/* Reference implementation of ACE16 AE1024, (0, 16)
 Written by:
 Yunjie Yi <yunjie.yi@uwaterloo.ca>
 */
#include "msp430.h"                     ; #define controlled include file

        NAME    main                    ; module name

        PUBLIC  main                    ; make the main label vissible
                                        ; outside this module
        ORG     0FFFEh
        DC16    init                    ; set reset vector to 'init' label

        RSEG    CSTACK                  ; pre-declaration of segment
        RSEG    CODE                    ; place program in 'CODE' segment

init:   MOV     #SFE(CSTACK), SP        ; set up stack

main:   NOP                             ; main program
        MOV.W   #WDTPW+WDTHOLD,&WDTCTL  ; Stop watchdog timer


;;begin main func
        ;;CALL #bgbg     
        
        
;;Pre-save t and SC to the memory(flash)
        MOV.B #0x07, &0xEE00
        MOV.B #0x53, &0xEE02
        MOV.B #0x43, &0xEE04
        MOV.B #0x0a, &0xEE06
        MOV.B #0x5d, &0xEE08
        MOV.B #0xe4, &0xEE0A
        MOV.B #0x9b, &0xEE0C
        MOV.B #0x49, &0xEE0E
        MOV.B #0x5e, &0xEE10
        MOV.B #0xe0, &0xEE12
        MOV.B #0x7f, &0xEE14
        MOV.B #0xcc, &0xEE16
        MOV.B #0xd1, &0xEE18
        MOV.B #0xbe, &0xEE1A
        MOV.B #0x32, &0xEE1C
        MOV.B #0x1a, &0xEE1E
        
        MOV.B #0x1d, &0xEE20
        MOV.B #0x4e, &0xEE22
        MOV.B #0x22, &0xEE24
        MOV.B #0x28, &0xEE26
        MOV.B #0x75, &0xEE28
        MOV.B #0xf7, &0xEE2A
        MOV.B #0x6c, &0xEE2C
        MOV.B #0x25, &0xEE2E
        MOV.B #0x62, &0xEE30
        MOV.B #0x82, &0xEE32
        MOV.B #0xfd, &0xEE34
        MOV.B #0x96, &0xEE36
        MOV.B #0x47, &0xEE38
        MOV.B #0xf9, &0xEE3A
        MOV.B #0x71, &0xEE3C
        MOV.B #0x6b, &0xEE3E

        MOV.B #0x76, &0xEE40
        MOV.B #0xaa, &0xEE42
        MOV.B #0x88, &0xEE44
        MOV.B #0xa0, &0xEE46
        MOV.B #0x2b, &0xEE48
        MOV.B #0xdc, &0xEE4A
        MOV.B #0xb0, &0xEE4C
        MOV.B #0xe9, &0xEE4E
        MOV.B #0x8b, &0xEE50
        MOV.B #0x09, &0xEE52
        MOV.B #0xcf, &0xEE54
        MOV.B #0x59, &0xEE56
        MOV.B #0x1e, &0xEE58
        MOV.B #0xb7, &0xEE5A
        MOV.B #0xc6, &0xEE5C
        MOV.B #0xad, &0xEE5E
;;;    input SC then:    
        MOV.B #0x50, &0xEF00
        MOV.B #0x28, &0xEF02
        MOV.B #0x14, &0xEF04
        MOV.B #0x5c, &0xEF06
        MOV.B #0xae, &0xEF08
        MOV.B #0x57, &0xEF0A
        MOV.B #0x91, &0xEF0C
        MOV.B #0x48, &0xEF0E
        MOV.B #0x24, &0xEF10
        MOV.B #0x8d, &0xEF12
        MOV.B #0xc6, &0xEF14
        MOV.B #0x63, &0xEF16        
        MOV.B #0x53, &0xEF18
        MOV.B #0xa9, &0xEF1A
        MOV.B #0x54, &0xEF1C
        MOV.B #0x60, &0xEF1E

        MOV.B #0x30, &0xEF20
        MOV.B #0x18, &0xEF22
        MOV.B #0x68, &0xEF24
        MOV.B #0x34, &0xEF26
        MOV.B #0x9a, &0xEF28
        MOV.B #0xe1, &0xEF2A
        MOV.B #0x70, &0xEF2C
        MOV.B #0x38, &0xEF2E
        MOV.B #0xf6, &0xEF30
        MOV.B #0x7b, &0xEF32
        MOV.B #0xbd, &0xEF34
        MOV.B #0x9d, &0xEF36
        MOV.B #0xce, &0xEF38
        MOV.B #0x67, &0xEF3A
        MOV.B #0x40, &0xEF3C
        MOV.B #0x20, &0xEF3E
        
