[9fans] An acid-question

[9fans] An acid-question

[Sascha Retzki]

Hi folks,

I got a function which goes like this:
void
m00(...)
{
	int i,j = 0;

	if(foo) {

		for(i=0;i<=n;i++) {
			some_data[j++] = srcdata[i]
		}

	} else {
		...
	}
}

(ignore the errors)

I get a fault write. Jippie. Starting to acid arround, I came across this:

acid: *m00:j
0x00000000
acid: *m00:i
<stdin>:3: (error) colon: local variable not found
acid:

so obviously acid does not 'see' i because it is first used in a new block (the if()), at least I guess so. How do I display the value of i in acid?


Mfg, Sascha

Re: [9fans] An acid-question

[erik quanstrom]

i doesn't exist -- it was optimized away because your code is equivalent to this

void
m00(...)
{
	int i;

	if(foo){
		for(i = 0; i <= n; i++){
			some_data[i] = srcdata[i]
		}

	} else {
		...
	}
}

because i == j.

- erik

On Sat Jul 15 05:55:56 CDT 2006, sretzki@gmx.de wrote:
>
> Hi folks,
>
> I got a function which goes like this:
> void
> m00(...)
> {
> 	int i,j = 0;
>
> 	if(foo) {
>
> 		for(i=0;i<=n;i++) {
> 			some_data[j++] = srcdata[i]
> 		}
>
> 	} else {
> 		...
> 	}
> }
>
> (ignore the errors)
>
> I get a fault write. Jippie. Starting to acid arround, I came across this:
>
> acid: *m00:j
> 0x00000000
> acid: *m00:i
> <stdin>:3: (error) colon: local variable not found
> acid:
>
> so obviously acid does not 'see' i because it is first used in a new block (the if()), at least I guess so. How do I display the value of i in acid?
>
>
> Mfg, Sascha

Re: [9fans] An acid-question

[Sascha Retzki]

> i doesn't exist -- it was optimized away because your code is equivalent to this
>

Okay, I will never write pseudo-like code to demonstrate what I mean :/

fmt is a structure holding all meta-data of the audio-input. convertchannels() is called by a function which reads fmt->byterate audio-data from stdin, and provides another array big enough to hold a copy of the input-data (datacopy).

void
convertchannels(short *data, short *datacopy)
{
	int i, j = 0;
	int inbyterate = fmt->samplerate * fmt->channels * 2;


	memcpy(datacopy,data,inbyterate * sizeof(short));


	if(fmt->channels == 1) {

		for(i=0;i <= inbyterate;i++) {

			/* mono to left ... */
			data[j++] = datacopy[i];

			/* mono to right ... */
			data[j++] = datacopy[i];

		}
	}
	else
		sysfatal("convertchannels(): I can just convert from mono to stereo :-/\n");


>
> - erik
>
> On Sat Jul 15 05:55:56 CDT 2006, sretzki@gmx.de wrote:
>>
>> Hi folks,
>>
>> I got a function which goes like this:
>> void
>> m00(...)
>> {
>> 	int i,j = 0;
>>
>> 	if(foo) {
>>
>> 		for(i=0;i<=n;i++) {
>> 			some_data[j++] = srcdata[i]
>> 		}
>>
>> 	} else {
>> 		...
>> 	}
>> }
>>
>> (ignore the errors)
>>
>> I get a fault write. Jippie. Starting to acid arround, I came across this:
>>
>> acid: *m00:j
>> 0x00000000
>> acid: *m00:i
>> <stdin>:3: (error) colon: local variable not found
>> acid:
>>
>> so obviously acid does not 'see' i because it is first used in a new block (the if()), at least I guess so. How do I display the value of i in acid?
>>
>>
>> Mfg, Sascha

Re: [9fans] An acid-question

[erik quanstrom]

your loop test has an obiwan error:

		for(i=0;i <= inbyterate;i++) {

should be
		for(i = 0; i < inbyterate; i++);

