English Amiga Board

English Amiga Board (http://eab.abime.net/index.php)
-   Coders. C/C++ (http://eab.abime.net/forumdisplay.php?f=118)
-   -   Wanted: C source for converting gcc (linux) float to FFP (http://eab.abime.net/showthread.php?t=103570)

alkis 15 August 2020 18:12
Wanted: C source for converting gcc (linux) float to FFP
 
I've looked through the internets and I only managed to find the reverse (get ieee from ffp)
I need to get ffp from ieee (I think).

Context: I've cross-compiled some ACE Basic programs from linux to amiga and I've noticed that floats are broken. ACE uses FFP math. Now linux doesn't know anything about FFP so it provides the code generator with normal floats. I need to change those with FFP ones.

Here is stuff, I've found.

Code:

/// https://marc.info/?l=classiccmp&m=145550087725800&w=2

// val is a FFP float, returns IEEE float format
// basically a direct conversion of the motorola FFPTIEEE function.
// comments are from there.
// you are not expected to understand this horrendous mess.
float ffptieee(const uint32_t val)
{
        uint32_t x = val;
        union {
                float f;
                uint32_t i;
        } _fcast;

        x = x + x;                // delete mantissa high bit
        if (x == 0) {
                // if zero, branch zero as finished
                return (float)x;
        }

        uint8_t k = x & 0xff;

        k ^= 0x80;        // to two's complement exponent
        k >>= 1;        // form 8-bit exponent
        k -= 0x82;        // adjust 64 to 127 and excessize

        x = (x & ~0xff) | k;

        x = (x << 16) | (x >> 16);        // swap for high byte placement
        x <<= 7;        // set sign+exp in high byte

        _fcast.i = x;

        return _fcast.f;
}

/// https://www.lysator.liu.se/~celeborn/sync/atari/68000/FFP.ZIP

/*
        TTL      FFPIEEE CONVERSIONS TO/FROM FFP (FFPTIEEE,FFPFIEEE)
******************************************
*  (C)  COPYRIGHT 1981 BY MOTOROLA INC.  *
******************************************
 
****************************************************
*        FFPTIEEE AND FFPFIEEE SUBROUTINES        *
*                                                  *
*  THIS MODULE CONTAINS SINGLE-PRECISION          *
*  CONVERSION ROUTINES FOR IEEE FORMAT FLOATING  *
*  POINT (DRAFT 8.0) TO AND FROM MOTOROLA FAST    *
*  FLOATING POINT (FFP) VALUES.                  *
*  THESE CAN BE USED WHEN LARGE GROUPS OF NUMBERS *
*  NEED TO BE CONVERTED BETWEEN FORMATS.  SEE    *
*  THE MC68344 USER'S GUIDE FOR A FULLER          *
*  EXPLANATION OF THE VARIOUS METHODS FOR IEEE    *
*  FORMAT SUPPORT.                                *
*                                                  *
*  THE FAST FLOATING POINT (NON-IEEE FORMAT)      *
*  PROVIDES RESULTS AS PRECISE AS THOSE REQUIRED  *
*  BY THE IEEE SPECIFICATION.  HOWEVER, THIS      *
*  FORMAT HAS SOME MINOR DIFFERENCES:            *
*    1) IF THE TRUE RESULT OF AN OPERATION        *
*      IS EXACTLY BETWEEN REPRESENTABLE          *
*      VALUES, THE FFP ROUND-TO-NEAREST          *
*      FUNCTION MAY ROUND TO EITHER EVEN OR ODD.  *
*    2) THE FFP EXPONENT ALLOWS HALF OF THE RANGE  *
*      THAT THE SINGLE-PRECISION IEEE FORMAT      *
*      PROVIDES (APPROX. 10 TO THE +-19 DECIMAL). *
*    3) THE IEEE FORMAT SPECIFIES INFINITY,        *
*      NOT-A-NUMBER, AND DENORMALIZED DATA        *
*      TYPES THAT ARE NOT DIRECTLY SUPPORTED      *
*      BY THE FFP FORMAT.  HOWEVER, THEY MAY BE  *
*      ADDED VIA CUSTOMIZING OR USED VIA THE IEEE *
*      FORMAT EQUIVALENT COMPATIBLE CALLS        *
*      DESCRIBED IN THE MC68344 USER'S GUIDE.    *
*    4) ALL ZEROES ARE CONSIDERED POSITIVE        *
*      IN FFP FORMAT.                            *
*    5) THE SLIGHTLY HIGHER PRECISION MULTIPLY    *
*      ROUTINE "FFPMUL2" SHOULD BE SUBSTITUTED    *
*      FOR THE DEFAULT ROUTINE "FFPMUL" FOR      *
*      COMPLETELY EQUIVALENT PRECISION.          *
*                                                  *
****************************************************
        PAGE
FFPIEEE  IDNT  1,1  FFP CONVERSIONS TO/FROM IEEE FORMAT
 
