/ / (c) copyright 1983 by the Vrije Universiteit, Amsterdam, The Netherlands. / See the copyright notice in the ACK home directory, in the file "Copyright". /------------------------------------------------------------------------------ / / This is an interpreter for EM programs with no virtual memory / which is adapted from an EM1 interpreter by Hans van Staveren / by Evert Wattel / Vrije Universiteit / Amsterdam / / Memory layout: / / interpreter em-text pd global tables heap unused stack / __________________________________________________________________ / | | | | | | | | | | / | | | | | | | | | | / | 1 | 2 | 3 | 4 | 5 | 6 | | 7 | 8| / | | | | | | | | | | / |____________|_______|____|_______|_______|_____|______|______|__| / / 1: Interpreter text+data+bss / 2: EM text / 3: EM procedure descriptors / 4: EM global data area / 5: flow, count and profile tables / 6: EM heap area / 7: EM local data and stack / 8: Arguments to interpreter / / Assembly time flags: / .test : controls checking for undefined variables,nil pointers, / array indices, etc.... / .prof : controls generation of a runtime profile / .opfreq: controls runtime frequency count per opcode / .flow : controls generation of a flow bitmap / .count : controls generation of a flow count / .last : controls generation of file with last 16 / consecutive blocks of lines executed / / Register layout: / pcx = EM programcounter / lb = EM base-address / nxt = address of start of interpreter loop / / The general structure of this interpreter is as follows: / The opcode byte of the instruction is placed in r0 / with sign-extension and multiplied by 2. / If .opfreq is nonzero each occurence of each opcode is counted. / Then, if .prof is nonzero an estimation of the time required / to execute the instruction is added to a counter associated / with the source-line number. This estimation is roughly the / number of memory-cycles needed. At the end of this accounting / loprof points to the loword of the double precision counter. / This can be used by individual execution routines to add some / more to the counter depending on their operand. / / NOTE: This interpreter can be loaded in separate I and D space / / /------------------------------------------------------------------------------ / Declaring of some constants /------------------------------------------------------------------------------ nxt = r4 pcx = r3 lb = r2 statd = -8. unixextra= 1280. / extra memory asked by heap overflow und = 100000 / undefined memory pattern signext = 177400 / high bits for signextension EINVAL = 22. / UNIX error code for bad signal / Interpreter options .float = 1 .opfreq = 0 .last = 1 V7 = 1 V6 = 0 VPLUS = 0 HARDWARE_FP = 1 /------------------------------------------------------------------------------ / EM1 machine errors (in the range 0-63) /------------------------------------------------------------------------------ EARRAY = 0. ERANGE = 1. ESET = 2. EIOVFL = 3. EFOVFL = 4. EFUNFL = 5. EIDIVZ = 6. EFDIVZ = 7. EIUND = 8. EFUND = 9. ECONV = 10. ESTACK = 16. EHEAP = 17. EILLINS = 18. EODDZ = 19. ECASE = 20. EMEMFLT = 21. EBADPTR = 22. EBADPC = 23. EBADLAE = 24. EBADMON = 25. EBADLIN = 26. EBADGTO = 27. /------------------------------------------------------------------------------ / Declaring of some instructions unknown to the assembler /------------------------------------------------------------------------------ next = 10407 / = mov nxt,pc; jump to decode loop rti = 2 / return from interrupt iot = 4 / force core dump stst = 170300^tst / store floating point status indir = 0 / for sys indir exit = 1 fork = 2 read = 3 write = 4 open = 5 close = 6 creat = 8. break = 17. alarm = 27. pause = 29. sleep = 35. signal = 48. /------------------------------------------------------------------------------ / External references /------------------------------------------------------------------------------ .globl _end / /------------------------------------------------------------------------------ / Now the real program starts /------------------------------------------------------------------------------ startoff: mov sp,r0 mov sp,ml mov sp,filb add $2,filb / pointer to argv in filb for error message dec (r0) mov (r0)+,argc / pass to userprogram later bgt 0f / go for argument mov $emfile,forward+2 / e.out is load file default mov $forward+2,argv br 1f 0: tst (r0)+ / skip interpreter name mov r0,argv / pass to userprogram later mov (r0),forward+2 / argv filename to open call 1: .if V7 tst (r0)+ / increment r0 look for last arg bne 1b mov r0,environ .endif sys indir;forward .data forward: sys open;0;0 emfile: .even .text jes badarg mov r0,saver0 / save filedescriptor mov r0,r5 / duplicate filedescriptor sys read;header;16. / skip first header jes badarg / failed mov r5,r0 / recall fildes sys read;header;16. / read second header jes badarg / failed cmp r0,$16. / long enough ? jne badarg / no. mov $_end,r0 / Bottom em-text mov r0,pb / program base add txtsiz,r0 / reserve space for text mov r0,pd / set up proc.descr base mov nprocs, r3 / number of procs ash $2,r3 / proc. descr is 4 bytes .if .count +.prof + .flow mul $3,r3 / or 12 bytes .endif add r3,r0 / reserve space mov r0,eb / top of pd space mov r0,r3 / base for data fill add szdata,r0 / size of external data jcs toolarge / too much text and data mov r0,globmax / maximum global add $1280.,r0 / bit extra core for setup mov r0,sybreak+2 / set up for core claim sys indir;sybreak / ask for the core jes toolarge / too much, sorry mov txtsiz,leescal+4 / set up for text read mov pb,leescal+2 / start address text read mov r5,r0 / file descriptor input sys indir;leescal / read!! .data leescal: 1: sys read;0;0 / read call .text lblread: / hier is nu dus de tekst ingelezen. De sybreak voor de / tabellen en de data moet opnieuw gebeuren. .if .last mov $47.,r0 mov $lasttab,r5 3: clr (r5)+ sob r0,3b mov $-1,(r5) sub $96.,r5 mov r5,linused .endif lblfloat: .if .float sys signal;8.;sig8 / catch floating exception ldfps $7600 movf $50200,fr3 / load 2^32 in fr3 for conversion / unsigned to float .endif sys signal;11.;sig11 / catch segmentation violation sys signal;12.;sig12 / catch bad system calls / We make a 1024 buffer for reading in / data descriptors. When the number of / bytes in the buffer is less than 512 / we read another block. Descriptors of / more than 512 bytes are not allowed. / This is no restriction since they do / not fit in the descriptor format. lblbuf: sub $02000,sp / new buffer bottom tst (sp) / ask for core mov sp,r4 / pointer in descriptor mov saver0,r0 / recall fildes clr r1 / clear registers for byte clr r2 / format instruction and data mov sp,r5 / copy mov r5,leescal+2 / set up for read mov $02000,leescal+4 / idem sys indir;leescal / read jes badarg / read failed cmp $02000,r0 / not yet eof? bgt 0f / eof encountered add $01000,r5 / buffer middle mov r5,saver1 / save buffermiddle to compare br datloop / start data initialization 0: add r0,r5 / now pointer at top of file mov r5,saver1 / still set up for compare datloop: cmp r4,saver1 / descriptor over middle? blt 9f / no? go ahead jsr pc,blshift / yes? shift block down, read next 9: dec ndatad / still data to initialize? blt finito / no? go on movb (r4)+,r1 / opcode descriptor beq 0f / if 0 then go there mov r3,r5 / copy data pointer clr r2 / unsigned integer byte bisb (r4)+,r2 / "ored" in for data size asl r1 / make opcode even mov datswi(r1),pc / main data swich .data datswi: 0; dat1; dat2; dat3; dat4; dat5; dat6; dat6; dofloat .text dat3: asl r2 / multiply with 2 dat2: 2: movb (r4)+,(r3)+ / copy byte from buffer to data sob r2,2b / until r2 is 0 br datloop / next datadescriptor dat4: mov eb,r0 / external base should be added br 2f / for data pointers dat5: mov pb,r0 / and program base for procedures 2: movb (r4)+,(r3) / move in first byte of pointer movb (r4)+,1(r3) / move in second byte of pointer add r0,(r3)+ / add pointer base sob r2,2b / jump back if there is more br datloop / next data descriptor dat1: mov $und,(r3)+ / reserve words with undefineds sob r2,dat1 / jump back if more br datloop / next data descriptor 0: mov r3,r1 / copy data pointer (odd register) sub r5,r1 / subtract previous pointer movb (r4)+,(r3) / copy first byte of operand movb (r4)+,1(r3) / copy second byte mul (r3),r1 / the number of bytes to copy 1: movb (r5)+,(r3)+ / is the product of the operand sob r1,1b / and the number of bytes in the br datloop / previous operation dat6: add r2,r3 / new data pointer, the old is mov r3,r0 / still in r5 asr r2 / divide by 2 beq 6f / case 1 byte is special sub $2,r0 / this is the least significant / byte in PDP11-standard 2: movb (r4)+,(r0)+ / copy low byte movb (r4)+,(r0) / copy high byte sub $3,r0 / next lowest byte sob r2,2b / jump if not ready br datloop / next descriptor 6: movb (r4)+,(r5) / copy one byte br datloop / next descriptor blshift: mov saver1,r1 / bottom of top half mov r1,r2 / set up bottom sub $01000,r2 mov $1000,r0 / number to copy mov r0,leescal+4 / amount to read sub r0,r4 / decrease pointer asr r0 / 512 bytes is 256 words 3: mov (r1)+,(r2)+ / copy top half in bottom half sob r0,3b mov saver1,leescal+2 / set up for read blockr: mov saver0,r0 / filedescriptor sys indir;leescal jes badarg clr r1 / clear registers which contain clr r2 / descriptor bytes later cmp $01000,r0 / look if eof is encountered beq 3f / yes? go on add r0,saver1 / no extra read necessary 3: rts pc finito: cmp globmax,r3 / test if data size ok jne badarg / load file error mov eb,filb add $4,filb mov nprocs,r5 / set up for procdesc read mov pd,r3 / proc descriptor base asl r5 / multiply with 4 because asl r5 / procdes is 4 bytes 1: mov saver1,r1 / look what is available sub r4,r1 / in buffer to be read add $3,r1 / let it be a multiple bic $3,r1 / of four sub r1,r5 / subtract what can be read asr r1; asr r1; / divide by four 0: movb (r4)+,(r3)+ / copy byte movb (r4)+,(r3)+ / copy byte movb (r4)+,(r3)+ / copy byte movb (r4)+,(r3)+ / copy byte add pb,-2(r3) / change em-address in pdp-address .if .count + .prof + .flow clr (r3)+ clr (r3)+ clr (r3)+ clr (r3)+ .endif sob r1,0b / look if there is more tst r5 / is there still a descriptor ble 2f; / no? go on jsr pc,blshift / yes? read again br 1b 2: cmp eb,r3 / test if procdes is ok jne badarg / load file error mov saver0,r0 / fildes in r0 sys close / close input load file mov ml,sp / fresh stack mov 2(sp),*filb .if .flow + .count + .prof / |==================| / Here we fill the fields in the procedure | bytes for locals | / descriptor with table information. The |------------------| / procedure descriptor has six fields, | start address | / like described in this picture. We |------------------| / construct a linked list of the proc. | count pointer | / descriptors, such that the defined |------------------| / order of procedures is compatible | first line nr | / with the text order. Thereafter we |------------------| / scan the text for line information to | link next proc | / fill the countpointer and startline |------------------| / field. The link to the first proc. | current file name| / is in firstp, links are descriptor |==================| / start addresses. The last procedure / links to the external base. All lines in the text get a count / number, lines of a procedure get consecutive count numbers, / the procedure count pointer gives the number of the first line. / Count pointer zero is reserved for the case that no line number / is yet defined. makelink: mov pd,r0 / first descriptor mov r0,r3 / points to first proc mov r0,r4 / pd in register mov eb,r5 / eb in register 0: mov r0,r1 / copy old descriptor bottom add $12.,r0 / next descriptor cmp r0,r5 / top of descriptor space bhis 4f / ready? continue 1: cmp 2(r0),2(r1) / compare start addresses bhis 2f / 2(r0) large? follow link sub $12.,r1 / 2(r0) small? previous descriptor cmp r1,r4 / is r1 smaller than pd? bhis 1b / no? try again mov r3,8.