Re: [Qemu-devel] [PATCH 01/17] tcg: extend max tcg opcodes on 32bit

[Aurelien Jarno]

On Fri, Apr 15, 2011 at 05:32:42PM +0200, Alexander Graf wrote:
> When running on a 32 bit host, we tend to use more TCG ops than on
> a 64 bit host. Reflect that in the reserved opcode amount constant.
> 
> Signed-off-by: Alexander Graf <address@hidden>
> ---
>  exec-all.h |    4 ++++
>  1 files changed, 4 insertions(+), 0 deletions(-)

I really like this solution as it is actually better than incrementing
this value for both 32- and 64-bit hosts. That said, I think it should
only be done for 64-bit guests on 32-bit hosts. 32-bit guests on 32-bit
hosts should not use more ops than 32-bit guests on 64-bit hosts.

I think we should do the same for TCG_MAX_OP_SIZE, as it tends the be 
the same issue, ie emulating 64-bit load/store on a 32-bit hosts use a
lot of host instructions. That said in that case we should probably
rollback to the previous value on 64-bit hosts. You can do it in the
next version of the patch, but it's nothing mandatory for s390x.

> diff --git a/exec-all.h b/exec-all.h
> index 496c001..d23a36d 100644
> --- a/exec-all.h
> +++ b/exec-all.h
> @@ -43,7 +43,11 @@ typedef ram_addr_t tb_page_addr_t;
>  typedef struct TranslationBlock TranslationBlock;
>  
>  /* XXX: make safe guess about sizes */
> +#if HOST_LONG_BITS == 32
> +#define MAX_OP_PER_INSTR 128
> +#else
>  #define MAX_OP_PER_INSTR 96
> +#endif
>  
>  #if HOST_LONG_BITS == 32
>  #define MAX_OPC_PARAM_PER_ARG 2
> -- 
> 1.6.0.2
> 
> 
> 

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
address@hidden                 http://www.aurel32.net