        MOV.B #0x10, &0xEF40
        MOV.B #0x4f, &0xEF42
        MOV.B #0x27, &0xEF44
        MOV.B #0x13, &0xEF46
        MOV.B #0xbe, &0xEF48
        MOV.B #0x5f, &0xEF4A
        MOV.B #0x2f, &0xEF4C
        MOV.B #0x5b, &0xEF4E
        MOV.B #0xad, &0xEF50
        MOV.B #0xd6, &0xEF52
        MOV.B #0xe9, &0xEF54
        MOV.B #0x74, &0xEF56        
        MOV.B #0xba, &0xEF58
        MOV.B #0x7f, &0xEF5A
        MOV.B #0x3f, &0xEF5C
        MOV.B #0x1f, &0xEF5E
        
;;finish pre-save

;;load initial value to registerat locations from 0xEdA6 to 0xEd80
    
        MOV.W #0x0000, &0xEDA6  ;;most left data in the fig.
        MOV.W #0x0000, &0xEDA4
        MOV.W #0x0000, &0xEDA2
        MOV.W #0x0000, &0xEDA0

        MOV.W #0x0000, &0xED9E
        MOV.W #0x0000, &0xED9C
        MOV.W #0x0000, &0xED9A
        MOV.W #0x0000, &0xED98

        MOV.W #0x0000, &0xED96
        MOV.W #0x0000, &0xED94
        MOV.W #0x0000, &0xED92
        MOV.W #0x0000, &0xED90

        MOV.W #0x0000, &0xED8E
        MOV.W #0x0000, &0xED8C
        MOV.W #0x0000, &0xED8A
        MOV.W #0x0000, &0xED88

        MOV.W #0x0000, &0xED86
        MOV.W #0x0000, &0xED84
        MOV.W #0x0000, &0xED82
        MOV.W #0x0000, &0xED80  ;;most right data in the fig. When read from mem, it is from bottom to up


        ;;;;
        CALL #FFUNAE
;;End main func
JMP $                           ; jump to current location '$'
                                        ; (endless loop)

;;this is the mode 
FFUNAE:
        CALL #FFUN3    ;;1 initia
        CALL #FFUNLOAD
        CALL #FFUNKEY1
        CALL #FFUNSAVE
        
        CALL #FFUN3   ;2  initia
        CALL #FFUNLOAD
        CALL #FFUNKEY2
        CALL #FFUNSAVE

        CALL #FFUN3

        MOV.W #0x0000, R15
        
WHIL8:  CMP.W #0x0010, R15
        JGE NEXT8


        CALL #FFUNLOAD
        CALL #FFUNM
        CALL #FFUNSAVE
        CALL #FFUN3
        
        INC R15
        JMP WHIL8
NEXT8:

        CALL #FFUNLOAD
        CALL #FFUNKEY1
        CALL #FFUNSAVE
        
        CALL #FFUN3  ;9  Final

        CALL #FFUNLOAD
        CALL #FFUNKEY2
        CALL #FFUNSAVE
        
        CALL #FFUN3 ;10 Final
        
        
RET

;;this is the AD absorption part in the mode
FFUNLOAD:
        MOV.B &0xEDA7, R12  ;;LOAD A[7] - A[4]
        MOV.B &0xEDA6, R11
        MOV.B &0xEDA5, R10
        MOV.B &0xEDA4, R9
        
        MOV.B &0xED97, R8 ;;LOAD C[7] - C[4]
        MOV.B &0xED96, R7
        MOV.B &0xED95, R6
        MOV.B &0xED94, R5
        
        MOV.B &0xED80, R4 ;LOAD E[0]
RET

;;this is the key absorption part 1 in the mode
FFUNKEY1: ;;(R12 - R4)
        XOR #0x00, R12 ;;XOR with key
        XOR #0x00, R11
        XOR #0x00, R10
        XOR #0x00, R9
        
        XOR #0x00, R8
        XOR #0x00, R7
        XOR #0x00, R6
        XOR #0x00, R5
RET
;;this is the key absorption part 2 in the mode
FFUNKEY2: ;;(R12 - R4)
        XOR #0x00, R12 ;;XOR with key
        XOR #0x00, R11
        XOR #0x00, R10
        XOR #0x00, R9
        