(r0) / yes? then r0 has small text address mov r0,r3 / now r3 again points to first proc br 0b / next descriptor 2: mov 8.(r1),r2 / follow link to compare with 2(r0) beq 3f / if 0 then no link defined cmp 2(r0),2(r2) / compare start addresses blo 3f / r0 between r1 and r2 mov r2,r1 / r0 above r2, br 2b / look again. 3: mov r0,8.(r1) / link of r1 points to r0 mov r2,8.(r0) / link of r0 points to r2 br 0b / next descriptor 4: mov r3,firstp / firstp links to first procedure procinf: mov $1,maxcount / countptr for first proc mov r3,r4 / points to first proc 0: mov r3,-(sp) / stack current procedure mov $-1,r1 / minimal line number 0177777 clr r5 / maximum line number on 0 mov 8.(r3),r4 / bottom address next descriptor beq 6f / if 0 last procedure mov 2(r4),r4 / top of current procedure br 2f / start looking for lines 6: mov pd,r4 / top of last procedure 2: mov 2(r3),r3 / start text address procedure 8: movb (r3)+,r2 / first opcode for scanning cmp $-2,r2 / case escape beq 1f / escape treated at label 1 cmp $-106.,r2 / case lni blt 7f / ordinary skip at label 7 beq 2f / lni treated at label 2 cmp $-108.,r2 / case lin.l bgt 7f / ordinary skip at label 7 beq 3f / lin.l at label 3 clr r0 / lin.s0 treated here bisb (r3)+,r0 / line number in r0 br 4f / compares at label 4 2: inc r0 / lni increases line number br 4f / compares at label 4 3: jsr pc,wrdoff / get 2 byte number 4: cmp r1,r0 / look if r0 less than minimum blo 5f / nothing to declare mov r0,r1 / r0 new minimum 5: cmp r0,r5 / look if r0 more than maximum blo 9f / nothing spectacular mov r0,r5 / r0 new maximum br 9f / line processed 1: clr r2 bisb (r3)+,r2 / escaped instruction opcode add $128.,r2 / ready for table entry 7: movb skipdisp(r2),r2 / skip the required number of bytes add r2,r3 9: cmp r3,r4 / still more text in this proc? blt 8b / yes? again filpd: mov (sp)+,r3 / get bottom descriptor back sub r1,r5 / number of lines encountered bcs 9f / no lines then no file information mov maxcount,4(r3) / this is the count pointer mov r1,6(r3) / minimum line in descriptor inc r5 add r5,maxcount / this is the new maximum 9: mov 8.(r3),r3 / follow link to next procedure bne 0b / restart .data .byte 2; .byte 2; .byte 0; .byte 0; .byte 1; .byte 1; .byte 1; .byte 0; .byte 0; .byte 2; .byte 1; .byte 0; .byte 1; .byte 0; .byte 0; .byte 1; .byte 1; .byte 1; .byte 0; .byte 0; .byte 2; .byte 1; .byte 0; .byte 2; .byte 0; .byte 1; .byte 1; .byte 2; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 2; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 2; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 2; .byte 2; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 1; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 1; .byte 1; .byte 1; .byte 1; .byte 0; .byte 2; .byte 1; .byte 1; .byte 1; .byte 2; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 2; .byte 2; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 0; .byte 2; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 2; .byte 1; .byte 0; .byte 0; .byte 1; .byte 2; .byte 7; .byte 5; skipdisp: .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 2; .byte 0; .byte 0; .byte 1; .byte 1; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 2; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 2; .byte 1; .byte 1; .byte 1; .byte 1; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 1; .byte 1; .byte 0; .byte 1; .byte 0; .byte 2; .byte 0; .byte 2; .byte 1; .byte 0; .byte 0; .byte 0; .byte 1; .byte 1; .byte 0; .byte 1; .byte 2; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; .byte 1; /escaped opcodes .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 0; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 0; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 0; .byte 0; .byte 0; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 1; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 0; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 0; .byte 1; .byte 0; .byte 0; .byte 0; .byte 2; .byte 0; .byte 2; .byte 0; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 2; .byte 0; .byte 2; .byte 0; .byte 1; .byte 2; .byte 0; .byte 0; .byte 2; .text mov globmax,r3 / bottom of table space mov r3,r5 / copy .if .prof mov r3,ltime / set up pointer to base mov r3,hiprof mov r3,loprof add $2,loprof mov maxcount,r0 / number of lines inc r0 asl r0 / four byter per prof count asl r0 add r0,r3 mov r0,profsiz .endif .if .flow mov r3,lflow / set up pointer to base mov maxcount,r0 ash $-3,r0 / divide by 8 add $2,r0 bic $1,r0 / Rounded up to an integral number of words add r0,r3 mov r0,flowsiz .endif .if .count mov r3,lcount / set up pointer mov maxcount,r0 inc r0 ash $2,r0 / multiply by 4 add r0,r3 mov r0,countsiz .endif mov r3,tblmax cmp r3,sybreak+2 / core available for tables? blos 2f mov r3,sybreak+2 sys indir;sybreak / ask for core 2: sub r5,r3 / this amount of space required asr r3 2: clr (r5)+ / clear table space sob r3,2b .endif .if [1 - .count] * [1 - .flow] * [1 - .prof] mov globmax,tblmax .endif / start calling sequence here calseq: mov tblmax,hp mov pd,r3 / top of em-text and top of stack clr r2 / are dummy return values mov environ,-(sp) / setup environment pointer mov argv,-(sp) / setup argument pointer mov *argv,*filb / setup first file message mov argc,-(sp) / setup argument count mov entry.,-(sp) / start procedure to call precal: mov $loop,r4 / main loop address in r4 jbr cai.z / according to the cai noarg: mov r0,argv mov $0f,r0; jbr rude_error badarg: mov $1f,r0; jbr rude_error toolarge:mov $2f,r0; jbr rude_error .data 0: 1: 2: .even .text dofloat: jsr pc,atof mov r5,r3 / restore r3 / Assumed that the result is 8 / bytes Recall r2 and move the / amount of bytes asked for clr r1 / restore for opcode sub $8.,r2 / 8 bytes? beq 1f / yes! later, 4 bytes next movfo fr0,-(sp) / push result mov (sp)+,(r3)+ / write result in data mov (sp)+,(r3)+ / idem jbr datloop / next loop 1: movf fr0,-(sp) / push result mov (sp)+,(r3)+ / write result in data mov (sp)+,(r3)+ / write result in data mov (sp)+,(r3)+ / write result in data mov (sp)+,(r3)+ / write result in data jbr datloop atof: mov r2,-(sp) / save byte count clr -(sp) clrf fr0 clr r2 1: movb (r4)+,r0 / get byte cmp $' ,r0 bge 1b cmpb r0,$'+ beq 1f cmpb r0,$'- bne 2f inc (sp) 1: movb (r4)+,r0 / get next byte 2: sub $'0,r0 cmp r0,$9. bhi 2f jsr pc,digit br 1b inc r2 br 1b 2: cmpb r0,$'.-'0 bne 2f 1: movb (r4)+,r0 / get next byte sub $'0,r0 cmp r0,$9. bhi 2f jsr pc,digit dec r2 br 1b 2: cmpb r0,$'E-'0 beq 3f cmpb r0,$'e-'0 bne 1f 3: clr r3 clr r1 movb (r4)+,r0 / get next byte cmpb r0,$'+ beq 3f cmpb r0,$'- bne 5f inc r3 3: movb (r4)+,r0 / get next byte 5: sub $'0,r0 cmp r0,$9. bhi 3f mul $10.,r1 add r0,r1 br 3b 3: tst r3 bne 3f neg r1 3: sub r1,r2 1: movf $one,fr1 movf $one,fr2 mov r2,-(sp) beq 2f bgt 1f neg r2 1: mulf $twohalf,fr1 mulf $four,fr2 sob r2,1b 2: tst (sp)+ bge 1f divf fr1,fr0 divf fr2,fr0 br 2f 1: mulf fr1,fr0 mulf fr2,fr0 2: tst (sp)+ beq 1f negf fr0 1: mov (sp)+,r2 rts pc digit: cmpf $big,fr0 cfcc blt 1f mulf $ten,fr0 movif r0,fr1 addf fr1,fr0 rts pc 1: add $2,(sp) rts pc / / one = 40200 twohalf = 40440 four = 40600 ten = 41040 big = 56200 huge = 77777 /------------------------------------------------------------------------------ /------------------------------------------------------------------------------ / Main loop of interpreter starts here /------------------------------------------------------------------------------ loop: movb (pcx)+,r0 / pickup opcode + sign extend 9: asl r0 / opcode now -256 .. 254 & even .if .opfreq mov r0,r1 asl r1 / multiply by two again add $1,counttab+514.(r1) / cannot be inc adc counttab+512.(r1) / double precision counters .endif .if .prof add timeinf(r0),*loprof adc *hiprof / double precision .endif mov dispat(r0),pc / fast dispatch /------------------------------------------------------------------------------ / Two byte opcodes come here for decoding of second byte /------------------------------------------------------------------------------ escape1: clr r0 bisb (pcx)+,r0 / fetch second byte no sign extend asl r0 / 0 to 512 & even cmp $0500,r0 / look for righ range jlt e.illins .if .opfreq mov r0,r1 asl r1 / multiply by two again add $1,counttab+1026.(r1) / cannot be inc adc counttab+1024.