From mboxrd@z Thu Jan  1 00:00:00 1970
MIME-Version: 1.0
In-Reply-To: <dd6fe68a0908312010j2fb6258dib81feb724ab715a9@mail.gmail.com>
References: <10afea42a3af8daf8abd3b8e8775850e@quanstro.net>
	<775b8d190908311851t701d860av587b7433bea29fcf@mail.gmail.com>
	<dd6fe68a0908312010j2fb6258dib81feb724ab715a9@mail.gmail.com>
Date: Tue,  1 Sep 2009 13:24:11 +1000
Message-ID: <775b8d190908312024q3a0167a0xf7f686a6f2408eb2@mail.gmail.com>
From: Bruce Ellis <bruce.ellis@gmail.com>
To: Fans of the OS Plan 9 from Bell Labs <9fans@9fans.net>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
Subject: Re: [9fans] 8c vlong bug
Topicbox-Message-UUID: 5dd09086-ead5-11e9-9d60-3106f5b1d025

will fix.

On Tue, Sep 1, 2009 at 1:10 PM, Russ Cox<rsc@swtch.com> wrote:
>> temporarly out of time on this one. =A0it appears
>> from the assembly output that 8c multiplies by
>> 0 and not 1 when computing z a second time.
>> nonetheless, i haven't yet seen the problem.
>
> not quite that simple.
> it's a register allocation
> (really register management) bug.
>
> On Mon, Aug 31, 2009 at 6:51 PM, Bruce Ellis<bruce.ellis@gmail.com> wrote=
:
>> if it's my fault i'll fix it. it did screw up mod in a subtle way but
>> that's been fixed.
>
> it looks like the cgen64 multiply case is
> not managing the registers correctly
> when AX is not where it expects.
> more details for you:
>
> int three =3D 3;
> int one =3D 1;
> uvlong twelve =3D 12;
>
> /* ok */
> uvlong
> mul1(void)
> {
> =A0 =A0 =A0 =A0return one * (twelve - three);
> }
>
> TEXT =A0 =A0mul1+0(SB),0,$12
> =A0 =A0 =A0 =A0MOVL =A0 =A0three+0(SB),AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,CX
> =A0 =A0 =A0 =A0SARL =A0 =A0$31,CX
> =A0 =A0 =A0 =A0// three is now in CX:AX
>
> =A0 =A0 =A0 =A0LEAL =A0 =A0twelve+0(SB),DX
> =A0 =A0 =A0 =A0MOVL =A0 =A0(DX),BX
> =A0 =A0 =A0 =A0SUBL =A0 =A0AX,BX
> =A0 =A0 =A0 =A0MOVL =A0 =A04(DX),AX
> =A0 =A0 =A0 =A0SBBL =A0 =A0CX,AX
> =A0 =A0 =A0 =A0// twelve - three is now in AX:BX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,CX
> =A0 =A0 =A0 =A0// twelve - three is now in CX:BX
>
> =A0 =A0 =A0 =A0MOVL =A0 =A0one+0(SB),AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,DX
> =A0 =A0 =A0 =A0SARL =A0 =A0$31,DX
> =A0 =A0 =A0 =A0// one is now in DX:AX
>
> =A0 =A0 =A0 =A0// 64-bit multiply CX:BX * DX:AX -> DX:AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0CX,BP
> =A0 =A0 =A0 =A0ORL =A0 =A0 DX,BP
> =A0 =A0 =A0 =A0CMPL =A0 =A0BP,$0
> =A0 =A0 =A0 =A0JNE =A0 =A0 ,3(PC)
> =A0 =A0 =A0 =A0MULL =A0 =A0BX,
> =A0 =A0 =A0 =A0JMP =A0 =A0 ,8(PC)