        XOR #0x00, R8
        XOR #0x00, R7
        XOR #0x00, R6
        XOR #0x00, R5
RET

;;this is the message absorption part in the mode 
FFUNM: ;;(R12 - R4)
        XOR #0x00, R12 ;;XOR with key
        XOR #0x00, R11
        XOR #0x00, R10
        XOR #0x00, R9
        
        XOR #0x00, R8
        XOR #0x00, R7
        XOR #0x00, R6
        XOR #0x00, R5
        
        XOR #0x0002, R4
RET


FFUNSAVE:
        MOV.B R12, &0xEDA7   ;;save A[7] - A[4]
        MOV.B R11, &0xEDA6
        MOV.B R10, &0xEDA5
        MOV.B R9, &0xEDA4
        
        MOV.B R8, &0xED97   ;;save C[7] - C[4]
        MOV.B R7, &0xED96
        MOV.B R6, &0xED95
        MOV.B R5, &0xED94
        
        MOV.B R4, &0xED80 ;;save E[0]
RET

;;this is permutation box 
FFUN3:   ;;LOOP the sliscp-light

        PUSH R4
        PUSH R5
        PUSH R6
        PUSH R7
        PUSH R8
        PUSH R9
        PUSH R10
        PUSH R11
        PUSH R12
        PUSH R13
        PUSH R14
        PUSH R15
        
        MOV.W #0xEE00, &0xEE80
        MOV.W #0xEF00, &0xEE84
        MOV.W #0x0001, &0xEE88

        MOV.W #0x0000, R14
        
WHIL3:  CMP.W #0x0010, R14
        JGE NEXTC
        CALL #FFUN2
        INC R14
        JMP WHIL3
NEXTC:

        POP R15
        POP R14
        POP R13
        POP R12
        POP R11
        POP R10
        POP R9
        POP R8
        POP R7
        POP R6
        POP R5
        POP R4

RET


;;this is permutation round
FFUN2:  ;; PASS data(&FD3C-&FD00), and R14(t)(&FB00-FB48) with Vector (&FC80), 
          ;;R15(sc)(&FC00-FC48) with Vector (&FC84), Indicator(&FC88)
            ;;...($FDBC-FD80) is used as temperaor ram data to save odd number operation data
        PUSH R14
        
        MOV.W &0xEE80, R12  ;;Get the current address of the t
        MOV.B @R12, R14  ;;save the t to r14
        MOV.W &0xEE84, R12  ;;Get the current address of the SC
        MOV.B @R12, R15       ;; sacve the sc to r15
        
        MOV.W &0xEE88, R13
        CLRC
        RRA.W R13
        JNC SaveToMemR  ;;if R13=0, then jmp to SaveToMem1

SaveToMemL:
        MOV.W &0xEDA6, R7  ;;LOAD A DATE_____
        MOV.W &0xEDA4, R6
        MOV.W &0xEDA2, R5
        MOV.W &0xEDA0, R4
        
        CALL #FFUN ;;finish the h on the left side        
        
        MOV.W R7, &0xEDD6
        MOV.W R6, &0xEDD4
        MOV.W R5, &0xEDD2
        MOV.W R4, &0xEDD0  ;;Finished A DATE_____
        

        MOV.W &0xED9E, R11  ;;;;LOAD B AND C DATE_____
        MOV.W &0xED9C, R10
        MOV.W &0xED9A, R9
        MOV.W &0xED98, R8
        MOV.W &0xED96, R7
        MOV.W &0xED94, R6
        MOV.W &0xED92, R5
        MOV.W &0xED90, R4

        ADD.W #0x0002, &0xEE80 ;;ADD the t's index by 2
        MOV.W &0xEE80, R12  ;;load the index of current t
        MOV.B @R12, R14  ;;load 2nd t
        
        CALL #FFUN ;;finish the h on the left side
        
        XOR #0xFFFF, R11
        XOR #0xFFFF, R10
        XOR #0xFFFF, R9
        XOR #0xFF00, R8
        XOR R15, R8
        
        XOR R7, R11
        XOR R6, R10
        XOR R5, R9
        XOR R4, R8
        
        MOV.W R11, &0xEDC6
        MOV.W R10, &0xEDC4
        MOV.W R9, &0xEDC2
        MOV.W R8, &0xEDC0
        MOV.W R7, &0xEDDE
        MOV.W R6, &0xEDDC
        MOV.W R5, &0xEDDA
        MOV.W R4, &0xEDD8  ;;;;Finished B AND C DATE_____
        
