diff options
author | cinap_lenrek <cinap_lenrek@gmx.de> | 2012-07-30 19:11:16 +0200 |
---|---|---|
committer | cinap_lenrek <cinap_lenrek@gmx.de> | 2012-07-30 19:11:16 +0200 |
commit | 4f33c88a51587681b7be1ae57cfbc43b627c6bc4 (patch) | |
tree | 25560404dc80007e5dc268811242c9071f6a1017 /sys/src/cmd/6c | |
parent | fcc5e75d07e5bc6cb3ddac6d9a437e7ec62d0d95 (diff) |
import updated compilers from sources
Diffstat (limited to 'sys/src/cmd/6c')
-rw-r--r-- | sys/src/cmd/6c/6.out.h | 820 | ||||
-rw-r--r-- | sys/src/cmd/6c/cgen.c | 1942 | ||||
-rw-r--r-- | sys/src/cmd/6c/div.c | 206 | ||||
-rw-r--r-- | sys/src/cmd/6c/enam.c | 669 | ||||
-rw-r--r-- | sys/src/cmd/6c/gc.h | 381 | ||||
-rw-r--r-- | sys/src/cmd/6c/list.c | 337 | ||||
-rw-r--r-- | sys/src/cmd/6c/machcap.c | 78 | ||||
-rw-r--r-- | sys/src/cmd/6c/mkenam | 15 | ||||
-rw-r--r-- | sys/src/cmd/6c/mkfile | 38 | ||||
-rw-r--r-- | sys/src/cmd/6c/mul.c | 428 | ||||
-rw-r--r-- | sys/src/cmd/6c/peep.c | 846 | ||||
-rw-r--r-- | sys/src/cmd/6c/reg.c | 1356 | ||||
-rw-r--r-- | sys/src/cmd/6c/sgen.c | 435 | ||||
-rw-r--r-- | sys/src/cmd/6c/swt.c | 530 | ||||
-rw-r--r-- | sys/src/cmd/6c/sys.c | 106 | ||||
-rw-r--r-- | sys/src/cmd/6c/txt.c | 1515 | ||||
-rw-r--r-- | sys/src/cmd/6c/vlrt.c | 746 |
17 files changed, 10448 insertions, 0 deletions
diff --git a/sys/src/cmd/6c/6.out.h b/sys/src/cmd/6c/6.out.h new file mode 100644 index 000000000..0766ea3d2 --- /dev/null +++ b/sys/src/cmd/6c/6.out.h @@ -0,0 +1,820 @@ +#define NSYM 50 +#define NSNAME 8 +#define NOPROF (1<<0) +#define DUPOK (1<<1) + +/* + * amd64 + */ + +enum as +{ + AXXX, + AAAA, + AAAD, + AAAM, + AAAS, + AADCB, + AADCL, + AADCW, + AADDB, + AADDL, + AADDW, + AADJSP, + AANDB, + AANDL, + AANDW, + AARPL, + ABOUNDL, + ABOUNDW, + ABSFL, + ABSFW, + ABSRL, + ABSRW, + ABTL, + ABTW, + ABTCL, + ABTCW, + ABTRL, + ABTRW, + ABTSL, + ABTSW, + ABYTE, + ACALL, + ACLC, + ACLD, + ACLI, + ACLTS, + ACMC, + ACMPB, + ACMPL, + ACMPW, + ACMPSB, + ACMPSL, + ACMPSW, + ADAA, + ADAS, + ADATA, + ADECB, + ADECL, + ADECQ, + ADECW, + ADIVB, + ADIVL, + ADIVW, + AENTER, + AGLOBL, + AGOK, + AHISTORY, + AHLT, + AIDIVB, + AIDIVL, + AIDIVW, + AIMULB, + AIMULL, + AIMULW, + AINB, + AINL, + AINW, + AINCB, + AINCL, + AINCQ, + AINCW, + AINSB, + AINSL, + AINSW, + AINT, + AINTO, + AIRETL, + AIRETW, + AJCC, + AJCS, + AJCXZ, + AJEQ, + AJGE, + AJGT, + AJHI, + AJLE, + AJLS, + AJLT, + AJMI, + AJMP, + AJNE, + AJOC, + AJOS, + AJPC, + AJPL, + AJPS, + ALAHF, + ALARL, + ALARW, + ALEAL, + ALEAW, + ALEAVEL, + ALEAVEW, + ALOCK, + ALODSB, + ALODSL, + ALODSW, + ALONG, + ALOOP, + ALOOPEQ, + ALOOPNE, + ALSLL, + ALSLW, + AMOVB, + AMOVL, + AMOVW, + AMOVBLSX, + AMOVBLZX, + AMOVBQSX, + AMOVBQZX, + AMOVBWSX, + AMOVBWZX, + AMOVWLSX, + AMOVWLZX, + AMOVWQSX, + AMOVWQZX, + AMOVSB, + AMOVSL, + AMOVSW, + AMULB, + AMULL, + AMULW, + ANAME, + ANEGB, + ANEGL, + ANEGW, + ANOP, + ANOTB, + ANOTL, + ANOTW, + AORB, + AORL, + AORW, + AOUTB, + AOUTL, + AOUTW, + AOUTSB, + AOUTSL, + AOUTSW, + APOPAL, + APOPAW, + APOPFL, + APOPFW, + APOPL, + APOPW, + APUSHAL, + APUSHAW, + APUSHFL, + APUSHFW, + APUSHL, + APUSHW, + ARCLB, + ARCLL, + ARCLW, + ARCRB, + ARCRL, + ARCRW, + AREP, + AREPN, + ARET, + AROLB, + AROLL, + AROLW, + ARORB, + ARORL, + ARORW, + ASAHF, + ASALB, + ASALL, + ASALW, + ASARB, + ASARL, + ASARW, + ASBBB, + ASBBL, + ASBBW, + ASCASB, + ASCASL, + ASCASW, + ASETCC, + ASETCS, + ASETEQ, + ASETGE, + ASETGT, + ASETHI, + ASETLE, + ASETLS, + ASETLT, + ASETMI, + ASETNE, + ASETOC, + ASETOS, + ASETPC, + ASETPL, + ASETPS, + ACDQ, + ACWD, + ASHLB, + ASHLL, + ASHLW, + ASHRB, + ASHRL, + ASHRW, + ASTC, + ASTD, + ASTI, + ASTOSB, + ASTOSL, + ASTOSW, + ASUBB, + ASUBL, + ASUBW, + ASYSCALL, + ATESTB, + ATESTL, + ATESTW, + ATEXT, + AVERR, + AVERW, + AWAIT, + AWORD, + AXCHGB, + AXCHGL, + AXCHGW, + AXLAT, + AXORB, + AXORL, + AXORW, + + AFMOVB, + AFMOVBP, + AFMOVD, + AFMOVDP, + AFMOVF, + AFMOVFP, + AFMOVL, + AFMOVLP, + AFMOVV, + AFMOVVP, + AFMOVW, + AFMOVWP, + AFMOVX, + AFMOVXP, + + AFCOMB, + AFCOMBP, + AFCOMD, + AFCOMDP, + AFCOMDPP, + AFCOMF, + AFCOMFP, + AFCOML, + AFCOMLP, + AFCOMW, + AFCOMWP, + AFUCOM, + AFUCOMP, + AFUCOMPP, + + AFADDDP, + AFADDW, + AFADDL, + AFADDF, + AFADDD, + + AFMULDP, + AFMULW, + AFMULL, + AFMULF, + AFMULD, + + AFSUBDP, + AFSUBW, + AFSUBL, + AFSUBF, + AFSUBD, + + AFSUBRDP, + AFSUBRW, + AFSUBRL, + AFSUBRF, + AFSUBRD, + + AFDIVDP, + AFDIVW, + AFDIVL, + AFDIVF, + AFDIVD, + + AFDIVRDP, + AFDIVRW, + AFDIVRL, + AFDIVRF, + AFDIVRD, + + AFXCHD, + AFFREE, + + AFLDCW, + AFLDENV, + AFRSTOR, + AFSAVE, + AFSTCW, + AFSTENV, + AFSTSW, + + AF2XM1, + AFABS, + AFCHS, + AFCLEX, + AFCOS, + AFDECSTP, + AFINCSTP, + AFINIT, + AFLD1, + AFLDL2E, + AFLDL2T, + AFLDLG2, + AFLDLN2, + AFLDPI, + AFLDZ, + AFNOP, + AFPATAN, + AFPREM, + AFPREM1, + AFPTAN, + AFRNDINT, + AFSCALE, + AFSIN, + AFSINCOS, + AFSQRT, + AFTST, + AFXAM, + AFXTRACT, + AFYL2X, + AFYL2XP1, + + AEND, + + ADYNT, + AINIT, + + ASIGNAME, + + /* extra 32-bit operations */ + ACMPXCHGB, + ACMPXCHGL, + ACMPXCHGW, + ACMPXCHG8B, + ACPUID, + AINVD, + AINVLPG, + ALFENCE, + AMFENCE, + AMOVNTIL, + ARDMSR, + ARDPMC, + ARDTSC, + ARSM, + ASFENCE, + ASYSRET, + AWBINVD, + AWRMSR, + AXADDB, + AXADDL, + AXADDW, + + /* conditional move */ + ACMOVLCC, + ACMOVLCS, + ACMOVLEQ, + ACMOVLGE, + ACMOVLGT, + ACMOVLHI, + ACMOVLLE, + ACMOVLLS, + ACMOVLLT, + ACMOVLMI, + ACMOVLNE, + ACMOVLOC, + ACMOVLOS, + ACMOVLPC, + ACMOVLPL, + ACMOVLPS, + ACMOVQCC, + ACMOVQCS, + ACMOVQEQ, + ACMOVQGE, + ACMOVQGT, + ACMOVQHI, + ACMOVQLE, + ACMOVQLS, + ACMOVQLT, + ACMOVQMI, + ACMOVQNE, + ACMOVQOC, + ACMOVQOS, + ACMOVQPC, + ACMOVQPL, + ACMOVQPS, + ACMOVWCC, + ACMOVWCS, + ACMOVWEQ, + ACMOVWGE, + ACMOVWGT, + ACMOVWHI, + ACMOVWLE, + ACMOVWLS, + ACMOVWLT, + ACMOVWMI, + ACMOVWNE, + ACMOVWOC, + ACMOVWOS, + ACMOVWPC, + ACMOVWPL, + ACMOVWPS, + + /* 64-bit */ + AADCQ, + AADDQ, + AANDQ, + ABSFQ, + ABSRQ, + ABTCQ, + ABTQ, + ABTRQ, + ABTSQ, + ACMPQ, + ACMPSQ, + ACMPXCHGQ, + ACQO, + ADIVQ, + AIDIVQ, + AIMULQ, + AIRETQ, + ALEAQ, + ALEAVEQ, + ALODSQ, + AMOVQ, + AMOVLQSX, + AMOVLQZX, + AMOVNTIQ, + AMOVSQ, + AMULQ, + ANEGQ, + ANOTQ, + AORQ, + APOPFQ, + APOPQ, + APUSHFQ, + APUSHQ, + ARCLQ, + ARCRQ, + AROLQ, + ARORQ, + AQUAD, + ASALQ, + ASARQ, + ASBBQ, + ASCASQ, + ASHLQ, + ASHRQ, + ASTOSQ, + ASUBQ, + ATESTQ, + AXADDQ, + AXCHGQ, + AXORQ, + + /* media */ + AADDPD, + AADDPS, + AADDSD, + AADDSS, + AANDNPD, + AANDNPS, + AANDPD, + AANDPS, + ACMPPD, + ACMPPS, + ACMPSD, + ACMPSS, + ACOMISD, + ACOMISS, + ACVTPD2PL, + ACVTPD2PS, + ACVTPL2PD, + ACVTPL2PS, + ACVTPS2PD, + ACVTPS2PL, + ACVTSD2SL, + ACVTSD2SQ, + ACVTSD2SS, + ACVTSL2SD, + ACVTSL2SS, + ACVTSQ2SD, + ACVTSQ2SS, + ACVTSS2SD, + ACVTSS2SL, + ACVTSS2SQ, + ACVTTPD2PL, + ACVTTPS2PL, + ACVTTSD2SL, + ACVTTSD2SQ, + ACVTTSS2SL, + ACVTTSS2SQ, + ADIVPD, + ADIVPS, + ADIVSD, + ADIVSS, + AEMMS, + AFXRSTOR, + AFXRSTOR64, + AFXSAVE, + AFXSAVE64, + ALDMXCSR, + AMASKMOVOU, + AMASKMOVQ, + AMAXPD, + AMAXPS, + AMAXSD, + AMAXSS, + AMINPD, + AMINPS, + AMINSD, + AMINSS, + AMOVAPD, + AMOVAPS, + AMOVOU, + AMOVHLPS, + AMOVHPD, + AMOVHPS, + AMOVLHPS, + AMOVLPD, + AMOVLPS, + AMOVMSKPD, + AMOVMSKPS, + AMOVNTO, + AMOVNTPD, + AMOVNTPS, + AMOVNTQ, + AMOVO, + AMOVQOZX, + AMOVSD, + AMOVSS, + AMOVUPD, + AMOVUPS, + AMULPD, + AMULPS, + AMULSD, + AMULSS, + AORPD, + AORPS, + APACKSSLW, + APACKSSWB, + APACKUSWB, + APADDB, + APADDL, + APADDQ, + APADDSB, + APADDSW, + APADDUSB, + APADDUSW, + APADDW, + APANDB, + APANDL, + APANDSB, + APANDSW, + APANDUSB, + APANDUSW, + APANDW, + APAND, + APANDN, + APAVGB, + APAVGW, + APCMPEQB, + APCMPEQL, + APCMPEQW, + APCMPGTB, + APCMPGTL, + APCMPGTW, + APEXTRW, + APFACC, + APFADD, + APFCMPEQ, + APFCMPGE, + APFCMPGT, + APFMAX, + APFMIN, + APFMUL, + APFNACC, + APFPNACC, + APFRCP, + APFRCPIT1, + APFRCPI2T, + APFRSQIT1, + APFRSQRT, + APFSUB, + APFSUBR, + APINSRW, + APMADDWL, + APMAXSW, + APMAXUB, + APMINSW, + APMINUB, + APMOVMSKB, + APMULHRW, + APMULHUW, + APMULHW, + APMULLW, + APMULULQ, + APOR, + APSADBW, + APSHUFHW, + APSHUFL, + APSHUFLW, + APSHUFW, + APSLLO, + APSLLL, + APSLLQ, + APSLLW, + APSRAL, + APSRAW, + APSRLO, + APSRLL, + APSRLQ, + APSRLW, + APSUBB, + APSUBL, + APSUBQ, + APSUBSB, + APSUBSW, + APSUBUSB, + APSUBUSW, + APSUBW, + APSWAPL, + APUNPCKHBW, + APUNPCKHLQ, + APUNPCKHQDQ, + APUNPCKHWL, + APUNPCKLBW, + APUNPCKLLQ, + APUNPCKLQDQ, + APUNPCKLWL, + APXOR, + ARCPPS, + ARCPSS, + ARSQRTPS, + ARSQRTSS, + ASHUFPD, + ASHUFPS, + ASQRTPD, + ASQRTPS, + ASQRTSD, + ASQRTSS, + ASTMXCSR, + ASUBPD, + ASUBPS, + ASUBSD, + ASUBSS, + AUCOMISD, + AUCOMISS, + AUNPCKHPD, + AUNPCKHPS, + AUNPCKLPD, + AUNPCKLPS, + AXORPD, + AXORPS, + + APF2IW, + APF2IL, + API2FW, + API2FL, + ARETFW, + ARETFL, + ARETFQ, + ASWAPGS, + + AMODE, + + ALAST +}; + +enum +{ + + D_AL = 0, + D_CL, + D_DL, + D_BL, + D_SPB, + D_BPB, + D_SIB, + D_DIB, + D_R8B, + D_R9B, + D_R10B, + D_R11B, + D_R12B, + D_R13B, + D_R14B, + D_R15B, + + D_AX = 16, + D_CX, + D_DX, + D_BX, + D_SP, + D_BP, + D_SI, + D_DI, + D_R8, + D_R9, + D_R10, + D_R11, + D_R12, + D_R13, + D_R14, + D_R15, + + D_AH = 32, + D_CH, + D_DH, + D_BH, + + D_F0 = 36, + + D_M0 = 44, + + D_X0 = 52, + + D_CS = 68, + D_SS, + D_DS, + D_ES, + D_FS, + D_GS, + + D_GDTR, /* global descriptor table register */ + D_IDTR, /* interrupt descriptor table register */ + D_LDTR, /* local descriptor table register */ + D_MSW, /* machine status word */ + D_TASK, /* task register */ + + D_CR = 79, + D_DR = 95, + D_TR = 103, + + D_NONE = 111, + + D_BRANCH = 112, + D_EXTERN = 113, + D_STATIC = 114, + D_AUTO = 115, + D_PARAM = 116, + D_CONST = 117, + D_FCONST = 118, + D_SCONST = 119, + D_ADDR = 120, + + D_FILE, + D_FILE1, + + D_INDIR, /* additive */ + + T_TYPE = 1<<0, + T_INDEX = 1<<1, + T_OFFSET = 1<<2, + T_FCONST = 1<<3, + T_SYM = 1<<4, + T_SCONST = 1<<5, + T_64 = 1<<6, + + REGARG = D_BP, /* MIGHT CHANGE */ + REGRET = D_AX, + FREGRET = D_X0, + REGSP = D_SP, + REGTMP = D_DI, + REGEXT = D_R15, /* compiler allocates external registers R15 down */ + FREGMIN = D_X0+5, /* first register variable */ + FREGEXT = D_X0+7 /* first external register */ +}; + +/* + * this is the ranlib header + */ +#define SYMDEF "__.SYMDEF" + +/* + * this is the simulated IEEE floating point + */ +typedef struct ieee Ieee; +struct ieee +{ + long l; /* contains ls-man 0xffffffff */ + long h; /* contains sign 0x80000000 + exp 0x7ff00000 + ms-man 0x000fffff */ +}; diff --git a/sys/src/cmd/6c/cgen.c b/sys/src/cmd/6c/cgen.c new file mode 100644 index 000000000..5e6b10b33 --- /dev/null +++ b/sys/src/cmd/6c/cgen.c @@ -0,0 +1,1942 @@ +#include "gc.h" + +/* ,x/^(print|prtree)\(/i/\/\/ */ +int castup(Type*, Type*); + +void +cgen(Node *n, Node *nn) +{ + Node *l, *r, *t; + Prog *p1; + Node nod, nod1, nod2, nod3, nod4; + int o, hardleft; + long v, curs; + vlong c; + + if(debug['g']) { + prtree(nn, "cgen lhs"); + prtree(n, "cgen"); + } + if(n == Z || n->type == T) + return; + if(typesu[n->type->etype]) { + sugen(n, nn, n->type->width); + return; + } + l = n->left; + r = n->right; + o = n->op; + if(n->addable >= INDEXED) { + if(nn == Z) { + switch(o) { + default: + nullwarn(Z, Z); + break; + case OINDEX: + nullwarn(l, r); + break; + } + return; + } + gmove(n, nn); + return; + } + curs = cursafe; + + if(l->complex >= FNX) + if(r != Z && r->complex >= FNX) + switch(o) { + default: + if(cond(o) && typesu[l->type->etype]) + break; + + regret(&nod, r); + cgen(r, &nod); + + regsalloc(&nod1, r); + gmove(&nod, &nod1); + + regfree(&nod); + nod = *n; + nod.right = &nod1; + + cgen(&nod, nn); + return; + + case OFUNC: + case OCOMMA: + case OANDAND: + case OOROR: + case OCOND: + case ODOT: + break; + } + + hardleft = l->addable < INDEXED || l->complex >= FNX; + switch(o) { + default: + diag(n, "unknown op in cgen: %O", o); + break; + + case ONEG: + case OCOM: + if(nn == Z) { + nullwarn(l, Z); + break; + } + regalloc(&nod, l, nn); + cgen(l, &nod); + gopcode(o, n->type, Z, &nod); + gmove(&nod, nn); + regfree(&nod); + break; + + case OAS: + if(l->op == OBIT) + goto bitas; + if(!hardleft) { + if(nn != Z || r->addable < INDEXED || hardconst(r)) { + if(r->complex >= FNX && nn == Z) + regret(&nod, r); + else + regalloc(&nod, r, nn); + cgen(r, &nod); + gmove(&nod, l); + if(nn != Z) + gmove(&nod, nn); + regfree(&nod); + } else + gmove(r, l); + break; + } + if(l->complex >= r->complex) { + if(l->op == OINDEX && immconst(r)) { + gmove(r, l); + break; + } + reglcgen(&nod1, l, Z); + if(r->addable >= INDEXED && !hardconst(r)) { + gmove(r, &nod1); + if(nn != Z) + gmove(r, nn); + regfree(&nod1); + break; + } + regalloc(&nod, r, nn); + cgen(r, &nod); + } else { + regalloc(&nod, r, nn); + cgen(r, &nod); + reglcgen(&nod1, l, Z); + } + gmove(&nod, &nod1); + regfree(&nod); + regfree(&nod1); + break; + + bitas: + n = l->left; + regalloc(&nod, r, nn); + if(l->complex >= r->complex) { + reglcgen(&nod1, n, Z); + cgen(r, &nod); + } else { + cgen(r, &nod); + reglcgen(&nod1, n, Z); + } + regalloc(&nod2, n, Z); + gmove(&nod1, &nod2); + bitstore(l, &nod, &nod1, &nod2, nn); + break; + + case OBIT: + if(nn == Z) { + nullwarn(l, Z); + break; + } + bitload(n, &nod, Z, Z, nn); + gmove(&nod, nn); + regfree(&nod); + break; + + case OLSHR: + case OASHL: + case OASHR: + if(nn == Z) { + nullwarn(l, r); + break; + } + if(r->op == OCONST) { + if(r->vconst == 0) { + cgen(l, nn); + break; + } + regalloc(&nod, l, nn); + cgen(l, &nod); + if(o == OASHL && r->vconst == 1) + gopcode(OADD, n->type, &nod, &nod); + else + gopcode(o, n->type, r, &nod); + gmove(&nod, nn); + regfree(&nod); + break; + } + + /* + * get nod to be D_CX + */ + if(nodreg(&nod, nn, D_CX)) { + regsalloc(&nod1, n); + gmove(&nod, &nod1); + cgen(n, &nod); /* probably a bug */ + gmove(&nod, nn); + gmove(&nod1, &nod); + break; + } + reg[D_CX]++; + if(nn->op == OREGISTER && nn->reg == D_CX) + regalloc(&nod1, l, Z); + else + regalloc(&nod1, l, nn); + if(r->complex >= l->complex) { + cgen(r, &nod); + cgen(l, &nod1); + } else { + cgen(l, &nod1); + cgen(r, &nod); + } + gopcode(o, n->type, &nod, &nod1); + gmove(&nod1, nn); + regfree(&nod); + regfree(&nod1); + break; + + case OADD: + case OSUB: + case OOR: + case OXOR: + case OAND: + if(nn == Z) { + nullwarn(l, r); + break; + } + if(typefd[n->type->etype]) + goto fop; + if(r->op == OCONST) { + if(r->vconst == 0 && o != OAND) { + cgen(l, nn); + break; + } + } + if(n->op == OADD && l->op == OASHL && l->right->op == OCONST + && (r->op != OCONST || r->vconst < -128 || r->vconst > 127)) { + c = l->right->vconst; + if(c > 0 && c <= 3) { + if(l->left->complex >= r->complex) { + regalloc(&nod, l->left, nn); + cgen(l->left, &nod); + if(r->addable < INDEXED) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + genmuladd(&nod, &nod, 1 << c, &nod1); + regfree(&nod1); + } + else + genmuladd(&nod, &nod, 1 << c, r); + } + else { + regalloc(&nod, r, nn); + cgen(r, &nod); + regalloc(&nod1, l->left, Z); + cgen(l->left, &nod1); + genmuladd(&nod, &nod1, 1 << c, &nod); + regfree(&nod1); + } + gmove(&nod, nn); + regfree(&nod); + break; + } + } + if(r->addable >= INDEXED && !hardconst(r)) { + regalloc(&nod, l, nn); + cgen(l, &nod); + gopcode(o, n->type, r, &nod); + gmove(&nod, nn); + regfree(&nod); + break; + } + if(l->complex >= r->complex) { + regalloc(&nod, l, nn); + cgen(l, &nod); + regalloc(&nod1, r, Z); + cgen(r, &nod1); + gopcode(o, n->type, &nod1, &nod); + } else { + regalloc(&nod1, r, nn); + cgen(r, &nod1); + regalloc(&nod, l, Z); + cgen(l, &nod); + gopcode(o, n->type, &nod1, &nod); + } + gmove(&nod, nn); + regfree(&nod); + regfree(&nod1); + break; + + case OLMOD: + case OMOD: + case OLMUL: + case OLDIV: + case OMUL: + case ODIV: + if(nn == Z) { + nullwarn(l, r); + break; + } + if(typefd[n->type->etype]) + goto fop; + if(r->op == OCONST && typechl[n->type->etype]) { /* TO DO */ + SET(v); + switch(o) { + case ODIV: + case OMOD: + c = r->vconst; + if(c < 0) + c = -c; + v = log2(c); + if(v < 0) + break; + /* fall thru */ + case OMUL: + case OLMUL: + regalloc(&nod, l, nn); + cgen(l, &nod); + switch(o) { + case OMUL: + case OLMUL: + mulgen(n->type, r, &nod); + break; + case ODIV: + sdiv2(r->vconst, v, l, &nod); + break; + case OMOD: + smod2(r->vconst, v, l, &nod); + break; + } + gmove(&nod, nn); + regfree(&nod); + goto done; + case OLDIV: + c = r->vconst; + if((c & 0x80000000) == 0) + break; + regalloc(&nod1, l, Z); + cgen(l, &nod1); + regalloc(&nod, l, nn); + zeroregm(&nod); + gins(ACMPL, &nod1, nodconst(c)); + gins(ASBBL, nodconst(-1), &nod); + regfree(&nod1); + gmove(&nod, nn); + regfree(&nod); + goto done; + } + } + + if(o == OMUL) { + if(l->addable >= INDEXED) { + t = l; + l = r; + r = t; + } + /* should favour AX */ + regalloc(&nod, l, nn); + cgen(l, &nod); + if(r->addable < INDEXED || hardconst(r)) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + gopcode(OMUL, n->type, &nod1, &nod); + regfree(&nod1); + }else + gopcode(OMUL, n->type, r, &nod); /* addressible */ + gmove(&nod, nn); + regfree(&nod); + break; + } + + /* + * get nod to be D_AX + * get nod1 to be D_DX + */ + if(nodreg(&nod, nn, D_AX)) { + regsalloc(&nod2, n); + gmove(&nod, &nod2); + v = reg[D_AX]; + reg[D_AX] = 0; + + if(isreg(l, D_AX)) { + nod3 = *n; + nod3.left = &nod2; + cgen(&nod3, nn); + } else + if(isreg(r, D_AX)) { + nod3 = *n; + nod3.right = &nod2; + cgen(&nod3, nn); + } else + cgen(n, nn); + + gmove(&nod2, &nod); + reg[D_AX] = v; + break; + } + if(nodreg(&nod1, nn, D_DX)) { + regsalloc(&nod2, n); + gmove(&nod1, &nod2); + v = reg[D_DX]; + reg[D_DX] = 0; + + if(isreg(l, D_DX)) { + nod3 = *n; + nod3.left = &nod2; + cgen(&nod3, nn); + } else + if(isreg(r, D_DX)) { + nod3 = *n; + nod3.right = &nod2; + cgen(&nod3, nn); + } else + cgen(n, nn); + + gmove(&nod2, &nod1); + reg[D_DX] = v; + break; + } + reg[D_AX]++; + + if(r->op == OCONST && (o == ODIV || o == OLDIV) && immconst(r) && typechl[r->type->etype]) { + reg[D_DX]++; + if(l->addable < INDEXED) { + regalloc(&nod2, l, Z); + cgen(l, &nod2); + l = &nod2; + } + if(o == ODIV) + sdivgen(l, r, &nod, &nod1); + else + udivgen(l, r, &nod, &nod1); + gmove(&nod1, nn); + if(l == &nod2) + regfree(l); + goto freeaxdx; + } + + if(l->complex >= r->complex) { + cgen(l, &nod); + reg[D_DX]++; + if(o == ODIV || o == OMOD) + gins(typechl[l->type->etype]? ACDQ: ACQO, Z, Z); + if(o == OLDIV || o == OLMOD) + zeroregm(&nod1); + if(r->addable < INDEXED || r->op == OCONST) { + regalloc(&nod3, r, Z); + cgen(r, &nod3); + gopcode(o, n->type, &nod3, Z); + regfree(&nod3); + } else + gopcode(o, n->type, r, Z); + } else { + regsalloc(&nod3, r); + cgen(r, &nod3); + cgen(l, &nod); + reg[D_DX]++; + if(o == ODIV || o == OMOD) + gins(typechl[l->type->etype]? ACDQ: ACQO, Z, Z); + if(o == OLDIV || o == OLMOD) + zeroregm(&nod1); + gopcode(o, n->type, &nod3, Z); + } + if(o == OMOD || o == OLMOD) + gmove(&nod1, nn); + else + gmove(&nod, nn); + freeaxdx: + regfree(&nod); + regfree(&nod1); + break; + + case OASLSHR: + case OASASHL: + case OASASHR: + if(r->op == OCONST) + goto asand; + if(l->op == OBIT) + goto asbitop; + if(typefd[n->type->etype]) + goto asand; /* can this happen? */ + + /* + * get nod to be D_CX + */ + if(nodreg(&nod, nn, D_CX)) { + regsalloc(&nod1, n); + gmove(&nod, &nod1); + cgen(n, &nod); + if(nn != Z) + gmove(&nod, nn); + gmove(&nod1, &nod); + break; + } + reg[D_CX]++; + + if(r->complex >= l->complex) { + cgen(r, &nod); + if(hardleft) + reglcgen(&nod1, l, Z); + else + nod1 = *l; + } else { + if(hardleft) + reglcgen(&nod1, l, Z); + else + nod1 = *l; + cgen(r, &nod); + } + + gopcode(o, l->type, &nod, &nod1); + regfree(&nod); + if(nn != Z) + gmove(&nod1, nn); + if(hardleft) + regfree(&nod1); + break; + + case OASAND: + case OASADD: + case OASSUB: + case OASXOR: + case OASOR: + asand: + if(l->op == OBIT) + goto asbitop; + if(typefd[l->type->etype] || typefd[r->type->etype]) + goto asfop; + if(l->complex >= r->complex) { + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + if(!immconst(r)) { + regalloc(&nod1, r, nn); + cgen(r, &nod1); + gopcode(o, l->type, &nod1, &nod); + regfree(&nod1); + } else + gopcode(o, l->type, r, &nod); + } else { + regalloc(&nod1, r, nn); + cgen(r, &nod1); + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + gopcode(o, l->type, &nod1, &nod); + regfree(&nod1); + } + if(nn != Z) + gmove(&nod, nn); + if(hardleft) + regfree(&nod); + break; + + asfop: + if(l->complex >= r->complex) { + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + if(r->addable < INDEXED){ + regalloc(&nod1, r, nn); + cgen(r, &nod1); + }else + nod1 = *r; + regalloc(&nod2, r, Z); + gmove(&nod, &nod2); + gopcode(o, r->type, &nod1, &nod2); + gmove(&nod2, &nod); + regfree(&nod2); + if(r->addable < INDEXED) + regfree(&nod1); + } else { + regalloc(&nod1, r, nn); + cgen(r, &nod1); + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + if(o != OASMUL && o != OASADD || !typefd[l->type->etype]) { + regalloc(&nod2, r, Z); + gmove(&nod, &nod2); + gopcode(o, r->type, &nod1, &nod2); + regfree(&nod1); + gmove(&nod2, &nod); + regfree(&nod2); + } else { + gopcode(o, r->type, &nod, &nod1); + gmove(&nod1, &nod); + regfree(&nod1); + } + } + if(nn != Z) + gmove(&nod, nn); + if(hardleft) + regfree(&nod); + break; + + case OASLMUL: + case OASLDIV: + case OASLMOD: + case OASMUL: + case OASDIV: + case OASMOD: + if(l->op == OBIT) + goto asbitop; + if(typefd[n->type->etype] || typefd[r->type->etype]) + goto asfop; + if(r->op == OCONST && typechl[n->type->etype]) { + SET(v); + switch(o) { + case OASDIV: + case OASMOD: + c = r->vconst; + if(c < 0) + c = -c; + v = log2(c); + if(v < 0) + break; + /* fall thru */ + case OASMUL: + case OASLMUL: + if(hardleft) + reglcgen(&nod2, l, Z); + else + nod2 = *l; + regalloc(&nod, l, nn); + cgen(&nod2, &nod); + switch(o) { + case OASMUL: + case OASLMUL: + mulgen(n->type, r, &nod); + break; + case OASDIV: + sdiv2(r->vconst, v, l, &nod); + break; + case OASMOD: + smod2(r->vconst, v, l, &nod); + break; + } + havev: + gmove(&nod, &nod2); + if(nn != Z) + gmove(&nod, nn); + if(hardleft) + regfree(&nod2); + regfree(&nod); + goto done; + case OASLDIV: + c = r->vconst; + if((c & 0x80000000) == 0) + break; + if(hardleft) + reglcgen(&nod2, l, Z); + else + nod2 = *l; + regalloc(&nod1, l, nn); + cgen(&nod2, &nod1); + regalloc(&nod, l, nn); + zeroregm(&nod); + gins(ACMPL, &nod1, nodconst(c)); + gins(ASBBL, nodconst(-1), &nod); + regfree(&nod1); + goto havev; + } + } + + if(o == OASMUL) { + /* should favour AX */ + regalloc(&nod, l, nn); + if(r->complex >= FNX) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + r = &nod1; + } + if(hardleft) + reglcgen(&nod2, l, Z); + else + nod2 = *l; + cgen(&nod2, &nod); + if(r->addable < INDEXED || hardconst(r)) { + if(r->complex < FNX) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + } + gopcode(OASMUL, n->type, &nod1, &nod); + regfree(&nod1); + } + else + gopcode(OASMUL, n->type, r, &nod); + if(r == &nod1) + regfree(r); + gmove(&nod, &nod2); + if(nn != Z) + gmove(&nod, nn); + regfree(&nod); + if(hardleft) + regfree(&nod2); + break; + } + + /* + * get nod to be D_AX + * get nod1 to be D_DX + */ + if(nodreg(&nod, nn, D_AX)) { + regsalloc(&nod2, n); + gmove(&nod, &nod2); + v = reg[D_AX]; + reg[D_AX] = 0; + + if(isreg(l, D_AX)) { + nod3 = *n; + nod3.left = &nod2; + cgen(&nod3, nn); + } else + if(isreg(r, D_AX)) { + nod3 = *n; + nod3.right = &nod2; + cgen(&nod3, nn); + } else + cgen(n, nn); + + gmove(&nod2, &nod); + reg[D_AX] = v; + break; + } + if(nodreg(&nod1, nn, D_DX)) { + regsalloc(&nod2, n); + gmove(&nod1, &nod2); + v = reg[D_DX]; + reg[D_DX] = 0; + + if(isreg(l, D_DX)) { + nod3 = *n; + nod3.left = &nod2; + cgen(&nod3, nn); + } else + if(isreg(r, D_DX)) { + nod3 = *n; + nod3.right = &nod2; + cgen(&nod3, nn); + } else + cgen(n, nn); + + gmove(&nod2, &nod1); + reg[D_DX] = v; + break; + } + reg[D_AX]++; + reg[D_DX]++; + + if(l->complex >= r->complex) { + if(hardleft) + reglcgen(&nod2, l, Z); + else + nod2 = *l; + cgen(&nod2, &nod); + if(r->op == OCONST && typechl[r->type->etype]) { + switch(o) { + case OASDIV: + sdivgen(&nod2, r, &nod, &nod1); + goto divdone; + case OASLDIV: + udivgen(&nod2, r, &nod, &nod1); + divdone: + gmove(&nod1, &nod2); + if(nn != Z) + gmove(&nod1, nn); + goto freelxaxdx; + } + } + if(o == OASDIV || o == OASMOD) + gins(typechl[l->type->etype]? ACDQ: ACQO, Z, Z); + if(o == OASLDIV || o == OASLMOD) + zeroregm(&nod1); + if(r->addable < INDEXED || r->op == OCONST || + !typeil[r->type->etype]) { + regalloc(&nod3, r, Z); + cgen(r, &nod3); + gopcode(o, l->type, &nod3, Z); + regfree(&nod3); + } else + gopcode(o, n->type, r, Z); + } else { + regalloc(&nod3, r, Z); + cgen(r, &nod3); + if(hardleft) + reglcgen(&nod2, l, Z); + else + nod2 = *l; + cgen(&nod2, &nod); + if(o == OASDIV || o == OASMOD) + gins(typechl[l->type->etype]? ACDQ: ACQO, Z, Z); + if(o == OASLDIV || o == OASLMOD) + zeroregm(&nod1); + gopcode(o, l->type, &nod3, Z); + regfree(&nod3); + } + if(o == OASMOD || o == OASLMOD) { + gmove(&nod1, &nod2); + if(nn != Z) + gmove(&nod1, nn); + } else { + gmove(&nod, &nod2); + if(nn != Z) + gmove(&nod, nn); + } + freelxaxdx: + if(hardleft) + regfree(&nod2); + regfree(&nod); + regfree(&nod1); + break; + + fop: + if(l->complex >= r->complex) { + regalloc(&nod, l, nn); + cgen(l, &nod); + if(r->addable < INDEXED) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + gopcode(o, n->type, &nod1, &nod); + regfree(&nod1); + } else + gopcode(o, n->type, r, &nod); + } else { + /* TO DO: could do better with r->addable >= INDEXED */ + regalloc(&nod1, r, Z); + cgen(r, &nod1); + regalloc(&nod, l, nn); + cgen(l, &nod); + gopcode(o, n->type, &nod1, &nod); + regfree(&nod1); + } + gmove(&nod, nn); + regfree(&nod); + break; + + asbitop: + regalloc(&nod4, n, nn); + if(l->complex >= r->complex) { + bitload(l, &nod, &nod1, &nod2, &nod4); + regalloc(&nod3, r, Z); + cgen(r, &nod3); + } else { + regalloc(&nod3, r, Z); + cgen(r, &nod3); + bitload(l, &nod, &nod1, &nod2, &nod4); + } + gmove(&nod, &nod4); + + { /* TO DO: check floating point source */ + Node onod; + + /* incredible grot ... */ + onod = nod3; + onod.op = o; + onod.complex = 2; + onod.addable = 0; + onod.type = tfield; + onod.left = &nod4; + onod.right = &nod3; + cgen(&onod, Z); + } + regfree(&nod3); + gmove(&nod4, &nod); + regfree(&nod4); + bitstore(l, &nod, &nod1, &nod2, nn); + break; + + case OADDR: + if(nn == Z) { + nullwarn(l, Z); + break; + } + lcgen(l, nn); + break; + + case OFUNC: + if(l->complex >= FNX) { + if(l->op != OIND) + diag(n, "bad function call"); + + regret(&nod, l->left); + cgen(l->left, &nod); + regsalloc(&nod1, l->left); + gmove(&nod, &nod1); + regfree(&nod); + + nod = *n; + nod.left = &nod2; + nod2 = *l; + nod2.left = &nod1; + nod2.complex = 1; + cgen(&nod, nn); + + return; + } + o = reg[REGARG]; + gargs(r, &nod, &nod1); + if(l->addable < INDEXED) { + reglcgen(&nod, l, nn); + nod.op = OREGISTER; + gopcode(OFUNC, n->type, Z, &nod); + regfree(&nod); + } else + gopcode(OFUNC, n->type, Z, l); + if(REGARG) + if(o != reg[REGARG]) + reg[REGARG]--; + if(nn != Z) { + regret(&nod, n); + gmove(&nod, nn); + regfree(&nod); + } + break; + + case OIND: + if(nn == Z) { + nullwarn(l, Z); + break; + } + regialloc(&nod, n, nn); + r = l; + while(r->op == OADD) + r = r->right; + if(sconst(r)) { + v = r->vconst; + r->vconst = 0; + cgen(l, &nod); + nod.xoffset += v; + r->vconst = v; + } else + cgen(l, &nod); + regind(&nod, n); + gmove(&nod, nn); + regfree(&nod); + break; + + case OEQ: + case ONE: + case OLE: + case OLT: + case OGE: + case OGT: + case OLO: + case OLS: + case OHI: + case OHS: + if(nn == Z) { + nullwarn(l, r); + break; + } + boolgen(n, 1, nn); + break; + + case OANDAND: + case OOROR: + boolgen(n, 1, nn); + if(nn == Z) + patch(p, pc); + break; + + case ONOT: + if(nn == Z) { + nullwarn(l, Z); + break; + } + boolgen(n, 1, nn); + break; + + case OCOMMA: + cgen(l, Z); + cgen(r, nn); + break; + + case OCAST: + if(nn == Z) { + nullwarn(l, Z); + break; + } + /* + * convert from types l->n->nn + */ + if(nocast(l->type, n->type) && nocast(n->type, nn->type)) { + /* both null, gen l->nn */ + cgen(l, nn); + break; + } + if(ewidth[n->type->etype] < ewidth[l->type->etype]){ + if(l->type->etype == TIND && typechlp[n->type->etype]) + warn(n, "conversion of pointer to shorter integer"); + }else if(0){ + if(nocast(n->type, nn->type) || castup(n->type, nn->type)){ + if(typefd[l->type->etype] != typefd[nn->type->etype]) + regalloc(&nod, l, nn); + else + regalloc(&nod, nn, nn); + cgen(l, &nod); + gmove(&nod, nn); + regfree(&nod); + break; + } + } + regalloc(&nod, l, nn); + cgen(l, &nod); + regalloc(&nod1, n, &nod); + gmove(&nod, &nod1); + gmove(&nod1, nn); + regfree(&nod1); + regfree(&nod); + break; + + case ODOT: + sugen(l, nodrat, l->type->width); + if(nn == Z) + break; + warn(n, "non-interruptable temporary"); + nod = *nodrat; + if(!r || r->op != OCONST) { + diag(n, "DOT and no offset"); + break; + } + nod.xoffset += (long)r->vconst; + nod.type = n->type; + cgen(&nod, nn); + break; + + case OCOND: + bcgen(l, 1); + p1 = p; + cgen(r->left, nn); + gbranch(OGOTO); + patch(p1, pc); + p1 = p; + cgen(r->right, nn); + patch(p1, pc); + break; + + case OPOSTINC: + case OPOSTDEC: + v = 1; + if(l->type->etype == TIND) + v = l->type->link->width; + if(o == OPOSTDEC) + v = -v; + if(l->op == OBIT) + goto bitinc; + if(nn == Z) + goto pre; + + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + + gmove(&nod, nn); + if(typefd[n->type->etype]) { + regalloc(&nod1, l, Z); + gmove(&nod, &nod1); + if(v < 0) + gopcode(OSUB, n->type, nodfconst(-v), &nod1); + else + gopcode(OADD, n->type, nodfconst(v), &nod1); + gmove(&nod1, &nod); + regfree(&nod1); + } else + gopcode(OADD, n->type, nodconst(v), &nod); + if(hardleft) + regfree(&nod); + break; + + case OPREINC: + case OPREDEC: + v = 1; + if(l->type->etype == TIND) + v = l->type->link->width; + if(o == OPREDEC) + v = -v; + if(l->op == OBIT) + goto bitinc; + + pre: + if(hardleft) + reglcgen(&nod, l, Z); + else + nod = *l; + if(typefd[n->type->etype]) { + regalloc(&nod1, l, Z); + gmove(&nod, &nod1); + if(v < 0) + gopcode(OSUB, n->type, nodfconst(-v), &nod1); + else + gopcode(OADD, n->type, nodfconst(v), &nod1); + gmove(&nod1, &nod); + regfree(&nod1); + } else + gopcode(OADD, n->type, nodconst(v), &nod); + if(nn != Z) + gmove(&nod, nn); + if(hardleft) + regfree(&nod); + break; + + bitinc: + if(nn != Z && (o == OPOSTINC || o == OPOSTDEC)) { + bitload(l, &nod, &nod1, &nod2, Z); + gmove(&nod, nn); + gopcode(OADD, tfield, nodconst(v), &nod); + bitstore(l, &nod, &nod1, &nod2, Z); + break; + } + bitload(l, &nod, &nod1, &nod2, nn); + gopcode(OADD, tfield, nodconst(v), &nod); + bitstore(l, &nod, &nod1, &nod2, nn); + break; + } +done: + cursafe = curs; +} + +void +reglcgen(Node *t, Node *n, Node *nn) +{ + Node *r; + long v; + + regialloc(t, n, nn); + if(n->op == OIND) { + r = n->left; + while(r->op == OADD) + r = r->right; + if(sconst(r)) { + v = r->vconst; + r->vconst = 0; + lcgen(n, t); + t->xoffset += v; + r->vconst = v; + regind(t, n); + return; + } + } + lcgen(n, t); + regind(t, n); +} + +void +lcgen(Node *n, Node *nn) +{ + Prog *p1; + Node nod; + + if(debug['g']) { + prtree(nn, "lcgen lhs"); + prtree(n, "lcgen"); + } + if(n == Z || n->type == T) + return; + if(nn == Z) { + nn = &nod; + regalloc(&nod, n, Z); + } + switch(n->op) { + default: + if(n->addable < INDEXED) { + diag(n, "unknown op in lcgen: %O", n->op); + break; + } + gopcode(OADDR, n->type, n, nn); + break; + + case OCOMMA: + cgen(n->left, n->left); + lcgen(n->right, nn); + break; + + case OIND: + cgen(n->left, nn); + break; + + case OCOND: + bcgen(n->left, 1); + p1 = p; + lcgen(n->right->left, nn); + gbranch(OGOTO); + patch(p1, pc); + p1 = p; + lcgen(n->right->right, nn); + patch(p1, pc); + break; + } +} + +void +bcgen(Node *n, int true) +{ + + if(n->type == T) + gbranch(OGOTO); + else + boolgen(n, true, Z); +} + +void +boolgen(Node *n, int true, Node *nn) +{ + int o; + Prog *p1, *p2; + Node *l, *r, nod, nod1; + long curs; + + if(debug['g']) { + prtree(nn, "boolgen lhs"); + prtree(n, "boolgen"); + } + curs = cursafe; + l = n->left; + r = n->right; + switch(n->op) { + + default: + o = ONE; + if(true) + o = OEQ; + /* bad, 13 is address of external that becomes constant */ + if(n->addable >= INDEXED && n->addable != 13) { + if(typefd[n->type->etype]) { + regalloc(&nod1, n, Z); + gmove(nodfconst(0.0), &nod1); /* TO DO: FREGZERO */ + gopcode(o, n->type, n, &nod1); + regfree(&nod1); + } else + gopcode(o, n->type, n, nodconst(0)); + goto com; + } + regalloc(&nod, n, nn); + cgen(n, &nod); + if(typefd[n->type->etype]) { + regalloc(&nod1, n, Z); + gmove(nodfconst(0.0), &nod1); /* TO DO: FREGZERO */ + gopcode(o, n->type, &nod, &nod1); + regfree(&nod1); + } else + gopcode(o, n->type, &nod, nodconst(0)); + regfree(&nod); + goto com; + + case OCONST: + o = vconst(n); + if(!true) + o = !o; + gbranch(OGOTO); + if(o) { + p1 = p; + gbranch(OGOTO); + patch(p1, pc); + } + goto com; + + case OCOMMA: + cgen(l, Z); + boolgen(r, true, nn); + break; + + case ONOT: + boolgen(l, !true, nn); + break; + + case OCOND: + bcgen(l, 1); + p1 = p; + bcgen(r->left, true); + p2 = p; + gbranch(OGOTO); + patch(p1, pc); + p1 = p; + bcgen(r->right, !true); + patch(p2, pc); + p2 = p; + gbranch(OGOTO); + patch(p1, pc); + patch(p2, pc); + goto com; + + case OANDAND: + if(!true) + goto caseor; + + caseand: + bcgen(l, true); + p1 = p; + bcgen(r, !true); + p2 = p; + patch(p1, pc); + gbranch(OGOTO); + patch(p2, pc); + goto com; + + case OOROR: + if(!true) + goto caseand; + + caseor: + bcgen(l, !true); + p1 = p; + bcgen(r, !true); + p2 = p; + gbranch(OGOTO); + patch(p1, pc); + patch(p2, pc); + goto com; + + case OEQ: + case ONE: + case OLE: + case OLT: + case OGE: + case OGT: + case OHI: + case OHS: + case OLO: + case OLS: + o = n->op; + if(true) + o = comrel[relindex(o)]; + if(l->complex >= FNX && r->complex >= FNX) { + regret(&nod, r); + cgen(r, &nod); + regsalloc(&nod1, r); + gmove(&nod, &nod1); + regfree(&nod); + nod = *n; + nod.right = &nod1; + boolgen(&nod, true, nn); + break; + } + if(immconst(l)) { + o = invrel[relindex(o)]; + /* bad, 13 is address of external that becomes constant */ + if(r->addable < INDEXED || r->addable == 13) { + regalloc(&nod, r, nn); + cgen(r, &nod); + gopcode(o, l->type, &nod, l); + regfree(&nod); + } else + gopcode(o, l->type, r, l); + goto com; + } + if(typefd[l->type->etype]) + o = invrel[relindex(logrel[relindex(o)])]; + if(l->complex >= r->complex) { + regalloc(&nod, l, nn); + cgen(l, &nod); + if(r->addable < INDEXED || hardconst(r) || typefd[l->type->etype]) { + regalloc(&nod1, r, Z); + cgen(r, &nod1); + gopcode(o, l->type, &nod, &nod1); + regfree(&nod1); + } else + gopcode(o, l->type, &nod, r); + regfree(&nod); + goto com; + } + regalloc(&nod, r, nn); + cgen(r, &nod); + if(l->addable < INDEXED || l->addable == 13 || hardconst(l)) { + regalloc(&nod1, l, Z); + cgen(l, &nod1); + if(typechl[l->type->etype] && ewidth[l->type->etype] <= ewidth[TINT]) + gopcode(o, types[TINT], &nod1, &nod); + else + gopcode(o, l->type, &nod1, &nod); + regfree(&nod1); + } else + gopcode(o, l->type, l, &nod); + regfree(&nod); + + com: + if(nn != Z) { + p1 = p; + gmove(nodconst(1L), nn); + gbranch(OGOTO); + p2 = p; + patch(p1, pc); + gmove(nodconst(0L), nn); + patch(p2, pc); + } + break; + } + cursafe = curs; +} + +void +sugen(Node *n, Node *nn, long w) +{ + Prog *p1; + Node nod0, nod1, nod2, nod3, nod4, *l, *r; + Type *t; + int c, mt, mo; + vlong o0, o1; + + if(n == Z || n->type == T) + return; + if(debug['g']) { + prtree(nn, "sugen lhs"); + prtree(n, "sugen"); + } + if(nn == nodrat) + if(w > nrathole) + nrathole = w; + switch(n->op) { + case OIND: + if(nn == Z) { + nullwarn(n->left, Z); + break; + } + + default: + goto copy; + + case OCONST: + goto copy; + + case ODOT: + l = n->left; + sugen(l, nodrat, l->type->width); + if(nn == Z) + break; + warn(n, "non-interruptable temporary"); + nod1 = *nodrat; + r = n->right; + if(!r || r->op != OCONST) { + diag(n, "DOT and no offset"); + break; + } + nod1.xoffset += (long)r->vconst; + nod1.type = n->type; + sugen(&nod1, nn, w); + break; + + case OSTRUCT: + /* + * rewrite so lhs has no fn call + */ + if(nn != Z && side(nn)) { + nod1 = *n; + nod1.type = typ(TIND, n->type); + regret(&nod2, &nod1); + lcgen(nn, &nod2); + regsalloc(&nod0, &nod1); + cgen(&nod2, &nod0); + regfree(&nod2); + + nod1 = *n; + nod1.op = OIND; + nod1.left = &nod0; + nod1.right = Z; + nod1.complex = 1; + + sugen(n, &nod1, w); + return; + } + + r = n->left; + for(t = n->type->link; t != T; t = t->down) { + l = r; + if(r->op == OLIST) { + l = r->left; + r = r->right; + } + if(nn == Z) { + cgen(l, nn); + continue; + } + /* + * hand craft *(&nn + o) = l + */ + nod0 = znode; + nod0.op = OAS; + nod0.type = t; + nod0.left = &nod1; + nod0.right = nil; + + nod1 = znode; + nod1.op = OIND; + nod1.type = t; + nod1.left = &nod2; + + nod2 = znode; + nod2.op = OADD; + nod2.type = typ(TIND, t); + nod2.left = &nod3; + nod2.right = &nod4; + + nod3 = znode; + nod3.op = OADDR; + nod3.type = nod2.type; + nod3.left = nn; + + nod4 = znode; + nod4.op = OCONST; + nod4.type = nod2.type; + nod4.vconst = t->offset; + + ccom(&nod0); + acom(&nod0); + xcom(&nod0); + nod0.addable = 0; + nod0.right = l; + + /* prtree(&nod0, "hand craft"); /* */ + cgen(&nod0, Z); + } + break; + + case OAS: + if(nn == Z) { + if(n->addable < INDEXED) + sugen(n->right, n->left, w); + break; + } + + sugen(n->right, nodrat, w); + warn(n, "non-interruptable temporary"); + sugen(nodrat, n->left, w); + sugen(nodrat, nn, w); + break; + + case OFUNC: + if(nn == Z) { + sugen(n, nodrat, w); + break; + } + if(nn->op != OIND) { + nn = new1(OADDR, nn, Z); + nn->type = types[TIND]; + nn->addable = 0; + } else + nn = nn->left; + n = new(OFUNC, n->left, new(OLIST, nn, n->right)); + n->type = types[TVOID]; + n->left->type = types[TVOID]; + cgen(n, Z); + break; + + case OCOND: + bcgen(n->left, 1); + p1 = p; + sugen(n->right->left, nn, w); + gbranch(OGOTO); + patch(p1, pc); + p1 = p; + sugen(n->right->right, nn, w); + patch(p1, pc); + break; + + case OCOMMA: + cgen(n->left, Z); + sugen(n->right, nn, w); + break; + } + return; + +copy: + if(nn == Z) { + switch(n->op) { + case OASADD: + case OASSUB: + case OASAND: + case OASOR: + case OASXOR: + + case OASMUL: + case OASLMUL: + + + case OASASHL: + case OASASHR: + case OASLSHR: + break; + + case OPOSTINC: + case OPOSTDEC: + case OPREINC: + case OPREDEC: + break; + + default: + return; + } + } + + if(n->complex >= FNX && nn != nil && nn->complex >= FNX) { + t = nn->type; + nn->type = types[TLONG]; + regialloc(&nod1, nn, Z); + lcgen(nn, &nod1); + regsalloc(&nod2, nn); + nn->type = t; + + gins(AMOVL, &nod1, &nod2); + regfree(&nod1); + + nod2.type = typ(TIND, t); + + nod1 = nod2; + nod1.op = OIND; + nod1.left = &nod2; + nod1.right = Z; + nod1.complex = 1; + nod1.type = t; + + sugen(n, &nod1, w); + return; + } + + if(w <= 32) { + c = cursafe; + if(n->left != Z && n->left->complex >= FNX + && n->right != Z && n->right->complex >= FNX) { + regsalloc(&nod1, n->right); + cgen(n->right, &nod1); + nod2 = *n; + nod2.right = &nod1; + cgen(&nod2, nn); + cursafe = c; + return; + } + if(w & 7) { + mt = TLONG; + mo = AMOVL; + } else { + mt = TVLONG; + mo = AMOVQ; + } + if(n->complex > nn->complex) { + t = n->type; + n->type = types[mt]; + regalloc(&nod0, n, Z); + if(!vaddr(n, 0)) { + reglcgen(&nod1, n, Z); + n->type = t; + n = &nod1; + } + else + n->type = t; + + t = nn->type; + nn->type = types[mt]; + if(!vaddr(nn, 0)) { + reglcgen(&nod2, nn, Z); + nn->type = t; + nn = &nod2; + } + else + nn->type = t; + } else { + t = nn->type; + nn->type = types[mt]; + regalloc(&nod0, nn, Z); + if(!vaddr(nn, 0)) { + reglcgen(&nod2, nn, Z); + nn->type = t; + nn = &nod2; + } + else + nn->type = t; + + t = n->type; + n->type = types[mt]; + if(!vaddr(n, 0)) { + reglcgen(&nod1, n, Z); + n->type = t; + n = &nod1; + } + else + n->type = t; + } + o0 = n->xoffset; + o1 = nn->xoffset; + w /= ewidth[mt]; + while(--w >= 0) { + gins(mo, n, &nod0); + gins(mo, &nod0, nn); + n->xoffset += ewidth[mt]; + nn->xoffset += ewidth[mt]; + } + n->xoffset = o0; + nn->xoffset = o1; + if(nn == &nod2) + regfree(&nod2); + if(n == &nod1) + regfree(&nod1); + regfree(&nod0); + return; + } + + /* botch, need to save in .safe */ + c = 0; + if(n->complex > nn->complex) { + t = n->type; + n->type = types[TLONG]; + nodreg(&nod1, n, D_SI); + if(reg[D_SI]) { + gins(APUSHQ, &nod1, Z); + c |= 1; + reg[D_SI]++; + } + lcgen(n, &nod1); + n->type = t; + + t = nn->type; + nn->type = types[TLONG]; + nodreg(&nod2, nn, D_DI); + if(reg[D_DI]) { +warn(Z, "DI botch"); + gins(APUSHQ, &nod2, Z); + c |= 2; + reg[D_DI]++; + } + lcgen(nn, &nod2); + nn->type = t; + } else { + t = nn->type; + nn->type = types[TLONG]; + nodreg(&nod2, nn, D_DI); + if(reg[D_DI]) { +warn(Z, "DI botch"); + gins(APUSHQ, &nod2, Z); + c |= 2; + reg[D_DI]++; + } + lcgen(nn, &nod2); + nn->type = t; + + t = n->type; + n->type = types[TLONG]; + nodreg(&nod1, n, D_SI); + if(reg[D_SI]) { + gins(APUSHQ, &nod1, Z); + c |= 1; + reg[D_SI]++; + } + lcgen(n, &nod1); + n->type = t; + } + nodreg(&nod3, n, D_CX); + if(reg[D_CX]) { + gins(APUSHQ, &nod3, Z); + c |= 4; + reg[D_CX]++; + } + gins(AMOVL, nodconst(w/SZ_INT), &nod3); + gins(ACLD, Z, Z); + gins(AREP, Z, Z); + gins(AMOVSL, Z, Z); + if(c & 4) { + gins(APOPQ, Z, &nod3); + reg[D_CX]--; + } + if(c & 2) { + gins(APOPQ, Z, &nod2); + reg[nod2.reg]--; + } + if(c & 1) { + gins(APOPQ, Z, &nod1); + reg[nod1.reg]--; + } +} + +/* + * TO DO + */ +void +layout(Node *f, Node *t, int c, int cv, Node *cn) +{ + Node t1, t2; + + while(c > 3) { + layout(f, t, 2, 0, Z); + c -= 2; + } + + regalloc(&t1, &lregnode, Z); + regalloc(&t2, &lregnode, Z); + if(c > 0) { + gmove(f, &t1); + f->xoffset += SZ_INT; + } + if(cn != Z) + gmove(nodconst(cv), cn); + if(c > 1) { + gmove(f, &t2); + f->xoffset += SZ_INT; + } + if(c > 0) { + gmove(&t1, t); + t->xoffset += SZ_INT; + } + if(c > 2) { + gmove(f, &t1); + f->xoffset += SZ_INT; + } + if(c > 1) { + gmove(&t2, t); + t->xoffset += SZ_INT; + } + if(c > 2) { + gmove(&t1, t); + t->xoffset += SZ_INT; + } + regfree(&t1); + regfree(&t2); +} + +/* + * constant is not vlong or fits as 32-bit signed immediate + */ +int +immconst(Node *n) +{ + long v; + + if(n->op != OCONST || !typechlpv[n->type->etype]) + return 0; + if(typechl[n->type->etype]) + return 1; + v = n->vconst; + return n->vconst == (vlong)v; +} + +/* + * if a constant and vlong, doesn't fit as 32-bit signed immediate + */ +int +hardconst(Node *n) +{ + return n->op == OCONST && !immconst(n); +} + +/* + * casting up to t2 covers an intermediate cast to t1 + */ +int +castup(Type *t1, Type *t2) +{ + int ft; + + if(!nilcast(t1, t2)) + return 0; + /* known to be small to large */ + ft = t1->etype; + switch(t2->etype){ + case TINT: + case TLONG: + return ft == TLONG || ft == TINT || ft == TSHORT || ft == TCHAR; + case TUINT: + case TULONG: + return ft == TULONG || ft == TUINT || ft == TUSHORT || ft == TUCHAR; + case TVLONG: + return ft == TLONG || ft == TINT || ft == TSHORT; + case TUVLONG: + return ft == TULONG || ft == TUINT || ft == TUSHORT; + } + return 0; +} + +void +zeroregm(Node *n) +{ + gins(AMOVL, nodconst(0), n); +} + +/* do we need to load the address of a vlong? */ +int +vaddr(Node *n, int a) +{ + switch(n->op) { + case ONAME: + if(a) + return 1; + return !(n->class == CEXTERN || n->class == CGLOBL || n->class == CSTATIC); + + case OCONST: + case OREGISTER: + case OINDREG: + return 1; + } + return 0; +} + +long +hi64v(Node *n) +{ + if(align(0, types[TCHAR], Aarg1)) /* isbigendian */ + return (long)(n->vconst) & ~0L; + else + return (long)((uvlong)n->vconst>>32) & ~0L; +} + +long +lo64v(Node *n) +{ + if(align(0, types[TCHAR], Aarg1)) /* isbigendian */ + return (long)((uvlong)n->vconst>>32) & ~0L; + else + return (long)(n->vconst) & ~0L; +} + +Node * +hi64(Node *n) +{ + return nodconst(hi64v(n)); +} + +Node * +lo64(Node *n) +{ + return nodconst(lo64v(n)); +} + +int +cond(int op) +{ + switch(op) { + case OANDAND: + case OOROR: + case ONOT: + return 1; + + case OEQ: + case ONE: + case OLE: + case OLT: + case OGE: + case OGT: + case OHI: + case OHS: + case OLO: + case OLS: + return 1; + } + return 0; +} diff --git a/sys/src/cmd/6c/div.c b/sys/src/cmd/6c/div.c new file mode 100644 index 000000000..cb5886789 --- /dev/null +++ b/sys/src/cmd/6c/div.c @@ -0,0 +1,206 @@ +#include "gc.h" + +/* + * Based on: Granlund, T.; Montgomery, P.L. + * "Division by Invariant Integers using Multiplication". + * SIGPLAN Notices, Vol. 29, June 1994, page 61. + */ + +#define TN(n) ((uvlong)1 << (n)) +#define T31 TN(31) +#define T32 TN(32) + +int +multiplier(ulong d, int p, uvlong *mp) +{ + int l; + uvlong mlo, mhi, tlo, thi; + + l = topbit(d - 1) + 1; + mlo = (((TN(l) - d) << 32) / d) + T32; + if(l + p == 64) + mhi = (((TN(l) + 1 - d) << 32) / d) + T32; + else + mhi = (TN(32 + l) + TN(32 + l - p)) / d; + /*assert(mlo < mhi);*/ + while(l > 0) { + tlo = mlo >> 1; + thi = mhi >> 1; + if(tlo == thi) + break; + mlo = tlo; + mhi = thi; + l--; + } + *mp = mhi; + return l; +} + +int +sdiv(ulong d, ulong *mp, int *sp) +{ + int s; + uvlong m; + + s = multiplier(d, 32 - 1, &m); + *mp = m; + *sp = s; + if(m >= T31) + return 1; + else + return 0; +} + +int +udiv(ulong d, ulong *mp, int *sp, int *pp) +{ + int p, s; + uvlong m; + + s = multiplier(d, 32, &m); + p = 0; + if(m >= T32) { + while((d & 1) == 0) { + d >>= 1; + p++; + } + s = multiplier(d, 32 - p, &m); + } + *mp = m; + *pp = p; + if(m >= T32) { + /*assert(p == 0);*/ + *sp = s - 1; + return 1; + } + else { + *sp = s; + return 0; + } +} + +void +sdivgen(Node *l, Node *r, Node *ax, Node *dx) +{ + int a, s; + ulong m; + vlong c; + + c = r->vconst; + if(c < 0) + c = -c; + a = sdiv(c, &m, &s); +//print("a=%d i=%ld s=%d m=%lux\n", a, (long)r->vconst, s, m); + gins(AMOVL, nodconst(m), ax); + gins(AIMULL, l, Z); + gins(AMOVL, l, ax); + if(a) + gins(AADDL, ax, dx); + gins(ASHRL, nodconst(31), ax); + gins(ASARL, nodconst(s), dx); + gins(AADDL, ax, dx); + if(r->vconst < 0) + gins(ANEGL, Z, dx); +} + +void +udivgen(Node *l, Node *r, Node *ax, Node *dx) +{ + int a, s, t; + ulong m; + Node nod; + + a = udiv(r->vconst, &m, &s, &t); +//print("a=%ud i=%ld p=%d s=%d m=%lux\n", a, (long)r->vconst, t, s, m); + if(t != 0) { + gins(AMOVL, l, ax); + gins(ASHRL, nodconst(t), ax); + gins(AMOVL, nodconst(m), dx); + gins(AMULL, dx, Z); + } + else if(a) { + if(l->op != OREGISTER) { + regalloc(&nod, l, Z); + gins(AMOVL, l, &nod); + l = &nod; + } + gins(AMOVL, nodconst(m), ax); + gins(AMULL, l, Z); + gins(AADDL, l, dx); + gins(ARCRL, nodconst(1), dx); + if(l == &nod) + regfree(l); + } + else { + gins(AMOVL, nodconst(m), ax); + gins(AMULL, l, Z); + } + if(s != 0) + gins(ASHRL, nodconst(s), dx); +} + +void +sext(Node *d, Node *s, Node *l) +{ + if(s->reg == D_AX && !nodreg(d, Z, D_DX)) { + reg[D_DX]++; + gins(ACDQ, Z, Z); + } + else { + regalloc(d, l, Z); + gins(AMOVL, s, d); + gins(ASARL, nodconst(31), d); + } +} + +void +sdiv2(long c, int v, Node *l, Node *n) +{ + Node nod; + + if(v > 0) { + if(v > 1) { + sext(&nod, n, l); + gins(AANDL, nodconst((1 << v) - 1), &nod); + gins(AADDL, &nod, n); + regfree(&nod); + } + else { + gins(ACMPL, n, nodconst(0x80000000)); + gins(ASBBL, nodconst(-1), n); + } + gins(ASARL, nodconst(v), n); + } + if(c < 0) + gins(ANEGL, Z, n); +} + +void +smod2(long c, int v, Node *l, Node *n) +{ + Node nod; + + if(c == 1) { + zeroregm(n); + return; + } + + sext(&nod, n, l); + if(v == 0) { + zeroregm(n); + gins(AXORL, &nod, n); + gins(ASUBL, &nod, n); + } + else if(v > 1) { + gins(AANDL, nodconst((1 << v) - 1), &nod); + gins(AADDL, &nod, n); + gins(AANDL, nodconst((1 << v) - 1), n); + gins(ASUBL, &nod, n); + } + else { + gins(AANDL, nodconst(1), n); + gins(AXORL, &nod, n); + gins(ASUBL, &nod, n); + } + regfree(&nod); +} diff --git a/sys/src/cmd/6c/enam.c b/sys/src/cmd/6c/enam.c new file mode 100644 index 000000000..ab8068f54 --- /dev/null +++ b/sys/src/cmd/6c/enam.c @@ -0,0 +1,669 @@ +char* anames[] = +{ + "XXX", + "AAA", + "AAD", + "AAM", + "AAS", + "ADCB", + "ADCL", + "ADCW", + "ADDB", + "ADDL", + "ADDW", + "ADJSP", + "ANDB", + "ANDL", + "ANDW", + "ARPL", + "BOUNDL", + "BOUNDW", + "BSFL", + "BSFW", + "BSRL", + "BSRW", + "BTL", + "BTW", + "BTCL", + "BTCW", + "BTRL", + "BTRW", + "BTSL", + "BTSW", + "BYTE", + "CALL", + "CLC", + "CLD", + "CLI", + "CLTS", + "CMC", + "CMPB", + "CMPL", + "CMPW", + "CMPSB", + "CMPSL", + "CMPSW", + "DAA", + "DAS", + "DATA", + "DECB", + "DECL", + "DECQ", + "DECW", + "DIVB", + "DIVL", + "DIVW", + "ENTER", + "GLOBL", + "GOK", + "HISTORY", + "HLT", + "IDIVB", + "IDIVL", + "IDIVW", + "IMULB", + "IMULL", + "IMULW", + "INB", + "INL", + "INW", + "INCB", + "INCL", + "INCQ", + "INCW", + "INSB", + "INSL", + "INSW", + "INT", + "INTO", + "IRETL", + "IRETW", + "JCC", + "JCS", + "JCXZ", + "JEQ", + "JGE", + "JGT", + "JHI", + "JLE", + "JLS", + "JLT", + "JMI", + "JMP", + "JNE", + "JOC", + "JOS", + "JPC", + "JPL", + "JPS", + "LAHF", + "LARL", + "LARW", + "LEAL", + "LEAW", + "LEAVEL", + "LEAVEW", + "LOCK", + "LODSB", + "LODSL", + "LODSW", + "LONG", + "LOOP", + "LOOPEQ", + "LOOPNE", + "LSLL", + "LSLW", + "MOVB", + "MOVL", + "MOVW", + "MOVBLSX", + "MOVBLZX", + "MOVBQSX", + "MOVBQZX", + "MOVBWSX", + "MOVBWZX", + "MOVWLSX", + "MOVWLZX", + "MOVWQSX", + "MOVWQZX", + "MOVSB", + "MOVSL", + "MOVSW", + "MULB", + "MULL", + "MULW", + "NAME", + "NEGB", + "NEGL", + "NEGW", + "NOP", + "NOTB", + "NOTL", + "NOTW", + "ORB", + "ORL", + "ORW", + "OUTB", + "OUTL", + "OUTW", + "OUTSB", + "OUTSL", + "OUTSW", + "POPAL", + "POPAW", + "POPFL", + "POPFW", + "POPL", + "POPW", + "PUSHAL", + "PUSHAW", + "PUSHFL", + "PUSHFW", + "PUSHL", + "PUSHW", + "RCLB", + "RCLL", + "RCLW", + "RCRB", + "RCRL", + "RCRW", + "REP", + "REPN", + "RET", + "ROLB", + "ROLL", + "ROLW", + "RORB", + "RORL", + "RORW", + "SAHF", + "SALB", + "SALL", + "SALW", + "SARB", + "SARL", + "SARW", + "SBBB", + "SBBL", + "SBBW", + "SCASB", + "SCASL", + "SCASW", + "SETCC", + "SETCS", + "SETEQ", + "SETGE", + "SETGT", + "SETHI", + "SETLE", + "SETLS", + "SETLT", + "SETMI", + "SETNE", + "SETOC", + "SETOS", + "SETPC", + "SETPL", + "SETPS", + "CDQ", + "CWD", + "SHLB", + "SHLL", + "SHLW", + "SHRB", + "SHRL", + "SHRW", + "STC", + "STD", + "STI", + "STOSB", + "STOSL", + "STOSW", + "SUBB", + "SUBL", + "SUBW", + "SYSCALL", + "TESTB", + "TESTL", + "TESTW", + "TEXT", + "VERR", + "VERW", + "WAIT", + "WORD", + "XCHGB", + "XCHGL", + "XCHGW", + "XLAT", + "XORB", + "XORL", + "XORW", + "FMOVB", + "FMOVBP", + "FMOVD", + "FMOVDP", + "FMOVF", + "FMOVFP", + "FMOVL", + "FMOVLP", + "FMOVV", + "FMOVVP", + "FMOVW", + "FMOVWP", + "FMOVX", + "FMOVXP", + "FCOMB", + "FCOMBP", + "FCOMD", + "FCOMDP", + "FCOMDPP", + "FCOMF", + "FCOMFP", + "FCOML", + "FCOMLP", + "FCOMW", + "FCOMWP", + "FUCOM", + "FUCOMP", + "FUCOMPP", + "FADDDP", + "FADDW", + "FADDL", + "FADDF", + "FADDD", + "FMULDP", + "FMULW", + "FMULL", + "FMULF", + "FMULD", + "FSUBDP", + "FSUBW", + "FSUBL", + "FSUBF", + "FSUBD", + "FSUBRDP", + "FSUBRW", + "FSUBRL", + "FSUBRF", + "FSUBRD", + "FDIVDP", + "FDIVW", + "FDIVL", + "FDIVF", + "FDIVD", + "FDIVRDP", + "FDIVRW", + "FDIVRL", + "FDIVRF", + "FDIVRD", + "FXCHD", + "FFREE", + "FLDCW", + "FLDENV", + "FRSTOR", + "FSAVE", + "FSTCW", + "FSTENV", + "FSTSW", + "F2XM1", + "FABS", + "FCHS", + "FCLEX", + "FCOS", + "FDECSTP", + "FINCSTP", + "FINIT", + "FLD1", + "FLDL2E", + "FLDL2T", + "FLDLG2", + "FLDLN2", + "FLDPI", + "FLDZ", + "FNOP", + "FPATAN", + "FPREM", + "FPREM1", + "FPTAN", + "FRNDINT", + "FSCALE", + "FSIN", + "FSINCOS", + "FSQRT", + "FTST", + "FXAM", + "FXTRACT", + "FYL2X", + "FYL2XP1", + "END", + "DYNT", + "INIT", + "SIGNAME", + "CMPXCHGB", + "CMPXCHGL", + "CMPXCHGW", + "CMPXCHG8B", + "CPUID", + "INVD", + "INVLPG", + "LFENCE", + "MFENCE", + "MOVNTIL", + "RDMSR", + "RDPMC", + "RDTSC", + "RSM", + "SFENCE", + "SYSRET", + "WBINVD", + "WRMSR", + "XADDB", + "XADDL", + "XADDW", + "CMOVLCC", + "CMOVLCS", + "CMOVLEQ", + "CMOVLGE", + "CMOVLGT", + "CMOVLHI", + "CMOVLLE", + "CMOVLLS", + "CMOVLLT", + "CMOVLMI", + "CMOVLNE", + "CMOVLOC", + "CMOVLOS", + "CMOVLPC", + "CMOVLPL", + "CMOVLPS", + "CMOVQCC", + "CMOVQCS", + "CMOVQEQ", + "CMOVQGE", + "CMOVQGT", + "CMOVQHI", + "CMOVQLE", + "CMOVQLS", + "CMOVQLT", + "CMOVQMI", + "CMOVQNE", + "CMOVQOC", + "CMOVQOS", + "CMOVQPC", + "CMOVQPL", + "CMOVQPS", + "CMOVWCC", + "CMOVWCS", + "CMOVWEQ", + "CMOVWGE", + "CMOVWGT", + "CMOVWHI", + "CMOVWLE", + "CMOVWLS", + "CMOVWLT", + "CMOVWMI", + "CMOVWNE", + "CMOVWOC", + "CMOVWOS", + "CMOVWPC", + "CMOVWPL", + "CMOVWPS", + "ADCQ", + "ADDQ", + "ANDQ", + "BSFQ", + "BSRQ", + "BTCQ", + "BTQ", + "BTRQ", + "BTSQ", + "CMPQ", + "CMPSQ", + "CMPXCHGQ", + "CQO", + "DIVQ", + "IDIVQ", + "IMULQ", + "IRETQ", + "LEAQ", + "LEAVEQ", + "LODSQ", + "MOVQ", + "MOVLQSX", + "MOVLQZX", + "MOVNTIQ", + "MOVSQ", + "MULQ", + "NEGQ", + "NOTQ", + "ORQ", + "POPFQ", + "POPQ", + "PUSHFQ", + "PUSHQ", + "RCLQ", + "RCRQ", + "ROLQ", + "RORQ", + "QUAD", + "SALQ", + "SARQ", + "SBBQ", + "SCASQ", + "SHLQ", + "SHRQ", + "STOSQ", + "SUBQ", + "TESTQ", + "XADDQ", + "XCHGQ", + "XORQ", + "ADDPD", + "ADDPS", + "ADDSD", + "ADDSS", + "ANDNPD", + "ANDNPS", + "ANDPD", + "ANDPS", + "CMPPD", + "CMPPS", + "CMPSD", + "CMPSS", + "COMISD", + "COMISS", + "CVTPD2PL", + "CVTPD2PS", + "CVTPL2PD", + "CVTPL2PS", + "CVTPS2PD", + "CVTPS2PL", + "CVTSD2SL", + "CVTSD2SQ", + "CVTSD2SS", + "CVTSL2SD", + "CVTSL2SS", + "CVTSQ2SD", + "CVTSQ2SS", + "CVTSS2SD", + "CVTSS2SL", + "CVTSS2SQ", + "CVTTPD2PL", + "CVTTPS2PL", + "CVTTSD2SL", + "CVTTSD2SQ", + "CVTTSS2SL", + "CVTTSS2SQ", + "DIVPD", + "DIVPS", + "DIVSD", + "DIVSS", + "EMMS", + "FXRSTOR", + "FXRSTOR64", + "FXSAVE", + "FXSAVE64", + "LDMXCSR", + "MASKMOVOU", + "MASKMOVQ", + "MAXPD", + "MAXPS", + "MAXSD", + "MAXSS", + "MINPD", + "MINPS", + "MINSD", + "MINSS", + "MOVAPD", + "MOVAPS", + "MOVOU", + "MOVHLPS", + "MOVHPD", + "MOVHPS", + "MOVLHPS", + "MOVLPD", + "MOVLPS", + "MOVMSKPD", + "MOVMSKPS", + "MOVNTO", + "MOVNTPD", + "MOVNTPS", + "MOVNTQ", + "MOVO", + "MOVQOZX", + "MOVSD", + "MOVSS", + "MOVUPD", + "MOVUPS", + "MULPD", + "MULPS", + "MULSD", + "MULSS", + "ORPD", + "ORPS", + "PACKSSLW", + "PACKSSWB", + "PACKUSWB", + "PADDB", + "PADDL", + "PADDQ", + "PADDSB", + "PADDSW", + "PADDUSB", + "PADDUSW", + "PADDW", + "PANDB", + "PANDL", + "PANDSB", + "PANDSW", + "PANDUSB", + "PANDUSW", + "PANDW", + "PAND", + "PANDN", + "PAVGB", + "PAVGW", + "PCMPEQB", + "PCMPEQL", + "PCMPEQW", + "PCMPGTB", + "PCMPGTL", + "PCMPGTW", + "PEXTRW", + "PFACC", + "PFADD", + "PFCMPEQ", + "PFCMPGE", + "PFCMPGT", + "PFMAX", + "PFMIN", + "PFMUL", + "PFNACC", + "PFPNACC", + "PFRCP", + "PFRCPIT1", + "PFRCPI2T", + "PFRSQIT1", + "PFRSQRT", + "PFSUB", + "PFSUBR", + "PINSRW", + "PMADDWL", + "PMAXSW", + "PMAXUB", + "PMINSW", + "PMINUB", + "PMOVMSKB", + "PMULHRW", + "PMULHUW", + "PMULHW", + "PMULLW", + "PMULULQ", + "POR", + "PSADBW", + "PSHUFHW", + "PSHUFL", + "PSHUFLW", + "PSHUFW", + "PSLLO", + "PSLLL", + "PSLLQ", + "PSLLW", + "PSRAL", + "PSRAW", + "PSRLO", + "PSRLL", + "PSRLQ", + "PSRLW", + "PSUBB", + "PSUBL", + "PSUBQ", + "PSUBSB", + "PSUBSW", + "PSUBUSB", + "PSUBUSW", + "PSUBW", + "PSWAPL", + "PUNPCKHBW", + "PUNPCKHLQ", + "PUNPCKHQDQ", + "PUNPCKHWL", + "PUNPCKLBW", + "PUNPCKLLQ", + "PUNPCKLQDQ", + "PUNPCKLWL", + "PXOR", + "RCPPS", + "RCPSS", + "RSQRTPS", + "RSQRTSS", + "SHUFPD", + "SHUFPS", + "SQRTPD", + "SQRTPS", + "SQRTSD", + "SQRTSS", + "STMXCSR", + "SUBPD", + "SUBPS", + "SUBSD", + "SUBSS", + "UCOMISD", + "UCOMISS", + "UNPCKHPD", + "UNPCKHPS", + "UNPCKLPD", + "UNPCKLPS", + "XORPD", + "XORPS", + "PF2IW", + "PF2IL", + "PI2FW", + "PI2FL", + "RETFW", + "RETFL", + "RETFQ", + "SWAPGS", + "MODE", + "LAST", +}; diff --git a/sys/src/cmd/6c/gc.h b/sys/src/cmd/6c/gc.h new file mode 100644 index 000000000..57d3a9994 --- /dev/null +++ b/sys/src/cmd/6c/gc.h @@ -0,0 +1,381 @@ +#include "../cc/cc.h" +#include "../6c/6.out.h" + +/* + * 6c/amd64 + * Intel 386 with AMD64 extensions + */ +#define SZ_CHAR 1 +#define SZ_SHORT 2 +#define SZ_INT 4 +#define SZ_LONG 4 +#define SZ_IND 8 +#define SZ_FLOAT 4 +#define SZ_VLONG 8 +#define SZ_DOUBLE 8 +#define FNX 100 + +typedef struct Adr Adr; +typedef struct Prog Prog; +typedef struct Case Case; +typedef struct C1 C1; +typedef struct Var Var; +typedef struct Reg Reg; +typedef struct Rgn Rgn; +typedef struct Renv Renv; + +EXTERN struct +{ + Node* regtree; + Node* basetree; + short scale; + short reg; + short ptr; +} idx; + +struct Adr +{ + vlong offset; + double dval; + char sval[NSNAME]; + + Sym* sym; + uchar type; + uchar index; + uchar etype; + uchar scale; /* doubles as width in DATA op */ +}; +#define A ((Adr*)0) + +#define INDEXED 9 +struct Prog +{ + Adr from; + Adr to; + Prog* link; + long lineno; + short as; +}; +#define P ((Prog*)0) + +struct Case +{ + Case* link; + vlong val; + long label; + char def; + char isv; +}; +#define C ((Case*)0) + +struct C1 +{ + vlong val; + long label; +}; + +struct Var +{ + vlong offset; + Sym* sym; + char name; + char etype; +}; + +struct Reg +{ + long pc; + long rpo; /* reverse post ordering */ + + Bits set; + Bits use1; + Bits use2; + + Bits refbehind; + Bits refahead; + Bits calbehind; + Bits calahead; + Bits regdiff; + Bits act; + + long regu; + long loop; /* could be shorter */ + + Reg* log5; + long active; + + Reg* p1; + Reg* p2; + Reg* p2link; + Reg* s1; + Reg* s2; + Reg* link; + Prog* prog; +}; +#define R ((Reg*)0) + +struct Renv +{ + int safe; + Node base; + Node* saved; + Node* scope; +}; + +#define NRGN 600 +struct Rgn +{ + Reg* enter; + short cost; + short varno; + short regno; +}; + +EXTERN long breakpc; +EXTERN long nbreak; +EXTERN Case* cases; +EXTERN Node constnode; +EXTERN Node fconstnode; +EXTERN Node vconstnode; +EXTERN long continpc; +EXTERN long curarg; +EXTERN long cursafe; +EXTERN Prog* firstp; +EXTERN Prog* lastp; +EXTERN long maxargsafe; +EXTERN int mnstring; +EXTERN Node* nodrat; +EXTERN Node* nodret; +EXTERN Node* nodsafe; +EXTERN long nrathole; +EXTERN long nstring; +EXTERN Prog* p; +EXTERN long pc; +EXTERN Node lregnode; +EXTERN Node qregnode; +EXTERN char string[NSNAME]; +EXTERN Sym* symrathole; +EXTERN Node znode; +EXTERN Prog zprog; +EXTERN int reg[D_NONE]; +EXTERN long exregoffset; +EXTERN long exfregoffset; +EXTERN uchar typechlpv[NTYPE]; + +#define BLOAD(r) band(bnot(r->refbehind), r->refahead) +#define BSTORE(r) band(bnot(r->calbehind), r->calahead) +#define LOAD(r) (~r->refbehind.b[z] & r->refahead.b[z]) +#define STORE(r) (~r->calbehind.b[z] & r->calahead.b[z]) + +#define bset(a,n) ((a).b[(n)/32]&(1L<<(n)%32)) + +#define CLOAD 5 +#define CREF 5 +#define CINF 1000 +#define LOOP 3 + +EXTERN Rgn region[NRGN]; +EXTERN Rgn* rgp; +EXTERN int nregion; +EXTERN int nvar; + +EXTERN Bits externs; +EXTERN Bits params; +EXTERN Bits consts; +EXTERN Bits addrs; + +EXTERN long regbits; +EXTERN long exregbits; + +EXTERN int change; +EXTERN int suppress; + +EXTERN Reg* firstr; +EXTERN Reg* lastr; +EXTERN Reg zreg; +EXTERN Reg* freer; +EXTERN Var var[NVAR]; +EXTERN long* idom; +EXTERN Reg** rpo2r; +EXTERN long maxnr; + +extern char* anames[]; + +/* + * sgen.c + */ +void codgen(Node*, Node*); +void gen(Node*); +void noretval(int); +void usedset(Node*, int); +void xcom(Node*); +void indx(Node*); +int bcomplex(Node*, Node*); + +/* + * cgen.c + */ +void zeroregm(Node*); +void cgen(Node*, Node*); +void reglcgen(Node*, Node*, Node*); +void lcgen(Node*, Node*); +void bcgen(Node*, int); +void boolgen(Node*, int, Node*); +void sugen(Node*, Node*, long); +int needreg(Node*, int); +int hardconst(Node*); +int immconst(Node*); + +/* + * cgen64.c + */ +int vaddr(Node*, int); +void loadpair(Node*, Node*); +int cgen64(Node*, Node*); +void testv(Node*, int); + +/* + * txt.c + */ +void ginit(void); +void gclean(void); +void nextpc(void); +void gargs(Node*, Node*, Node*); +void garg1(Node*, Node*, Node*, int, Node**); +Node* nodconst(long); +Node* nodfconst(double); +Node* nodgconst(vlong, Type*); +int nodreg(Node*, Node*, int); +int isreg(Node*, int); +void regret(Node*, Node*); +void regalloc(Node*, Node*, Node*); +void regfree(Node*); +void regialloc(Node*, Node*, Node*); +void regsalloc(Node*, Node*); +void regaalloc1(Node*, Node*); +void regaalloc(Node*, Node*); +void regind(Node*, Node*); +void gprep(Node*, Node*); +void naddr(Node*, Adr*); +void gcmp(int, Node*, vlong); +void gmove(Node*, Node*); +void gins(int a, Node*, Node*); +void gopcode(int, Type*, Node*, Node*); +int samaddr(Node*, Node*); +void gbranch(int); +void patch(Prog*, long); +int sconst(Node*); +void gpseudo(int, Sym*, Node*); + +/* + * swt.c + */ +int swcmp(const void*, const void*); +void doswit(Node*); +void swit1(C1*, int, long, Node*); +void casf(void); +void bitload(Node*, Node*, Node*, Node*, Node*); +void bitstore(Node*, Node*, Node*, Node*, Node*); +long outstring(char*, long); +void nullwarn(Node*, Node*); +void sextern(Sym*, Node*, long, long); +void gextern(Sym*, Node*, long, long); +void outcode(void); +void ieeedtod(Ieee*, double); + +/* + * list + */ +void listinit(void); +int Pconv(Fmt*); +int Aconv(Fmt*); +int Dconv(Fmt*); +int Sconv(Fmt*); +int Rconv(Fmt*); +int Xconv(Fmt*); +int Bconv(Fmt*); + +/* + * reg.c + */ +Reg* rega(void); +int rcmp(const void*, const void*); +void regopt(Prog*); +void addmove(Reg*, int, int, int); +Bits mkvar(Reg*, Adr*); +void prop(Reg*, Bits, Bits); +void loopit(Reg*, long); +void synch(Reg*, Bits); +ulong allreg(ulong, Rgn*); +void paint1(Reg*, int); +ulong paint2(Reg*, int); +void paint3(Reg*, int, long, int); +void addreg(Adr*, int); + +/* + * peep.c + */ +void peep(void); +void excise(Reg*); +Reg* uniqp(Reg*); +Reg* uniqs(Reg*); +int regtyp(Adr*); +int anyvar(Adr*); +int subprop(Reg*); +int copyprop(Reg*); +int copy1(Adr*, Adr*, Reg*, int); +int copyu(Prog*, Adr*, Adr*); + +int copyas(Adr*, Adr*); +int copyau(Adr*, Adr*); +int copysub(Adr*, Adr*, Adr*, int); +int copysub1(Prog*, Adr*, Adr*, int); + +long RtoB(int); +long FtoB(int); +int BtoR(long); +int BtoF(long); + +#define D_HI D_NONE +#define D_LO D_NONE + +#define isregtype(t) ((t)>= D_AX && (t)<=D_R15) + +/* + * bound + */ +void comtarg(void); + +/* + * com64 + */ +int cond(int); +int com64(Node*); +void com64init(void); +void bool64(Node*); +long lo64v(Node*); +long hi64v(Node*); +Node* lo64(Node*); +Node* hi64(Node*); + +/* + * div/mul + */ +void sdivgen(Node*, Node*, Node*, Node*); +void udivgen(Node*, Node*, Node*, Node*); +void sdiv2(long, int, Node*, Node*); +void smod2(long, int, Node*, Node*); +void mulgen(Type*, Node*, Node*); +void genmuladd(Node*, Node*, int, Node*); +void shiftit(Type*, Node*, Node*); + +#pragma varargck type "A" int +#pragma varargck type "B" Bits +#pragma varargck type "D" Adr* +#pragma varargck type "P" Prog* +#pragma varargck type "R" int +#pragma varargck type "S" char* + +#define D_X7 (D_X0+7) + +void fgopcode(int, Node*, Node*, int, int); diff --git a/sys/src/cmd/6c/list.c b/sys/src/cmd/6c/list.c new file mode 100644 index 000000000..73efdf139 --- /dev/null +++ b/sys/src/cmd/6c/list.c @@ -0,0 +1,337 @@ +#define EXTERN +#include "gc.h" + +void +listinit(void) +{ + + fmtinstall('A', Aconv); + fmtinstall('B', Bconv); + fmtinstall('P', Pconv); + fmtinstall('S', Sconv); + fmtinstall('D', Dconv); + fmtinstall('R', Rconv); +} + +int +Bconv(Fmt *fp) +{ + char str[STRINGSZ], ss[STRINGSZ], *s; + Bits bits; + int i; + + str[0] = 0; + bits = va_arg(fp->args, Bits); + while(bany(&bits)) { + i = bnum(bits); + if(str[0]) + strcat(str, " "); + if(var[i].sym == S) { + sprint(ss, "$%lld", var[i].offset); + s = ss; + } else + s = var[i].sym->name; + if(strlen(str) + strlen(s) + 1 >= STRINGSZ) + break; + strcat(str, s); + bits.b[i/32] &= ~(1L << (i%32)); + } + return fmtstrcpy(fp, str); +} + +int +Pconv(Fmt *fp) +{ + char str[STRINGSZ]; + Prog *p; + + p = va_arg(fp->args, Prog*); + if(p->as == ADATA) + sprint(str, " %A %D/%d,%D", + p->as, &p->from, p->from.scale, &p->to); + else if(p->as == ATEXT) + sprint(str, " %A %D,%d,%D", + p->as, &p->from, p->from.scale, &p->to); + else + sprint(str, " %A %D,%D", + p->as, &p->from, &p->to); + return fmtstrcpy(fp, str); +} + +int +Aconv(Fmt *fp) +{ + int i; + + i = va_arg(fp->args, int); + return fmtstrcpy(fp, anames[i]); +} + +int +Dconv(Fmt *fp) +{ + char str[40], s[20]; + Adr *a; + int i; + + a = va_arg(fp->args, Adr*); + i = a->type; + if(i >= D_INDIR) { + if(a->offset) + sprint(str, "%lld(%R)", a->offset, i-D_INDIR); + else + sprint(str, "(%R)", i-D_INDIR); + goto brk; + } + switch(i) { + + default: + if(a->offset) + sprint(str, "$%lld,%R", a->offset, i); + else + sprint(str, "%R", i); + break; + + case D_NONE: + str[0] = 0; + break; + + case D_BRANCH: + sprint(str, "%lld(PC)", a->offset-pc); + break; + + case D_EXTERN: + sprint(str, "%s+%lld(SB)", a->sym->name, a->offset); + break; + + case D_STATIC: + sprint(str, "%s<>+%lld(SB)", a->sym->name, + a->offset); + break; + + case D_AUTO: + sprint(str, "%s+%lld(SP)", a->sym->name, a->offset); + break; + + case D_PARAM: + if(a->sym) + sprint(str, "%s+%lld(FP)", a->sym->name, a->offset); + else + sprint(str, "%lld(FP)", a->offset); + break; + + case D_CONST: + sprint(str, "$%lld", a->offset); + break; + + case D_FCONST: + sprint(str, "$(%.17e)", a->dval); + break; + + case D_SCONST: + sprint(str, "$\"%S\"", a->sval); + break; + + case D_ADDR: + a->type = a->index; + a->index = D_NONE; + sprint(str, "$%D", a); + a->index = a->type; + a->type = D_ADDR; + goto conv; + } +brk: + if(a->index != D_NONE) { + sprint(s, "(%R*%d)", (int)a->index, (int)a->scale); + strcat(str, s); + } +conv: + return fmtstrcpy(fp, str); +} + +char* regstr[] = +{ + "AL", /* [D_AL] */ + "CL", + "DL", + "BL", + "SPB", + "BPB", + "SIB", + "DIB", + "R8B", + "R9B", + "R10B", + "R11B", + "R12B", + "R13B", + "R14B", + "R15B", + + "AX", /* [D_AX] */ + "CX", + "DX", + "BX", + "SP", + "BP", + "SI", + "DI", + "R8", + "R9", + "R10", + "R11", + "R12", + "R13", + "R14", + "R15", + + "AH", + "CH", + "DH", + "BH", + + "F0", /* [D_F0] */ + "F1", + "F2", + "F3", + "F4", + "F5", + "F6", + "F7", + + "M0", + "M1", + "M2", + "M3", + "M4", + "M5", + "M6", + "M7", + + "X0", + "X1", + "X2", + "X3", + "X4", + "X5", + "X6", + "X7", + "X8", + "X9", + "X10", + "X11", + "X12", + "X13", + "X14", + "X15", + + "CS", /* [D_CS] */ + "SS", + "DS", + "ES", + "FS", + "GS", + + "GDTR", /* [D_GDTR] */ + "IDTR", /* [D_IDTR] */ + "LDTR", /* [D_LDTR] */ + "MSW", /* [D_MSW] */ + "TASK", /* [D_TASK] */ + + "CR0", /* [D_CR] */ + "CR1", + "CR2", + "CR3", + "CR4", + "CR5", + "CR6", + "CR7", + "CR8", + "CR9", + "CR10", + "CR11", + "CR12", + "CR13", + "CR14", + "CR15", + + "DR0", /* [D_DR] */ + "DR1", + "DR2", + "DR3", + "DR4", + "DR5", + "DR6", + "DR7", + + "TR0", /* [D_TR] */ + "TR1", + "TR2", + "TR3", + "TR4", + "TR5", + "TR6", + "TR7", + + "NONE", /* [D_NONE] */ +}; + +int +Rconv(Fmt *fp) +{ + char str[20]; + int r; + + r = va_arg(fp->args, int); + if(r >= D_AL && r <= D_NONE) + sprint(str, "%s", regstr[r-D_AL]); + else + sprint(str, "gok(%d)", r); + + return fmtstrcpy(fp, str); +} + +int +Sconv(Fmt *fp) +{ + int i, c; + char str[30], *p, *a; + + a = va_arg(fp->args, char*); + p = str; + for(i=0; i<sizeof(double); i++) { + c = a[i] & 0xff; + if(c >= 'a' && c <= 'z' || + c >= 'A' && c <= 'Z' || + c >= '0' && c <= '9') { + *p++ = c; + continue; + } + *p++ = '\\'; + switch(c) { + default: + if(c < 040 || c >= 0177) + break; /* not portable */ + p[-1] = c; + continue; + case 0: + *p++ = 'z'; + continue; + case '\\': + case '"': + *p++ = c; + continue; + case '\n': + *p++ = 'n'; + continue; + case '\t': + *p++ = 't'; + continue; + } + *p++ = (c>>6) + '0'; + *p++ = ((c>>3) & 7) + '0'; + *p++ = (c & 7) + '0'; + } + *p = 0; + return fmtstrcpy(fp, str); +} diff --git a/sys/src/cmd/6c/machcap.c b/sys/src/cmd/6c/machcap.c new file mode 100644 index 000000000..68c7fa8d2 --- /dev/null +++ b/sys/src/cmd/6c/machcap.c @@ -0,0 +1,78 @@ +#include "gc.h" + +int +machcap(Node *n) +{ + + if(n == Z) + return 1; /* test */ + + switch(n->op) { + case OMUL: + case OLMUL: + case OASMUL: + case OASLMUL: + if(typechl[n->type->etype]) + return 1; + if(typev[n->type->etype]) { + return 1; + } + break; + + case OCOM: + case ONEG: + case OADD: + case OAND: + case OOR: + case OSUB: + case OXOR: + case OASHL: + case OLSHR: + case OASHR: + if(typechlv[n->left->type->etype]) + return 1; + break; + + case OCAST: + return 1; + + case OCOND: + case OCOMMA: + case OLIST: + case OANDAND: + case OOROR: + case ONOT: + return 1; + + case OASADD: + case OASSUB: + case OASAND: + case OASOR: + case OASXOR: + return 1; + + case OASASHL: + case OASASHR: + case OASLSHR: + return 1; + + case OPOSTINC: + case OPOSTDEC: + case OPREINC: + case OPREDEC: + return 1; + + case OEQ: + case ONE: + case OLE: + case OGT: + case OLT: + case OGE: + case OHI: + case OHS: + case OLO: + case OLS: + return 1; + } + return 0; +} diff --git a/sys/src/cmd/6c/mkenam b/sys/src/cmd/6c/mkenam new file mode 100644 index 000000000..32cb1d958 --- /dev/null +++ b/sys/src/cmd/6c/mkenam @@ -0,0 +1,15 @@ +ed - ../6c/6.out.h <<'!' +v/^ A/d +,s/^ A/ "/ +g/ .*$/s/// +,s/,*$/",/ +1i +char* anames[] = +{ +. +$a +}; +. +w enam.c +Q +! diff --git a/sys/src/cmd/6c/mkfile b/sys/src/cmd/6c/mkfile new file mode 100644 index 000000000..18dea3867 --- /dev/null +++ b/sys/src/cmd/6c/mkfile @@ -0,0 +1,38 @@ +</$objtype/mkfile + +TARG=6c +OFILES=\ + cgen.$O\ + enam.$O\ + list.$O\ + sgen.$O\ + swt.$O\ + txt.$O\ + reg.$O\ + peep.$O\ + pgen.$O\ + pswt.$O\ + machcap.$O\ + div.$O\ + mul.$O\ + +HFILES=\ + gc.h\ + 6.out.h\ + ../cc/cc.h\ + +LIB=../cc/cc.a$O + +BIN=/$objtype/bin +</sys/src/cmd/mkone + +$LIB: + cd ../cc + mk install + mk clean + +%.$O: ../cc/%.c + $CC $CFLAGS ../cc/$stem.c + +bound.$O: bound.h + diff --git a/sys/src/cmd/6c/mul.c b/sys/src/cmd/6c/mul.c new file mode 100644 index 000000000..a60908c64 --- /dev/null +++ b/sys/src/cmd/6c/mul.c @@ -0,0 +1,428 @@ +#include "gc.h" + +typedef struct Malg Malg; +typedef struct Mparam Mparam; + +struct Malg +{ + char vals[10]; +}; + +struct Mparam +{ + ulong value; + char alg; + char neg; + char shift; + char arg; + char off; +}; + +static Mparam multab[32]; +static int mulptr; + +static Malg malgs[] = +{ + {0, 100}, + {-1, 1, 100}, + {-9, -5, -3, 3, 5, 9, 100}, + {6, 10, 12, 18, 20, 24, 36, 40, 72, 100}, + {-8, -4, -2, 2, 4, 8, 100}, +}; + +/* + * return position of lowest 1 + */ +int +lowbit(ulong v) +{ + int s, i; + ulong m; + + s = 0; + m = 0xFFFFFFFFUL; + for(i = 16; i > 0; i >>= 1) { + m >>= i; + if((v & m) == 0) { + v >>= i; + s += i; + } + } + return s; +} + +void +genmuladd(Node *d, Node *s, int m, Node *a) +{ + Node nod; + + nod.op = OINDEX; + nod.left = a; + nod.right = s; + nod.scale = m; + nod.type = types[TIND]; + nod.xoffset = 0; + xcom(&nod); + gopcode(OADDR, d->type, &nod, d); +} + +void +mulparam(ulong m, Mparam *mp) +{ + int c, i, j, n, o, q, s; + int bc, bi, bn, bo, bq, bs, bt; + char *p; + long u; + ulong t; + + bc = bq = 10; + bi = bn = bo = bs = bt = 0; + for(i = 0; i < nelem(malgs); i++) { + for(p = malgs[i].vals, j = 0; (o = p[j]) < 100; j++) + for(s = 0; s < 2; s++) { + c = 10; + q = 10; + u = m - o; + if(u == 0) + continue; + if(s) { + o = -o; + if(o > 0) + continue; + u = -u; + } + n = lowbit(u); + t = (ulong)u >> n; + switch(i) { + case 0: + if(t == 1) { + c = s + 1; + q = 0; + break; + } + switch(t) { + case 3: + case 5: + case 9: + c = s + 1; + if(n) + c++; + q = 0; + break; + } + if(s) + break; + switch(t) { + case 15: + case 25: + case 27: + case 45: + case 81: + c = 2; + if(n) + c++; + q = 1; + break; + } + break; + case 1: + if(t == 1) { + c = 3; + q = 3; + break; + } + switch(t) { + case 3: + case 5: + case 9: + c = 3; + q = 2; + break; + } + break; + case 2: + if(t == 1) { + c = 3; + q = 2; + break; + } + break; + case 3: + if(s) + break; + if(t == 1) { + c = 3; + q = 1; + break; + } + break; + case 4: + if(t == 1) { + c = 3; + q = 0; + break; + } + break; + } + if(c < bc || (c == bc && q > bq)) { + bc = c; + bi = i; + bn = n; + bo = o; + bq = q; + bs = s; + bt = t; + } + } + } + mp->value = m; + if(bc <= 3) { + mp->alg = bi; + mp->shift = bn; + mp->off = bo; + mp->neg = bs; + mp->arg = bt; + } + else + mp->alg = -1; +} + +int +m0(int a) +{ + switch(a) { + case -2: + case 2: + return 2; + case -3: + case 3: + return 2; + case -4: + case 4: + return 4; + case -5: + case 5: + return 4; + case 6: + return 2; + case -8: + case 8: + return 8; + case -9: + case 9: + return 8; + case 10: + return 4; + case 12: + return 2; + case 15: + return 2; + case 18: + return 8; + case 20: + return 4; + case 24: + return 2; + case 25: + return 4; + case 27: + return 2; + case 36: + return 8; + case 40: + return 4; + case 45: + return 4; + case 72: + return 8; + case 81: + return 8; + } + diag(Z, "bad m0"); + return 0; +} + +int +m1(int a) +{ + switch(a) { + case 15: + return 4; + case 25: + return 4; + case 27: + return 8; + case 45: + return 8; + case 81: + return 8; + } + diag(Z, "bad m1"); + return 0; +} + +int +m2(int a) +{ + switch(a) { + case 6: + return 2; + case 10: + return 2; + case 12: + return 4; + case 18: + return 2; + case 20: + return 4; + case 24: + return 8; + case 36: + return 4; + case 40: + return 8; + case 72: + return 8; + } + diag(Z, "bad m2"); + return 0; +} + +void +shiftit(Type *t, Node *s, Node *d) +{ + long c; + + c = (long)s->vconst & 31; + switch(c) { + case 0: + break; + case 1: + gopcode(OADD, t, d, d); + break; + default: + gopcode(OASHL, t, s, d); + } +} + +static int +mulgen1(ulong v, Node *n) +{ + int i, o; + Mparam *p; + Node nod, nods; + + for(i = 0; i < nelem(multab); i++) { + p = &multab[i]; + if(p->value == v) + goto found; + } + + p = &multab[mulptr]; + if(++mulptr == nelem(multab)) + mulptr = 0; + + mulparam(v, p); + +found: +// print("v=%.lx a=%d n=%d s=%d g=%d o=%d \n", p->value, p->alg, p->neg, p->shift, p->arg, p->off); + if(p->alg < 0) + return 0; + + nods = *nodconst(p->shift); + + o = OADD; + if(p->alg > 0) { + regalloc(&nod, n, Z); + if(p->off < 0) + o = OSUB; + } + + switch(p->alg) { + case 0: + switch(p->arg) { + case 1: + shiftit(n->type, &nods, n); + break; + case 15: + case 25: + case 27: + case 45: + case 81: + genmuladd(n, n, m1(p->arg), n); + /* fall thru */ + case 3: + case 5: + case 9: + genmuladd(n, n, m0(p->arg), n); + shiftit(n->type, &nods, n); + break; + default: + goto bad; + } + if(p->neg == 1) + gins(ANEGL, Z, n); + break; + case 1: + switch(p->arg) { + case 1: + gmove(n, &nod); + shiftit(n->type, &nods, &nod); + break; + case 3: + case 5: + case 9: + genmuladd(&nod, n, m0(p->arg), n); + shiftit(n->type, &nods, &nod); + break; + default: + goto bad; + } + if(p->neg) + gopcode(o, n->type, &nod, n); + else { + gopcode(o, n->type, n, &nod); + gmove(&nod, n); + } + break; + case 2: + genmuladd(&nod, n, m0(p->off), n); + shiftit(n->type, &nods, n); + goto comop; + case 3: + genmuladd(&nod, n, m0(p->off), n); + shiftit(n->type, &nods, n); + genmuladd(n, &nod, m2(p->off), n); + break; + case 4: + genmuladd(&nod, n, m0(p->off), nodconst(0)); + shiftit(n->type, &nods, n); + goto comop; + default: + diag(Z, "bad mul alg"); + break; + comop: + if(p->neg) { + gopcode(o, n->type, n, &nod); + gmove(&nod, n); + } + else + gopcode(o, n->type, &nod, n); + } + + if(p->alg > 0) + regfree(&nod); + + return 1; + +bad: + diag(Z, "mulgen botch"); + return 1; +} + +void +mulgen(Type *t, Node *r, Node *n) +{ + if(!mulgen1(r->vconst, n)) + gopcode(OMUL, t, r, n); +} diff --git a/sys/src/cmd/6c/peep.c b/sys/src/cmd/6c/peep.c new file mode 100644 index 000000000..2511b6c7e --- /dev/null +++ b/sys/src/cmd/6c/peep.c @@ -0,0 +1,846 @@ +#include "gc.h" + +static int +needc(Prog *p) +{ + while(p != P) { + switch(p->as) { + case AADCL: + case AADCQ: + case ASBBL: + case ASBBQ: + case ARCRL: + case ARCRQ: + return 1; + case AADDL: + case AADDQ: + case ASUBL: + case ASUBQ: + case AJMP: + case ARET: + case ACALL: + return 0; + default: + if(p->to.type == D_BRANCH) + return 0; + } + p = p->link; + } + return 0; +} + +static Reg* +rnops(Reg *r) +{ + Prog *p; + Reg *r1; + + if(r != R) + for(;;){ + p = r->prog; + if(p->as != ANOP || p->from.type != D_NONE || p->to.type != D_NONE) + break; + r1 = uniqs(r); + if(r1 == R) + break; + r = r1; + } + return r; +} + +void +peep(void) +{ + Reg *r, *r1, *r2; + Prog *p, *p1; + int t; + + /* + * complete R structure + */ + t = 0; + for(r=firstr; r!=R; r=r1) { + r1 = r->link; + if(r1 == R) + break; + p = r->prog->link; + while(p != r1->prog) + switch(p->as) { + default: + r2 = rega(); + r->link = r2; + r2->link = r1; + + r2->prog = p; + r2->p1 = r; + r->s1 = r2; + r2->s1 = r1; + r1->p1 = r2; + + r = r2; + t++; + + case ADATA: + case AGLOBL: + case ANAME: + case ASIGNAME: + p = p->link; + } + } + + pc = 0; /* speculating it won't kill */ + +loop1: + + t = 0; + for(r=firstr; r!=R; r=r->link) { + p = r->prog; + switch(p->as) { + case AMOVL: + case AMOVQ: + case AMOVSS: + case AMOVSD: + if(regtyp(&p->to)) + if(regtyp(&p->from)) { + if(copyprop(r)) { + excise(r); + t++; + } else + if(subprop(r) && copyprop(r)) { + excise(r); + t++; + } + } + break; + + case AMOVBLZX: + case AMOVWLZX: + case AMOVBLSX: + case AMOVWLSX: + if(regtyp(&p->to)) { + r1 = rnops(uniqs(r)); + if(r1 != R) { + p1 = r1->prog; + if(p->as == p1->as && p->to.type == p1->from.type){ + p1->as = AMOVL; + t++; + } + } + } + break; + + case AMOVBQSX: + case AMOVBQZX: + case AMOVWQSX: + case AMOVWQZX: + case AMOVLQSX: + case AMOVLQZX: + if(regtyp(&p->to)) { + r1 = rnops(uniqs(r)); + if(r1 != R) { + p1 = r1->prog; + if(p->as == p1->as && p->to.type == p1->from.type){ + p1->as = AMOVQ; + t++; + } + } + } + break; + + case AADDL: + case AADDQ: + case AADDW: + if(p->from.type != D_CONST || needc(p->link)) + break; + if(p->from.offset == -1){ + if(p->as == AADDQ) + p->as = ADECQ; + else if(p->as == AADDL) + p->as = ADECL; + else + p->as = ADECW; + p->from = zprog.from; + } + else if(p->from.offset == 1){ + if(p->as == AADDQ) + p->as = AINCQ; + else if(p->as == AADDL) + p->as = AINCL; + else + p->as = AINCW; + p->from = zprog.from; + } + break; + + case ASUBL: + case ASUBQ: + case ASUBW: + if(p->from.type != D_CONST || needc(p->link)) + break; + if(p->from.offset == -1) { + if(p->as == ASUBQ) + p->as = AINCQ; + else if(p->as == ASUBL) + p->as = AINCL; + else + p->as = AINCW; + p->from = zprog.from; + } + else if(p->from.offset == 1){ + if(p->as == ASUBQ) + p->as = ADECQ; + else if(p->as == ASUBL) + p->as = ADECL; + else + p->as = ADECW; + p->from = zprog.from; + } + break; + } + } + if(t) + goto loop1; +} + +void +excise(Reg *r) +{ + Prog *p; + + p = r->prog; + p->as = ANOP; + p->from = zprog.from; + p->to = zprog.to; +} + +Reg* +uniqp(Reg *r) +{ + Reg *r1; + + r1 = r->p1; + if(r1 == R) { + r1 = r->p2; + if(r1 == R || r1->p2link != R) + return R; + } else + if(r->p2 != R) + return R; + return r1; +} + +Reg* +uniqs(Reg *r) +{ + Reg *r1; + + r1 = r->s1; + if(r1 == R) { + r1 = r->s2; + if(r1 == R) + return R; + } else + if(r->s2 != R) + return R; + return r1; +} + +int +regtyp(Adr *a) +{ + int t; + + t = a->type; + if(t >= D_AX && t <= D_R15) + return 1; + if(t >= D_X0 && t <= D_X0+15) + return 1; + return 0; +} + +/* + * the idea is to substitute + * one register for another + * from one MOV to another + * MOV a, R0 + * ADD b, R0 / no use of R1 + * MOV R0, R1 + * would be converted to + * MOV a, R1 + * ADD b, R1 + * MOV R1, R0 + * hopefully, then the former or latter MOV + * will be eliminated by copy propagation. + */ +int +subprop(Reg *r0) +{ + Prog *p; + Adr *v1, *v2; + Reg *r; + int t; + + p = r0->prog; + v1 = &p->from; + if(!regtyp(v1)) + return 0; + v2 = &p->to; + if(!regtyp(v2)) + return 0; + for(r=uniqp(r0); r!