(r1) / double precision counters .endif .if .prof add time2inf(r0),*loprof adc *hiprof / double precision .endif mov dispae1(r0),pc / fast dispatch /---------------------------------------------------------------------------- escape2: movb (pcx)+,r0 / fetch second byte and sign extend jne e.illins .if .opfreq add $1,counttab+1666. / cannot be inc adc counttab+1664. / double precision counters .endif jbr loc.f / fast dispatch /------------------------------------------------------------------------------ / dispatch tables, first the unescaped opcodes / / name convention is as follows: / each execution routine has as a name the name of the instruction / followed by a dot and a suffix. / suffix can be an integer (sometimes followed by a W), / an 's'or a 'w', followed by an integer, an 'l' ,a 'p' , / a 'n', sometimes followed by a 'w', or a 'z'. / loc.1 routine to execute loc 1 / zge.s0 routine to execute zge 0 thru 255 / lae.w1 routine to execute lae 1024 thru lae 2046 / lof.2W routine to execute lof 2*the word size / lol.pw routine to execute positive lol instructions / loe.l routine to execute all loe instructions / add.z routine to execute instruction without operand / or with operand on the stack. /------------------------------------------------------------------------------ .data lal.p; lal.n; lal.0; lal._1; lal.w0; lal.w_1; lal.w_2; lar.1W ldc.0; lde.lw; lde.w0; ldl.0; ldl.w_1; lfr.1W; lfr.2W; lfr.s0 lil.w_1; lil.w0; lil.0; lil.1W; lin.l; lin.s0; lni.z; loc.l loc._1; loc.s0; loc.s_1; loe.lw; loe.w0; loe.w1; loe.w2; loe.w3 loe.w4; lof.l; lof.1W; lof.2W; lof.3W; lof.4W; lof.s0; loi.l loi.1; loi.1W; loi.2W; loi.3W; loi.4W; loi.s0; lol.pw; lol.nw lol.0; lol.1W; lol.2W; lol.3W; lol._1W; lol._2W; lol._3W; lol._4W lol._5W; lol._6W; lol._7W; lol._8W; lol.w0; lol.w_1; lxa.1; lxl.1 lxl.2; mlf.s0; mli.1W; mli.2W; rck.1W; ret.0; ret.1W; ret.s0 rmi.1W; sar.1W; sbf.s0; sbi.1W; sbi.2W; sdl.w_1; set.s0; sil.w_1 sil.w0; sli.1W; ste.lw; ste.w0; ste.w1; ste.w2; stf.l; stf.1W stf.2W; stf.s0; sti.1; sti.1W; sti.2W; sti.3W; sti.4W; sti.s0 stl.pw; stl.nw; stl.0; stl.1W; stl._1W; stl._2W; stl._3W; stl._4W stl._5W; stl.w_1; teq.z; tgt.z; tlt.z; tne.z; zeq.l; zeq.s0 zeq.s1; zer.s0; zge.s0; zgt.s0; zle.s0; zlt.s0; zne.s0; zne.s_1 zre.lw; zre.w0; zrl._1W; zrl._2W; zrl.w_1; zrl.nw; escape1; escape2 dispat: / dispatch table for unescaped opcodes loc.0; loc.1; loc.2; loc.3; loc.4; loc.5; loc.6; loc.7 loc.8; loc.9; loc.10; loc.11; loc.12; loc.13; loc.14; loc.15 loc.16; loc.17; loc.18; loc.19; loc.20; loc.21; loc.22; loc.23 loc.24; loc.25; loc.26; loc.27; loc.28; loc.29; loc.30; loc.31 loc.32; loc.33; aar.1W; adf.s0; adi.1W; adi.2W; adp.l ; adp.1 adp.2; adp.s0; adp.s_1; ads.1W; and.1W; asp.1W; asp.2W; asp.3W asp.4W; asp.5W; asp.w0; beq.l; beq.s0; bge.s0; bgt.s0; ble.s0 blm.s0; blt.s0; bne.s0; bra.l; bra.s_1; bra.s_2; bra.s0; bra.s1 cal.1; cal.2; cal.3; cal.4; cal.5; cal.6; cal.7; cal.8 cal.9; cal.10; cal.11; cal.12; cal.13; cal.14; cal.15; cal.16 cal.17; cal.18; cal.19; cal.20; cal.21; cal.22; cal.23; cal.24 cal.25; cal.26; cal.27; cal.28; cal.s0; cff.z; cif.z; cii.z cmf.s0; cmi.1W; cmi.2W; cmp.z; cms.s0; csa.1W; csb.1W; dec.z dee.w0; del.w_1; dup.1W; dvf.s0; dvi.1W; fil.l; inc.z; ine.lw ine.w0; inl._1W; inl._2W; inl._3W; inl.w_1; inn.s0; ior.1W; ior.s0 lae.l; lae.w0; lae.w1; lae.w2; lae.w3; lae.w4; lae.w5; lae.w6 /------------------------------------------------------------------------------ / now dispatch table for escaped opcodes /------------------------------------------------------------------------------ dispae1: /dispatch escaped opcodes 1 aar.l; aar.z; adf.l; adf.z; adi.l; adi.z; ads.l; ads.z adu.l; adu.z; and.l; and.z; asp.lw; ass.l; ass.z; bge.l bgt.l; ble.l; blm.l; bls.l; bls.z; blt.l; bne.l; cai.z cal.l; cfi.z; cfu.z; ciu.z; cmf.l; cmf.z; cmi.l; cmi.z cms.l; cms.z; cmu.l; cmu.z; com.l; com.z; csa.l; csa.z csb.l; csb.z; cuf.z; cui.z; cuu.z; dee.lw; del.pw; del.nw dup.l; dus.l; dus.z; dvf.l; dvf.z; dvi.l; dvi.z; dvu.l dvu.z; fef.l; fef.z; fif.l; fif.z; inl.pw; inl.nw; inn.l inn.z; ior.l; ior.z; lar.l; lar.z; ldc.l; ldf.l; ldl.pw ldl.nw; lfr.l; lil.pw; lil.nw; lim.z; los.l; los.z; lor.s0 lpi.l; lxa.l; lxl.l; mlf.l; mlf.z; mli.l; mli.z; mlu.l mlu.z; mon.z; ngf.l; ngf.z; ngi.l; ngi.z; nop.z; rck.l rck.z; ret.l; rmi.l; rmi.z; rmu.l; rmu.z; rol.l; rol.z ror.l; ror.z; rtt.z; sar.l; sar.z; sbf.l; sbf.z; sbi.l sbi.z; sbs.l; sbs.z; sbu.l; sbu.z; sde.l; sdf.l; sdl.pw sdl.nw; set.l; set.z; sig.z; sil.pw; sil.nw; sim.z; sli.l sli.z; slu.l; slu.z; sri.l; sri.z; sru.l; sru.z; sti.l sts.l; sts.z; str.s0; tge.z; tle.z; trp.z; xor.l; xor.z zer.l; zer.z; zge.l; zgt.l; zle.l; zlt.l; zne.l; zrf.l zrf.z; zrl.pw; dch.z; exg.s0; exg.l; exg.z; lpb.z; gto.l /------------------------------------------------------------------------------ / timeinf tables, first the unescaped opcodes / these tables are parallel to the tables dispat and dispae1 / Each entry contains a reasonable estimate of / the number of memory-cycles needed to / execute that instruction. The exact amount cannot be / supplied, since this can depend rather heavily on the / size of the object in set, array case instructions etc. / The table timeinf also contains, added to each entry, / the number of memory-cycles needed to decode the instruction. / This number is currently 6. The number is computed for / the case that all check and runinf options are off. /------------------------------------------------------------------------------ .if .prof 23.; 23.; 12.; 12.; 18.; 17.; 19.; 61. 11.; 31.; 21.; 15.; 20.; 30.; 30.; 31. 20.; 18.; 18.; 19.; 29.; 18.; 13.; 20. 10.; 14.; 13.; 27.; 20.; 20.; 20.; 20. 20.; 23.; 16.; 16.; 16.; 16.; 17.; 38. 14.; 26.; 26.; 26.; 26.; 28.; 26.; 25. 11.; 11.; 11.; 11.; 11.; 11.; 11.; 11. 11.; 11.; 11.; 11.; 16.; 16.; 26.; 24. 24.; 53.; 25.; 25.; 18.; 27.; 44.; 54. 30.; 59.; 53.; 21.; 28.; 19.; 51.; 18. 18.; 21.; 27.; 19.; 20.; 18.; 25.; 16. 16.; 15.; 12.; 24.; 24.; 24.; 24.; 25. 26.; 25.; 15.; 13.; 11.; 11.; 11.; 11. 11.; 16.; 14.; 14.; 14.; 14.; 20.; 16. 16.; 21.; 16.; 16.; 16.; 16.; 16.; 16. 26.; 16.; 10.; 10.; 15.; 24.; 10.; 40. timeinf: 9.; 10.; 10.; 10.; 10.; 10.; 10.; 10. 10.; 10.; 10.; 10.; 10.; 10.; 10.; 10. 10.; 10.; 10.; 10.; 10.; 10.; 10.; 10. 10.; 10.; 10.; 10.; 10.; 10.; 10.; 10. 10.; 10.; 48.; 53.; 21.; 28.; 20.; 10. 10.; 12.; 13.; 11.; 44.; 11.; 11.; 11. 11.; 11.; 27.; 21.; 17.; 17.; 17.; 17. 81.; 17.; 17.; 21.; 12.; 12.; 11.; 12. 54.; 54.; 54.; 54.; 54.; 54.; 54.; 54. 54.; 54.; 54.; 54.; 54.; 54.; 54.; 54. 54.; 54.; 54.; 54.; 54.; 54.; 54.; 54. 54.; 54.; 54.; 54.; 54.; 41.; 49.; 37. 40.; 53.; 53.; 51.; 60.; 24.; 41.; 11. 20.; 19.; 10.; 53.; 30.; 29.; 11.; 30. 20.; 15.; 15.; 15.; 19.; 44.; 37.; 36. 25.; 19.; 19.; 19.; 19.; 19.; 19.; 19. /------------------------------------------------------------------------------ / time2inf table for escaped opcodes / cycles necessary for decoding is already accounted for in timeinf /------------------------------------------------------------------------------ time2inf: 57.; 46.; 61.; 50.; 37.; 26.; 30.; 19. 45.; 34.; 52.; 41.; 37.; 42.; 31.; 21. 21.; 21.; 91.; 108.; 97.; 21.; 21.; 53. 60.; 56.; 55.; 26.; 53.; 42.; 62.; 51. 72.; 61.; 72.; 61.; 38.; 27.; 40.; 29. 53.; 46.; 54.; 38.; 23.; 30.; 30.; 28. 36.; 45.; 34.; 61.; 50.; 39.; 28.; 44. 33.; 68.; 57.; 68.; 57.; 30.; 28.; 54. 45.; 44.; 33.; 70.; 59.; 22.; 27.; 28. 29.; 37.; 28.; 27.; 11.; 47.; 40.; 21. 20.; 35.; 33.; 61.; 50.; 34.; 23.; 39. 28.; 500.; 47.; 36.; 41.; 30.; 100.; 38. 27.; 62.; 39.; 28.; 44.; 33.; 88.; 77. 92.; 81.; 32.; 68.; 57.; 61.; 50.; 37. 26.; 33.; 22.; 45.; 34.; 29.; 28.; 30. 28.; 61.; 52.; 16.; 28.; 27.; 11.; 30. 19.; 36.; 25.; 32.; 21.; 36.; 25.; 31. 39.; 32.; 32.; 14.; 14.; 117.; 45.; 34. 31.; 22.; 20.; 20.; 20.; 20.; 20.; 27. 16.; 26.; 17.; 39.; 47.; 36.; 10.; 29. .endif .text /------------------------------------------------------------------------------ / LOAD CONSTANT, LOAD LOCAL, STORE LOCAL /------------------------------------------------------------------------------ loc.0: clr -(sp) next loc.1: loc.2: loc.3: loc.4: loc.5: loc.6: loc.7: loc.8: loc.9: loc.10: loc.11: loc.12: loc.13: loc.14: loc.15: loc.16: loc.17: loc.18: loc.19: loc.20: loc.21: loc.22: loc.23: loc.24: loc.25: loc.26: loc.27: loc.28: loc.29: loc.30: loc.31: loc.32: loc.33: asr r0 / make multiplication undone mov r0,-(sp) next loc._1: mov $-1,-(sp) next loc.s0: clr r0 br 2f loc.s_1: mov $-400,r0 2: bisb (pcx)+,r0 mov r0,-(sp) next lpi.l: / let op, dit is een pointer / zonder offset op het moment! loc.l: jsr pc,wrdoff mov r0,-(sp) next ldc.0: clr -(sp) clr -(sp) next ldc.l: jsr pc,wrdoff mov r0,-(sp) sxt -(sp) next loc.f: jsr pc,wrdoff; mov r0,r1 jsr pc,wrdoff; mov r0,-(sp) mov r1,-(sp); next /__________________________________________________________________________ lol.0: mov 010(r2),-(sp); next lol.1W: mov 012(r2),-(sp); next lol.2W: mov 014(r2),-(sp); next lol.3W: mov 016(r2),-(sp); next lol._1W: mov -02(r2),-(sp); next lol._2W: mov -04(r2),-(sp); next lol._3W: mov -06(r2),-(sp); next lol._4W: mov -010(r2),-(sp); next lol._5W: mov -012(r2),-(sp); next lol._6W: mov -014(r2),-(sp); next lol._7W: mov -016(r2),-(sp); next lol._8W: mov -020(r2),-(sp); next lol.w0: clr r0; bisb (pcx)+,r0 5: asl r0; add r2,r0 mov 010(r0),-(sp); next lol.w_1: mov $-400,r0; bisb (pcx)+,r0 2: asl r0; add r2,r0 mov (r0),-(sp); next lol.pw: jsr pc,wrdoff; br 5b lol.nw: jsr pc,wrdoff; br 2b /------------------------------------------------------------------------------ ldl.0: mov 10.(r2),-(sp); mov 8.(r2),-(sp); next ldl.w_1: mov $-400,r0; bisb (pcx)+,r0 2: asl r0; add r2,r0 mov 2(r0),-(sp); mov (r0),-(sp); next ldl.pw: jsr pc,wrdoff; asl r0 add r2,r0; mov 10.(r0),-(sp) mov 8.