 
        MOV.W &0xED8E, R11  ;;;;LOAD B AND C DATE_____
        MOV.W &0xED8C, R10
        MOV.W &0xED8A, R9
        MOV.W &0xED88, R8
        MOV.W &0xED86, R7
        MOV.W &0xED84, R6
        MOV.W &0xED82, R5
        MOV.W &0xED80, R4
        
        ADD.W #0x0002, &0xEE80
        ADD.W #0x0002, &0xEE84
        MOV.W &0xEE80, R12  ;;Get the current address of the t
        MOV.B @R12, R14  ;; get 3rd t
        MOV.W &0xEE84, R12  ;;Get the current address of the SC
        MOV.B @R12, R15       ;; get 2nd sc

        CALL #FFUN ;;finish the h on the left side
        
        XOR #0xFFFF, R11
        XOR #0xFFFF, R10
        XOR #0xFFFF, R9
        XOR #0xFF00, R8
        XOR R15, R8 
        
        XOR R7, R11
        XOR R6, R10
        XOR R5, R9
        XOR R4, R8

        ADD.W #0x0002, &0xEE84
        MOV.W &0xEE84, R12  ;;Get the current address of the SC
        MOV.B @R12, R15       ;; get 3nd sc

        XOR #0xFFFF, R7
        XOR #0xFFFF, R6
        XOR #0xFFFF, R5
        XOR #0xFF00, R4
        XOR R15, R4
        ;;;

        XOR &0xEDD6, R7
        XOR &0xEDD4, R6
        XOR &0xEDD2, R5
        XOR &0xEDD0, R4

        MOV.W R11, &0xEDE6
        MOV.W R10, &0xEDE4
        MOV.W R9, &0xEDE2
        MOV.W R8, &0xEDE0
        MOV.W R7, &0xEDCE
        MOV.W R6, &0xEDCC
        MOV.W R5, &0xEDCA
        MOV.W R4, &0xEDC8
        
        ADD.W #0x0002, &0xEE80
        ADD.W #0x0002, &0xEE84
        MOV.W #0x0000,&0xEE88
        JMP PermutFinished
        
SaveToMemR:
        MOV.W &0xEDE6, R7  ;;LOAD A DATE_____
        MOV.W &0xEDE4, R6
        MOV.W &0xEDE2, R5
        MOV.W &0xEDE0, R4
        
        CALL #FFUN ;;finish the h on the left side        
        
        MOV.W R7, &0xED96
        MOV.W R6, &0xED94
        MOV.W R5, &0xED92
        MOV.W R4, &0xED90  ;;Finished A DATE_____
        

        MOV.W &0xEDDE, R11  ;;;;LOAD B AND C DATE_____
        MOV.W &0xEDDC, R10
        MOV.W &0xEDDA, R9
        MOV.W &0xEDD8, R8
        MOV.W &0xEDD6, R7
        MOV.W &0xEDD4, R6
        MOV.W &0xEDD2, R5
        MOV.W &0xEDD0, R4

        ADD.W #0x0002, &0xEE80 ;;ADD the t's index by 2
        MOV.W &0xEE80, R12  ;;load the index of current t
        MOV.B @R12, R14  ;;load 2nd t
        
        CALL #FFUN ;;finish the h on the left side
        
        XOR #0xFFFF, R11
        XOR #0xFFFF, R10
        XOR #0xFFFF, R9
        XOR #0xFF00, R8
        XOR R15, R8
        
        XOR R7, R11
        XOR R6, R10
        XOR R5, R9
        XOR R4, R8
        
        MOV.W R11, &0xED86
        MOV.W R10, &0xED84
        MOV.W R9, &0xED82
        MOV.W R8, &0xED80
        MOV.W R7, &0xED9E
        MOV.W R6, &0xED9C
        MOV.W R5, &0xED9A
        MOV.W R4, &0xED98  ;;;;Finished B AND C DATE_____
        
 
        MOV.W &0xEDCE, R11  ;;;;LOAD B AND C DATE_____
        MOV.W &0xEDCC, R10
        MOV.W &0xEDCA, R9
        MOV.W &0xEDC8, R8
        MOV.W &0xEDC6, R7
        MOV.W &0xEDC4, R6
        MOV.W &0xEDC2, R5
        MOV.W &0xEDC0, R4
        
