AVRFreaks.net

uint64_t __muldi3_6 (uint64_t c, uint16_t a, uint16_t b)
{
    return c + 0x10000ULL * ((uint32_t) a * b);
}

#define __zero_reg__ r1
#define __tmp_reg__  r0

.macro DEFUN name
.global \name
.func \name
\name:
.endm

.macro ENDF name
.size \name, .-\name
.endfunc
.endm


.text

;; A[0..3]: In: Operand;
;; Out: Result.High (R25:R22)
#define A0  22
#define A1  A0+1
#define A2  A0+2
#define A3  A0+3

;; P[0..7]: Product
#define P0  18
#define P1  P0+1
#define P2  20
#define P3  P2+1
#define P4  26
#define P5  P4+1
#define P6  30
#define P7  P6+1

;; T[0..7]: Result
;; note T[0..3] are never referenced, as they are aliases for P[0..3]
#define T0  18
#define T1  T0+1
#define T2  20
#define T3  T2+1
#define T4  22
#define T5  T4+1
#define T6  24
#define T7  T6+1

#define zero    17

;; R25:R18 = (uint64_t) R25:R22 ^ 2
;; Ordinary ABI-Function
;; 79 cycles, 64 words (128 bytes)

DEFUN __umulsidi3
    ;; prologue
    push    zero

    ;; need a local zero register
    clr     zero
   
    ;; zero the 6 MSB of the product
    clr     P2
    clr     P3
    movw    P4, P2
    movw    P6, P2

    ;; 0 bytes offset
    mul A0,A0  $  movw P0,r0
    ;; P0 is now complete

    ;; 1 byte offset
    ;; R1 is <= 0xfe = 255*255/256  -> first ADC don't set Carry
    mul A0,A1  $  add P1,r0  $  adc P2,r1
                  add P1,r0  $  adc P2,r1  $ adc P3,zero
    ;; P1, P0 are now complete

    ;; 2 bytes offset
    ;; R1 is <= 0xfe and P3 <= 1  -> first ADC don't set Carry
    mul A0,A2  $  add P2,r0  $  adc P3,r1
               $  add P2,r0  $  adc P3,r1  $ adc P4,zero
    mul A1,A1  $  add P2,r0  $  adc P3,r1  $ adc P4,zero
    ;; P2, P1, P0 are now complete

    ;; 3 bytes offset
    mul A0,A3  $  add P3,r0  $  adc P4,r1  $  adc P5,zero
               $  add P3,r0  $  adc P4,r1  $  adc P5,zero
    mul A1,A2  $  add P3,r0  $  adc P4,r1  $  adc P5,zero
               $  add P3,r0  $  adc P4,r1  $  adc P5,zero
    ;; A0 is done with
    ;; P3, P2, P1, P0 are now complete

    ;; 4 bytes offset
    mul A1,A3  $  add P4,r0  $  adc P5,r1  $  adc P6,zero
               $  add P4,r0  $  adc P5,r1  $  adc P6,zero
    ;; A1, A0 are done with
    mul A2,A2  $  add P4,r0  $  adc P5,r1  $  adc P6,zero
    ;; P4, P3, P2, P1, P0 are now complete

    ;; 5 bytes offset
    mul A2,A3  $  add P5,r0  $  adc P6,r1  $  adc P7,zero
               $  add P5,r0  $  adc P6,r1  $  adc P7,zero
    ;; A2, A1, A0 are done with
    ;; P5, P4, P3, P2, P1, P0 are now complete

    ;; 6 bytes offset
    mul A3,A3  $  add P6,r0  $  adc P7,r1

    ;; A3, A2, A1, A0 are done with
    ;; P7, P6, P5, P4, P3, P2, P1, P0 are now complete

    ;; Move Result P[7..0] to T[7..0] (R25:R18) according to ABI
    ;; P[3..0] already in place
    movw    T4, P4
    movw    T6, P6

    ;; Epilogue
    pop     zero
    clr __zero_reg__
    ret
ENDF __umulsidi3

uint32_t divU4U2(uint32_t n, uint16_t d)
{
  uint8_t s=0;
  uint32_t temp;

  if (d==0) { return MAX_UINT4; }
  if (d==1) { return n; }

  // 'normalise' denominator
  while (d > 4096) { d >>= 1; s++; }

  temp = pgm_read_dword(&reciprocal[d]);  // 32-bit reciprocal
  temp = MulU4U4SH32RND(n, temp) >> s;
  return(temp);
}