[Qemu-devel] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Qemu-devel] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Richard Henderson]

Cc: qemu-ppc@nongnu.org
Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 target/ppc/translate.c | 23 ++++++++---------------
 1 file changed, 8 insertions(+), 15 deletions(-)

diff --git a/target/ppc/translate.c b/target/ppc/translate.c
index c0cd64d..9aa66f5 100644
--- a/target/ppc/translate.c
+++ b/target/ppc/translate.c
@@ -3419,7 +3419,7 @@ static inline bool use_goto_tb(DisasContext *ctx, target_ulong dest)
 }
 
 /***                                Branch                                 ***/
-static inline void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
+static void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
 {
     if (NARROW_MODE(ctx)) {
         dest = (uint32_t) dest;
@@ -3441,7 +3441,7 @@ static inline void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
                 gen_debug_exception(ctx);
             }
         }
-        tcg_gen_exit_tb(0);
+        tcg_gen_lookup_and_goto_ptr(cpu_nip);
     }
 }
 
@@ -3479,7 +3479,7 @@ static void gen_b(DisasContext *ctx)
 #define BCOND_CTR 2
 #define BCOND_TAR 3
 
-static inline void gen_bcond(DisasContext *ctx, int type)
+static void gen_bcond(DisasContext *ctx, int type)
 {
     uint32_t bo = BO(ctx->opcode);
     TCGLabel *l1;
@@ -3543,26 +3543,19 @@ static inline void gen_bcond(DisasContext *ctx, int type)
         } else {
             gen_goto_tb(ctx, 0, li);
         }
-        if ((bo & 0x14) != 0x14) {
-            gen_set_label(l1);
-            gen_goto_tb(ctx, 1, ctx->nip);
-        }
     } else {
         if (NARROW_MODE(ctx)) {
             tcg_gen_andi_tl(cpu_nip, target, (uint32_t)~3);
         } else {
             tcg_gen_andi_tl(cpu_nip, target, ~3);
         }
-        tcg_gen_exit_tb(0);
-        if ((bo & 0x14) != 0x14) {
-            gen_set_label(l1);
-            gen_update_nip(ctx, ctx->nip);
-            tcg_gen_exit_tb(0);
-        }
-    }
-    if (type == BCOND_LR || type == BCOND_CTR || type == BCOND_TAR) {
+        tcg_gen_lookup_and_goto_ptr(cpu_nip);
         tcg_temp_free(target);
     }
+    if ((bo & 0x14) != 0x14) {
+        gen_set_label(l1);
+        gen_goto_tb(ctx, 1, ctx->nip);
+    }
 }
 
 static void gen_bc(DisasContext *ctx)
-- 
2.9.4

Re: [Qemu-devel] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Emilio G. Cota]

On Fri, Jun 30, 2017 at 11:37:36 -0700, Richard Henderson wrote:
> Cc: qemu-ppc@nongnu.org
> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
>  target/ppc/translate.c | 23 ++++++++---------------
>  1 file changed, 8 insertions(+), 15 deletions(-)

Tested-by: Emilio G. Cota <cota@braap.org>

Just measured the speedup for linux-user:

