On Nov. 11, 2001, I sent in a C51 V6.20 bug report via e-mail to support.us@keil.com. I never received a response, so assumed Keil's policy was to silently introduce the bug fix in a subsequent update. After downloading V6.21, I see that the bug is still present. Now I'm wondering whether my e-mailed bug report was ever acted upon, so I'm going to try the forum as the means to report the bug. I have a section of code with pointer incrementation that compiles and executes correctly when the pointer is defined as a generic pointer or as an xdata pointer, or when the optimization setting is reduced from the default OT(8,SPEED) to OT(7,SPEED). However, when the pointer is defined as a data or idata pointer, the compiler does not emit code to increment the pointer with default optimization. I have provided example code demonstrating the problem. The specific statement is line 50 below.
unsigned char idata u64[8] = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF }; unsigned char idata s[21]; void U64ToStr( char idata *s, unsigned char idata *binP, unsigned char n ) { unsigned char bcd; unsigned char idata *bcdP; unsigned char bitMsk = 0x80; unsigned char carry; unsigned char i; /* Zero the target buffer (NUL-terminating the string). */ for (bcdP = s, i = n + 1; i != 0; *bcdP++ = 0, i--) ; /* Convert 64 bits of binary to unpacked BCD. For each bit, starting * with the MSBit, BCD = (BCD * 2) + bit. */ for (i = 64; i != 0; i--) { carry = (*binP & bitMsk) ? 1 : 0; for (bcdP = s + n; bcdP != s; /* Iter in body */) { bcd = *--bcdP * 2; bcd += carry; if (bcd < 10) carry = 0; else { carry = 1; bcd -= 10; } *bcdP = bcd; } if ((bitMsk >>= 1) == 0) { bitMsk = 0x80; binP++; } } /* ASCII'fy BCD. */ for (i = n; i != 0; i--) *bcdP++ += '0'; /* <-- C51 BUG: bcdP doesn't increment for (i)data */ } void main( void ) { for (;;) U64ToStr( s, u64, 6 ); }
It's not a bug, its a great optimization. Think about, does anyone one look at bcdP after this loop? No. So why emit code to update this variable? To better understand this add
i = *bcdP;
Huh??? Great optimization? So you are suggesting that rather than adding '0' (0x30) to ASCII'fy the first 6 digits in my string like I want, it's OK for the compiler to emit code that adds '0' 6 times to only first digit? I don't understand.
The ptr variable is store in a register and that is incremented.
The ptr variable is store in a register and that is incremented. Right and wrong. Jon, you really must try it to see. Here's an excerpt from the compiled code:
; SOURCE LINE # 15 ;---- Variable 'bcdP' assigned to Register 'R5' ---- ;=== MOST CODE SNIPPED === ; SOURCE LINE # 50 0063 AC01 MOV R4,AR1 0065 ?C0014: 0065 EC MOV A,R4 0066 600B JZ ?C0017 ; SOURCE LINE # 51 0068 AF05 MOV R7,AR5 006A A807 MOV R0,AR7 006C 7430 MOV A,#030H 006E 26 ADD A,@R0 006F F6 MOV @R0,A 0070 1C DEC R4 0071 80F2 SJMP ?C0014
Sorry, the file that I copied the code to compile had a optimization setting different than the project. I now get the same results as you.
Hey, no problem. Thanks for your consideration in even looking at the problem. You just had me sweatin' there for a little while. I would be embarassed to submit a false alarm. I just hope the report gets seen this time, since e-mailing it is (apparently) the wrong means. Thanks again, --Dan Henry
/* ASCII'fy BCD. */ for (i = n; i != 0; i--) *bcdP++ += '0'; /* <-- C51 BUG: bcdP doesn't increment for (i)data */
/* ASCII'fy BCD. */ for (i = n; i != 0; i--) { *bcdP += '0'; ++bcdP; }
for (i = n; i != 0; i--) { bcdP[i] += '0'; }
Thanks for your thoughts Andrew. Regarding "I have had problems before with packing too much stuff into a single line - particularly the pre/post increment/decrement operators.", the compiler handles the statement fine if I change "binP" to "s":
for (i = n; i != 0; i--) *s++ += '0'; /* Using 's' does work. */ ;---- Variable 's' assigned to Register 'R7' ---- ;---- Variable 'i' assigned to Register 'R4' ---- 0061 AC00 R MOV R4,n 0063 ?C0014: 0063 EC MOV A,R4 0064 600C JZ ?C0017 0066 AE07 MOV R6,AR7 0068 0F INC R7 ; <-- *** 0069 A806 MOV R0,AR6 006B 7430 MOV A,#030H 006D 26 ADD A,@R0 006E F6 MOV @R0,A 006F 1C DEC R4 0070 80F1 SJMP ?C0014
for (i = n; i != 0; i--) { *bcdP += '0'; ++bcdP; } ;---- Variable 'n' assigned to Register 'R1' ---- ;---- Variable 'bcdP' assigned to Register 'R5' ---- ;---- Variable 'i' assigned to Register 'R4' ---- 0063 AC01 MOV R4,AR1 0065 ?C0014: 0065 EC MOV A,R4 0066 6009 JZ ?C0017 0068 A805 MOV R0,AR5 006A 7430 MOV A,#030H 006C 26 ADD A,@R0 006D F6 MOV @R0,A 006E 1C DEC R4 006F 80F4 SJMP ?C0014 0071 ?C0017:
;---- Variable 'n' assigned to Register 'R1' ---- ;---- Variable 'bcdP' assigned to Register 'R5' ---- ;---- Variable 'i' assigned to Register 'R4' ---- 0064 AC01 MOV R4,AR1 0066 ?C0014: 0066 EC MOV A,R4 0067 600A JZ ?C0017 0069 A805 MOV R0,AR5 006B 7430 MOV A,#030H 006D 26 ADD A,@R0 006E F6 MOV @R0,A 006F 0D INC R5 ; <-- *** 0070 1C DEC R4 0071 80F3 SJMP ?C0014 0073 ?C0017:
Is the lack of response to my Forum-posted bug report on 12/7 due to my Dec. 2001 support expiration? This is certainly unfortunate timing -- trying to debug buggy client hardware, while trying to debug my software, while trying to help with an RTOS port to the Keil C51 toolchain. Now with unresolved compiler questions, it makes for an interesting exercise of "who done it?" when something does not work. I now realize that I've brought the lack of attention to my report upon myself by not anticipating the end date for my support and that I wouldn't be eligible for any bug fixes now (even though I originally reported the bug one month ago on 11/11), but I'd think the original e-mailed report, which was within my support date parameters, would have at least qualified for an acknowledgement of receiving the report. I'll be renewing support again after I dredge up records and locate my Sales contact, and will try reporting the bug again then (sigh). (Hint to Keil Sales: Professional C51 users [not kit level, but those who are in the embedded business] would likely welcome a kindly reminder that their support is expiring.) The lack of any response to both email-posted and Forum-posted bug reports still makes me wonder which is the proper means for reporting bugs. I seem to recall seeing a Keil response to someone's bug report on the Forum. I prefer to use private e-mail first, but has anybody had success e-mailing bug reports? Experiences/recommendations anyone? Frustratedly yours, --Dan Henry