> =A0 =A0 =A0 =A0IMULL =A0 BX,DX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,BP
> =A0 =A0 =A0 =A0IMULL =A0 CX,BP
> =A0 =A0 =A0 =A0ADDL =A0 =A0DX,BP
> =A0 =A0 =A0 =A0MOVL =A0 =A0BX,DX
> =A0 =A0 =A0 =A0MULL =A0 =A0DX,
> =A0 =A0 =A0 =A0ADDL =A0 =A0BP,DX
>
> =A0 =A0 =A0 =A0MOVL =A0 =A0.ret+0(FP),CX =A0 // probably LEAL
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,(CX)
> =A0 =A0 =A0 =A0MOVL =A0 =A0DX,4(CX)
> =A0 =A0 =A0 =A0RET =A0 =A0 ,
> =A0 =A0 =A0 =A0RET =A0 =A0 ,
>
> vs
>
> /* not ok */
>
> uvlong
> mul2(void)
> {
> =A0 =A0 =A0 =A0return (twelve - three) * one;
> }
>
> TEXT =A0 =A0mul2+0(SB),0,$12
> =A0 =A0 =A0 =A0MOVL =A0 =A0one+0(SB),AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,CX
> =A0 =A0 =A0 =A0SARL =A0 =A0$31,CX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,BX
> =A0 =A0 =A0 =A0// one is now in CX:BX
>
> =A0 =A0 =A0 =A0MOVL =A0 =A0three+0(SB),AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,DX
> =A0 =A0 =A0 =A0SARL =A0 =A0$31,DX
> =A0 =A0 =A0 =A0// three is now in DX:AX
>
> =A0 =A0 =A0 =A0LEAL =A0 =A0twelve+0(SB),BP
> =A0 =A0 =A0 =A0MOVL =A0 =A0(BP),SI
> =A0 =A0 =A0 =A0SUBL =A0 =A0AX,SI
> =A0 =A0 =A0 =A0MOVL =A0 =A04(BP),AX
> =A0 =A0 =A0 =A0SBBL =A0 =A0DX,AX
> =A0 =A0 =A0 =A0// twelve - three is now in AX:SI
>
> =A0 =A0 =A0 =A0// 64-bit multiply CX:BX * AX:SI
> =A0 =A0 =A0 =A0// saving AX and DX so they can be reused.
> =A0 =A0 =A0 =A0// DX doesn't even need saving.
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,.safe+-4(SP)
> =A0 =A0 =A0 =A0MOVL =A0 =A0DX,.safe+-8(SP)
> =A0 =A0 =A0 =A0MOVL =A0 =A0SI,DX
>
> =A0 =A0 =A0 =A0// 8c is very confused at this point.
> =A0 =A0 =A0 =A0// it is using AX for multiple purposes.
> =A0 =A0 =A0 =A0MOVL =A0 =A0CX,AX
> =A0 =A0 =A0 =A0ORL =A0 =A0 SI,AX
> =A0 =A0 =A0 =A0CMPL =A0 =A0AX,$0
> =A0 =A0 =A0 =A0JNE =A0 =A0 ,4(PC)
> =A0 =A0 =A0 =A0MOVL =A0 =A0BX,AX
> =A0 =A0 =A0 =A0MULL =A0 =A0AX,
> =A0 =A0 =A0 =A0JMP =A0 =A0 ,7(PC)
> =A0 =A0 =A0 =A0IMULL =A0 BX,DX
> =A0 =A0 =A0 =A0IMULL =A0 CX,AX
> =A0 =A0 =A0 =A0ADDL =A0 =A0DX,AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0BX,DX
> =A0 =A0 =A0 =A0MULL =A0 =A0DX,
> =A0 =A0 =A0 =A0ADDL =A0 =A0AX,DX
>
> =A0 =A0 =A0 =A0// restore old AX DX now that multiply is over.
> =A0 =A0 =A0 =A0// (forgot that multiply left result in DX:AX)
> =A0 =A0 =A0 =A0MOVL =A0 =A0.safe+-4(SP),AX
> =A0 =A0 =A0 =A0MOVL =A0 =A0.safe+-8(SP),DX
>
> =A0 =A0 =A0 =A0MOVL =A0 =A0.ret+0(FP),CX =A0 // probably LEAL
> =A0 =A0 =A0 =A0MOVL =A0 =A0AX,(CX)
> =A0 =A0 =A0 =A0MOVL =A0 =A0DX,4(CX)
> =A0 =A0 =A0 =A0RET =A0 =A0 ,
>
>