                                     SPECint06 (test set), ppc64le-linux-user. Host: IBM POWER8 @ 3.42 GHz                                  
                                                                                                                                            
  1.6 +-+------+---------+--------+--------+--------+---------+--------+--------+---------+--------+--------+--------+---------+------+-+   
      |                                                                                   |                                             |   
      |                                                                                   |                               after         |   
  1.5 +-+.................................................................................|...........................................+-+   
      |                                                                                   |                                             |   
      |                                                                                ******                                           |   
  1.4 +-+.....+++.....................................................+++..............*..|.*.....+++.................................+-+   
      |        |                                   +++                 |               *  | *      |                                    |   
      |        |                                    |                  |               *  | *    *****                                  |   
  1.3 +-+......|....................................|..................|...............*..|.*....*.|.*................................+-+   
      |        |                 +++              ******             *****             * +++*    * | *             ******               |   
      |      ******               |               * |  *             * | *             *    *    *+++*             *+++ *               |   
  1.2 +-+....*.|..*...............|...............*.|..*.....+++.....*.|.*.............*....*....*...*.............*....*.............+-+   
      |      * |  *               |               * |  *      |      *+++*             *    *    *   *             *    *   ******      |   
      |      * |  *             *****             *+++ *   ******    *   *             *    *    *   *             *    *   * +++*      |   
  1.1 +-+....*+++.*.............*.|.*.............*....*...*.+++*....*...*.............*....*....*...*.....+++.....*....*...*....*....+-+   
      |      *    *             * | *             *    *   *    *    *   *             *    *    *   *      |      *    *   *    *      |   
      |      *    *             * | *             *    *   *    *    *   *    ******   *    *    *   *      |      *    *   *    *      |   
    1 +-+....*....*.....+++.....*+++*.....+++.....*....*...*....*....*...*....*....*...*....*....*...*....*****....*....*...*....*....+-+   
      |      *    *      |      *   *      |      *    *   *    *    *   *    *    *   *    *    *   *    * | *    *    *   *    *      |   
      |      *    *      |      *   *    *****    *    *   *    *    *   *    *    *   *    *    *   *    * | *    *    *   *    *      |   
  0.9 +-+....*....*...******....*...*....*.|.*....*....*...*....*....*...*....*....*...*....*....*...*....*.|.*....*....*...*....*....+-+   
      |      *    *   *  | *    *   *    * | *    *    *   *    *    *   *    *    *   *    *    *   *    *+++*    *    *   *    *      |   
      |      *    *   * +++*    *   *    *+++*    *    *   *    *    *   *    *    *   *    *    *   *    *   *    *    *   *    *      |   
  0.8 +-+----******---******----*****----*****----******---******----*****----******---******----*****----*****----******---******----+-+   
  400.perlbench 401.bzip2  403.gcc  429.mcf445.gobmk 456.hmmer458.462.libquantu464.h264r471.omnetpp473.a483.xalancbmk     gmean             
png: http://imgur.com/a/twj6V

Thanks,

		Emilio

PS. If you have spec06, you can very easily generate these plots. See
  https://github.com/cota/runspec-simple/commit/14211b927

Re: [Qemu-devel] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Thomas Huth]

On 30.06.2017 20:37, Richard Henderson wrote:
> Cc: qemu-ppc@nongnu.org
> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
>  target/ppc/translate.c | 23 ++++++++---------------
>  1 file changed, 8 insertions(+), 15 deletions(-)
> 
> diff --git a/target/ppc/translate.c b/target/ppc/translate.c
> index c0cd64d..9aa66f5 100644
> --- a/target/ppc/translate.c
> +++ b/target/ppc/translate.c
> @@ -3419,7 +3419,7 @@ static inline bool use_goto_tb(DisasContext *ctx, target_ulong dest)
>  }
>  
>  /***                                Branch                                 ***/
> -static inline void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
> +static void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
>  {
>      if (NARROW_MODE(ctx)) {
>          dest = (uint32_t) dest;
> @@ -3441,7 +3441,7 @@ static inline void gen_goto_tb(DisasContext *ctx, int n, target_ulong dest)
>                  gen_debug_exception(ctx);
>              }
>          }
> -        tcg_gen_exit_tb(0);
> +        tcg_gen_lookup_and_goto_ptr(cpu_nip);
>      }
>  }
>  
> @@ -3479,7 +3479,7 @@ static void gen_b(DisasContext *ctx)
>  #define BCOND_CTR 2
>  #define BCOND_TAR 3
>  
> -static inline void gen_bcond(DisasContext *ctx, int type)
> +static void gen_bcond(DisasContext *ctx, int type)
>  {
>      uint32_t bo = BO(ctx->opcode);
>      TCGLabel *l1;
> @@ -3543,26 +3543,19 @@ static inline void gen_bcond(DisasContext *ctx, int type)
>          } else {
>              gen_goto_tb(ctx, 0, li);
>          }
> -        if ((bo & 0x14) != 0x14) {
> -            gen_set_label(l1);
> -            gen_goto_tb(ctx, 1, ctx->nip);
> -        }
>      } else {
>          if (NARROW_MODE(ctx)) {
>              tcg_gen_andi_tl(cpu_nip, target, (uint32_t)~3);
>          } else {
>              tcg_gen_andi_tl(cpu_nip, target, ~3);
>          }
> -        tcg_gen_exit_tb(0);
> -        if ((bo & 0x14) != 0x14) {
> -            gen_set_label(l1);
> -            gen_update_nip(ctx, ctx->nip);
> -            tcg_gen_exit_tb(0);
> -        }
> -    }
> -    if (type == BCOND_LR || type == BCOND_CTR || type == BCOND_TAR) {
> +        tcg_gen_lookup_and_goto_ptr(cpu_nip);
>          tcg_temp_free(target);
>      }
> +    if ((bo & 0x14) != 0x14) {
> +        gen_set_label(l1);
> +        gen_goto_tb(ctx, 1, ctx->nip);
> +    }
>  }
>  
>  static void gen_bc(DisasContext *ctx)
> 

