[v2,21/22] target-mips: use pointers referring to appropriate decoding function

After selecting CPU in QEMU the base ISA will not change. Therefore
introducing *_arch function pointers that are set in cpu_state_reset to
point at the appropriate SPECIAL and SPECIAL3 decoding functions, and avoid
unnecessary 'if' statements.

Signed-off-by: Leon Alrae <leon.alrae@imgtec.com>
---
 target-mips/translate.c |   32 +++++++++++++++++++++-----------
 1 files changed, 21 insertions(+), 11 deletions(-)

On Wed, Jun 11, 2014 at 04:19:51PM +0100, Leon Alrae wrote:
> After selecting CPU in QEMU the base ISA will not change. Therefore
> introducing *_arch function pointers that are set in cpu_state_reset to
> point at the appropriate SPECIAL and SPECIAL3 decoding functions, and avoid
> unnecessary 'if' statements.
> 
> Signed-off-by: Leon Alrae <leon.alrae@imgtec.com>
> ---
>  target-mips/translate.c |   32 +++++++++++++++++++++-----------
>  1 files changed, 21 insertions(+), 11 deletions(-)
> 
> diff --git a/target-mips/translate.c b/target-mips/translate.c
> index de35b77..7ff7829 100644
> --- a/target-mips/translate.c
> +++ b/target-mips/translate.c
> @@ -15634,6 +15634,13 @@ out:
>      tcg_temp_free(t1);
>  }
>  
> +/* Some instructions from MIPS32R6 and pre-MIPS32R6 have identical encoding.
> +
> +   decode_opc_*_arch are pointing at the appropriate decoding functions
> +   depending on a base ISA supported by selected MIPS CPU. */
> +static void (*decode_opc_special_arch) (CPUMIPSState*, DisasContext*);
> +static void (*decode_opc_special3_arch) (CPUMIPSState*, DisasContext*);
> +
>  static void decode_opc_special_r6(CPUMIPSState *env, DisasContext *ctx)
>  {
>      int rs, rt, rd, sa;
> @@ -16002,11 +16009,8 @@ static void decode_opc_special(CPUMIPSState *env, DisasContext *ctx)
>          break;
>  #endif
>      default:
> -        if (ctx->insn_flags & ISA_MIPS32R6) {
> -            decode_opc_special_r6(env, ctx);
> -        } else {
> -            decode_opc_special_legacy(env, ctx);
> -        }
> +        decode_opc_special_arch(env, ctx);
> +        break;
>      }
>  }
>  
> @@ -16799,12 +16803,9 @@ static void decode_opc_special3(CPUMIPSState *env, DisasContext *ctx)
>              tcg_temp_free(t0);
>          }
>          break;
> -    default:            /* Invalid */
> -        if (ctx->insn_flags & ISA_MIPS32R6) {
> -            decode_opc_special3_r6(env, ctx);
> -        } else {
> -            decode_opc_special3_legacy(env, ctx);
> -        }
> +    default:
> +        decode_opc_special3_arch(env, ctx);
> +        break;
>      }
>  }
>  
> @@ -17831,6 +17832,15 @@ void cpu_state_reset(CPUMIPSState *env)
>      env->active_fpu.fcr0 = env->cpu_model->CP1_fcr0;
>      env->insn_flags = env->cpu_model->insn_flags;
>  
> +    /* Select decoding functions appropriate for supported ISA */
> +    if (env->insn_flags & ISA_MIPS32R6) {
> +        decode_opc_special_arch = decode_opc_special_r6;
> +        decode_opc_special3_arch = decode_opc_special3_r6;
> +    } else {
> +        decode_opc_special_arch = decode_opc_special_legacy;
> +        decode_opc_special3_arch = decode_opc_special3_legacy;
> +    }
> +
>  #if defined(CONFIG_USER_ONLY)
>      env->CP0_Status = (MIPS_HFLAG_UM << CP0St_KSU);
>  # ifdef TARGET_MIPS64

I have mixed filling about that. While it clearly makes the code more
readable, I am not sure it would be any faster given that it might kill
branch prediction. Any opinion from others?

On 06/19/2014 03:18 PM, Aurelien Jarno wrote:
> I have mixed filling about that. While it clearly makes the code more
> readable, I am not sure it would be any faster given that it might kill
> branch prediction. Any opinion from others?

Given the function pointer is used only once, I don't think it's useful.  With
normal branch predictors we'll do much better with direct branches, and
probably inlining.


r~