulong cs_base, pc;
+    unsigned int hash = tb_jmp_cache_hash_func(addr);
     uint32_t flags;
 
-    tb = atomic_rcu_read(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(addr)]);
-    if (likely(tb)) {
-        cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
-        if (likely(tb->pc == addr && tb->cs_base == cs_base &&
-                   tb->flags == flags)) {
-            goto found;
-        }
+    tb = atomic_rcu_read(&cpu->tb_jmp_cache[hash]);
+    cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags);
+    if (unlikely(tb == NULL || tb->pc != addr || tb->cs_base != cs_base ||
+               tb->flags != flags)) {
         tb = tb_htable_lookup(cpu, addr, cs_base, flags);
-        if (likely(tb)) {
-            atomic_set(&cpu->tb_jmp_cache[tb_jmp_cache_hash_func(addr)], tb);
-            goto found;
+        if (unlikely(tb == NULL)) {
+            return tcg_ctx.code_gen_epilogue;
         }
+        atomic_set(&cpu->tb_jmp_cache[hash], tb);
     }
-    return tcg_ctx.code_gen_epilogue;
- found:
     qemu_log_mask_and_addr(CPU_LOG_EXEC, addr,
                            "Chain %p [%d: " TARGET_FMT_lx "] %s\n",
                            tb->tc_ptr, cpu->cpu_index, addr,