[avr-gcc-list] float sizes

[Theodore A. Roth]

Hi,

I'm doing some more fix-ups on my avr-gdb patch and need to know sizes for 
floats. Unless I hear otherwise, I'll use these assuptions:

  sizeof(float)       == 4 * sizeof(char)
  sizeof(double)      == 4 * sizeof(char)
  sizeof(long double) == 4 * sizeof(char)

Also, I'm assuming litle endian IEEE format for float.

As a test, I wrote this program and compiled for 8515:

      1 #include <stdlib.h>
      2 
      3 int main( void )
      4 {
      5     float       ff = 1.2;
      6     double      dd = 2.3;
      7     long double ld = 3.4;
      8 
      9     ff += 1.0;
     10     dd += 1.0;
     11     ld += 1.0;
     12 
     13     return(0);
     14 }

avr-gcc -g -Wall -mmcu=at90s8515   -c -o bug.o bug.c
avr-gcc -g -Wall -mmcu=at90s8515  -o bug bug.o
cp bug bug.elf
avr-objdump -h -S -D --disassemble-zeroes bug.elf > bug.lst

which generates this code (part of bug.lst):

     82 int main( void )
     83 {
     84   56:   c3 e5           ldi     r28, 0x53       ; 83
     85   58:   d2 e0           ldi     r29, 0x02       ; 2
     86   5a:   de bf           out     0x3e, r29       ; 62
     87   5c:   cd bf           out     0x3d, r28       ; 61
     88         float       ff = 1.2;
     89   5e:   8a e9           ldi     r24, 0x9A       ; 154
     90   60:   99 e9           ldi     r25, 0x99       ; 153
     91   62:   a9 e9           ldi     r26, 0x99       ; 153
     92   64:   bf e3           ldi     r27, 0x3F       ; 63
     93   66:   89 83           std     Y+1, r24        ; 0x01
     94   68:   9a 83           std     Y+2, r25        ; 0x02
     95   6a:   ab 83           std     Y+3, r26        ; 0x03
     96   6c:   bc 83           std     Y+4, r27        ; 0x04
     97         double      dd = 2.3;
     98   6e:   83 e3           ldi     r24, 0x33       ; 51
     99   70:   93 e3           ldi     r25, 0x33       ; 51
    100   72:   a3 e1           ldi     r26, 0x13       ; 19
    101   74:   b0 e4           ldi     r27, 0x40       ; 64
    102   76:   8d 83           std     Y+5, r24        ; 0x05
    103   78:   9e 83           std     Y+6, r25        ; 0x06
    104   7a:   af 83           std     Y+7, r26        ; 0x07
    105   7c:   b8 87           std     Y+8, r27        ; 0x08
    106         long double ld = 3.4;
    107   7e:   8a e9           ldi     r24, 0x9A       ; 154
    108   80:   99 e9           ldi     r25, 0x99       ; 153
    109   82:   a9 e5           ldi     r26, 0x59       ; 89
    110   84:   b0 e4           ldi     r27, 0x40       ; 64
    111   86:   89 87           std     Y+9, r24        ; 0x09
    112   88:   9a 87           std     Y+10, r25       ; 0x0a
    113   8a:   ab 87           std     Y+11, r26       ; 0x0b
    114   8c:   bc 87           std     Y+12, r27       ; 0x0c
    115 
    116         ff += 1.0;
    117   8e:   20 e0           ldi     r18, 0x00       ; 0
    118   90:   30 e0           ldi     r19, 0x00       ; 0
    119   92:   40 e8           ldi     r20, 0x80       ; 128
<snip>

This seems to confirm that my size assumptions are correct.

I know nothing about IEEE float format, so I still need confirmation on my 
assumption there.


Ted Roth


avr-gcc-list at http://avr1.org