(r0),-(sp); next ldl.nw: jsr pc,wrdoff; br 2b /------------------------------------------------------------------------------ loe.lw: jsr pc,wrdoff; br 2f loe.w0: loe.w1: loe.w2: loe.w3: loe.w4: asr r0; add $0144,r0 swab r0; bisb (pcx)+,r0 2: asl r0; add eb,r0 mov (r0),-(sp); next lde.lw: jsr pc,wrdoff; br 2f lde.w0: clr r0; bisb (pcx)+,r0 2: asl r0; add eb,r0 mov 2(r0),-(sp); mov (r0),-(sp); next /------------------------------------------------------------------------------ lil.0: clr r0; br 1f lil.1W: mov $1,r0; br 1f lil.pw: jsr pc,wrdoff; br 1f lil.w0: clr r0; bisb (pcx)+,r0 1: add $04,r0 2: asl r0; add r2,r0 mov (r0),-(sp); jsr pc,chckptr mov *(sp),(sp); next lil.w_1: mov $-400,r0; bisb (pcx)+,r0; br 2b lil.nw: jsr pc,wrdoff; br 2b /------------------------------------------------------------------------------ lof.l: jsr pc,wrdoff 1: jsr pc,chckptr; add (sp)+,r0; mov (r0),-(sp); next lof.1W: lof.2W: lof.3W: lof.4W: add $0276,r0; br 1b lof.s0: clr r0; bisb (pcx)+,r0; br 1b ldf.l: jsr pc,wrdoff; add (sp)+,r0 mov 2(r0),-(sp); mov (r0),-(sp); next /------------------------------------------------------------------------------ lal.p: jsr pc,wrdoff 5: add r2,r0; add $8.,r0 mov r0,-(sp); next lal.0: mov r2,-(sp); add $8.,(sp); next lal.w0: clr r0; bisb (pcx)+,r0 asl r0; br 5b lal.n: jsr pc,wrdoff; br 2f lal._1: mov $-1,r0 2: add r2,r0; mov r0,-(sp); next lal.w_1: mov $-400,r0 3: bisb (pcx)+,r0; asl r0; br 2b lal.w_2: mov $-1000,r0; br 3b lae.l: jsr pc,wrdoff; br 1f lae.w0: lae.w1: lae.w2: lae.w3: lae.w4: lae.w5: lae.w6: asr r0 sub $0171,r0; swab r0 bisb (pcx)+,r0; asl r0 1: add eb,r0; .if .test cmp globmax,r0; bhi 1f; jsr pc,e.badlae; .endif 1: mov r0,-(sp); next /------------------------------------------------------------------------------ lxl.1: mov $1,r0; br 1f lxl.2: mov $2,r0; br 1f lxl.l: jsr pc,wrdoff bgt 1f; jlt e.oddz mov r2,-(sp); next 1: mov r2,r1 2: mov 8(r1),r1; sob r0,2b mov r1,-(sp); next lxa.1: mov $1,r0; br 1f lxa.l: jsr pc,wrdoff; bgt 1f jlt e.oddz; mov r2,-(sp) add $10,(sp); next 1: mov r2,r1 2: mov 8(r1),r1; sob r0,2b add $10,r1; mov r1,-(sp); next /------------------------------------------------------------------------------ loi.l: jsr pc,wrdoff; br 2f loi.1W: loi.2W: loi.3W: loi.4W: add $260,r0; br 1f loi.s0: clr r0; bisb (pcx)+,r0 2: cmp $1,r0; beq loi.1 1: jsr pc,chckptr; mov (sp)+,r1; add r0,r1 asr r0; jcs e.oddz 1: mov -(r1),-(sp); sob r0,1b; next loi.1: jsr pc,chckptb; mov (sp),r1; clr r0 bisb (r1),r0; mov r0,(sp); next los.z: mov (sp)+,r0 br 0f los.l: jsr pc,wrdoff 0: cmp $04,r0 beq 4f cmp $02,r0; beq 3f jbr e.oddz 4: mov (sp)+,r0 3: mov (sp)+,r0; br 2b /------------------------------------------------------------------------------ /Store group /------------------------------------------------------------------------------ stl.pw: jsr pc,wrdoff; asl r0; br 0f stl.0: clr r0; br 0f stl.1W: mov $2,r0 0: add r2,r0; mov(sp)+,8.(r0); next stl.nw: jsr pc,wrdoff; br 0f stl.w_1: mov $-400,r0; bisb (pcx)+,r0 0: asl r0; add r2,r0 mov (sp)+,(r0); next stl._1W: mov (sp)+,-2(r2); next stl._2W: mov (sp)+,-4(r2); next stl._3W: mov (sp)+,-6(r2); next stl._4W: mov (sp)+,-10(r2); next stl._5W: mov (sp)+,-12(r2); next sdl.w_1: mov $-400,r0; bisb (pcx)+,r0 0: asl r0; add r2,r0 2: mov (sp)+,(r0)+; mov (sp)+,(r0); next sdl.nw: jsr pc,wrdoff; br 0b sdl.pw: jsr pc,wrdoff; asl r0 add r2,r0; add $8.,r0; br 2b /------------------------------------------------------------------------------ sde.l: jsr pc,wrdoff; add eb,r0 br 2b ste.lw: jsr pc,wrdoff; br 1f ste.w0: clr r0; br 0f ste.w1: mov $400,r0; br 0f ste.w2: mov $1000,r0 0: bisb (pcx)+,r0 1: asl r0; add eb,r0 mov (sp)+,(r0); next /------------------------------------------------------------------------------ stf.l: jsr pc,wrdoff; br 6f stf.1W: mov $2,r0; br 6f stf.2W: mov $4,r0; br 6f stf.s0: clr r0; bisb (pcx)+,r0 6: add (sp)+,r0; br 7f sdf.l: jsr pc,wrdoff; add (sp)+,r0 jbr 2b /------------------------------------------------------------------------------ sil.w0: clr r0; bisb (pcx)+,r0 5: asl r0; add r2,r0 mov 8.(r0),r0; br 7f sil.w_1: mov $-400,r0; bisb (pcx)+,r0 2: asl r0; add r2,r0 mov (r0),r0; 7: mov (sp),r1; mov r0,(sp); jsr pc,chckptr; mov r1,*(sp)+; next sil.pw: jsr pc,wrdoff; br 5b sil.nw: jsr pc,wrdoff; br 2b /------------------------------------------------------------------------------ sti.1: jsr pc,chckptb; mov (sp)+,r1; movb (sp)+,(r1); next sti.1W: sti.2W: sti.3W: sti.4W: add $114,r0; br 1f sti.s0: clr r0; bisb (pcx)+,r0; br 1f sti.l: jsr pc,wrdoff 1: asr r0; beq 3f jcs e.oddz; jsr pc,chckptr; mov (sp)+,r1 2: mov (sp)+,(r1)+; sob r0,2b; next 3: jcs sti.1; jbr e.oddz sts.l: jsr pc,wrdoff 0: cmp $2,r0; beq 2f cmp $4,r0; beq 4f; jbr e.oddz 4: mov (sp)+,r0 2: mov (sp)+,r0; br 1b sts.z: mov (sp)+,r0; br 0b /------------------------------------------------------------------------------ / POINTER ARITHMETIC /------------------------------------------------------------------------------ adp.l: jsr pc,wrdoff; add r0,(sp); next adp.1: add $1,(sp); next adp.2: add $2,(sp); next adp.s0: clr r0; bisb (pcx)+,r0 add r0,(sp); next adp.s_1: mov $-400,r0; bisb (pcx)+,r0 add r0,(sp); next ads.l: jsr pc,wrdoff; br 0f ads.z: mov (sp)+,r0 0: cmp $1,r0; beq 1f asr r0; jcs e.oddz 2: mov (sp)+,r1; sob r0,2b add r1,(sp); next ads.1W: mov (sp)+,r1; add r1,(sp); next 1: movb (sp)+,r1 add r1,(sp); next sbs.l: jsr pc,wrdoff; br 0f sbs.z: mov (sp)+,r0 0: mov (sp)+,r1; sub r1,(sp) beq 0f; mov $-1,r1 br 1f 0: clr r1 1: dec r0; beq 3f dec r0; beq 2f asr r0 4: mov r1,-(sp); sob r0,4b 2: next 3: clrb 1(sp); next /------------------------------------------------------------------------------ /------------------------------------------------------------------------------ / Clears, increments and decrements /------------------------------------------------------------------------------ inc.z: mov sp,r1; 4: .if .test cmp (r1),$und; jne 3f; jsr pc,e.iund; 3: .endif inc (r1); bvs 9f; next inl._1W: mov r2,r1; sub $2,r1; br 4b inl._2W: mov r2,r1; sub $4,r1; br 4b inl._3W: mov r2,r1; sub $6,r1; br 4b inl.w_1: mov $-400,r0; bisb (pcx)+,r0; 1: asl r0; mov r2,r1; add r0,r1; br 4b inl.pw: jsr pc,wrdoff; add $4,r0; br 1b; / !! proc frame 4 words inl.nw: jsr pc,wrdoff; br 1b ine.lw: jsr pc,wrdoff; br 1f ine.w0: clr r0; bisb (pcx)+,r0; 1: asl r0; add eb,r0; mov r0,r1; br 4b dec.z: mov sp,r1; 4: .if .test cmp (r1),$und; jne 3f; jsr pc,e.iund; 3: .endif dec (r1); bvs 9f; next del.w_1: mov $-400,r0; bisb (pcx)+,r0; 1: asl r0; mov r0,r1; add r2,r1; br 4b del.pw: jsr pc,wrdoff; add $4,r0; br 1b; / !proc frame 4 words del.nw: jsr pc,wrdoff; br 1b dee.w0: clr r0; bisb (pcx)+,r0; 1: asl r0; add eb,r0; mov r0,r1; br 4b dee.lw: jsr pc,wrdoff; br 1b; 9: jsr pc,e.iovfl; next / jump to an error routine for integer overflow zrl._1W: clr -2(r2); next zrl._2W: clr -4(r2); next zrl.w_1: mov $-400,r0; bisb (pcx)+,r0; 1: asl r0; add r2,r0; clr (r0); next zrl.nw: jsr pc,wrdoff; br 1b zrl.pw: jsr pc,wrdoff; add $4,r0; br 1b zre.lw: jsr pc,wrdoff; br 1f zre.w0: clr r0; bisb (pcx)+,r0; 1: asl r0; add eb,r0; clr (r0); next zrf.l: jsr pc,wrdoff; br 1f zrf.z: mov (sp)+,r0; 1: asr r0; 2: clr -(sp); sob r0,2b; next zer.s0: clr r0; bisb (pcx)+,r0; 3: bit $1,r0; jne e.illins / test if number of bytes is even br 1b zer.l: jsr pc,wrdoff; br 3b zer.z: mov (sp)+,r0; br 3b /------------------------------------------------------------------------------ / LOGICAL GROUP /------------------------------------------------------------------------------ and.1W: mov $1,r1; mov $2,r0; br lbland; and.l: jsr pc,wrdoff; br 0f and.z: mov (sp)+,r0; 0: ble 9f; mov r0,r1; asr r1; bcs 9f; lbland: add sp,r0; 1: mov (sp)+,r5; com r5; bic r5,(r0)+; sob r1,1b; next ior.1W: mov $1,r1; mov $2,r0; br 0f ior.s0: clr r0; bisb (pcx)+,r0; br 0f ior.l: jsr pc,wrdoff; br 0f ior.z: mov (sp)+,r0; lblior: 0: ble 9f; bit $1,r0; bne 9f; mov r0,r1; mov sp,r5; add r0,r5; asr r1; 1: bis (sp)+,(r5)+; sob r1,1b; next xor.l: jsr pc,wrdoff; br 0f; xor.z: mov (sp)+,r0; 0: ble 9f; bit $1,r0; bne 9f; mov r0,r1; mov sp,r5; add r0,r5; asr r1 1: mov (sp)+,r0; xor r0,(r5)+; sob r1,1b; next com.l: jsr pc,wrdoff; br 1f com.z: mov (sp)+,r0; 1: bit $1,r0; bne 9f mov r0,r1; asr r1 add sp,r0; 2: com -(r0); sob r1,2b next rol.l: jsr pc,wrdoff; br 3f rol.z: mov (sp)+,r0; 3: clr r4; mov (sp)+,r5; ash $3,r0; div r0,r4; mov r5,r4; bge 1f; add r0,r4; 1: ash $-3,r0; mov sp,r1; cmp r0,$1; beq 1f; add r0,r1; mov r1,r5; asr r0; jcs 9f mov r3,saver0; mov r0,r3; 4: mov r3,r0; mov r5,r1; 2: rol -(r1); sob r0,2b; adc -2(r5); sob r4,4b; mov saver0,r3; mov $loop,r4; next 1: rolb (r1)+; adc (r1); sob r4,1b; mov saver1,r4; next ror.l: jsr pc,wrdoff; neg (sp); br 3b ror.z: mov (sp)+,r0; neg (sp); br 3b 9: jsr pc,e.oddz /error codes for odd or /negative number of bytes /------------------------------------------------------------------------------ / SET GROUP /------------------------------------------------------------------------------ set.s0: clr r0; bisb (pcx)+,r0 1: .if .test bgt 9f; jsr pc,e.set 9: .endif mov (sp)+,r1 jsr pc,settest; inc r0 asr r0; / if r0 odd choose next even 2: clr -(sp); sob r0,2b; / empty set mov r1,r0; ash $-3,r0; add sp,r0; bic $177770,r1; bisb bits(r1),(r0); next set.l: jsr pc,wrdoff; br 1b set.z: mov (sp)+,r0; br 1b inn.s0: clr r0; bisb (pcx)+,r0 1: .if .test bgt 9f; jsr pc,e.set 9: .endif mov sp,r5; add r0,r5; mov (sp)+,r1; jsr pc,settest; mov r1,r0 ash $-3,r0; add sp,r0; clr -(sp); bic $177770,r1; bitb bits(r1),(r0) beq 2f; mov r5,sp; mov $1,(sp); next 2: mov r5,sp; clr (sp); next inn.l: jsr pc,wrdoff; br 1b inn.z: mov (sp)+,r0; br 1b .data bits: .byte 1 .byte 2 .byte 4 .byte 10 .byte 20 .byte 40 .byte 100 .byte 200 .even .text settest: mov r0,-(sp); clc ash $3,r0; sub r1,r0; .if .test bgt 3f; jsr pc,e.set .endif 3: mov (sp)+,r0; rts pc /------------------------------------------------------------------------------ / ARRAY GROUP /------------------------------------------------------------------------------ lar.1W: mov $2,r0; br 1f lar.l: jsr pc,wrdoff; br 1f lar.z: mov (sp)+,r0; 1: jsr pc,calcarr; clr -2(sp); sub r5,sp; bic $1,sp; mov sp,r0; 2: movb (r1)+,(r0)+; sob r5,2b; next sar.1W: mov $2,r0; br 1f sar.l: jsr pc,wrdoff; br 1f sar.z: mov (sp)+,r0; 1: jsr pc,calcarr; mov sp,r0; add r5,sp; inc sp; bic $1,sp; 2: movb (r0)+,(r1)+; sob r5,2b; next aar.1W: mov $2,r0; br 1f aar.l: jsr pc,wrdoff; br 1f aar.z: mov (sp)+,r0; 1: jsr pc,calcarr; mov r1,-(sp); next calcarr: sub $02,r0; beq 0f; jsr pc,e.oddz; 0: tst (sp)+; mov (sp)+,r0; mov (sp)+,r1; sub (r0)+,r1; bge 9f jsr pc,e.array 9: cmp (r0)+,r1; bge 9f jsr pc,e.array 9: mov (r0),r5; mul r5,r1; add (sp)+,r1; mov -010(sp),-(sp); rts pc; /------------------------------------------------------------------------------ /-------------------------------------------------------------- / CONVERT GROUP /-------------------------------------------------------------- cii.z: / convert int to int / 1 byte -> ? : sign extension mov (sp)+,r0 inc r0 / dest 1 byte = dest 1 word bic $1,r0 .if .test cmp (sp),$2 / if size 2 then trap for undefined bne 7f cmp 2(sp),$und bne 7f jsr pc,e.iund / this is the trap 7: .endif sub (sp)+,r0 0: blt 1f asr r0 bcc 2f movb (sp),r1 mov r1,(sp) 2: tst r0 beq 3f tst (sp) 4: sxt -(sp) sob r0,4b 3: next 1: sub r0,sp .if .test mov sp,r1 neg r0 asr r0 tst (sp) blt 3f 5: tst -(r1) bne 9f sob r0,5b next 3: cmp -(r1),$-1 bne 9f sob r0,3b .endif next /------- cui.z: mov (sp)+,r0 sub (sp)+,r0 clr -(sp) add $-2,r0 br 0b cif.z: mov (sp)+,r0 jsr pc,setfloat mov (sp)+,r0 .if .test cmp r0,$2 / trap if size 2 undefined integer bne 7f cmp (sp),$und bne 7f jsr pc,e.iund / trap for undefined integer 7: .endif jsr pc,setint movif (sp)+,fr0 movf fr0,-(sp) next cuf.z: mov (sp)+,r0 jsr pc,setfloat mov (sp)+,r0 cmp r0,$02 bne 1b clr -(sp) mov $04,r0 jsr pc,setint movif (sp)+,fr0 cfcc bge 1f addf fr3,fr0 1: movf fr0,-(sp) next /------- cff.z: mov (sp)+,r0 cmp (sp)+,r0 beq 1f jsr pc,setfloat movof (sp)+,fr0 movf fr0,-(sp) 1: next /------- ciu.z: mov (sp)+,r0 .if .test cmp (sp),$2 / trap undefined of size 2 bne 7f cmp 2(sp),$und bne 7f jsr pc,e.iund / this is the trap 7: .endif sub (sp)+,r0 asr r0 bcc 2f clrb 1(sp) 2: 6: tst r0 beq 3f blt 5f 4: clr -(sp) sob r0,4b 3: next 9: jsr pc,e.conv; next 5: neg r0 4: tst (sp)+ sob r0,4b next cuu.z: mov (sp)+,r0 sub (sp)+,r0 asr r0 jbr 6b /------- cfu.z: mov (sp)+,r0 jsr pc,setint mov (sp)+,r0 jsr pc,setfloat movf (sp)+,fr0 movfi fr0,-(sp) .if .test jcs 9b jlt 9b .endif next /------- cfi.z: mov (sp)+,r0 jsr pc,setint mov (sp)+,r0 jsr pc,setfloat movf (sp)+,fr0 movfi fr0,-(sp) jcs e.conv next /-------------------------------------------------------------- / INTEGER ARITHMETIC /-------------------------------------------------------------- adi.l: jsr pc,wrdoff; br 0f adi.z: mov (sp)+,r0 0: cmp r0,$04 bgt 1f cmp r0,$02 bgt 2f bne 1f adi.1W: .if .test cmp (sp),$und / trap undefineds of size 2 beq 6f cmp 2(sp),$und bne 7f 6: jsr pc,e.iund / this is the trap 7: .endif add (sp)+,(sp) .if .test bvs 9f .endif next adi.2W: 2: add (sp)+,02(sp) .if .test bvc 2f jsr pc,e.iovfl 2: .endif add (sp)+,02(sp) adc (sp) .if .test bvs 9f .endif next 1: jsr pc,e.oddz ; next /------- sbi.l: jsr pc,wrdoff; br 0f sbi.z: mov (sp)+,r0 0: cmp r0,$04 bgt 1b cmp r0,$02 bgt 2f bne 1b sbi.1W: .if .test cmp (sp),$und / trap for size 2 undefineds beq 6f cmp 2(sp),$und bne 7f 6: jsr pc,e.iund / this is the trap 7: .endif sub (sp)+,(sp) .if .test bvs 9f .endif next sbi.2W: 2: sub (sp)+,02(sp) .if .test bvc 2f jsr pc,e.iovfl 2: .endif sub (sp)+,02(sp) sbc (sp) .if .test bvs 9f next 9: jsr pc,e.iovfl .endif next /------ mli.l: jsr pc,wrdoff; br 0f mli.z: mov (sp)+,r0 0: cmp r0,$04 bgt 1f beq mli4 cmp r0,$02 bne 1f mli.1W: mov (sp)+,r1 .if .test cmp r1,$und / trap for undefineds of size 2 beq 6f cmp (sp),$und bne 7f 6: jsr pc,e.iund / this is the trap 7: .endif mul (sp)+,r1 .if .test bcc 9f / overflow jsr pc,e.iovfl 9: .endif mov r1,-(sp) next 1: jmp e.oddz /------ mli.2W: mli4: .if .prof add $91.,*loprof adc *hiprof .endif jsr pc,regsave tst 02(sp) sxt r0 sub (sp),r0 tst 06(sp) sxt r2 sub 04(sp),r2 mov r0,r4 mul r2,r4 mul 06(sp),r0 .if .test bge 2f inc r4 2: .endif mul 02(sp),r2 .if .test bge 2f inc r4 2: sub r2,r5 sbc r4 sub r0,r5 sbc r4 add r1,r3 sbc r5 sbc r4 .endif mov 02(sp),r0 mul 06(sp),r0 .if .test bge 2f sub $1,r5 sbc r4 .endif 2: sub r3,r0 .if .test sxt r2 sbc r5 sbc r4 cmp r2,r4 bne 2f cmp r2,r5 beq 9f 2: jsr pc,e.iovfl 9: .endif add $010,sp mov r1,-(sp); mov r0,-(sp); jsr pc,regretu; next /------- dvi.l: jsr pc,wrdoff; br 0f dvi.z: mov (sp)+,r0 0: cmp r0,$04 bgt 1f beq dvi4 cmp r0,$02 bne 1f dvi.1W: mov 02(sp),r1 sxt r0 .if .test cmp r1,$und / trap for undifined of size 2 beq 6f cmp (sp),$und bne 7f 6: jsr pc,e.iund / this is the trap 7: .endif div (sp)+,r0 jcs 9f mov r0,(sp) next 1: jmp e.oddz /------- dvi4: .if .prof add $100.,*loprof adc *hiprof .endif jsr pc,regsave mov 02(sp),r3 bne 1f tst (sp) bne 1f 9: jsr pc,e.idivz 1: sxt r4 bpl 1f neg r3 1: cmp r4,(sp) bne hardldiv mov 06(sp),r2 mov 04(sp),r1 bge 2f neg r1 neg r2 sbc r1 com r4 2: mov r4,-(sp) clr r0 div r3,r0 mov r0,-(sp) mov r1,r0 mov r1,r4 mov r2,r1 div r3,r0 bvc 3f mov r2,r1 mov r4,r0 sub r3,r0 div r3,r0 tst r1 sxt r1 add r1,r0 3: mov r0,r1 mov (sp)+,r0 br 4f hardldiv: .if .prof add $75.,*loprof adc *hiprof .endif clr -(sp) mov 010(sp),r2 mov 06(sp),r1 bpl 5f com (sp) neg r1 neg r2 sbc r1 5: clr r0 mov 02(sp),r3 bge 6f neg r3 neg 04(sp) sbc r3 com (sp) 6: mov $16.,r4 9: clc rol r2 rol r1 rol r0 cmp r3,r0 bhi 7f bcs 8f cmp 04(sp),r1 blos 8f 7: sob r4,9b br 1f 8: sub 04(sp),r1 sbc r0 sub r3,r0 inc r2 sob r4,9b 1: mov r2,r1 clr r0 4: tst (sp)+ beq 1f neg r0 neg r1 sbc r0 1: add $010,sp mov r1,-(sp); mov r0,-(sp); jsr pc,regretu; next /------- rmi.l: jsr pc,wrdoff; br 0f rmi.z: mov (sp)+,r0 0: cmp r0,$04 bgt 1f beq rmi4 cmp r0,$02 bne 1f rmi.1W: mov 02(sp),r1 sxt r0 .if .test cmp r1,$und / trap for undefineds of size 2 beq 6f cmp (sp),$und bne 7f 6: jsr pc,e.iund / this is the trap 7: .endif div (sp)+,r0 bcs 9f mov r1,(sp) next 1: jmp e.oddz /------- rmi4: .if .prof add $100.,*loprof adc *hiprof .endif jsr pc,regsave mov 02(sp),r3 bne 1f tst (sp) bne 1f 9: jsr pc,e.idivz 1: tst r3 sxt r4 bpl 1f neg r3 1: cmp r4,(sp) bne hardrmi4 mov 06(sp),r2 mov 04(sp),r1 mov r1,r4 bge 2f neg r1 neg r2 sbc r1 2: mov r4,-(sp) clr r0 div r3,r0 mov r1,r0 mov r1,r4 mov r2,r1 div r3,r0 bvc 3f mov r2,r1 mov r4,r0 sub r3,r0 div r3,r0 tst r1 beq 3f add r3,r1 3: tst (sp)+ bpl 4f neg r1 4: sxt r0 br 9f hardrmi4: .if .prof add $75.,*loprof adc *hiprof .endif mov 06(sp),r2 mov 04(sp),r1 bpl 5f neg r1 neg r2 sbc r1 5: clr r0 mov (sp),r3 bge 6f neg r3 neg 02(sp) sbc r3 6: mov $16.,r4 1: clc rol r2 rol r1 rol r0 cmp r3,r0 bhi 7f bcs 8f cmp 02(sp),r1 blos 8f 7: sob r4,1b br 2f 8: sub 02(sp),r1 sbc r0 sub r3,r0 sob r4,1b 2: tst 04(sp) bge 9f neg r0 neg r1 sbc r0 9: add $010,sp mov r1,-(sp) mov r0,-(sp) jsr pc,regretu; next /------- ngi.l: jsr pc,wrdoff; br 1f ngi.z: mov (sp)+,r0 1: lblngi: cmp r0,$02 bgt 1f .if .test cmp (sp),$und / trap for undefineds of size 2 bne 7f jsr pc,e.iund 7: .endif neg (sp) 3: next 1: cmp r0,$04 bgt 2f mov (sp),r0 neg (sp) mov 02(sp),r1 neg 02(sp) sbc (sp) cmp r0,(sp) .if .test bne 3b cmp r1,02(sp) bne 3b 2: jsr pc,e.iovfl .endif next /------- sli.l: jsr pc,wrdoff; br 0f sli.z: mov (sp)+,r0 0: cmp r0,$02 bgt 1f sli.1W: mov (sp)+,r1 mov (sp)+,r0 .if .test cmp r0,$und / trap for undefined size 2 bne 7f jsr pc,e.iund 7: .endif ash r1,r0 .if .test bvc 7f jsr pc,e.iovfl 7: .endif mov r0,-(sp) next 1: cmp r0,$04 bgt 2f mov 02(sp),r0 mov 04(sp),r1 ashc (sp)+,r0 .if .test bvc 7f jsr pc,e.iovfl 7: .endif mov r0,(sp) mov r1,02(sp) next 2: jmp e.oddz /------- sri.l: jsr pc,wrdoff; br 0f sri.z: mov (sp)+,r0 0: cmp r0,$02 bgt 1f mov (sp)+,r1 mov (sp)+,r0 .if .test cmp r0,$und / trap for undefined size 2 bne 7f jsr pc,e.iund 7: .endif neg r1 ash r1,r0 mov r0,-(sp) next 1: cmp r0,$04 bgt 2f mov 02(sp),r0 mov 04(sp),r1 neg (sp) ashc (sp)+,r0 mov r0,(sp) mov r1,02(sp) next 2: jmp e.oddz /-------------------------------------------------------------- adu.l: jsr pc,wrdoff; br 0f adu.z: mov (sp)+,r0 0: jsr pc,tstr0; add r0,sp mov sp,r1; add r0,sp; asr r0 2: adc -(sp); add -(r1),(sp); sob r0,2b next sbu.l: jsr pc,wrdoff; br 0f sbu.z: mov (sp)+,r0 0: jsr pc,tstr0; add r0,sp mov sp,r1; add r0,sp; asr r0; 2: sbc -(sp); sub -(r1),(sp); sob r0,2b next mlu.l: jsr pc,wrdoff; br 0f mlu.z: mov (sp)+,r0 0: jsr pc,tstr0 cmp r0,$04 bgt 1f beq mlu4 mov (sp)+,r1 mul (sp)+,r1 mov r1,-(sp) next 1: jmp e.oddz mlu4: .if .prof add $90.,*loprof adc *hiprof .endif jsr pc,regsave clr r0 mov 02(sp),r1 mov 06(sp),r3 mul r3,r0 tst r3 bge 1f ashc $15.,r0 1: mov 02(sp),r3 clr r2 mul 04(sp),r2 add r3,r0 mov 06(sp),r3 clr r2 mul (sp),r2 add r3,r0 add $010,sp mov r1,-(sp) mov r0,-(sp) jsr pc,regretu; next 9: jmp e.oddz /------- dvu.l: jsr pc,wrdoff; br 0f dvu.z: mov (sp)+,r0 0: clr saver0; cmp r0,$04 bgt 9b beq dvu4 clr r0 mov 02(sp),r1 tst (sp) blt 1f div (sp)+,r0 mov r0,(sp); next 1: mov (sp),-(sp); clr 02(sp); clr -(sp); mov $1,saver0; dvu4: .if .prof add $95.,*loprof adc *hiprof .endif jsr pc,regsave clr r0 tst (sp) bne harddvu4 tst 02(sp) blt harddvu4 mov 06(sp),r2 mov 04(sp),r1 mov 02(sp),r3 div r3,r0 mov r0,-(sp) mov r1,r0 mov r1,r4 mov r2,r1 div r3,r0 bvc 1f mov r2,r1 mov r4,r0 sub r3,r0 div r3,r0 tst r1 sxt r1 add r1,r0 1: mov r0,r1 mov (sp)+,r0 br 2f harddvu4: .if .prof add $75.,*loprof adc *hiprof .endif mov 06(sp),r2 mov 04(sp),r1 mov (sp),r3 mov $17.,r4 br 3f 6: rol r2 rol r1 rol r0 3: cmp r3,r0 bhi 4f blo 5f cmp 02(sp),r1 blos 5f 4: clc sob r4,6b br 7f 5: sub 02(sp),r1 sbc r0 sub r3,r0 sec sob r4,6b 7: rol r2 bcc 8f mov $01,r0 br 9f 8: clr r0 9: mov r2,r1 2: add $010,sp mov r1,-(sp) mov r0,-(sp) jsr pc,regretu tst saver0; beq 2f; tst (sp)+ 2: next /------- 9: jbr e.oddz rmu.l: jsr pc,wrdoff; br 0f rmu.z: mov (sp)+,r0 0: clr saver0; cmp r0,$04 bgt 9b beq rmu4 cmp r0,$02 bne 9b mov $1,saver0; mov 02(sp),r1 tst (sp) blt 1f clr r0 div (sp)+,r0 mov r1,(sp); next 1: mov (sp),-(sp) clr 02(sp) clr -(sp) rmu4: .if .prof add $95.,*loprof adc *hiprof .endif jsr pc,regsave clr r0 tst (sp) bne hardrmu4 tst 02(sp) blt hardrmu4 mov 06(sp),r2 mov 04(sp),r1 mov 02(sp),r3 div r3,r0 mov r1,r0 mov r1,r4 mov r2,r1 div r3,r0 bvc 1f mov r2,r1 mov r4,r0 sub r3,r0 div r3,r0 tst r1 beq 1f add r3,r1 1: clr r0 br 2f hardrmu4: .if .prof add $75.,*loprof adc *hiprof .endif mov 06(sp),r2 mov 04(sp),r1 mov (sp),r3 mov $17.,r4 br 3f 6: clc rol r2 rol r1 rol r0 3: cmp r3,r0 bhi 4f bcs 5f cmp 02(sp),r1 blos 5f 4: sob r4,6b br 2f 5: sub 02(sp),r1 sbc r0 sub r3,r0 sob r4,6b 2: add $010,sp mov r1,-(sp) mov r0,-(sp) jsr pc,regretu tst saver0 beq 2f; tst (sp)+; 2: next /------- slu.l: jsr pc,wrdoff; br 0f slu.z: mov (sp)+,r0 0: cmp r0,$02 bgt 1f mov (sp)+,r1 mov (sp)+,r0 ash r1,r0 mov r0,-(sp) next 1: cmp r0,$04 bgt 2f mov 02(sp),r0 mov 04(sp),r1 ashc (sp)+,r0 mov r0,(sp) mov r1,02(sp) next 2: jmp e.oddz /------- sru.l: jsr pc,wrdoff; br 0f sru.z: mov (sp)+,r0 0: cmp r0,$02 bgt 1f mov 2(sp),r1 neg (sp) clr r0 ashc (sp)+,r0 2: mov r1,-(sp) next 1: cmp r0,$04 bgt 3f mov 02(sp),r0 mov 04(sp),r1 neg (sp) beq 4f ashc $-1,r0 bic $0100000,r0 inc (sp) beq 4f ashc (sp)+,r0 4: mov r0,(sp) mov r1,02(sp) next 3: jmp e.oddz /------- /-------------------------------------------------------------- / FLOATING POINT INSTRUCTIONS /-------------------------------------------------------------- adf.s0: clr r0; bisb (pcx)+,r0; br 0f adf.l: jsr pc,wrdoff; br 0f adf.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp)+,fr0 addf (sp)+,fr0 movf fr0,-(sp) next /------- sbf.s0: clr r0; bisb (pcx)+,r0; br 0f sbf.l: jsr pc,wrdoff; br 0f sbf.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp)+,fr0 subf (sp)+,fr0 negf fr0 movf fr0,-(sp) next /------- mlf.s0: clr r0; bisb (pcx)+,r0; br 0f mlf.l: jsr pc,wrdoff; br 0f mlf.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp)+,fr0 mulf (sp)+,fr0 movf fr0,-(sp) next /------- dvf.s0: clr r0; bisb (pcx)+,r0; br 0f dvf.l: jsr pc,wrdoff; br 0f dvf.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp)+,fr0 movf (sp)+,fr1 divf fr0,r1 movf fr1,-(sp) next /------- ngf.l: jsr pc,wrdoff; br 0f ngf.z: mov (sp)+,r0 0: jsr pc,setfloat negf (sp) next /------- fif.l: jsr pc,wrdoff; br 0f fif.