        SECTION  9
 
        XDEF      FFPTIEEE,FFPFIEEE
 
****************************************************
*              FFPTIEEE                          *
*                                                  *
*  FAST FLOATING POINT TO IEEE FORMAT              *
*                                                  *
*  INPUT:  D7 - FAST FLOATING POINT VALUE        *
*                                                  *
*  OUTPUT:  D7 - IEEE FORMAT FLOATING POINT VALUE *
*                                                  *
*  CONDITION CODES:                              *
*            N - SET IF THE RESULT IS NEGATIVE    *
*            Z - SET IF THE RESULT IS ZERO        *
*            V - UNDEFINED                        *
*            C - UNDEFINED                        *
*            X - UNDEFINED                        *
*                                                  *
*  NOTES:                                        *
*    1) NO WORK STORAGE OR REGISTERS REQUIRED.    *
*    2) ALL ZEROES WILL BE CONVERTED POSITIVE.    *
*    3) NO NOT-A-NUMBER, INIFINITY, DENORMALIZED, *
*        OR INDEFINITES GENERATED. (UNLESS        *
*        USER CUSTOMIZED.)                        *
*                                                  *
*  TIMES (ASSUMING IN-LINE CODE):                *
*          VALUE ZERO      18 CYCLES              *
*          VALUE NOT ZERO  66 CYCLES              *
*                                                  *
****************************************************
 
FFPTIEEE EQU      *
 
        ADD.L    D7,D7    DELETE MANTISSA HIGH BIT
        BEQ.S    DONE1    BRANCH ZERO AS FINISHED
        EOR.B    #$80,D7  TO TWOS COMPLEMENT EXPONENT
        ASR.B    #1,D7    FORM 8-BIT EXPONENT
        SUB.B    #$82,D7  ADJUST 64 TO 127 AND EXCESSIZE
        SWAP.W    D7        SWAP FOR HIGH BYTE PLACEMENT
        ROL.L    #7,D7    SET SIGN+EXP IN HIGH BYTE
DONE1    EQU      *
 
        RTS                RETURN TO CALLER
        PAGE
************************************************************
*                    FFPFIEEE                            *
*        FAST FLOATING POINT FROM IEEE FORMAT            *
*                                                          *
*  INPUT:  D7 - IEEE FORMAT FLOATING POINT VALUE        *
*  OUTPUT:  D7 - FFP FORMAT FLOATING POINT VALUE          *
*                                                          *
*  CONDITION CODES:                                      *
*            N - UNDEFINED                                *
*            Z - SET IF THE RESULT IS ZERO                *
*            V - SET IF RESULT OVERFLOWED FFP FORMAT      *
*            C - UNDEFINED                                *
*            X - UNDEFINED                                *
*                                                          *
*  NOTES:                                                *
*    1) REGISTER D5 IS USED FOR WORK STORAGE AND NOT      *
*        TRANSPARENT.                                      *
*    2) NOT-A-NUMBER, INIFINITY, AND DENORMALIZED        *
*        TYPES AS WELL AS AN EXPONENT OUTSIDE OF FFP RANGE *
*        GENERATE A BRANCH TO A SPECIFIC PART OF THE      *
*        ROUTINE.  CUSTOMIZING MAY EASILY BE DONE THERE.  *
*                                                          *
*        THE DEFAULT ACTIONS FOR THE VARIOUS TYPES ARE:    *
*      LABEL      TYPE        DEFAULT SUBSTITUTION        *
*      -----      ----        --------------------        *
*      NAN    NOT-A-NUMBER    ZERO                        *
*      INF    INFINITY        LARGEST FFP VALUE SAME SIGN *
*                              ("V" SET IN CCR)            *
*      DENOR  DENORMALIZED    ZERO                        *
*      EXPHI  EXP TOO LARGE  LARGEST FFP VALUE SAME SIGN *
*                              ("V" SET IN CCR)            *
*      EXPLO  EXP TOO SMALL  ZERO                        *
*                                                          *
*  TIMES (ASSUMING IN-LINE CODE):                        *
*          VALUE ZERO      78 CYCLES                      *
*          VALUE NOT ZERO  72 CYCLES                      *
*                                                          *
************************************************************
 