- erik

On Sat Jul 15 06:42:57 CDT 2006, sretzki@gmx.de wrote:
> Okay, I will never write pseudo-like code to demonstrate what I mean :/
>
> fmt is a structure holding all meta-data of the audio-input. convertchannels() is called by a function which reads fmt->byterate audio-data from stdin, and provides another array big enough to hold a copy of the input-data (datacopy).
>
> void
> convertchannels(short *data, short *datacopy)
> {
> 	int i, j = 0;
> 	int inbyterate = fmt->samplerate * fmt->channels * 2;
>
>
> 	memcpy(datacopy,data,inbyterate * sizeof(short));
>
>
> 	if(fmt->channels == 1) {
>
> 		for(i=0;i <= inbyterate;i++) {
>
> 			/* mono to left ... */
> 			data[j++] = datacopy[i];
>
> 			/* mono to right ... */
> 			data[j++] = datacopy[i];
>
> 		}
> 	}
> 	else
> 		sysfatal("convertchannels(): I can just convert from mono to stereo :-/\n");

Re: [9fans] An acid-question

[Sascha Retzki]

> your loop test has an obiwan error:
>
> 		for(i=0;i <= inbyterate;i++) {
>
> should be
> 		for(i = 0; i < inbyterate; i++);
>

heh.. 'oups' ;) .. Thanks Erik. Nevertheless, that is not the problem. Do you have an idea why acid does not 'see' i ? Or how I can investigate i?

> - erik
>
> On Sat Jul 15 06:42:57 CDT 2006, sretzki@gmx.de wrote:
>> Okay, I will never write pseudo-like code to demonstrate what I mean :/
>>
>> fmt is a structure holding all meta-data of the audio-input. convertchannels() is called by a function which reads fmt->byterate audio-data from stdin, and provides another array big enough to hold a copy of the input-data (datacopy).
>>
>> void
>> convertchannels(short *data, short *datacopy)
>> {
>> 	int i, j = 0;
>> 	int inbyterate = fmt->samplerate * fmt->channels * 2;
>>
>>
>> 	memcpy(datacopy,data,inbyterate * sizeof(short));
>>
>>
>> 	if(fmt->channels == 1) {
>>
>> 		for(i=0;i <= inbyterate;i++) {
>>
>> 			/* mono to left ... */
>> 			data[j++] = datacopy[i];
>>
>> 			/* mono to right ... */
>> 			data[j++] = datacopy[i];
>>
>> 		}
>> 	}
>> 	else
>> 		sysfatal("convertchannels(): I can just convert from mono to stereo :-/\n");

Re: [9fans] An acid-question

[erik quanstrom]

On Sat Jul 15 07:37:29 CDT 2006, sretzki@gmx.de wrote:
> > your loop test has an obiwan error:
> >
> > 		for(i=0;i <= inbyterate;i++) {
> >
> > should be
> > 		for(i = 0; i < inbyterate; i++);
> >
>
> heh.. 'oups' ;) .. Thanks Erik. Nevertheless, that is not the problem. Do you have an idea why acid does not 'see' i ? Or how I can investigate i?

i = j/2.  i'm still pretty sure that kenc is optimizing i away.  here's what i'd
do.  if samplerate is large (~10000), you need to make sure that data and datacopy
are not on the stack.  then run acid on the broken process and make sure you're
actually on the line expected and use i = j/2 to figure out what i is.

- erik

Re: [9fans] An acid-question

[jmk]

On Sat Jul 15 08:37:19 EDT 2006, sretzki@gmx.de wrote:
> ...
> Thanks Erik. Nevertheless, that is not the problem.
> Do you have an idea why acid does not 'see' i ? Or how
> I can investigate i?

Erik is right that 'i' is optimised. It's not optimised away,
but it does not appear in the symbol table because its entire
use lifetime it is held in a register; acid therefore can't
find it.