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp)+,fr0 modf (sp)+,fr0 movf fr0,-(sp) movf fr1,-(sp) next /------- fef.l: jsr pc,wrdoff; br 0f fef.z: mov (sp)+,r0 0: jsr pc,setfloat movf (sp),fr0 movei fr0,-(sp) movie $0,fr0 movf fr0,02(sp) next /---------------------------------------------------------------------------- / TEST AND BRANCH GROUP /---------------------------------------------------------------------------- tlt.z: tst (sp)+; blt true; clr -(sp); next tle.z: tst (sp)+; ble true; clr -(sp); next teq.z: tst (sp)+; beq true; clr -(sp); next tne.z: tst (sp)+; bne true; clr -(sp); next tge.z: tst (sp)+; bge true; clr -(sp); next tgt.z: tst (sp)+; bgt true; clr -(sp); next /---------------------------------------------------------------------------- zlt.s0: tst (sp)+; blt yesbr2; br nobr2 zlt.l: tst (sp)+; blt yesbr3; br nobr3 zle.s0: tst (sp)+; ble yesbr2; br nobr2 zle.l: tst (sp)+; ble yesbr3; br nobr3 zeq.s0: tst (sp)+; beq yesbr2; br nobr2 zeq.s1: tst (sp)+; beq yesbr4; br nobr2 zeq.l: tst (sp)+; beq yesbr3; br nobr3 zne.s0: tst (sp)+; bne yesbr2; br nobr2 zne.l: tst (sp)+; bne yesbr3; br nobr3 zne.s_1: tst (sp)+; bne yesbr5; br nobr2 zge.s0: tst (sp)+; bge yesbr2; br nobr2 zge.l: tst (sp)+; bge yesbr3; br nobr3 zgt.s0: tst (sp)+; bgt yesbr2; br nobr2 zgt.l: tst (sp)+; bgt yesbr3; br nobr3 great: true: mov $1,-(sp) next /------------------------------------------------------------------------------ blt.s0: cmp (sp)+,(sp)+; bgt yesbr2; br nobr2 blt.l: cmp (sp)+,(sp)+; bgt yesbr3; br nobr3 ble.s0: cmp (sp)+,(sp)+; bge yesbr2; br nobr2 ble.l: cmp (sp)+,(sp)+; bge yesbr3; br nobr3 beq.l: cmp (sp)+,(sp)+; beq yesbr3; br nobr3 beq.s0: cmp (sp)+,(sp)+; beq yesbr2; br nobr2 bne.s0: cmp (sp)+,(sp)+; bne yesbr2; br nobr2 bne.l: cmp (sp)+,(sp)+; bne yesbr3; br nobr3 bge.s0: cmp (sp)+,(sp)+; ble yesbr2; br nobr2 bge.l: cmp (sp)+,(sp)+; ble yesbr3; br nobr3 bgt.s0: cmp (sp)+,(sp)+; blt yesbr2; br nobr2 bgt.l: cmp (sp)+,(sp)+; blt yesbr3; br nobr3 bra.s0: yesbr2: clr r0; 5: bisb (pcx)+,r0 1: add r0,pcx next bra.l: yesbr3: jsr pc,wrdoff br 1b bra.s1: yesbr4: mov $400,r0; br 5b bra.s_1: yesbr5: mov $-400,r0; br 5b bra.s_2: mov $-800,r0; br 5b nobr2: tstb (pcx)+ next nobr3: cmpb (pcx)+,(pcx)+ next / /------------------------------------------------------------------------------ / Compare group /------------------------------------------------------------------------------ calccomp: mov (sp)+,saver1; / old pc mov sp,r1; add r0,r1; mov r1,r5; add r0,r1; mov r1,saver0; / new sp mov r0,r1; asr r1; 2: cmp (r5)+,(sp)+; blt 3f; bgt 4f; dec r1; beq 5f 2: cmp (r5)+,(sp)+; blo 3f; bhi 4f; sob r1,2b; 5: mov saver0,sp; clr -(sp); br 5f 4: mov saver0,sp; mov $1,-(sp); br 5f 3: mov saver0,sp; mov $-1,-(sp); br 5f 5: mov saver1,-(sp); rts pc cmi.1W: mov $2,r0; br 1f cmi.2W: mov $4,r0; br 1f cmi.l: jsr pc,wrdoff; br 1f cmi.z: mov (sp)+,r0; 1: jsr pc,calccomp; next cms.s0: clr r0; bisb (pcx)+,r0 1: jsr pc,calccomp; tst (sp)+; bne great; clr -(sp); next cms.l: jsr pc,wrdoff; br 1b cms.z: mov (sp)+,r0; br 1b cmu.l: jsr pc,wrdoff; br 1f cmu.z: mov (sp)+,r0; br 1f cmp.z: mov $2,r0; 1: jsr pc,calccomp; mov (sp)+,r1; jlo less; jhi great; clr -(sp); next cmf.s0: clr r0; bisb (pcx)+,r0; 1: cmp $4,r0; beq 3f cmp $8,r0; beq 3f jsr pc,e.oddz; 3: jsr pc,setfloat movf (sp)+,fr0; cmpf (sp)+,fr0; 4: cfcc; jlt less; jgt great; clr -(sp); next cmf.l: jsr pc,wrdoff; br 1b cmf.z: mov (sp)+,r0; br 1b less: mov $-1,-(sp); next /------------------------------------------------------------------------------ /------------------------------------------------------------------------------ /------------------------------------------------------------------------------ / call and return group /------------------------------------------------------------------------------ / / Frame format on the stack is: / / | Parameter 1 | / | - - - - - - - - - - - - | / | Static link = param 0 | < AB / |_________________________| ____ / |-------------------------| | P / | Saved line number | | R / |-------------------------| | O / | Saved file address | | C / |-------------------------| | F / | Return address | | R / |-------------------------| | A / | Dynamic link | < LB | M / |_________________________| ____| E / |-------------------------| / | | / | local variable -1 | / |-------------------------| / | | / / / The static link and the parameters are set by the / calling procedure; the frame is set by cal which reserves / also the space for local variables. /------------------------------------------------------------------------------ cal.1: cal.2: cal.3: cal.4: cal.5: cal.6: cal.7: cal.8: cal.9: cal.10: cal.11: cal.12: cal.13: cal.14: cal.15: cal.16: cal.17: cal.18: cal.19: cal.20: cal.21: cal.22: cal.23: cal.24: cal.25: cal.26: cal.27: cal.28: asr r0; sub $077,r0; br 1f cal.s0: clr r0; bisb (pcx)+,r0; br 1f cai.z: mov (sp)+,r0; br 1f cal.l: jsr pc,wrdoff lblcal: 1: cmp nprocs,r0 jlos e.badpc 1: asl r0; asl r0; .if .flow + .count + .prof mov r0,r5; asl r5 add r5,r0 / procdes 12 bytes in this case .endif add pd,r0; / now r0 points to the bottom / of the proc descriptor mov *eb,-(sp) / save line number mov *filb,-(sp) / save file address .if .prof + .count + .flow tst 10.(r0) / yet a filename defined in this proc? bne 4f / yes? continue mov (sp),10.(r0) / no? Then take the old filename 4: mov curproc,(sp) / save old procdescriptor mov r0,curproc / continue with the new one mov 4(r0),countfld / count pointer minus line number sub 6(r0),countfld / to add at line numbers .endif mov pcx,-(sp) / return address mov r2,-(sp) / old local base mov sp,r2 / new local base mov (r0)+,r1 / number of bytes for locals beq 3f / no locals inc r1; asr r1 / make even 2: mov $und,-(sp) / make undefined sob r1,2b / for r1 words local 3: mov (r0),pcx / em start address next / ready /------------------------------------------------------------------------------ ret.l: jsr pc,wrdoff; br 1f ret.1W: mov $2,r0; br 1f ret.s0: clr r0; bisb (pcx)+,r0; jlt e.oddz / bad result size beq ret.0 / no result cmp $32.,r0; bge 1f; jsr pc,e.badlfr 1: inc r0; asr r0 / make even mov r0,retsize / result size mov $retarea,r1 / must point to retarea 2: mov (sp)+,(r1)+ / load result sob r0,2b / continue when there is more br 3f retcount: ret.0: clr retsize / no function result 3: mov r2,sp / set sp to bottom frame mov (sp)+,r2 / restore (dynamic) local base jeq lblhalt; / if zero then main procedure returns mov (sp)+,pcx / return address in program counter mov eb,r1 / must point to external base mov (sp)+,r0 / file or procdesbase in r0 .if .count + .flow + .prof mov r0,curproc / save procdesbase current proc mov 4(r0),countfld / restore pointer to beq 5f / countfield 0 non interesting sub 6(r0),countfld / table entries mov countfld,countptr add (sp),countptr .if .prof tst (sp) beq 5f mov countptr,r5 asl r5 asl r5 add ltime,r5 mov r5,hiprof add $2,r5 mov r5,loprof .endif 5: mov 10.(r0),r0 / pointer to file name in r0 .endif retlast: .if .last tst 4(r1) / file 0 not stacked beq 4f cmp r0,4(r1) / same file name? beq 4f / yes continue jsr pc,nexttab / next line table entry in r5 clr (r5) / 0 indicates file pointer follows jsr pc,nexttab / next entry in r5 mov 4(r1),(r5) / store pointer to file name 5: mov (sp),(r1) / restored line number in *eb beq 4f / do not put 0 in last table jsr pc,nexttab / next entry in r5 mov (sp),(r5) / line number in table 4: cmp (sp),(r1) / line number different? bne 5b / yes? put it in table .endif mov r0,4(r1) / old file address mov (sp)+,(r1) / reset line number next / main loop again lfr.l: jsr pc,wrdoff; 8: bit $1,r0; jne e.illins cmp r0,retsize; jeq 7f; jsr pc,e.badlfr 7: asr r0; / number of words tst r0; beq 0f; mov $retarea,r1; / must point to retarea add r0,r1; add r0,r1; / top of result 2: mov -(r1),-(sp); / move word of result sob r0,2b; / look for more 0: next lfr.1W: mov $2,r0; br 8b lfr.2W: mov $4,r0; br 8b lfr.s0: clr r0; bisb (pcx)+,r0; br 8b / /------------------------------------------------------------------------------ / miscellaneous /------------------------------------------------------------------------------ exg.s0: clr r0; bisb (pcx)+,r0; br 1f exg.l: jsr pc,wrdoff; br 1f exg.z: mov (sp)+,r0; 1: cmp r0,$1; beq 0f; blt 9f; bit $1,r0; bne 9f; add r0,sp; mov r0,r1; asr r1; add sp,r0; 2: mov -(sp),r5; mov -(r0),(sp); mov r5,(r0); sob r1,2b; next 0: swab (sp); next 9: jsr pc,e.oddz; next /------------------------------------------------------------------------------ dup.1W: mov (sp),-(sp); next dup.l: jsr pc,wrdoff; br 1f; dus.l: jsr pc,wrdoff; br 0f; dus.z: mov (sp)+,r0; 0: ble 9b; bit $1,r0; bne 9b; add r0,sp; tst -(sp); mov (sp)+,r0; 1: ble 9b; inc r0; bic $1,r0; mov r0,r1; mov sp,r5; add r0,r5; asr r1; 2: mov -(r5),-(sp); sob r1,2b; next nop.z: next /------------------------------------------------------------------------------ fil.l: jsr pc,wrdoff; add eb,r0 cmp r0,*filb; beq 1f; .if .last clr *eb / new file asks for new line jsr pc,nexttab; clr (r5); jsr pc,nexttab; mov *filb,(r5); .endif .if .flow + .count + .prof mov curproc,r1 / current proc descriptor mov 4(r1),countfld sub 6(r1),countfld / start countptr for this proc mov r0,10.