        ADD.W #0x0002, &0xEE80
        ADD.W #0x0002, &0xEE84
        MOV.W &0xEE80, R12 
        MOV.B @R12, R14  
        MOV.W &0xEE84, R12 
        MOV.B @R12, R15      

        CALL #FFUN ;;finish the h on the left side
        
        XOR #0xFFFF, R11
        XOR #0xFFFF, R10
        XOR #0xFFFF, R9
        XOR #0xFF00, R8
        XOR R15, R8 
        
        XOR R7, R11
        XOR R6, R10
        XOR R5, R9
        XOR R4, R8

        ADD.W #0x0002, &0xEE84
        MOV.W &0xEE84, R12  
        MOV.B @R12, R15 

        XOR #0xFFFF, R7
        XOR #0xFFFF, R6
        XOR #0xFFFF, R5
        XOR #0xFF00, R4
        XOR R15, R4
        ;;;

        XOR &0xED96, R7
        XOR &0xED94, R6
        XOR &0xED92, R5
        XOR &0xED90, R4

        MOV.W R11, &0xEDA6
        MOV.W R10, &0xEDA4
        MOV.W R9, &0xEDA2
        MOV.W R8, &0xEDA0
        MOV.W R7, &0xED8E
        MOV.W R6, &0xED8C
        MOV.W R5, &0xED8A
        MOV.W R4, &0xED88
        
        ADD.W #0x0002, &0xEE80
        ADD.W #0x0002, &0xEE84
        MOV.W #0x0001,&0xEE88
        
PermutFinished: 
        POP R14
RET

;;this is Simeck box
FFUN: ;(R7, R6, R5, R4, R14)
        PUSH R8
        PUSH R9
        PUSH R10
        PUSH R11
        PUSH R12
        
        MOV.W #0x0000, R12
WHIL2:  CMP.W #0x0008, R12 ; compare R12 and 8
        JGE NEXTB ; When R13>=8, then branch
        call #SimeckRF
        INC R12
        JMP WHIL2 ; Go back to beginning of WHILE loop
NEXTB:
        POP R12
        POP R11
        POP R10
        POP R9
        POP R8
RET


;; this is Simeck round
SimeckRF: ;(R7, R6, R5, R4, R14)
        PUSH R12
        
        MOV.W R7,R9
        MOV.W R6,R8

        MOV.W #0x0005, R12 ;SET A
        CALL #LCircularShift ; pass r7, R6, R12 into the function.
                                                ; r30, r31 will be changed

        AND R9, R7 ;;first AND operation in the graph
        AND R8, R6

        MOV.W R7, R11 ;;move r7,r6 to save
        MOV.W R6, R10 

        MOV.W R9, R7 ;; retore R7 and R6 to original value
        MOV.W R8, R6



        MOV.W #0x0001, R12;SET c
        CALL #LCircularShift ; pass r19, r18, r17, r16, and r29 into the function.
                                                ; r30, r31 will be changed

        XOR R11, R7
        XOR R10, R6
        XOR R5, R7
        XOR R4, R6 ; then we need to xor rc with r7 and r6, then finish

        XOR #0xFFFF, R7
        MOV.W R14, R12
        BIS #0xFFFE, R12
        XOR R12, R6
        RRA.W R14
        
        MOV.W R9, R5
        MOV.W R8, R4

        POP R12
        
RET


;;circular shift function
LCircularShift: ;(R7, R6, R12), where R12 is the number of shifts in bit
        PUSH R13

	MOV.W #0x0000,R13
        
WHIL:	CMP.W R12, R13 ; compare R12 and R13
        JGE NEXTA ; When R14>=R12, then branch
                      ;circular left shift begin for 2 Registers
        CLRC ; clear the carry
        RLA.W R7  ; shift R5 to the left by 1 bit
        JNC carry5 ; if carry is 0, then branch

        RLA.W R6
        BIS #0x0001, R6
        JNC carryFinish
        BIS #0x0001, R7
        JMP carryFinish
carry5:
        RLA.W R6 ; shift R
        JNC carryFinish ; if carry is 0, then branch
        BIS #0x0001, R7
carryFinish:
        
;finished the circular left shift for 4 reginsters R19:R18:R17:R16
        inc R13 ; n++
        JMP WHIL ; Go back to beginning of WHILE loop
NEXTA:
        POP R13
RET
        
;;End sub function        
        END