=R; r=uniqp(r)) { + if(uniqs(r) == R) + break; + p = r->prog; + switch(p->as) { + case ACALL: + return 0; + + case AIMULL: + case AIMULQ: + case AIMULW: + if(p->to.type != D_NONE) + break; + + case ADIVB: + case ADIVL: + case ADIVQ: + case ADIVW: + case AIDIVB: + case AIDIVL: + case AIDIVQ: + case AIDIVW: + case AIMULB: + case AMULB: + case AMULL: + case AMULQ: + case AMULW: + + case AROLB: + case AROLL: + case AROLQ: + case AROLW: + case ARORB: + case ARORL: + case ARORQ: + case ARORW: + case ASALB: + case ASALL: + case ASALQ: + case ASALW: + case ASARB: + case ASARL: + case ASARQ: + case ASARW: + case ASHLB: + case ASHLL: + case ASHLQ: + case ASHLW: + case ASHRB: + case ASHRL: + case ASHRQ: + case ASHRW: + + case AREP: + case AREPN: + + case ACWD: + case ACDQ: + case ACQO: + + case AMOVSL: + case AMOVSQ: + return 0; + + case AMOVL: + case AMOVQ: + if(p->to.type == v1->type) + goto gotit; + break; + } + if(copyau(&p->from, v2) || + copyau(&p->to, v2)) + break; + if(copysub(&p->from, v1, v2, 0) || + copysub(&p->to, v1, v2, 0)) + break; + } + return 0; + +gotit: + copysub(&p->to, v1, v2, 1); + if(debug['P']) { + print("gotit: %D->%D\n%P", v1, v2, r->prog); + if(p->from.type == v2->type) + print(" excise"); + print("\n"); + } + for(r=uniqs(r); r!=r0; r=uniqs(r)) { + p = r->prog; + copysub(&p->from, v1, v2, 1); + copysub(&p->to, v1, v2, 1); + if(debug['P']) + print("%P\n", r->prog); + } + t = v1->type; + v1->type = v2->type; + v2->type = t; + if(debug['P']) + print("%P last\n", r->prog); + return 1; +} + +/* + * The idea is to remove redundant copies. + * v1->v2 F=0 + * (use v2 s/v2/v1/)* + * set v1 F=1 + * use v2 return fail + * ----------------- + * v1->v2 F=0 + * (use v2 s/v2/v1/)* + * set v1 F=1 + * set v2 return success + */ +int +copyprop(Reg *r0) +{ + Prog *p; + Adr *v1, *v2; + Reg *r; + + p = r0->prog; + v1 = &p->from; + v2 = &p->to; + if(copyas(v1, v2)) + return 1; + for(r=firstr; r!=R; r=r->link) + r->active = 0; + return copy1(v1, v2, r0->s1, 0); +} + +int +copy1(Adr *v1, Adr *v2, Reg *r, int f) +{ + int t; + Prog *p; + + if(r->active) { + if(debug['P']) + print("act set; return 1\n"); + return 1; + } + r->active = 1; + if(debug['P']) + print("copy %D->%D f=%d\n", v1, v2, f); + for(; r != R; r = r->s1) { + p = r->prog; + if(debug['P']) + print("%P", p); + if(!f && uniqp(r) == R) { + f = 1; + if(debug['P']) + print("; merge; f=%d", f); + } + t = copyu(p, v2, A); + switch(t) { + case 2: /* rar, cant split */ + if(debug['P']) + print("; %D rar; return 0\n", v2); + return 0; + + case 3: /* set */ + if(debug['P']) + print("; %D set; return 1\n", v2); + return 1; + + case 1: /* used, substitute */ + case 4: /* use and set */ + if(f) { + if(!debug['P']) + return 0; + if(t == 4) + print("; %D used+set and f=%d; return 0\n", v2, f); + else + print("; %D used and f=%d; return 0\n", v2, f); + return 0; + } + if(copyu(p, v2, v1)) { + if(debug['P']) + print("; sub fail; return 0\n"); + return 0; + } + if(debug['P']) + print("; sub %D/%D", v2, v1); + if(t == 4) { + if(debug['P']) + print("; %D used+set; return 1\n", v2); + return 1; + } + break; + } + if(!f) { + t = copyu(p, v1, A); + if(!f && (t == 2 || t == 3 || t == 4)) { + f = 1; + if(debug['P']) + print("; %D set and !f; f=%d", v1, f); + } + } + if(debug['P']) + print("\n"); + if(r->s2) + if(!copy1(v1, v2, r->s2, f)) + return 0; + } + return 1; +} + +/* + * return + * 1 if v only used (and substitute), + * 2 if read-alter-rewrite + * 3 if set + * 4 if set and used + * 0 otherwise (not touched) + */ +int +copyu(Prog *p, Adr *v, Adr *s) +{ + + switch(p->as) { + + default: + if(debug['P']) + print("unknown op %A\n", p->as); + /* SBBL; ADCL; FLD1; SAHF */ + return 2; + + + case ANEGB: + case ANEGW: + case ANEGL: + case ANEGQ: + case ANOTB: + case ANOTW: + case ANOTL: + case ANOTQ: + if(copyas(&p->to, v)) + return 2; + break; + + case ALEAL: /* lhs addr, rhs store */ + case ALEAQ: + if(copyas(&p->from, v)) + return 2; + + + case ANOP: /* rhs store */ + case AMOVL: + case AMOVQ: + case AMOVBLSX: + case AMOVBLZX: + case AMOVBQSX: + case AMOVBQZX: + case AMOVLQSX: + case AMOVLQZX: + case AMOVWLSX: + case AMOVWLZX: + case AMOVWQSX: + case AMOVWQZX: + + case AMOVSS: + case AMOVSD: + case ACVTSD2SL: + case ACVTSD2SQ: + case ACVTSD2SS: + case ACVTSL2SD: + case ACVTSL2SS: + case ACVTSQ2SD: + case ACVTSQ2SS: + case ACVTSS2SD: + case ACVTSS2SL: + case ACVTSS2SQ: + case ACVTTSD2SL: + case ACVTTSD2SQ: + case ACVTTSS2SL: + case ACVTTSS2SQ: + if(copyas(&p->to, v)) { + if(s != A) + return copysub(&p->from, v, s, 1); + if(copyau(&p->from, v)) + return 4; + return 3; + } + goto caseread; + + case AROLB: + case AROLL: + case AROLQ: + case AROLW: + case ARORB: + case ARORL: + case ARORQ: + case ARORW: + case ASALB: + case ASALL: + case ASALQ: + case ASALW: + case ASARB: + case ASARL: + case ASARQ: + case ASARW: + case ASHLB: + case ASHLL: + case ASHLQ: + case ASHLW: + case ASHRB: + case ASHRL: + case ASHRQ: + case ASHRW: + if(copyas(&p->to, v)) + return 2; + if(copyas(&p->from, v)) + if(p->from.type == D_CX) + return 2; + goto caseread; + + case AADDB: /* rhs rar */ + case AADDL: + case AADDQ: + case AADDW: + case AANDB: + case AANDL: + case AANDQ: + case AANDW: + case ADECL: + case ADECQ: + case ADECW: + case AINCL: + case AINCQ: + case AINCW: + case ASUBB: + case ASUBL: + case ASUBQ: + case ASUBW: + case AORB: + case AORL: + case AORQ: + case AORW: + case AXORB: + case AXORL: + case AXORQ: + case AXORW: + case AMOVB: + case AMOVW: + + case AADDSD: + case AADDSS: + case ACMPSD: + case ACMPSS: + case ADIVSD: + case ADIVSS: + case AMAXSD: + case AMAXSS: + case AMINSD: + case AMINSS: + case AMULSD: + case AMULSS: + case ARCPSS: + case ARSQRTSS: + case ASQRTSD: + case ASQRTSS: + case ASUBSD: + case ASUBSS: + case AXORPD: + if(copyas(&p->to, v)) + return 2; + goto caseread; + + case ACMPL: /* read only */ + case ACMPW: + case ACMPB: + case ACMPQ: + + case ACOMISD: + case ACOMISS: + case AUCOMISD: + case AUCOMISS: + caseread: + if(s != A) { + if(copysub(&p->from, v, s, 1)) + return 1; + return copysub(&p->to, v, s, 1); + } + if(copyau(&p->from, v)) + return 1; + if(copyau(&p->to, v)) + return 1; + break; + + case AJGE: /* no reference */ + case AJNE: + case AJLE: + case AJEQ: + case AJHI: + case AJLS: + case AJMI: + case AJPL: + case AJGT: + case AJLT: + case AJCC: + case AJCS: + + case AADJSP: + case AWAIT: + case ACLD: + break; + + case AIMULL: + case AIMULQ: + case AIMULW: + if(p->to.type != D_NONE) { + if(copyas(&p->to, v)) + return 2; + goto caseread; + } + + case ADIVB: + case ADIVL: + case ADIVQ: + case ADIVW: + case AIDIVB: + case AIDIVL: + case AIDIVQ: + case AIDIVW: + case AIMULB: + case AMULB: + case AMULL: + case AMULQ: + case AMULW: + + case ACWD: + case ACDQ: + case ACQO: + if(v->type == D_AX || v->type == D_DX) + return 2; + goto caseread; + + case AMOVSL: + case AMOVSQ: + case AREP: + case AREPN: + if(v->type == D_CX || v->type == D_DI || v->type == D_SI) + return 2; + goto caseread; + + case AJMP: /* funny */ + if(s != A) { + if(copysub(&p->to, v, s, 1)) + return 1; + return 0; + } + if(copyau(&p->to, v)) + return 1; + return 0; + + case ARET: /* funny */ + if(v->type == REGRET || v->type == FREGRET) + return 2; + if(s != A) + return 1; + return 3; + + case ACALL: /* funny */ + if(REGEXT && v->type <= REGEXT && v->type > exregoffset) + return 2; + if(REGARG && v->type == REGARG) + return 2; + + if(s != A) { + if(copysub(&p->to, v, s, 1)) + return 1; + return 0; + } + if(copyau(&p->to, v)) + return 4; + return 3; + + case ATEXT: /* funny */ + if(REGARG && v->type == REGARG) + return 3; + return 0; + } + return 0; +} + +/* + * direct reference, + * could be set/use depending on + * semantics + */ +int +copyas(Adr *a, Adr *v) +{ + if(a->type != v->type) + return 0; + if(regtyp(v)) + return 1; + if(v->type == D_AUTO || v->type == D_PARAM) + if(v->offset == a->offset) + return 1; + return 0; +} + +/* + * either direct or indirect + */ +int +copyau(Adr *a, Adr *v) +{ + + if(copyas(a, v)) + return 1; + if(regtyp(v)) { + if(a->type-D_INDIR == v->type) + return 1; + if(a->index == v->type) + return 1; + } + return 0; +} + +/* + * substitute s for v in a + * return failure to substitute + */ +int +copysub(Adr *a, Adr *v, Adr *s, int f) +{ + int t; + + if(copyas(a, v)) { + t = s->type; + if(t >= D_AX && t <= D_R15 || t >= D_X0 && t <= D_X0+15) { + if(f) + a->type = t; + } + return 0; + } + if(regtyp(v)) { + t = v->type; + if(a->type == t+D_INDIR) { + if((s->type == D_BP || s->type == D_R13) && a->index != D_NONE) + return 1; /* can't use BP-base with index */ + if(f) + a->type = s->type+D_INDIR; +// return 0; + } + if(a->index == t) { + if(f) + a->index = s->type; + return 0; + } + return 0; + } + return 0; +} diff --git a/sys/src/cmd/6c/reg.c b/sys/src/cmd/6c/reg.c new file mode 100644 index 000000000..32b9abf5d --- /dev/null +++ b/sys/src/cmd/6c/reg.c @@ -0,0 +1,1356 @@ +#include "gc.h" + +Reg* +rega(void) +{ + Reg *r; + + r = freer; + if(r == R) { + r = alloc(sizeof(*r)); + } else + freer = r->link; + + *r = zreg; + return r; +} + +int +rcmp(const void *a1, const void *a2) +{ + Rgn *p1, *p2; + int c1, c2; + + p1 = (Rgn*)a1; + p2 = (Rgn*)a2; + c1 = p2->cost; + c2 = p1->cost; + if(c1 -= c2) + return c1; + return p2->varno - p1->varno; +} + +void +regopt(Prog *p) +{ + Reg *r, *r1, *r2; + Prog *p1; + int i, z; + long initpc, val, npc; + ulong vreg; + Bits bit; + struct + { + long m; + long c; + Reg* p; + } log5[6], *lp; + + firstr = R; + lastr = R; + nvar = 0; + regbits = RtoB(D_SP) | RtoB(D_AX) | RtoB(D_X0); + for(z=0; z<BITS; z++) { + externs.b[z] = 0; + params.b[z] = 0; + consts.b[z] = 0; + addrs.b[z] = 0; + } + + /* + * pass 1 + * build aux data structure + * allocate pcs + * find use and set of variables + */ + val = 5L * 5L * 5L * 5L * 5L; + lp = log5; + for(i=0; i<5; i++) { + lp->m = val; + lp->c = 0; + lp->p = R; + val /= 5L; + lp++; + } + val = 0; + for(; p != P; p = p->link) { + switch(p->as) { + case ADATA: + case AGLOBL: + case ANAME: + case ASIGNAME: + continue; + } + r = rega(); + if(firstr == R) { + firstr = r; + lastr = r; + } else { + lastr->link = r; + r->p1 = lastr; + lastr->s1 = r; + lastr = r; + } + r->prog = p; + r->pc = val; + val++; + + lp = log5; + for(i=0; i<5; i++) { + lp->c--; + if(lp->c <= 0) { + lp->c = lp->m; + if(lp->p != R) + lp->p->log5 = r; + lp->p = r; + (lp+1)->c = 0; + break; + } + lp++; + } + + r1 = r->p1; + if(r1 != R) + switch(r1->prog->as) { + case ARET: + case AJMP: + case AIRETL: + case AIRETQ: + r->p1 = R; + r1->s1 = R; + } + + bit = mkvar(r, &p->from); + if(bany(&bit)) + switch(p->as) { + /* + * funny + */ + case ALEAL: + case ALEAQ: + for(z=0; z<BITS; z++) + addrs.b[z] |= bit.b[z]; + break; + + /* + * left side read + */ + default: + for(z=0; z<BITS; z++) + r->use1.b[z] |= bit.b[z]; + break; + } + + bit = mkvar(r, &p->to); + if(bany(&bit)) + switch(p->as) { + default: + diag(Z, "reg: unknown op: %A", p->as); + break; + + /* + * right side read + */ + case ACMPB: + case ACMPL: + case ACMPQ: + case ACMPW: + case ACOMISS: + case ACOMISD: + case AUCOMISS: + case AUCOMISD: + for(z=0; z<BITS; z++) + r->use2.b[z] |= bit.b[z]; + break; + + /* + * right side write + */ + case ANOP: + case AMOVL: + case AMOVQ: + case AMOVB: + case AMOVW: + case AMOVBLSX: + case AMOVBLZX: + case AMOVBQSX: + case AMOVBQZX: + case AMOVLQSX: + case AMOVLQZX: + case AMOVWLSX: + case AMOVWLZX: + case AMOVWQSX: + case AMOVWQZX: + + case AMOVSS: + case AMOVSD: + case ACVTSD2SL: + case ACVTSD2SQ: + case ACVTSD2SS: + case ACVTSL2SD: + case ACVTSL2SS: + case ACVTSQ2SD: + case ACVTSQ2SS: + case ACVTSS2SD: + case ACVTSS2SL: + case ACVTSS2SQ: + case ACVTTSD2SL: + case ACVTTSD2SQ: + case ACVTTSS2SL: + case ACVTTSS2SQ: + for(z=0; z<BITS; z++) + r->set.b[z] |= bit.b[z]; + break; + + /* + * right side read+write + */ + case AADDB: + case AADDL: + case AADDQ: + case AADDW: + case AANDB: + case AANDL: + case AANDQ: + case AANDW: + case ASUBB: + case ASUBL: + case ASUBQ: + case ASUBW: + case AORB: + case AORL: + case AORQ: + case AORW: + case AXORB: + case AXORL: + case AXORQ: + case AXORW: + case ASALB: + case ASALL: + case ASALQ: + case ASALW: + case ASARB: + case ASARL: + case ASARQ: + case ASARW: + case AROLB: + case AROLL: + case AROLQ: + case AROLW: + case ARORB: + case ARORL: + case ARORQ: + case ARORW: + case ASHLB: + case ASHLL: + case ASHLQ: + case ASHLW: + case ASHRB: + case ASHRL: + case ASHRQ: + case ASHRW: + case AIMULL: + case AIMULQ: + case AIMULW: + case ANEGL: + case ANEGQ: + case ANOTL: + case ANOTQ: + case AADCL: + case AADCQ: + case ASBBL: + case ASBBQ: + + case AADDSD: + case AADDSS: + case ACMPSD: + case ACMPSS: + case ADIVSD: + case ADIVSS: + case AMAXSD: + case AMAXSS: + case AMINSD: + case AMINSS: + case AMULSD: + case AMULSS: + case ARCPSS: + case ARSQRTSS: + case ASQRTSD: + case ASQRTSS: + case ASUBSD: + case ASUBSS: + case AXORPD: + for(z=0; z<BITS; z++) { + r->set.b[z] |= bit.b[z]; + r->use2.b[z] |= bit.b[z]; + } + break; + + /* + * funny + */ + case ACALL: + for(z=0; z<BITS; z++) + addrs.b[z] |= bit.b[z]; + break; + } + + switch(p->as) { + case AIMULL: + case AIMULQ: + case AIMULW: + if(p->to.type != D_NONE) + break; + + case AIDIVB: + case AIDIVL: + case AIDIVQ: + case AIDIVW: + case AIMULB: + case ADIVB: + case ADIVL: + case ADIVQ: + case ADIVW: + case AMULB: + case AMULL: + case AMULQ: + case AMULW: + + case ACWD: + case ACDQ: + case ACQO: + r->regu |= RtoB(D_AX) | RtoB(D_DX); + break; + + case AREP: + case AREPN: + case ALOOP: + case ALOOPEQ: + case ALOOPNE: + r->regu |= RtoB(D_CX); + break; + + case AMOVSB: + case AMOVSL: + case AMOVSQ: + case AMOVSW: + case ACMPSB: + case ACMPSL: + case ACMPSQ: + case ACMPSW: + r->regu |= RtoB(D_SI) | RtoB(D_DI); + break; + + case ASTOSB: + case ASTOSL: + case ASTOSQ: + case ASTOSW: + case ASCASB: + case ASCASL: + case ASCASQ: + case ASCASW: + r->regu |= RtoB(D_AX) | RtoB(D_DI); + break; + + case AINSB: + case AINSL: + case AINSW: + case AOUTSB: + case AOUTSL: + case AOUTSW: + r->regu |= RtoB(D_DI) | RtoB(D_DX); + break; + } + } + if(firstr == R) + return; + initpc = pc - val; + npc = val; + + /* + * pass 2 + * turn branch references to pointers + * build back pointers + */ + for(r = firstr; r != R; r = r->link) { + p = r->prog; + if(p->to.type == D_BRANCH) { + val = p->to.offset - initpc; + r1 = firstr; + while(r1 != R) { + r2 = r1->log5; + if(r2 != R && val >= r2->pc) { + r1 = r2; + continue; + } + if(r1->pc == val) + break; + r1 = r1->link; + } + if(r1 == R) { + nearln = p->lineno; + diag(Z, "ref not found\n%P", p); + continue; + } + if(r1 == r) { + nearln = p->lineno; + diag(Z, "ref to self\n%P", p); + continue; + } + r->s2 = r1; + r->p2link = r1->p2; + r1->p2 = r; + } + } + if(debug['R']) { + p = firstr->prog; + print("\n%L %D\n", p->lineno, &p->from); + } + + /* + * pass 2.5 + * find looping structure + */ + for(r = firstr; r != R; r = r->link) + r->active = 0; + change = 0; + loopit(firstr, npc); + if(debug['R'] && debug['v']) { + print("\nlooping