(r1) / file pointer in procdes .endif mov r0,*filb 1: next lni.z: inc *eb; .if .count + .flow + .prof inc countptr .if .prof add $4,hiprof add $4,loprof .endif .endif br 8f lin.l: jsr pc,wrdoff; br 1f lin.s0: clr r0; bisb (pcx)+,r0 1: cmp *eb,r0; beq 9f; mov r0,*eb .if .count + .flow + .prof mov countfld,r5 add r0,r5 / this is the right countptr mov r5,countptr .if .prof asl r5 asl r5 add ltime,r5 mov r5,hiprof add $2,r5 mov r5,loprof .endif .endif 8: .if .last jsr pc,nexttab / next entry in lasttab mov *eb,(r5) / set endline .endif .if .count mov countptr,r1 / line number in r1 asl r1 asl r1 / multiply by 4 add lcount,r1 / r1 is pointer to long integer now add $1,2(r1) / cannot be inc adc (r1) / that was all .endif .if .flow mov countptr,r1 / line number in r1 clr r0 / get ready for divide div $8.,r0 / r0 = byte offset; r1 = bit offset add lflow,r0 / r0 is byte pointer now bisb bits(r1),(r0) / set bit .endif 9: next /------------------------------------------------------------------------------ bls.l: jsr pc,wrdoff; br 0f; bls.z: mov (sp)+,r0; 0: asr r0 bhi 1f jbr e.oddz 1: mov (sp)+,r1; sob r0,1b; mov r1,r0; br 1f; blm.l: jsr pc,wrdoff tst r0 br 1f blm.s0: clr r0; bisb (pcx)+,r0 1: jsr pc,regsave; jsr pc,chckptr mov (sp)+,r2; jsr pc,chckptr mov (sp)+,r3 mov r0,r1 asr r0 beq 2f / Now avoid wrong copy. The pieces may overlap ! cmp r3,r2 beq 2f blt 3f 1: mov (r3)+,(r2)+ sob r0,1b 2: jsr pc,regretu; next 3: add r1,r3; add r1,r2; 1: mov -(r3),-(r2); sob r0,1b; br 2b /------------------------------------------------------------------------------ / Range check and case instructions /------------------------------------------------------------------------------ csa.l: jsr pc,wrdoff; br 1f; csa.z: mov (sp)+,r0; 1: sub $2,r0; jne e.illins; csa.1W: mov (sp)+,r0; mov (sp)+,r1; sub 2(r0),r1; blt 6f; cmp 4(r0),r1; blo 6f; asl r1; add $6,r1; add r0,r1; 5: mov (r1),pcx; jeq e.case; next 6: mov r0,r1; br 5b; csb.z: mov (sp)+,r0; br 1f; csb.l: jsr pc,wrdoff; 1: sub $2,r0; jne e.illins; csb.1W: mov (sp)+,r0; mov (sp)+,r1; mov r0,pcx; mov 2(r0),r5 /use pcx as ordinary register ble 3f 2: add $4,r0; cmp (r0),r1; beq 4f; sob r5,2b; 3: mov (pcx),pcx; jeq e.case; next 4: mov 2(r0),pcx; jeq e.case; next rck.l: jsr pc,wrdoff; br 1f; rck.z: mov (sp)+,r0; 1: sub $2,r0; beq rck.1W; sub $2,r0; jne e.oddz; mov (sp)+,r1; cmp (sp),(r1); blt 9f; cmp (sp),4(r1); bgt 9f; next rck.1W: mov (sp)+,r1; cmp (sp),(r1); blt 9f; cmp (sp),2(r1); bgt 9f; next 9: mov $ERANGE,-(sp); jmp trp.z; /------------------------------------------------------------------------------ / Load and store register /------------------------------------------------------------------------------ lbllor: lor.s0: clr r0; bisb (pcx)+,r0 cmp r0,$2; jhi e.illins asl r0; jmp 1f(r0) 1: br 2f; br 3f; br 4f 2: mov lb,-(sp); next 3: mov sp,r1; mov r1,-(sp); next 4: mov hp,-(sp); next lblstr: str.s0: clr r0; bisb (pcx)+,r0 cmp r0,$2; jhi e.illins asl r0; jmp 1f(r0) 1: br 2f; br 3f; br 4f 2: mov (sp)+,lb; next 3: mov (sp)+,r1; mov r1,sp; next 4: mov (sp)+,r1; 5: cmp r1,sybreak+2; blos 5f; add $unixextra,sybreak+2; sys indir;sybreak / ask for more core jec 5b; 1: jsr pc,e.heap; / core claim failed next 5: cmp r1,globmax; jlo 1b mov r1,hp; next /------------------------------------------------------------------------------ ass.l: jsr pc,wrdoff; br 1f ass.z: mov (sp)+,r0; 1: cmp $2,r0; beq 2f cmp $4,r0; jne e.oddz mov (sp)+,r0; 2: mov (sp)+,r0; br 3f asp.lw: jsr pc,wrdoff; br 2f asp.w0: clr r0; bisb (pcx)+,r0; 2: asl r0; br 3f asp.1W: asp.2W: asp.3W: asp.4W: asp.5W: sub $88.,r0; 3: blt 4f; add r0,sp; next 4: neg r0; asr r0; 2: mov $und,-(sp); sob r0,2b; next /------------------------------------------------------------------------------ dch.z: mov (sp)+,r1; mov (r1),-(sp); cmp (sp),ml; jge e.badptr; next / dch adjusts local base to / dynamically previous local base lpb.z: add $8.,(sp); next / static link 8 bytes from lb /------------------------------------------------------------------------------ gto.l: jsr pc,wrdoff; / get descriptor address add eb,r0 / descriptor is in external address space mov 4(r0),r2; / new local base after jump mov 2(r0),sp; / new stack pointer after jump mov (r0),pcx; / new program counter .if .prof + .flow + .count mov firstp,r0 1: mov 8.(r0),r1 cmp r3,2(r1) blos 2f mov r1,r0 br 1b 2: mov r0,curproc / procdesbase current proc mov 4(r0),countfld / restore pointer to sub 6(r0),countfld / table entries mov 10.(r0),*filb / file name restored .endif next /------------------------------------------------------------------------------ / Ignore mask /------------------------------------------------------------------------------ lim.z: mov ignmask,-(sp); next / load ignore mask sim.z: mov (sp)+,ignmask; next / store ignore mask / for trap routines sig.z: mov (sp),r1; / exchange previous mov uerrorp,(sp); / and stacked error mov r1,uerrorp; / procedure pointer next /------------------------------------------------------------------------------ / Signals generated by UNIX /------------------------------------------------------------------------------ sig1: mov $sig.trp+0.,-(sp); br 1f sig2: mov $sig.trp+2.,-(sp); br 1f sig3: mov $sig.trp+4.,-(sp); br 1f sig13: mov $sig.trp+24.,-(sp); br 1f sig14: mov $sig.trp+26.,-(sp); br 1f sig15: mov $sig.trp+28.,-(sp); br 1f sig16: mov $sig.trp+30.,-(sp) / push address trap number 1: mov *(sp),-2(sp); / save trap number mov $-2,*(sp) / set defold trap number mov $retutrap,(sp) / set return address to rti tst -(sp) / trap number position on stack jbr error sig12: mov r0,-(sp) mov $2,r0; / fildes standard error sys 0; 9b / call write message sys signal;12.;sig12 / this is a mon-error jsr pc,e.badmon mov (sp)+,r0 rti sig11: mov r0,saver1 /save r0 mov sybreak+2,r0 sub sp,r0 jgt e.memflt /real trap mov $7f,r0 mov argv,sp /setup a new stack jbr rude_error .data 7: .even sybreak:sys break;_end /For indirect calling of break .text sig8: mov r0,saver1; sys signal;8.;sig8 mov $ECONV,fpperr+6 stfps r0 bit $100,r0 beq 1f mov $ECONV,fpperr+6 1: stst r0 mov $retutrap,-(sp) mov fpperr(r0),-(sp) mov saver1,r0 jbr error retutrap: rti .data fpperr: EILLINS; EILLINS; EFDIVZ; ECONV; EFOVFL; EFUNFL; EFUND; EILLINS .text /------------------------------------------------------------------------------ / Errors,traps and all their friends /------------------------------------------------------------------------------ savereg: mov r1,-(sp) / stack r1 so r1 scratch register mov 2(sp),r1 / now r1 contains return address mov r0,2(sp) / save r0 mov r2,-(sp) / save r2 mov r3,-(sp) / save r3 mov r4,-(sp) / save r4 mov r5,-(sp) / save r5 mov 12.(sp),r0 / copy error number or param 0 movf fr0,-(sp) / save float registers movf fr1,-(sp) / fr0 and fr1 stfps -(sp) / and status mov $retsize+26.,r5 mov $13.,r2 8: mov -(r5),-(sp); sob r2,8b mov r0,-(sp) / extra errno (param 0) on stack mov r1,-(sp) / set return address rts pc restoreg: mov (sp)+,r1 / return address in r1 add $2,sp / skip error number (param 0) mov $13.,r2; mov $retsize,r5; 8: mov (sp)+,(r5)+; sob r2,8b ldfps (sp)+ / restore floating point status movf (sp)+,fr1 / restore float registers movf (sp)+,fr0 / fr0 and fr1 mov (sp)+,r5 / restore r5 mov (sp)+,r4 / restore r4 mov (sp)+,r3 / restore r3 mov (sp)+,r2 / restore r2 mov 2(sp),r0 / restore r0 mov r1,2(sp) / reset return address mov (sp)+,r1 / restore r1 rts pc /------------------------------------------------------------------------------ / Various error entries /------------------------------------------------------------------------------ e.badlfr: mov r0,r5; mov $2,r0; mov $blfr,9f+2 sys 0;9f; mov r5,r0; rts pc .data 9: sys write;7;0; blfr: .even .text e.iund: mov $EIUND,-(sp); br error e.iovfl: mov $EIOVFL,-(sp); br error e.idivz: mov $EIDIVZ,-(sp); br error e.range: mov $ERANGE,-(sp); br error e.set: mov $ESET,-(sp); br error e.badptr: mov $EBADPTR,-(sp); br fatal e.fovfl: mov $EFOVFL,-(sp); br error e.funfl: mov $EFUNFL,-(sp); br error e.fdivz: mov $EFDIVZ,-(sp); br error e.fund: mov $EFUND,-(sp); br error e.conv: mov $ECONV,-(sp); br error e.stack: mov $ESTACK,-(sp); br fatal e.badpc: mov $EBADPC,-(sp); br fatal e.badlin: mov $EBADLIN,-(sp); br error e.badlae: mov $EBADLAE,-(sp); br error e.array: mov $EARRAY,-(sp); br error e.badmon: mov $EBADMON,-(sp); br error e.case: mov $ECASE,-(sp); br fatal e.oddz: mov $EODDZ,-(sp); br fatal e.illins: mov $EILLINS,-(sp); br fatal e.heap: mov $EHEAP,-(sp); br error e.memflt: mov $EMEMFLT,-(sp); br fatal e.badgto: mov $EBADGTO,-(sp); br error /------------------------------------------------------------------------------ fatal: mov $hlt.z,-(sp) / return from fatal halts mov 2(sp),-(sp) / dup error number error: jsr pc,savereg / save old register contents cmp $16.,r0; ble 9f; mov $1,r1 ash r0,r1 bic ignmask,r1 bne 9f jsr pc,restoreg add $2,sp / remove error number from stack rts pc 9: cmp uerrorp,$-1 / has the user defined a trapprocedure ? beq notrap / if not kill it off fast mov uerrorp,-(sp) / go for cal mov $-1,uerrorp / user must set trap again jbr precal / call trap routine /------------------------------------------------------------------------------ rtt.z: mov r2,sp / sp to bottom frame add $8,sp / set to top frame jsr pc,restoreg / restore status and registers add $2,sp / skip error number rts pc /------------------------------------------------------------------------------ trp.z: mov (sp),-(sp); / error number one down mov r4,2(sp); / set return address to main loop jbr error / call error routine /------------------------------------------------------------------------------ notrap: mov (sp),r1 / error number mov $num+9.,r0 mov r1,retarea / set error number for exit status jsr pc,itoa / make string mov $num,r0 rude_error: mov r0,8f+4 mov *filb,8f / pointer to file name mov *eb,r1 / line number mov $line,8f+6 / line message mov $line+21.,r0 jsr pc,itoa / convert to string 1: mov $8f,r4 2: mov (r4)+,r0 / next string beq 5f 3: mov r0,9f+2 4: tstb (r0)+ / find zero byte bne 4b dec r0 sub 9f+2,r0 / string length mov r0,9f+4 mov $2,r0 / diagnostic output sys 0; 9f jbr 2b 5: / no file mov $-1,argc jbr hlt.z itoa: mov r5,-(sp) mov r0,r5 1: clr r0 div $10.,r0 add $'0,r1 movb r1,-(r5) mov r0,r1 bne 1b mov r5,r0 mov (sp)+,r5 rts pc .data uerrorp: -1 / undefined trap procedure sep: <: \0> line: < on source line \n\0> num: .even 8: 0 / name of text file sep / separator 0 / error line+21. / line number if present 0 / end of list lblwri: 9: sys write;0;0 .text /------------------------------------------------------------------------------ / Exit instruction, with all it's crud /------------------------------------------------------------------------------ hlt.z: mov (sp)+,retarea .if .count + .flow + .prof br 9f .endif bne 9f clr r0 sys exit 9: lblhalt: sys creat;1f;666 .data 1: 2: .even .text bec 2f 3: mov $2b,lblwri+2 mov $19.,lblwri+4 mov $2.,r0 sys indir;lblwri br 9f 2: mov r0,saver0 mov $hp,r1 mov r1,ndatad mov $txtsiz,r5 mov r5,txtsiz sub r5,r1 mov r5,lblwri+2 mov r1,lblwri+4 clr r1 .if .last add $1,r1 .endif .if .opfreq add $2,r1 .endif .if .count add $4,r1 .endif .if .flow add $8.,r1 .endif .if .prof add $16.,r1 .endif mov r1,entry. sys indir;lblwri bes 3b mov pd,lblwri+2 mov tblmax,lblwri+4 sub pd,lblwri+4 mov saver0,r0 sys indir;lblwri bes 3b mov ml,lblwri+2 mov ml,lblwri+4 neg lblwri+4 mov saver0,r0 sys indir;lblwri bes 3b 9: mov retarea,r0 / set exit status 2: sys exit /------------------------------------------------------------------------------ / System call interface routine /------------------------------------------------------------------------------ R0_IN = 0200 R1_IN = 0100 R0_OUT = 040 R1_OUT = 020 CBIT = 010 lblmon: mon.z: mov (sp)+,r0; / call number from stack cmp r0,$1 / sys exit ? jeq hlt.z / go