VBIT    EQU      $02      CONDITION CODE REGISTER "V" BIT MASK
 
FFPFIEEE EQU      *
 
        SWAP.W    D7        SWAP WORD HALVES
        ROR.L    #7,D7    EXPONENT TO LOW BYTE
        MOVE.L    #-128,D5  LOAD $80 MASK IN WORK REGISTER
        EOR.B    D5,D7    CONVERT FROM EXCESS 127 TO TWO'S-COMPLEMENT
        ADD.B    D7,D7    FROM 8 TO 7 BIT EXPONENT
        BVS.S    FFPOVF    BRANCH WILL NOT FIT
        ADD.B    #2<<1+1,D7 ADJUST EXCESS 127 TO 64 AND SET MANTISSA HIGH BIT
        BVS.S    EXPHI    BRANCH EXPONENT TOO LARGE (OVERFLOW)
        EOR.B    D5,D7    BACK TO EXCESS 64
        ROR.L    #1,D7    TO FAST FLOAT REPRESENTATION ("V" CLEARED)
DONE2    EQU      *
 
        RTS                RETURN TO CALLER
        PAGE
* OVERFLOW DETECTED - CAUSED BY ONE OF THE FOLLOWING:
*        - FALSE OVERFLOW DUE TO DIFFERENCE BETWEEN EXCESS 127 AND 64 FORMAT
*        - EXPONENT TOO HIGH OR LOW TO FIT IN 7 BITS (EXPONENT OVER/UNDERFLOW)
*        - AN EXPONENT OF $FF REPRESENTING AN INFINITY
*        - AN EXPONENT OF $00 REPRESENTING A ZERO, NAN, OR DENORMALIZED VALUE
FFPOVF  BCC.S    FFPOVLW  BRANCH IF OVERFLOW (EXPONENT $FF OR TOO LARGE)
* OVERFLOW - CHECK FOR POSSIBLE FALSE OVERFLOW DUE TO DIFFERENT EXCESS FORMATS
        CMP.B    #$7C,D7  ? WAS ORIGINAL ARGUMENT REALLY IN RANGE
        BEQ.S    FFPOVFLS  YES, BRANCH FALSE OVERFLOW
        CMP.B    #$7E,D7  ? WAS ORIGINAL ARGUMENT REALLY IN RANGE
        BNE.S    FFPTOVF  NO, BRANCH TRUE OVERFLOW
FFPOVFLS ADD.B    #$80+2<<1+1,D7  EXCESS 64 ADJUSTMENT AND MANTISSA HIGH BIT
        ROR.L    #1,D7    FINALIZE TO FAST FLOATING POINT FORMAT
        TST.B    D7        INSURE NO ILLEGAL ZERO SIGN+EXPONENT BYTE
        BNE.S    DONE2    DONE IF DOES NOT SET S+EXP ALL ZEROES
        BRA.S    EXPLO    TREAT AS UNDERFLOWED EXPONENT OTHERWISE
* EXPONENT LOW - CHECK FOR ZERO, DENORMALIZED VALUE, OR TOO SMALL AN EXPONENT
FFPTOVF  AND.W    #$FEFF,D7 CLEAR SIGN BIT OUT
        TST.L    D7        ? ENTIRE VALUE NOW ZERO
        BEQ.S    DONE2    BRANCH IF VALUE IS ZERO
        TST.B    D7        ? DENORMALIZED NUMBER (SIGNIFICANT#0, EXP=0)
        BEQ.S    DENOR    BRANCH IF DENORMALIZED
 
