This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH, i386]: Unify TARGET_SSE_MATH for float* patterns


On Mon, Dec 13, 2004 at 10:13:52AM +0100, Uros Bizjak wrote:
> +(define_insn "*floathisf2_i387"
>    [(set (match_operand:SF 0 "register_operand" "=f,f")
> -	(float:SF (match_operand:HI 1 "nonimmediate_operand" "m,r")))]
> -  "TARGET_80387 && (!TARGET_SSE || !TARGET_SSE_MATH)"
> +	(float:SF (match_operand:HI 1 "nonimmediate_operand" "m,?r")))]
> +  "TARGET_80387"

It's wrong to enable this pattern with TARGET_SSE_MATH.  There's
no floathisf2_{mixed,sse} patterns to shadow it.

> +(define_insn "*floathidf2_i387"
>    [(set (match_operand:DF 0 "register_operand" "=f,f")
> -	(float:DF (match_operand:HI 1 "nonimmediate_operand" "m,r")))]
> -  "TARGET_80387 && (!TARGET_SSE2 || !TARGET_SSE_MATH)"
> +	(float:DF (match_operand:HI 1 "nonimmediate_operand" "m,?r")))]
> +  "TARGET_80387"

Similarly, except obviously we can enable it if !SSE2, since all DFmode
arithmetic will be going through the 387 stack in that case.

Otherwise ok.


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]