Assuming this code was in file a.c, after you have done
	8c -FTVw a.c
then
	nm -a a.8
will show a line for 'j':
       8 a j
indicating it is an automatic variable, but there is no
output for 'i'. If you look at the code with acid (e.g.
'asm(convertchannels)' followed by 'casm()' until you get
to the code you are interested in you will see something
like this:

ethel% acid 8.out
8.out:386 plan 9 executable

/sys/lib/acid/port
/sys/lib/acid/386
lib/acid
acid: asm(convertchannels)
convertchannels 0x00001020	SUBL	$0x20,SP
convertchannels+0x3 0x00001023	MOVL	fmt(SB),BX
convertchannels+0x9 0x00001029	MOVL	$0x0,j+0x18(SP)
convertchannels+0x11 0x00001031	MOVL	0x0(BX),AX
convertchannels+0x13 0x00001033	MOVL	0x4(BX),CX
convertchannels+0x16 0x00001036	IMULL	AX,CX
convertchannels+0x18 0x00001038	ADDL	AX,AX
convertchannels+0x1a 0x0000103a	MOVL	AX,CX
convertchannels+0x1c 0x0000103c	MOVL	datacopy+0x4(FP),AX
convertchannels+0x20 0x00001040	MOVL	AX,0x0(SP)
convertchannels+0x23 0x00001043	MOVL	data+0x0(FP),AX
convertchannels+0x27 0x00001047	MOVL	AX,0x4(SP)
convertchannels+0x2b 0x0000104b	MOVL	CX,AX
convertchannels+0x2d 0x0000104d	MOVL	CX,inbyterate+0x14(SP)
convertchannels+0x31 0x00001051	ADDL	AX,AX
convertchannels+0x33 0x00001053	MOVL	AX,0x8(SP)
convertchannels+0x37 0x00001057	CALL	memcpy(SB)
convertchannels+0x3c 0x0000105c	MOVL	data+0x0(FP),DI
convertchannels+0x40 0x00001060	MOVL	j+0x18(SP),SI
convertchannels+0x44 0x00001064	MOVL	fmt(SB),AX
convertchannels+0x4a 0x0000106a	MOVL	0x4(AX),AX
convertchannels+0x4d 0x0000106d	CMPL	AX,$0x1
convertchannels+0x50 0x00001070	JNE	convertchannels+0x87(SB)
convertchannels+0x52 0x00001072	XORL	BP,BP
convertchannels+0x54 0x00001074	CMPL	inbyterate+0x14(SP),BP
convertchannels+0x58 0x00001078	JGT	convertchannels+0x83(SB)
convertchannels+0x5a 0x0000107a	MOVL	SI,AX
convertchannels+0x5c 0x0000107c	INCL	SI
convertchannels+0x5d 0x0000107d	LEAL	0x0(DI)(AX*2),AX
convertchannels+0x60 0x00001080	MOVL	datacopy+0x4(FP),BX
acid: casm()
convertchannels+0x64 0x00001084	MOVWSX	0x0(BX)(BP*2),CX
convertchannels+0x68 0x00001088	MOVW	CX,0x0(AX)
convertchannels+0x6b 0x0000108b	MOVL	SI,AX
convertchannels+0x6d 0x0000108d	INCL	SI
convertchannels+0x6e 0x0000108e	LEAL	0x0(DI)(AX*2),AX
convertchannels+0x71 0x00001091	MOVL	datacopy+0x4(FP),BX
convertchannels+0x75 0x00001095	MOVWSX	0x0(BX)(BP*2),CX
convertchannels+0x79 0x00001099	MOVW	CX,0x0(AX)
convertchannels+0x7c 0x0000109c	INCL	BP
convertchannels+0x7d 0x0000109d	CMPL	inbyterate+0x14(SP),BP
convertchannels+0x81 0x000010a1	JLE	convertchannels+0x5a(SB)
convertchannels+0x83 0x000010a3	ADDL	$0x20,SP
convertchannels+0x86 0x000010a6	RET
convertchannels+0x87 0x000010a7	MOVL	$.string(SB),AX
convertchannels+0x8c 0x000010ac	MOVL	AX,0x0(SP)
convertchannels+0x8f 0x000010af	CALL	sysfatal(SB)
convertchannels+0x94 0x000010b4	JMP	convertchannels+0x83(SB)
acid:
ethel%

The loop using 'i' starts at convertchannels+0x52
and 'i is held in register BP. You can use the 'regs()'
or 'gpr()' commands to examine BP in a broken proc.

Re: [9fans] An acid-question

[Sascha Retzki]

> If you look at the code with acid (e.g.
> 'asm(convertchannels)' followed by 'casm()' until you get
> to the code you are interested in you will see something
> like this:

Nice to know, thanks.


Sascha