***************
*****EXPLO - EXPONENT TO SMALL FOR FFP FORMAT
***************
*  THE SIGN BIT WILL BE BIT 8.
EXPLO    MOVE.L    #0,D7    DEFAULT ZERO FOR THIS CASE ("V" CLEARED)
        BRA.S    DONE2    RETURN TO MAINLINE
 
***************
*****DENOR - DENORMALIZED NUMBER
***************
DENOR    MOVE.L    #0,D7    DEFAULT IS TO RETURN A ZERO ("V" CLEARED)
        BRA.S    DONE2    RETURN TO MAINLINE
 
* EXPONENT HIGH - CHECK FOR EXPONENT TOO HIGH, INFINITY, OR NAN
FFPOVLW  CMP.B    #$FE,D7  ? WAS ORIGINAL EXPONENT $FF
        BNE.S    EXPHI    NO, BRANCH EXPONENT TOO LARGE
        LSR.L    #8,D7    SHIFT OUT EXPONENT
        LSR.L    #1,D7    SHIFT OUT SIGN
        BNE.S    NAN      BRANCH NOT-A-NUMBER
 
***************
*****INF - INFINITY
***************
*  THE CARRY AND X BIT REPRESENT THE SIGN
INF      MOVE.L    #-1,D7    SETUP MAXIMUM FFP VALUE
        ROXR.B    #1,D7    SHIFT IN SIGN
        OR.B      #VBIT,SR  SHOW OVERFLOW OCCURED
        BRA.S    DONE2    RETURN WITH MAXIMUM SAME SIGN TO MAINLINE
 
***************
*****EXPHI - EXPONENT TO LARGE FOR FFP FORMAT
***************
*  THE SIGN BIT WILL BE BIT 8.
EXPHI    LSL.W    #8,D7    SET X BIT TO SIGN
        MOVE.L    #-1,D7    SETUP MAXIMUM NUMBER
        ROXR.B    #1,D7    SHIFT IN SIGN
        OR.B      #VBIT,SR  SHOW OVERFLOW OCURRED
        BRA.S    DONE2    RETURN MAXIMUM SAME SIGN TO MAINLINE
 
***************
*****NAN - NOT-A-NUMBER
***************
* BITS 0 THRU 22 CONTAIN THE NAN DATA FIELD
NAN      MOVE.L    #0,D7    DEFAULT TO A ZERO ("V" BIT CLEARED)
        BRA.S    DONE2    RETURN TO MAINLINE
 
        END


 */

Jobbo 15 August 2020 19:15
The 68000 code you posted has the same ieee to ffp conversion code I've used, which I can confirm works well.

Can you port that to C code yourself?

alkis 15 August 2020 19:28
Quote:
Originally Posted by Jobbo (Post 1420976)
The 68000 code you posted has the same ieee to ffp conversion code I've used, which I can confirm works well.

Can you port that to C code yourself?
If there is no any already used/tested function lying around in some folk's hard drive, I could give it a shot. :)

Docent 16 August 2020 01:38
You can use SPFieee function from mathtrans.library - it will convert a number from ieee to ffp format.

alkis 16 August 2020 02:35
Quote:
Originally Posted by Docent (Post 1421041)
You can use SPFieee function from mathtrans.library - it will convert a number from ieee to ffp format.
Thanks, but the code has to run on linux.

alkis 16 August 2020 02:39
Ok, this is a working draft
Very very lightly tested, seems to work. Also fixed a bug or two on the initial C code that I found.

