Re: [PATCH 13/29] tcg/i386: Support TCG_COND_TST{EQ,NE}

[Richard Henderson]

On 10/26/23 04:29, Paolo Bonzini wrote:
> On 10/26/23 02:14, Richard Henderson wrote:
> > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > ---
>
> Also, a TST{EQ,NE} with a one-bit immediate argument can be changed to:
>
> - a TEST reg, reg + js/jns (or sets/setns, or cmovs/cmovns) when testing bits 
> 7, 15 or 31
>
> - a BT reg, imm + jc/jnc (or setc/setnc, or cmovc/cmovnc) when testing other bits in the 
> 8..63 range.
>
> I will take a look at using this to get rid of the mask field in CCPrepare, but I would 
> not mind if someone else took a look at these code generation optimizations in tcg/i386.

I thought about that while working on this series, and is part of the reason why 
tcg_out_cmp now returns a JCC_* value rather than having the caller look it up.

I thought I'd start simpler before adding these optimizations.


r~