structure:\n"); + for(r = firstr; r != R; r = r->link) { + print("%ld:%P", r->loop, r->prog); + for(z=0; z<BITS; z++) + bit.b[z] = r->use1.b[z] | + r->use2.b[z] | + r->set.b[z]; + if(bany(&bit)) { + print("\t"); + if(bany(&r->use1)) + print(" u1=%B", r->use1); + if(bany(&r->use2)) + print(" u2=%B", r->use2); + if(bany(&r->set)) + print(" st=%B", r->set); + } + print("\n"); + } + } + + /* + * pass 3 + * iterate propagating usage + * back until flow graph is complete + */ +loop1: + change = 0; + for(r = firstr; r != R; r = r->link) + r->active = 0; + for(r = firstr; r != R; r = r->link) + if(r->prog->as == ARET) + prop(r, zbits, zbits); +loop11: + /* pick up unreachable code */ + i = 0; + for(r = firstr; r != R; r = r1) { + r1 = r->link; + if(r1 && r1->active && !r->active) { + prop(r, zbits, zbits); + i = 1; + } + } + if(i) + goto loop11; + if(change) + goto loop1; + + + /* + * pass 4 + * iterate propagating register/variable synchrony + * forward until graph is complete + */ +loop2: + change = 0; + for(r = firstr; r != R; r = r->link) + r->active = 0; + synch(firstr, zbits); + if(change) + goto loop2; + + + /* + * pass 5 + * isolate regions + * calculate costs (paint1) + */ + r = firstr; + if(r) { + for(z=0; z<BITS; z++) + bit.b[z] = (r->refahead.b[z] | r->calahead.b[z]) & + ~(externs.b[z] | params.b[z] | addrs.b[z] | consts.b[z]); + if(bany(&bit)) { + nearln = r->prog->lineno; + warn(Z, "used and not set: %B", bit); + if(debug['R'] && !debug['w']) + print("used and not set: %B\n", bit); + } + } + if(debug['R'] && debug['v']) + print("\nprop structure:\n"); + for(r = firstr; r != R; r = r->link) + r->act = zbits; + rgp = region; + nregion = 0; + for(r = firstr; r != R; r = r->link) { + if(debug['R'] && debug['v']) { + print("%P\t", r->prog); + if(bany(&r->set)) + print("s:%B ", r->set); + if(bany(&r->refahead)) + print("ra:%B ", r->refahead); + if(bany(&r->calahead)) + print("ca:%B ", r->calahead); + print("\n"); + } + for(z=0; z<BITS; z++) + bit.b[z] = r->set.b[z] & + ~(r->refahead.b[z] | r->calahead.b[z] | addrs.b[z]); + if(bany(&bit)) { + nearln = r->prog->lineno; + warn(Z, "set and not used: %B", bit); + if(debug['R']) + print("set and not used: %B\n", bit); + excise(r); + } + for(z=0; z<BITS; z++) + bit.b[z] = LOAD(r) & ~(r->act.b[z] | addrs.b[z]); + while(bany(&bit)) { + i = bnum(bit); + rgp->enter = r; + rgp->varno = i; + change = 0; + if(debug['R'] && debug['v']) + print("\n"); + paint1(r, i); + bit.b[i/32] &= ~(1L<<(i%32)); + if(change <= 0) { + if(debug['R']) + print("%L$%d: %B\n", + r->prog->lineno, change, blsh(i)); + continue; + } + rgp->cost = change; + nregion++; + if(nregion >= NRGN) { + warn(Z, "too many regions"); + goto brk; + } + rgp++; + } + } +brk: + qsort(region, nregion, sizeof(region[0]), rcmp); + + /* + * pass 6 + * determine used registers (paint2) + * replace code (paint3) + */ + rgp = region; + for(i=0; i<nregion; i++) { + bit = blsh(rgp->varno); + vreg = paint2(rgp->enter, rgp->varno); + vreg = allreg(vreg, rgp); + if(debug['R']) { + print("%L$%d %R: %B\n", + rgp->enter->prog->lineno, + rgp->cost, + rgp->regno, + bit); + } + if(rgp->regno != 0) + paint3(rgp->enter, rgp->varno, vreg, rgp->regno); + rgp++; + } + /* + * pass 7 + * peep-hole on basic block + */ + if(!debug['R'] || debug['P']) + peep(); + + /* + * pass 8 + * recalculate pc + */ + val = initpc; + for(r = firstr; r != R; r = r1) { + r->pc = val; + p = r->prog; + p1 = P; + r1 = r->link; + if(r1 != R) + p1 = r1->prog; + for(; p != p1; p = p->link) { + switch(p->as) { + default: + val++; + break; + + case ANOP: + case ADATA: + case AGLOBL: + case ANAME: + case ASIGNAME: + break; + } + } + } + pc = val; + + /* + * fix up branches + */ + if(debug['R']) + if(bany(&addrs)) + print("addrs: %B\n", addrs); + + r1 = 0; /* set */ + for(r = firstr; r != R; r = r->link) { + p = r->prog; + if(p->to.type == D_BRANCH) + p->to.offset = r->s2->pc; + r1 = r; + } + + /* + * last pass + * eliminate nops + * free aux structures + */ + for(p = firstr->prog; p != P; p = p->link){ + while(p->link && p->link->as == ANOP) + p->link = p->link->link; + } + if(r1 != R) { + r1->link = freer; + freer = firstr; + } +} + +/* + * add mov b,rn + * just after r + */ +void +addmove(Reg *r, int bn, int rn, int f) +{ + Prog *p, *p1; + Adr *a; + Var *v; + + p1 = alloc(sizeof(*p1)); + *p1 = zprog; + p = r->prog; + + p1->link = p->link; + p->link = p1; + p1->lineno = p->lineno; + + v = var + bn; + + a = &p1->to; + a->sym = v->sym; + a->offset = v->offset; + a->etype = v->etype; + a->type = v->name; + + p1->as = AMOVL; + if(v->etype == TCHAR || v->etype == TUCHAR) + p1->as = AMOVB; + if(v->etype == TSHORT || v->etype == TUSHORT) + p1->as = AMOVW; + if(v->etype == TVLONG || v->etype == TUVLONG || v->etype == TIND) + p1->as = AMOVQ; + if(v->etype == TFLOAT) + p1->as = AMOVSS; + if(v->etype == TDOUBLE) + p1->as = AMOVSD; + + p1->from.type = rn; + if(!f) { + p1->from = *a; + *a = zprog.from; + a->type = rn; + if(v->etype == TUCHAR) + p1->as = AMOVB; + if(v->etype == TUSHORT) + p1->as = AMOVW; + } + if(debug['R']) + print("%P\t.a%P\n", p, p1); +} + +ulong +doregbits(int r) +{ + ulong b; + + b = 0; + if(r >= D_INDIR) + r -= D_INDIR; + if(r >= D_AX && r <= D_R15) + b |= RtoB(r); + else + if(r >= D_AL && r <= D_R15B) + b |= RtoB(r-D_AL+D_AX); + else + if(r >= D_AH && r <= D_BH) + b |= RtoB(r-D_AH+D_AX); + else + if(r >= D_X0 && r <= D_X0+15) + b |= FtoB(r); + return b; +} + +Bits +mkvar(Reg *r, Adr *a) +{ + Var *v; + int i, t, n, et, z; + long o; + Bits bit; + Sym *s; + + /* + * mark registers used + */ + t = a->type; + r->regu |= doregbits(t); + r->regu |= doregbits(a->index); + + switch(t) { + default: + goto none; + case D_ADDR: + a->type = a->index; + bit = mkvar(r, a); + for(z=0; z<BITS; z++) + addrs.b[z] |= bit.b[z]; + a->type = t; + goto none; + case D_EXTERN: + case D_STATIC: + case D_PARAM: + case D_AUTO: + n = t; + break; + } + s = a->sym; + if(s == S) + goto none; + if(s->name[0] == '.') + goto none; + et = a->etype; + o = a->offset; + v = var; + for(i=0; i<nvar; i++) { + if(s == v->sym) + if(n == v->name) + if(o == v->offset) + goto out; + v++; + } + if(nvar >= NVAR) { + if(debug['w'] > 1 && s) + warn(Z, "variable not optimized: %s", s->name); + goto none; + } + i = nvar; + nvar++; + v = &var[i]; + v->sym = s; + v->offset = o; + v->name = n; + v->etype = et; + if(debug['R']) + print("bit=%2d et=%2d %D\n", i, et, a); + +out: + bit = blsh(i); + if(n == D_EXTERN || n == D_STATIC) + for(z=0; z<BITS; z++) + externs.b[z] |= bit.b[z]; + if(n == D_PARAM) + for(z=0; z<BITS; z++) + params.b[z] |= bit.b[z]; + if(v->etype != et || !(typechlpfd[et] || typev[et])) /* funny punning */ + for(z=0; z<BITS; z++) + addrs.b[z] |= bit.b[z]; + return bit; + +none: + return zbits; +} + +void +prop(Reg *r, Bits ref, Bits cal) +{ + Reg *r1, *r2; + int z; + + for(r1 = r; r1 != R; r1 = r1->p1) { + for(z=0; z<BITS; z++) { + ref.b[z] |= r1->refahead.b[z]; + if(ref.b[z] != r1->refahead.b[z]) { + r1->refahead.b[z] = ref.b[z]; + change++; + } + cal.b[z] |= r1->calahead.b[z]; + if(cal.b[z] != r1->calahead.b[z]) { + r1->calahead.b[z] = cal.b[z]; + change++; + } + } + switch(r1->prog->as) { + case ACALL: + for(z=0; z<BITS; z++) { + cal.b[z] |= ref.b[z] | externs.b[z]; + ref.b[z] = 0; + } + break; + + case ATEXT: + for(z=0; z<BITS; z++) { + cal.b[z] = 0; + ref.b[z] = 0; + } + break; + + case ARET: + for(z=0; z<BITS; z++) { + cal.b[z] = externs.b[z]; + ref.b[z] = 0; + } + } + for(z=0; z<BITS; z++) { + ref.b[z] = (ref.b[z] & ~r1->set.b[z]) | + r1->use1.b[z] | r1->use2.b[z]; + cal.b[z] &= ~(r1->set.b[z] | r1->use1.b[z] | r1->use2.b[z]); + r1->refbehind.b[z] = ref.b[z]; + r1->calbehind.b[z] = cal.b[z]; + } + if(r1->active) + break; + r1->active = 1; + } + for(; r != r1; r = r->p1) + for(r2 = r->p2; r2 != R; r2 = r2->p2link) + prop(r2, r->refbehind, r->calbehind); +} + +/* + * find looping structure + * + * 1) find reverse postordering + * 2) find approximate dominators, + * the actual dominators if the flow graph is reducible + * otherwise, dominators plus some other non-dominators. + * See Matthew S. Hecht and Jeffrey D. Ullman, + * "Analysis of a Simple Algorithm for Global Data Flow Problems", + * Conf. Record of ACM Symp. on Principles of Prog. Langs, Boston, Massachusetts, + * Oct. 1-3, 1973, pp. 207-217. + * 3) find all nodes with a predecessor dominated by the current node. + * such a node is a loop head. + * recursively, all preds with a greater rpo number are in the loop + */ +long +postorder(Reg *r, Reg **rpo2r, long n) +{ + Reg *r1; + + r->rpo = 1; + r1 = r->s1; + if(r1 && !r1->rpo) + n = postorder(r1, rpo2r, n); + r1 = r->s2; + if(r1 && !r1->rpo) + n = postorder(r1, rpo2r, n); + rpo2r[n] = r; + n++; + return n; +} + +long +rpolca(long *idom, long rpo1, long rpo2) +{ + long t; + + if(rpo1 == -1) + return rpo2; + while(rpo1 != rpo2){ + if(rpo1 > rpo2){ + t = rpo2; + rpo2 = rpo1; + rpo1 = t; + } + while(rpo1 < rpo2){ + t = idom[rpo2]; + if(t >= rpo2) + fatal(Z, "bad idom"); + rpo2 = t; + } + } + return rpo1; +} + +int +doms(long *idom, long r, long s) +{ + while(s > r) + s = idom[s]; + return s == r; +} + +int +loophead(long *idom, Reg *r) +{ + long src; + + src = r->rpo; + if(r->p1 != R && doms(idom, src, r->p1->rpo)) + return 1; + for(r = r->p2; r != R; r = r->p2link) + if(doms(idom, src, r->rpo)) + return 1; + return 0; +} + +void +loopmark(Reg **rpo2r, long head, Reg *r) +{ + if(r->rpo < head || r->active == head) + return; + r->active = head; + r->loop += LOOP; + if(r->p1 != R) + loopmark(rpo2r, head, r->p1); + for(r = r->p2; r != R; r = r->p2link) + loopmark(rpo2r, head, r); +} + +void +loopit(Reg *r, long nr) +{ + Reg *r1; + long i, d, me; + + if(nr > maxnr) { + rpo2r = alloc(nr * sizeof(Reg*)); + idom = alloc(nr * sizeof(long)); + maxnr = nr; + } + + d = postorder(r, rpo2r, 0); + if(d > nr) + fatal(Z, "too many reg nodes"); + nr = d; + for(i = 0; i < nr / 2; i++){ + r1 = rpo2r[i]; + rpo2r[i] = rpo2r[nr - 1 - i]; + rpo2r[nr - 1 - i] = r1; + } + for(i = 0; i < nr; i++) + rpo2r[i]->rpo = i; + + idom[0] = 0; + for(i = 0; i < nr; i++){ + r1 = rpo2r[i]; + me = r1->rpo; + d = -1; + if(r1->p1 != R && r1->p1->rpo < me) + d = r1->p1->rpo; + for(r1 = r1->p2; r1 != nil; r1 = r1->p2link) + if(r1->rpo < me) + d = rpolca(idom, d, r1->rpo); + idom[i] = d; + } + + for(i = 0; i < nr; i++){ + r1 = rpo2r[i]; + r1->loop++; + if(r1->p2 != R && loophead(idom, r1)) + loopmark(rpo2r, i, r1); + } +} + +void +synch(Reg *r, Bits dif) +{ + Reg *r1; + int z; + + for(r1 = r; r1 != R; r1 = r1->s1) { + for(z=0; z<BITS; z++) { + dif.b[z] = (dif.b[z] & + ~(~r1->refbehind.b[z] & r1->refahead.b[z])) | + r1->set.b[z] | r1->regdiff.b[z]; + if(dif.b[z] != r1->regdiff.b[z]) { + r1->regdiff.b[z] = dif.b[z]; + change++; + } + } + if(r1->active) + break; + r1->active = 1; + for(z=0; z<BITS; z++) + dif.b[z] &= ~(~r1->calbehind.b[z] & r1->calahead.b[z]); + if(r1->s2 != R) + synch(r1->s2, dif); + } +} + +ulong +allreg(ulong b, Rgn *r) +{ + Var *v; + int i; + + v = var + r->varno; + r->regno = 0; + switch(v->etype) { + + default: + diag(Z, "unknown etype %d/%d", bitno(b), v->etype); + break; + + case TCHAR: + case TUCHAR: + case TSHORT: + case TUSHORT: + case TINT: + case TUINT: + case TLONG: + case TULONG: + case TVLONG: + case TUVLONG: + case TIND: + case TARRAY: + i = BtoR(~b); + if(i && r->cost > 0) { + r->regno = i; + return RtoB(i); + } + break; + + case TDOUBLE: + case TFLOAT: + i = BtoF(~b); + if(i && r->cost > 0) { + r->regno = i; + return FtoB(i); + } + break; + } + return 0; +} + +void +paint1(Reg *r, int bn) +{ + Reg *r1; + Prog *p; + int z; + ulong bb; + + z = bn/32; + bb = 1L<<(bn%32); + if(r->act.b[z] & bb) + return; + for(;;) { + if(!(r->refbehind.b[z] & bb)) + break; + r1 = r->p1; + if(r1 == R) + break; + if(!(r1->refahead.b[z] & bb)) + break; + if(r1->act.b[z] & bb) + break; + r = r1; + } + + if(LOAD(r) & ~(r->set.b[z]&~(r->use1.b[z]|r->use2.b[z])) & bb) { + change -= CLOAD * r->loop; + if(debug['R'] && debug['v']) + print("%ld%P\tld %B $%d\n", r->loop, + r->prog, blsh(bn), change); + } + for(;;) { + r->act.b[z] |= bb; + p = r->prog; + + if(r->use1.b[z] & bb) { + change += CREF * r->loop; + if(debug['R'] && debug['v']) + print("%ld%P\tu1 %B $%d\n", r->loop, + p, blsh(bn), change); + } + + if((r->use2.b[z]|r->set.b[z]) & bb) { + change += CREF * r->loop; + if(debug['R'] && debug['v']) + print("%ld%P\tu2 %B $%d\n", r->loop, + p, blsh(bn), change); + } + + if(STORE(r) & r->regdiff.b[z] & bb) { + change -= CLOAD * r->loop; + if(debug['R'] && debug['v']) + print("%ld%P\tst %B $%d\n", r->loop, + p, blsh(bn), change); + } + + if(r->refbehind.b[z] & bb) + for(r1 = r->p2; r1 != R; r1 = r1->p2link) + if(r1->refahead.b[z] & bb) + paint1(r1, bn); + + if(!(r->refahead.b[z] & bb)) + break; + r1 = r->s2; + if(r1 != R) + if(r1->refbehind.b[z] & bb) + paint1(r1, bn); + r = r->s1; + if(r == R) + break; + if(r->act.b[z] & bb) + break; + if(!(r->refbehind.b[z] & bb)) + break; + } +} + +ulong +regset(Reg *r, ulong bb) +{ + ulong b, set; + Adr v; + int c; + + set = 0; + v = zprog.from; + while(b = bb & ~(bb-1)) { + v.type = b & 0xFFFF? BtoR(b): BtoF(b); + if(v.type == 0) + diag(Z, "zero v.type for %#lux", b); + c = copyu(r->prog, &v, A); + if(c == 3) + set |= b; + bb &= ~b; + } + return set; +} + +ulong +reguse(Reg *r, ulong bb) +{ + ulong b, set; + Adr v; + int c; + + set = 0; + v = zprog.from; + while(b = bb & ~(bb-1)) { + v.type = b & 0xFFFF? BtoR(b): BtoF(b); + c = copyu(r->prog, &v, A); + if(c == 1 || c == 2 || c == 4) + set |= b; + bb &= ~b; + } + return set; +} + +ulong +paint2(Reg *r, int bn) +{ + Reg *r1; + int z; + ulong bb, vreg, x; + + z = bn/32; + bb = 1L << (bn%32); + vreg = regbits; + if(!(r->act.b[z] & bb)) + return vreg; + for(;;) { + if(!(r->refbehind.b[z] & bb)) + break; + r1 = r->p1; + if(r1 == R) + break; + if(!(r1->refahead.b[z] & bb)) + break; + if(!(r1->act.b[z] & bb)) + break; + r = r1; + } + for(;;) { + r->act.b[z] &= ~bb; + + vreg |= r->regu; + + if(r->refbehind.b[z] & bb) + for(r1 = r->p2; r1 != R; r1 = r1->p2link) + if(r1->refahead.b[z] & bb) + vreg |= paint2(r1, bn); + + if(!(r->refahead.b[z] & bb)) + break; + r1 = r->s2; + if(r1 != R) + if(r1->refbehind.b[z] & bb) + vreg |= paint2(r1, bn); + r = r->s1; + if(r == R) + break; + if(!(r->act.b[z] & bb)) + break; + if(!(r->refbehind.b[z] & bb)) + break; + } + + bb = vreg; + for(; r; r=r->s1) { + x = r->regu & ~bb; + if(x) { + vreg |= reguse(r, x); + bb |= regset(r, x); + } + } + return vreg; +} + +void +paint3(Reg *r, int bn, long rb, int rn) +{ + Reg *r1; + Prog *p; + int z; + ulong bb; + + z = bn/32; + bb = 1L << (bn%32); + if(r->act.b[z] & bb) + return; + for(;;) { + if(!(r->refbehind.b[z] & bb)) + break; + r1 = r->p1; + if(r1 == R) + break; + if(!(r1->refahead.b[z] & bb)) + break; + if(r1->act.b[z] & bb) + break; + r = r1; + } + + if(LOAD(r) & ~(r->set.b[z] & ~(r->use1.b[z]|r->use2.b[z])) & bb) + addmove(r, bn, rn, 0); + for(;;) { + r->act.b[z] |= bb; + p = r->prog; + + if(r->use1.b[z] & bb) { + if(debug['R']) + print("%P", p); + addreg(&p->from, rn); + if(debug['R']) + print("\t.c%P\n", p); + } + if((r->use2.b[z]|r->set.b[z]) & bb) { + if(debug['R']) + print("%P", p); + addreg(&p->to, rn); + if(debug['R']) + print("\t.c%P\n", p); + } + + if(STORE(r) & r->regdiff.b[z] & bb) + addmove(r, bn, rn, 1); + r->regu |= rb; + + if(r->refbehind.b[z] & bb) + for(r1 = r->p2; r1 != R; r1 = r1->p2link) + if(r1->refahead.b[z] & bb) + paint3(r1, bn, rb, rn); + + if(!(r->refahead.b[z] & bb)) + break; + r1 = r->s2; + if(r1 != R) + if(r1->refbehind.b[z] & bb) + paint3(r1, bn, rb, rn); + r = r->s1; + if(r == R) + break; + if(r->act.b[z] & bb) + break; + if(!