Code:
#include <stdio.h>
#include <stdint.h>


uint32_t rol(uint32_t x, uint32_t n) {
  //return (x >> n % 32) | (x << (32-n) % 32);
  return (x << n ) | (x >> (32-n));
}

uint32_t ror(uint32_t x, uint32_t n) {
  //return (x >> n % 32) | (x << (32-n) % 32);
  return (x >> n ) | (x << (32-n));
}


float ffptieee(const uint32_t val)
{
        uint32_t x = val;
        union {
                float f;
                uint32_t i;
        } _fcast;

        x = x + x;                // delete mantissa high bit
        if (x == 0) {
                // if zero, branch zero as finished
          _fcast.i = 0;
          return _fcast.f;
        }

        uint8_t k = x & 0xff;

        k ^= 0x80;        // to two's complement exponent
        k >>= 1;        // form 8-bit exponent
        if (k & 0x40)  // do an actual ASR here
          k |= 0x80;    // if previous leftmost bit was 1, do the now leftmost bit 1 as well
       
       
        k -= 0x82;        // adjust 64 to 127 and excessize

        x = (x & ~0xff) | k;

        x = (x << 16) | (x >> 16);        // swap for high byte placement
        x = rol(x, 7);        // set sign+exp in high byte
        _fcast.i = x;

        return _fcast.f;
}


uint32_t ffpfieee(float val) {
  union {
    float f;
    uint32_t i;
  } _fcast;

  _fcast.f = val;

  uint32_t x = _fcast.i;

  x = (x << 16) | (x >> 16);  // swap
  x = ror(x, 7);

  uint8_t k = x & 0xff;

  k ^= 0x80;

  uint8_t orig_k=k;
  uint8_t b = k & 0x80;
  k+=k;
  uint8_t a = k & 0x80;
  if (a != b) {
    //printf("Overflow? (a=%d b=%d)\n", a, b);
    if (k < orig_k) {
      //printf("Carry must be set\n");
      if ((k != 0x7C) && (k != 0x7E)) {
        //printf("Need to implement FFPTOVF (k=%02X)\n", k);
        //x &= 0xFFFFFEFF;  // clear sign bit out
        //if (!x)... no need for code, all branches lead to 0
        return 0;
      } else {
        k+= 0x80+(2<<1)+1;
        x = ror(x, 1);
        if (k)
          return x;
        return 0;
      }
    } else {
      // FFPOVLW, EXPONENT HIGH - CHECK FOR EXPONENT TOO HIGH, INFINITY, OR NAN
      if (k == 0xFE) {
        x >>= 9;
        if (x) {
          // NaN
          return 0;
        } else {
          // INF
          if (x & 0x100)
            x = 0xFFFFFFFF;
          else
            x = 0xFFFFFF7F;
          return x;
        }
      } else {
        // EXPHI
        if (x & 0x100)
          x = 0xFFFFFFFF;
        else
          x = 0xFFFFFF7F;
        return x;
      }
    }
  }
  // BVS.S    FFPOVF    BRANCH WILL NOT FIT

  k+=(2<<1)+1;
  b = k & 0x80;
  if (a == 0 && b) {
    // EXPHI
    if (x & 0x100)
      x = 0xFFFFFFFF;
    else
      x = 0xFFFFFF7F;
    return x;
  }
  // BVS.S    EXPHI    BRANCH EXPONENT TOO LARGE (OVERFLOW)

  k ^= 0x80;

  x = (x & 0xFFFFFF00) | k;
  x = ror(x, 1);
  return x;
}

int main(int argc, char *argv[]) {
  union {
    float f;
    uint32_t i;
  } _fcast, g;

  for (int i = -5; i <= 5; ++i) {
    float f = i*1;
    _fcast.f = f;
    uint32_t ffp = ffpfieee(f);
    g.f = ffptieee(ffp);
    printf("%f %f FFP: %08X %08X (%08X) ", f, ffptieee(ffp), ffp, _fcast.i, g.i);
    if (_fcast.i != g.i)
      printf(" Mismatch\n");
    else
      printf("\n");
  }
  /* for (int i = 0; i < 10; ++i) { */
  /*  float f = i/10.f; */
  /*  _fcast.f = f; */
  /*  printf("$%08X, ", _fcast.i); */
  /* } */
  /* printf("\n"); */
 
  return 0;
}


All times are GMT +2. The time now is 13:59.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, vBulletin Solutions Inc.

Page generated in 0.04397 seconds with 11 queries