there bit $177700,r0; / range 0-63? bne mon.bad; / no? bad call movb r0,call; / move call number in system call movb tab(r0),r5; / table lookup call conditions cmp r5,$-1; / compare for special context beq mon.special; / yes? jump to special context monmon: mov r5,r4; / save call conditions rolb r5 / R0_IN bcc 1f / check if bit 7 is on mov (sp)+,r0; / call argument in r0 1: rolb r5 / R1_IN bcc 1f / check if bit 6 is on mov (sp)+,r1; / argument in r1 1: bic $177770,r4 / clear all exept bits 2,1,0 beq 2f / if 0 forget about args mov r3,saver1 / save r3 mov $call+2,r3 / base of args for call 1: mov (sp)+,(r3)+ / move argument sob r4,1b / look for more mov saver1,r3 / restore r3 2: sys indir;call / this is it folks bcc 1f / no error set? forward mov r0,r4 / copy error in r4 1: rolb r5 / R0_OUT bcc 1f / check original bit 5 mov r0,-(sp) / stack r0 from errno 1: rolb r5 / R1_OUT bcc 1f / check original bit 4 mov r1,-(sp) / stack r1 1: rolb r5 / CBIT bcc mon.end / no c-bit then ready mon.cbit: mov r4,-(sp) / stack errno beq mon.end / only once if no error mov r4,-(sp) / stack errno twice when error mon.end: mov $loop,r4 / restore r4 next / ready mon.special: / special calls decoded here cmp r0,$fork / fork? beq mon.fork cmp r0,$signal / signal? beq mon.signal mon.bad: / otherwise undecodable mov saver0,r4 / restore r4 jsr pc,e.badmon / mon call error routine next mon.fork: clr r5 clr r4 sys fork inc r5 bcc 1f mov r0,r4 1: mov r0,-(sp) mov r5,-(sp) br mon.cbit mon.signal: msign: mov (sp)+,r1 / trap number mov (sp)+,r0 / signal number cmp r0,$16. / only 1 - 16 legal bhi sig.bad mov r0,call+2 / move signal number into call beq sig.bad / 0 illegal asl r0 / make 2-32 and even mov sig.trp-2(r0),r5 / previous trap number cmp r1,$256. / values -1 and -2 special bhis 1f mov sig.adr-2(r0),r4 / zero label means illegal signal bne 2f sig.bad: mov $EINVAL,r4 / bad signal jbr mon.cbit / and continue 1: cmp r1,$-3 / -2: reset default, -3: ignore blo sig.bad mov r1,r4 / trap number in r4 inc r4 inc r4 / -2 -> 0, -3 -> -1 2: mov r1,sig.trp-2(r0) / new trap number / -3 if ignored; -2 if default action mov r4,call+4 / translated trap number in call clr r4 sys indir;call bcs sig.bad / unlikely to happen asr r0 / special if old label odd bcc 1f mov $-3,r5 / set ignore signal 1: mov r5,-(sp) / push trap number jbr mon.cbit .data call: sys 0; 0; 0; 0; 0 sig.trp: -2; -2; -2; -2; -2; -2; -2; -2 -2; -2; 21.; 25.; -2; -2; -2; -2 sig.adr: sig1; sig2; sig3; 0; 0; 0; 0; sig8 0; 0; sig11; sig12; sig13; sig14; sig15; sig16 tab: .if V6 .byte -1 / 0 = indir .byte -1 / 1 = exit .byte -1 / 2 = fork .byte 2 +R0_IN +R0_OUT +CBIT / 3 = read .byte 2 +R0_IN +R0_OUT +CBIT / 4 = write .byte 2 +R0_OUT +CBIT / 5 = open .byte 0 +R0_IN +CBIT / 6 = close .byte 0 +R0_OUT +R1_OUT +CBIT / 7 = wait .byte 2 +R0_OUT +CBIT / 8 = creat .byte 2 +CBIT / 9 = link .byte 1 +CBIT / 10 = unlink .byte 2 +CBIT / 11 = exec .byte 1 +CBIT / 12 = chdir .byte 0 +R0_OUT +R1_OUT / 13 = time .byte 3 +CBIT / 14 = mknod .byte 2 +CBIT / 15 = chmod .byte 2 +CBIT / 16 = chown .byte -1 / 17 = break .byte 2 +CBIT / 18 = stat .byte 2 +R0_IN +CBIT / 19 = seek .byte 0 +R0_OUT / 20 = getpid .byte 3 +CBIT / 21 = mount .byte 1 +CBIT / 22 = umount .byte 0 +R0_IN +CBIT / 23 = setuid .byte 0 +R0_OUT / 24 = getuid .byte 0 +R1_IN +R0_IN +CBIT / 25 = stime .byte 3 +R0_IN +R0_OUT +CBIT / 26 = ptrace .byte -1 / 27 = x .byte 1 +R0_IN +CBIT / 28 = fstat .byte -1 / 29 = x .byte -1 / 30 = x .byte 1 +R0_IN +CBIT / 31 = stty .byte 1 +R0_IN +CBIT / 32 = gtty .byte -1 / 33 = x .byte 0 +R0_IN +CBIT / 34 = nice .byte 0 +R0_IN / 35 = sleep .byte 0 / 36 = sync .byte 1 +R0_IN +CBIT / 37 = kill .byte 0 +R0_OUT / 38 = csw .byte -1 / 39 = x .byte -1 / 40 = x .byte 0 +R0_IN +R0_OUT +CBIT / 41 = dup .byte 0 +R0_OUT +R1_OUT +CBIT / 42 = pipe .byte 1 / 43 = times .byte 4 / 44 = prof .byte -1 / 45 = x .byte 0 +R0_IN +CBIT / 46 = setgid .byte 0 +R0_OUT / 47 = getgid .byte -1 / 48 = signal .byte -1 / 49 = reserved for USG .byte -1 / 50 = reserved for USG .byte -1 / 51 = x .byte -1 / 52 = x .byte -1 / 53 = x .byte -1 / 54 = x .byte -1 / 55 = x .byte -1 / 56 = x .byte -1 / 57 = x .byte -1 / 58 = x .byte -1 / 59 = x .byte -1 / 60 = x .byte -1 / 61 = x .byte -1 / 62 = x .byte -1 / 63 = x .endif .if VPLUS .byte -1 / 0 = indir .byte -1 / 1 = exit .byte -1 / 2 = fork .byte 2 +R0_IN +R0_OUT +CBIT / 3 = read .byte 2 +R0_IN +R0_OUT +CBIT / 4 = write .byte 2 +R0_OUT +CBIT / 5 = open .byte 0 +R0_IN +CBIT / 6 = close .byte 0 +R0_OUT +R1_OUT +CBIT / 7 = wait .byte 2 +R0_OUT +CBIT / 8 = creat .byte 2 +CBIT / 9 = link .byte 1 +CBIT / 10 = unlink .byte 2 +CBIT / 11 = exec .byte 1 +CBIT / 12 = chdir .byte 0 +R0_OUT +R1_OUT / 13 = time .byte 3 +CBIT / 14 = mknod .byte 2 +CBIT / 15 = chmod .byte 2 +CBIT / 16 = chown .byte -1 / 17 = break .byte 2 +CBIT / 18 = stat .byte 2 +R0_IN +CBIT / 19 = seek .byte 0 +R0_OUT / 20 = getpid .byte 3 +CBIT / 21 = mount .byte 1 +CBIT / 22 = umount .byte 0 +R0_IN +CBIT / 23 = setuid .byte 0 +R0_OUT / 24 = getuid .byte 0 +R1_IN +R0_IN +CBIT / 25 = stime .byte 3 +R0_IN +R0_OUT +CBIT / 26 = ptrace .byte 0 +R0_IN +R0_OUT / 27 = alarm .byte 1 +R0_IN +CBIT / 28 = fstat .byte 0 / 29 = pause .byte -1 / 30 = x .byte 1 +R0_IN +CBIT / 31 = stty .byte 1 +R0_IN +CBIT / 32 = gtty .byte 2 +CBIT / 33 = access .byte 0 +R0_IN +CBIT / 34 = nice .byte 0 +R0_IN / 35 = sleep .byte 0 / 36 = sync .byte 1 +R0_IN +CBIT / 37 = kill .byte 0 +R0_OUT / 38 = csw .byte -1 / 39 = x .byte 0 +R0_IN +R0_OUT +R1_OUT +CBIT / 40 = tell .byte 0 +R0_IN +R0_OUT +CBIT / 41 = dup .byte 0 +R0_OUT +R1_OUT +CBIT / 42 = pipe .byte 1 / 43 = times .byte 4 / 44 = prof .byte -1 / 45 = x .byte 0 +R0_IN +CBIT / 46 = setgid .byte 0 +R0_OUT / 47 = getgid .byte -1 / 48 = signal .byte -1 / 49 = reserved for USG .byte -1 / 50 = reserved for USG .byte 1 +CBIT / 51 = acct .byte -1 / 52 = x .byte -1 / 53 = x .byte -1 / 54 = x .byte -1 / 55 = x .byte -1 / 56 = x .byte -1 / 57 = x .byte -1 / 58 = x .byte -1 / 59 = x .byte -1 / 60 = x .byte -1 / 61 = x .byte -1 / 62 = x .byte -1 / 63 = x .endif .if V7 .byte -1 / 0 = indir .byte 0 / 1 = exit .byte -1 / 2 = fork .byte 2 +R0_IN +R0_OUT +CBIT / 3 = read .byte 2 +R0_IN +R0_OUT +CBIT / 4 = write .byte 2 +R0_OUT +CBIT / 5 = open .byte 0 +R0_IN +CBIT / 6 = close .byte 0 +R0_OUT +R1_OUT +CBIT / 7 = wait .byte 2 +R0_OUT +CBIT / 8 = creat .byte 2 +CBIT / 9 = link .byte 1 +CBIT / 10 = unlink .byte -1 / 11 = x no exec in em code .byte 1 +CBIT / 12 = chdir .byte -1 / 13 = x time is obsolete .byte 3 +CBIT / 14 = mknod .byte 2 +CBIT / 15 = chmod .byte 3 +CBIT / 16 = chown .byte -1 / 17 = break .byte 2 +CBIT / 18 = stat .byte 3 +R0_IN +R0_OUT +R1_OUT +CBIT / 19 = lseek .byte 0 +R0_OUT / 20 = getpid .byte 3 +CBIT / 21 = mount .byte 1 +CBIT / 22 = umount .byte 0 +R0_IN +CBIT / 23 = setuid .byte 0 +R0_OUT +R1_OUT / 24 = getuid .byte 0 +R1_IN +R0_IN +CBIT / 25 = stime .byte 3 +R0_IN +R0_OUT +CBIT / 26 = ptrace .byte 0 +R0_IN +R0_OUT / 27 = alarm .byte 1 +R0_IN +CBIT / 28 = fstat .byte 0 / 29 = pause .byte 2 +CBIT / 30 = utime .byte -1 / 31 = x .byte -1 / 32 = x .byte 2 +CBIT / 33 = access .byte 0 +R0_IN +CBIT / 34 = nice .byte 1 / 35 = ftime .byte 0 / 36 = sync .byte 1 +R0_IN +CBIT / 37 = kill .byte -1 / 38 = x .byte -1 / 39 = x .byte -1 / 40 = x .byte 0 +R1_IN +R0_IN +R0_OUT +CBIT / 41 = dup .byte 0 +R0_OUT +R1_OUT +CBIT / 42 = pipe .byte 1 / 43 = times .byte 4 / 44 = prof .byte -1 / 45 = x .byte 0 +R0_IN +CBIT / 46 = setgid .byte 0 +R0_OUT +R1_OUT / 47 = getgid .byte -1 / 48 = signal .byte -1 / 49 = reserved for USG .byte -1 / 50 = reserved for USG .byte 1 +CBIT / 51 = acct .byte 3 +CBIT / 52 = phys .byte 1 +CBIT / 53 = lock .byte 3 +CBIT / 54 = ioctl .byte -1 / 55 = x .byte 2 +CBIT / 56 = mpxcall .byte -1 / 57 = x .byte -1 / 58 = x .byte 3 +CBIT / 59 = exece .byte 1 +CBIT / 60 = umask .byte 1 +CBIT / 61 = chroot .byte -1 / 62 = x .byte -1 / 63 = x .endif .text /------------------------------------------------------------------------------ / General subroutines /------------------------------------------------------------------------------ wrdoff: movb (pcx)+,r0 /get first byte swab r0 /put it in high byte clrb r0 /clear low byte of r0 bisb (pcx)+,r0 /"or" second byte in rts pc /done /------------------------------------------------------------------------------ tstr0: cmp r0,$04; jgt e.oddz; cmp r0,$02; jne e.oddz; rts pc chckptr: / this routine traps a pointer outside / the globals, the stack or the heap bit $1,2(sp); bne 8f chckptb: mov 2(sp),r5; cmp r5,sp; bhis 9f cmp r5,hp; bhis 8f .if .count + .prof + .flow cmp r5,tblmax; bhis 9f cmp r5,globmax; bhis 8f .endif cmp r5,eb; bhis 9f 8: jsr pc,e.badptr 9: rts pc .if .last nexttab: mov linused,r5; add $2,r5 / increment lasttab cmp r5,$linused / top of table reached? blo 1f sub $96.,r5 1: mov r5,linused rts pc .endif regsave: mov r5,savearea mov $[savearea+2],r5 mov r4,(r5)+ mov r3,(r5)+ mov r2,(r5) rts pc regretu: mov $[savearea+6],r5 mov (r5),r2 mov -(r5),r3 mov -(r5),r4 mov -(r5),r5 rts pc setfloat: cmp r0,$8. bne 1f setd rts pc 1: cmp r0,$04 bne 3f setf 2: rts pc 3: jmp e.oddz setint: cmp r0,$04 bne 4f setl rts pc 4: cmp r0,$02 bne 3b seti 5: rts pc /------------------------------------------------------------------------------ / Leftover data /------------------------------------------------------------------------------ .bss filb: .=.+2 curproc:.=.+2 linmax: .=.+2 countptr:.=.+2 countfld:.=.+2 hiprof: .=.+2 loprof: .=.+2 ignmask:.=.+2 / ignore mask for trap retsize:.=.+2 / size of return value of function retarea:.=.+8 / return area for function value savearea: .=.+8 / save register area saver0: .=.+2 saver1: .=.+2 header: txtsiz: .=.+2 / program textsize in bytes ndatad: .=.+2 / number of loadfile descriptors nprocs: .=.+2 / number of entries in procedure descriptors option: entry.: .=.+2 / procedure number to start nlines: .=.+2 / maximum sorceline number szdata: .=.+2 / address of lowest uninitialized byte firstp: .=.+2 / descriptor address first basic block of text maxcount: .=.+2 / total number of processable source lines argc: .=.+2 argv: .=.+2 environ: .=.+2 pb: .=.+2 pd: .=.+2 eb: .=.+2 globmax: .=.+2 tblmax: .=.+2 ml: .=.+2 .if .last lasttab:.=.+96. / 16 descriptors of integers + index at the end linused:.=.+2 .endif .if .opfreq counttab: .=.+1664. .endif .if .count lcount: .=.+2 countsiz:.=.+2 .endif .if .flow lflow: .=.+2 flowsiz:.=.+2 .endif .if .prof ltime: .=.+2 profsiz:.=.+2 .endif hp: .=.+2