Using this patch increases the speed of the SLOF firmware boot quite a
bit! The following simple test finishes at least 4 seconds faster when
I've got this patch applied:

time ppc64-softmmu/qemu-system-ppc64 -nographic -vga none \
 -prom-env 'boot-command=power-off'

I also can still run my big endian and my little endian pseries guest
images successfully, so feel free to add my:

Tested-by: Thomas Huth <thuth@redhat.com>

Re: [Qemu-devel] [Qemu-ppc] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Thomas Huth]

On 30.06.2017 20:37, Richard Henderson wrote:
> Cc: qemu-ppc@nongnu.org
> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
>  target/ppc/translate.c | 23 ++++++++---------------
>  1 file changed, 8 insertions(+), 15 deletions(-)

*ping*

Richard, David, any chance that we can still get this in for 2.11? I
think this would be quite a nice performance improvement for PPC TCG...

 Thomas

Re: [Qemu-devel] [Qemu-ppc] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[Philippe Mathieu-Daudé]

>> Signed-off-by: Richard Henderson <rth@twiddle.net>
>> ---
>>  target/ppc/translate.c | 23 ++++++++---------------
>>  1 file changed, 8 insertions(+), 15 deletions(-)
> 
> *ping*
> 
> Richard, David, any chance that we can still get this in for 2.11? I
> think this would be quite a nice performance improvement for PPC TCG...

This still miss a Reviewed-by tag ;)

Re: [Qemu-devel] [Qemu-ppc] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 728 bytes --]

On Thu, Oct 26, 2017 at 02:22:57PM +0200, Thomas Huth wrote:
> On 30.06.2017 20:37, Richard Henderson wrote:
> > Cc: qemu-ppc@nongnu.org
> > Signed-off-by: Richard Henderson <rth@twiddle.net>
> > ---
> >  target/ppc/translate.c | 23 ++++++++---------------
> >  1 file changed, 8 insertions(+), 15 deletions(-)
> 
> *ping*
> 
> Richard, David, any chance that we can still get this in for 2.11? I
> think this would be quite a nice performance improvement for PPC TCG...

Uh.. sorry, I don't even remember this patch.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Qemu-devel] [Qemu-ppc] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 1076 bytes --]

On Thu, Oct 26, 2017 at 11:12:00PM +0200, David Gibson wrote:
> On Thu, Oct 26, 2017 at 02:22:57PM +0200, Thomas Huth wrote:
> > On 30.06.2017 20:37, Richard Henderson wrote:
> > > Cc: qemu-ppc@nongnu.org
> > > Signed-off-by: Richard Henderson <rth@twiddle.net>
> > > ---
> > >  target/ppc/translate.c | 23 ++++++++---------------
> > >  1 file changed, 8 insertions(+), 15 deletions(-)
> > 
> > *ping*
> > 
> > Richard, David, any chance that we can still get this in for 2.11? I
> > think this would be quite a nice performance improvement for PPC TCG...
> 
> Uh.. sorry, I don't even remember this patch.

Managed to dig it out of my archives.  Looks like the problem was it
was just sent to the lists, not CCed to me.  I'm not generally able to
keep up with the lists, and I must have missed it on my periodic scans
through the list.  I've applied it now.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [Qemu-devel] [Qemu-ppc] [PATCH] target/ppc: Use tcg_gen_lookup_and_goto_ptr

[David Gibson]

[-- Attachment #1: Type: text/plain, Size: 1285 bytes --]

On Fri, Oct 27, 2017 at 03:44:22PM +0200, David Gibson wrote:
> On Thu, Oct 26, 2017 at 11:12:00PM +0200, David Gibson wrote:
> > On Thu, Oct 26, 2017 at 02:22:57PM +0200, Thomas Huth wrote:
> > > On 30.06.2017 20:37, Richard Henderson wrote:
> > > > Cc: qemu-ppc@nongnu.org
> > > > Signed-off-by: Richard Henderson <rth@twiddle.net>
> > > > ---
> > > >  target/ppc/translate.c | 23 ++++++++---------------
> > > >  1 file changed, 8 insertions(+), 15 deletions(-)
> > > 
> > > *ping*
> > > 
> > > Richard, David, any chance that we can still get this in for 2.11? I
> > > think this would be quite a nice performance improvement for PPC TCG...
> > 
> > Uh.. sorry, I don't even remember this patch.
> 
> Managed to dig it out of my archives.  Looks like the problem was it
> was just sent to the lists, not CCed to me.  I'm not generally able to
> keep up with the lists, and I must have missed it on my periodic scans
> through the list.  I've applied it now.

.. and now I've unapplied it, because it doesn't compile.  Richard,
can you resend a current version?

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]