(r->refbehind.b[z] & bb)) + break; + } +} + +void +addreg(Adr *a, int rn) +{ + + a->sym = 0; + a->offset = 0; + a->type = rn; +} + +long +RtoB(int r) +{ + + if(r < D_AX || r > D_R15) + return 0; + return 1L << (r-D_AX); +} + +int +BtoR(long b) +{ + + b &= 0xffffL; + if(b == 0) + return 0; + return bitno(b) + D_AX; +} + +/* + * bit reg + * 16 X5 + * 17 X6 + * 18 X7 + */ +long +FtoB(int f) +{ + if(f < FREGMIN || f > FREGEXT) + return 0; + return 1L << (f - FREGMIN + 16); +} + +int +BtoF(long b) +{ + + b &= 0x70000L; + if(b == 0) + return 0; + return bitno(b) - 16 + FREGMIN; +} diff --git a/sys/src/cmd/6c/sgen.c b/sys/src/cmd/6c/sgen.c new file mode 100644 index 000000000..a7d751c0f --- /dev/null +++ b/sys/src/cmd/6c/sgen.c @@ -0,0 +1,435 @@ +#include "gc.h" + +void +noretval(int n) +{ + + if(n & 1) { + gins(ANOP, Z, Z); + p->to.type = REGRET; + } + if(n & 2) { + gins(ANOP, Z, Z); + p->to.type = FREGRET; + } +} + +/* welcome to commute */ +static void +commute(Node *n) +{ + Node *l, *r; + + l = n->left; + r = n->right; + if(r->complex > l->complex) { + n->left = r; + n->right = l; + } +} + +void +indexshift(Node *n) +{ + int g; + + if(!typechlpv[n->type->etype]) + return; + simplifyshift(n); + if(n->op == OASHL && n->right->op == OCONST){ + g = vconst(n->right); + if(g >= 0 && g <= 3) + n->addable = 7; + } +} + +/* + * calculate addressability as follows + * NAME ==> 10/11 name+value(SB/SP) + * REGISTER ==> 12 register + * CONST ==> 20 $value + * *(20) ==> 21 value + * &(10) ==> 13 $name+value(SB) + * &(11) ==> 1 $name+value(SP) + * (13) + (20) ==> 13 fold constants + * (1) + (20) ==> 1 fold constants + * *(13) ==> 10 back to name + * *(1) ==> 11 back to name + * + * (20) * (X) ==> 7 multiplier in indexing + * (X,7) + (13,1) ==> 8 adder in indexing (addresses) + * (8) ==> &9(OINDEX) index, almost addressable + * + * calculate complexity (number of registers) + */ +void +xcom(Node *n) +{ + Node *l, *r; + int g; + + if(n == Z) + return; + l = n->left; + r = n->right; + n->complex = 0; + n->addable = 0; + switch(n->op) { + case OCONST: + n->addable = 20; + break; + + case ONAME: + n->addable = 10; + if(n->class == CPARAM || n->class == CAUTO) + n->addable = 11; + break; + + case OREGISTER: + n->addable = 12; + break; + + case OINDREG: + n->addable = 12; + break; + + case OADDR: + xcom(l); + if(l->addable == 10) + n->addable = 13; + else + if(l->addable == 11) + n->addable = 1; + break; + + case OADD: + xcom(l); + xcom(r); + if(n->type->etype != TIND) + break; + + switch(r->addable) { + case 20: + switch(l->addable) { + case 1: + case 13: + commadd: + l->type = n->type; + *n = *l; + l = new(0, Z, Z); + *l = *(n->left); + l->xoffset += r->vconst; + n->left = l; + r = n->right; + goto brk; + } + break; + + case 1: + case 13: + case 10: + case 11: + /* l is the base, r is the index */ + if(l->addable != 20) + n->addable = 8; + break; + } + switch(l->addable) { + case 20: + switch(r->addable) { + case 13: + case 1: + r = n->left; + l = n->right; + n->left = l; + n->right = r; + goto commadd; + } + break; + + case 13: + case 1: + case 10: + case 11: + /* r is the base, l is the index */ + if(r->addable != 20) + n->addable = 8; + break; + } + if(n->addable == 8 && !side(n)) { + indx(n); + l = new1(OINDEX, idx.basetree, idx.regtree); + l->scale = idx.scale; + l->addable = 9; + l->complex = l->right->complex; + l->type = l->left->type; + n->op = OADDR; + n->left = l; + n->right = Z; + n->addable = 8; + break; + } + break; + + case OINDEX: + xcom(l); + xcom(r); + n->addable = 9; + break; + + case OIND: + xcom(l); + if(l->op == OADDR) { + l = l->left; + l->type = n->type; + *n = *l; + return; + } + switch(l->addable) { + case 20: + n->addable = 21; + break; + case 1: + n->addable = 11; + break; + case 13: + n->addable = 10; + break; + } + break; + + case OASHL: + xcom(l); + xcom(r); + indexshift(n); + break; + + case OMUL: + case OLMUL: + xcom(l); + xcom(r); + g = vlog(l); + if(g >= 0) { + n->left = r; + n->right = l; + l = r; + r = n->right; + } + g = vlog(r); + if(g >= 0) { + n->op = OASHL; + r->vconst = g; + r->type = types[TINT]; + indexshift(n); + break; + } + commute(n); + break; + + case OASLDIV: + xcom(l); + xcom(r); + g = vlog(r); + if(g >= 0) { + n->op = OASLSHR; + r->vconst = g; + r->type = types[TINT]; + } + break; + + case OLDIV: + xcom(l); + xcom(r); + g = vlog(r); + if(g >= 0) { + n->op = OLSHR; + r->vconst = g; + r->type = types[TINT]; + indexshift(n); + break; + } + break; + + case OASLMOD: + xcom(l); + xcom(r); + g = vlog(r); + if(g >= 0) { + n->op = OASAND; + r->vconst--; + } + break; + + case OLMOD: + xcom(l); + xcom(r); + g = vlog(r); + if(g >= 0) { + n->op = OAND; + r->vconst--; + } + break; + + case OASMUL: + case OASLMUL: + xcom(l); + xcom(r); + g = vlog(r); + if(g >= 0) { + n->op = OASASHL; + r->vconst = g; + } + break; + + case OLSHR: + case OASHR: + xcom(l); + xcom(r); + indexshift(n); + break; + + default: + if(l != Z) + xcom(l); + if(r != Z) + xcom(r); + break; + } +brk: + if(n->addable >= 10) + return; + if(l != Z) + n->complex = l->complex; + if(r != Z) { + if(r->complex == n->complex) + n->complex = r->complex+1; + else + if(r->complex > n->complex) + n->complex = r->complex; + } + if(n->complex == 0) + n->complex++; + + switch(n->op) { + + case OFUNC: + n->complex = FNX; + break; + + case OCAST: + if(l->type->etype == TUVLONG && typefd[n->type->etype]) + n->complex += 2; + break; + + case OLMOD: + case OMOD: + case OLMUL: + case OLDIV: + case OMUL: + case ODIV: + case OASLMUL: + case OASLDIV: + case OASLMOD: + case OASMUL: + case OASDIV: + case OASMOD: + if(r->complex >= l->complex) { + n->complex = l->complex + 3; + if(r->complex > n->complex) + n->complex = r->complex; + } else { + n->complex = r->complex + 3; + if(l->complex > n->complex) + n->complex = l->complex; + } + break; + + case OLSHR: + case OASHL: + case OASHR: + case OASLSHR: + case OASASHL: + case OASASHR: + if(r->complex >= l->complex) { + n->complex = l->complex + 2; + if(r->complex > n->complex) + n->complex = r->complex; + } else { + n->complex = r->complex + 2; + if(l->complex > n->complex) + n->complex = l->complex; + } + break; + + case OADD: + case OXOR: + case OAND: + case OOR: + /* + * immediate operators, make const on right + */ + if(l->op == OCONST) { + n->left = r; + n->right = l; + } + break; + + case OEQ: + case ONE: + case OLE: + case OLT: + case OGE: + case OGT: + case OHI: + case OHS: + case OLO: + case OLS: + /* + * compare operators, make const on left + */ + if(r->op == OCONST) { + n->left = r; + n->right = l; + n->op = invrel[relindex(n->op)]; + } + break; + } +} + +void +indx(Node *n) +{ + Node *l, *r; + + if(debug['x']) + prtree(n, "indx"); + + l = n->left; + r = n->right; + if(l->addable == 1 || l->addable == 13 || r->complex > l->complex) { + n->right = l; + n->left = r; + l = r; + r = n->right; + } + if(l->addable != 7) { + idx.regtree = l; + idx.scale = 1; + } else + if(l->right->addable == 20) { + idx.regtree = l->left; + idx.scale = 1 << l->right->vconst; + } else + if(l->left->addable == 20) { + idx.regtree = l->right; + idx.scale = 1 << l->left->vconst; + } else + diag(n, "bad index"); + + idx.basetree = r; + if(debug['x']) { + print("scale = %d\n", idx.scale); + prtree(idx.regtree, "index"); + prtree(idx.basetree, "base"); + } +} diff --git a/sys/src/cmd/6c/swt.c b/sys/src/cmd/6c/swt.c new file mode 100644 index 000000000..b201c9cae --- /dev/null +++ b/sys/src/cmd/6c/swt.c @@ -0,0 +1,530 @@ +#include "gc.h" + +void +swit1(C1 *q, int nc, long def, Node *n) +{ + C1 *r; + int i; + Prog *sp; + + if(nc < 5) { + for(i=0; i<nc; i++) { + if(debug['W']) + print("case = %.8llux\n", q->val); + gcmp(OEQ, n, q->val); + patch(p, q->label); + q++; + } + gbranch(OGOTO); + patch(p, def); + return; + } + i = nc / 2; + r = q+i; + if(debug['W']) + print("case > %.8llux\n", r->val); + gcmp(OGT, n, r->val); + sp = p; + gbranch(OGOTO); + p->as = AJEQ; + patch(p, r->label); + swit1(q, i, def, n); + + if(debug['W']) + print("case < %.8llux\n", r->val); + patch(sp, pc); + swit1(r+1, nc-i-1, def, n); +} + +void +bitload(Node *b, Node *n1, Node *n2, Node *n3, Node *nn) +{ + int sh; + long v; + Node *l; + + /* + * n1 gets adjusted/masked value + * n2 gets address of cell + * n3 gets contents of cell + */ + l = b->left; + if(n2 != Z) { + regalloc(n1, l, nn); + reglcgen(n2, l, Z); + regalloc(n3, l, Z); + gmove(n2, n3); + gmove(n3, n1); + } else { + regalloc(n1, l, nn); + cgen(l, n1); + } + if(b->type->shift == 0 && typeu[b->type->etype]) { + v = ~0 + (1L << b->type->nbits); + gopcode(OAND, tfield, nodconst(v), n1); + } else { + sh = 32 - b->type->shift - b->type->nbits; + if(sh > 0) + gopcode(OASHL, tfield, nodconst(sh), n1); + sh += b->type->shift; + if(sh > 0) + if(typeu[b->type->etype]) + gopcode(OLSHR, tfield, nodconst(sh), n1); + else + gopcode(OASHR, tfield, nodconst(sh), n1); + } +} + +void +bitstore(Node *b, Node *n1, Node *n2, Node *n3, Node *nn) +{ + long v; + Node nod; + int sh; + + regalloc(&nod, b->left, Z); + v = ~0 + (1L << b->type->nbits); + gopcode(OAND, types[TLONG], nodconst(v), n1); + gmove(n1, &nod); + if(nn != Z) + gmove(n1, nn); + sh = b->type->shift; + if(sh > 0) + gopcode(OASHL, types[TLONG], nodconst(sh), &nod); + v <<= sh; + gopcode(OAND, types[TLONG], nodconst(~v), n3); + gopcode(OOR, types[TLONG], n3, &nod); + gmove(&nod, n2); + + regfree(&nod); + regfree(n1); + regfree(n2); + regfree(n3); +} + +long +outstring(char *s, long n) +{ + long r; + + if(suppress) + return nstring; + r = nstring; + while(n) { + string[mnstring] = *s++; + mnstring++; + nstring++; + if(mnstring >= NSNAME) { + gpseudo(ADATA, symstring, nodconst(0L)); + p->from.offset += nstring - NSNAME; + p->from.scale = NSNAME; + p->to.type = D_SCONST; + memmove(p->to.sval, string, NSNAME); + mnstring = 0; + } + n--; + } + return r; +} + +void +sextern(Sym *s, Node *a, long o, long w) +{ + long e, lw; + + for(e=0; e<w; e+=NSNAME) { + lw = NSNAME; + if(w-e < lw) + lw = w-e; + gpseudo(ADATA, s, nodconst(0L)); + p->from.offset += o+e; + p->from.scale = lw; + p->to.type = D_SCONST; + memmove(p->to.sval, a->cstring+e, lw); + } +} + +void +gextern(Sym *s, Node *a, long o, long w) +{ + if(0 && a->op == OCONST && typev[a->type->etype]) { + gpseudo(ADATA, s, lo64(a)); + p->from.offset += o; + p->from.scale = 4; + gpseudo(ADATA, s, hi64(a)); + p->from.offset += o + 4; + p->from.scale = 4; + return; + } + gpseudo(ADATA, s, a); + p->from.offset += o; + p->from.scale = w; + switch(p->to.type) { + default: + p->to.index = p->to.type; + p->to.type = D_ADDR; + case D_CONST: + case D_FCONST: + case D_ADDR: + break; + } +} + +void zname(Biobuf*, Sym*, int); +void zaddr(Biobuf*, Adr*, int); +void outhist(Biobuf*); + +void +outcode(void) +{ + struct { Sym *sym; short type; } h[NSYM]; + Prog *p; + Sym *s; + int f, sf, st, t, sym; + Biobuf b; + + if(debug['S']) { + for(p = firstp; p != P; p = p->link) + if(p->as != ADATA && p->as != AGLOBL) + pc--; + for(p = firstp; p != P; p = p->link) { + print("%P\n", p); + if(p->as != ADATA && p->as != AGLOBL) + pc++; + } + } + f = open(outfile, OWRITE); + if(f < 0) { + diag(Z, "cannot open %s", outfile); + return; + } + Binit(&b, f, OWRITE); + Bseek(&b, 0L, 2); + outhist(&b); + for(sym=0; sym<NSYM; sym++) { + h[sym].sym = S; + h[sym].type = 0; + } + sym = 1; + for(p = firstp; p != P; p = p->link) { + jackpot: + sf = 0; + s = p->from.sym; + while(s != S) { + sf = s->sym; + if(sf < 0 || sf >= NSYM) + sf = 0; + t = p->from.type; + if(t == D_ADDR) + t = p->from.index; + if(h[sf].type == t) + if(h[sf].sym == s) + break; + s->sym = sym; + zname(&b, s, t); + h[sym].sym = s; + h[sym].type = t; + sf = sym; + sym++; + if(sym >= NSYM) + sym = 1; + break; + } + st = 0; + s = p->to.sym; + while(s != S) { + st = s->sym; + if(st < 0 || st >= NSYM) + st = 0; + t = p->to.type; + if(t == D_ADDR) + t = p->to.index; + if(h[st].type == t) + if(h[st].sym == s) + break; + s->sym = sym; + zname(&b, s, t); + h[sym].sym = s; + h[sym].type = t; + st = sym; + sym++; + if(sym >= NSYM) + sym = 1; + if(st == sf) + goto jackpot; + break; + } + Bputc(&b, p->as); + Bputc(&b, p->as>>8); + Bputc(&b, p->lineno); + Bputc(&b, p->lineno>>8); + Bputc(&b, p->lineno>>16); + Bputc(&b, p->lineno>>24); + zaddr(&b, &p->from, sf); + zaddr(&b, &p->to, st); + } + Bflush(&b); + close(f); + firstp = P; + lastp = P; +} + +void +outhist(Biobuf *b) +{ + Hist *h; + char *p, *q, *op, c; + Prog pg; + int n; + + pg = zprog; + pg.as = AHISTORY; + c = pathchar(); + for(h = hist; h != H; h = h->link) { + p = h->name; + op = 0; + /* on windows skip drive specifier in pathname */ + if(systemtype(Windows) && p && p[1] == ':'){ + p += 2; + c = *p; + } + if(p && p[0] != c && h->offset == 0 && pathname){ + /* on windows skip drive specifier in pathname */ + if(systemtype(Windows) && pathname[1] == ':') { + op = p; + p = pathname+2; + c = *p; + } else if(pathname[0] == c){ + op = p; + p = pathname; + } + } + while(p) { + q = utfrune(p, c); + if(q) { + n = q-p; + if(n == 0){ + n = 1; /* leading "/" */ + *p = '/'; /* don't emit "\" on windows */ + } + q++; + } else { + n = strlen(p); + q = 0; + } + if(n) { + Bputc(b, ANAME); + Bputc(b, ANAME>>8); + Bputc(b, D_FILE); + Bputc(b, 1); + Bputc(b, '<'); + Bwrite(b, p, n); + Bputc(b, 0); + } + p = q; + if(p == 0 && op) { + p = op; + op = 0; + } + } + pg.lineno = h->line; + pg.to.type = zprog.to.type; + pg.to.offset = h->offset; + if(h->offset) + pg.to.type = D_CONST; + + Bputc(b, pg.as); + Bputc(b, pg.as>>8); + Bputc(b, pg.lineno); + Bputc(b, pg.lineno>>8); + Bputc(b, pg.lineno>>16); + Bputc(b, pg.lineno>>24); + zaddr(b, &pg.from, 0); + zaddr(b, &pg.to, 0); + } +} + +void +zname(Biobuf *b, Sym *s, int t) +{ + char *n; + ulong sig; + + if(debug['T'] && t == D_EXTERN && s->sig != SIGDONE && s->type != types[TENUM] && s != symrathole){ + sig = sign(s); + Bputc(b, ASIGNAME); + Bputc(b, ASIGNAME>>8); + Bputc(b, sig); + Bputc(b, sig>>8); + Bputc(b, sig>>16); + Bputc(b, sig>>24); + s->sig = SIGDONE; + } + else{ + Bputc(b, ANAME); /* as */ + Bputc(b, ANAME>>8); /* as */ + } + Bputc(b, t); /* type */ + Bputc(b, s->sym); /* sym */ + n = s->name; + while(*n) { + Bputc(b, *n); + n++; + } + Bputc(b, 0); +} + +void +zaddr(Biobuf *b, Adr *a, int s) +{ + long l; + int i, t; + char *n; + Ieee e; + + t = 0; + if(a->index != D_NONE || a->scale != 0) + t |= T_INDEX; + if(s != 0) + t |= T_SYM; + + switch(a->type) { + default: + t |= T_TYPE; + case D_NONE: + if(a->offset != 0) { + t |= T_OFFSET; + l = a->offset; + if((vlong)l != a->offset) + t |= T_64; + } + break; + case D_FCONST: + t |= T_FCONST; + break; + case D_SCONST: + t |= T_SCONST; + break; + } + Bputc(b, t); + + if(t & T_INDEX) { /* implies index, scale */ + Bputc(b, a->index); + Bputc(b, a->scale); + } + if(t & T_OFFSET) { /* implies offset */ + l = a->offset; + Bputc(b, l); + Bputc(b, l>>8); + Bputc(b, l>>16); + Bputc(b, l>>24); + if(t & T_64) { + l = a->offset>>32; + Bputc(b, l); + Bputc(b, l>>8); + Bputc(b, l>>16); + Bputc(b, l>>24); + } + } + if(t & T_SYM) /* implies sym */ + Bputc(b, s); + if(t & T_FCONST) { + ieeedtod(&e, a->dval); + l = e.l; + Bputc(b, l); + Bputc(b, l>>8); + Bputc(b, l>>16); + Bputc(b, l>>24); + l = e.h; + Bputc(b, l); + Bputc(b, l>>8); + Bputc(b, l>>16); + Bputc(b, l>>24); + return; + } + if(t & T_SCONST) { + n = a->sval; + for(i=0; i<NSNAME; i++) { + Bputc(b, *n); + n++; + } + return; + } + if(t & T_TYPE) + Bputc(b, a->type); +} + +long +align(long i, Type *t, int op) +{ + long o; + Type *v; + int w; + + o = i; + w = 1; + switch(op) { + default: + diag(Z, "unknown align opcode %d", op); + break; + + case Asu2: /* padding at end of a struct */ + w = SZ_VLONG; + if(packflg) + w = packflg; + break; + + case Ael1: /* initial align of struct element */ + for(v=t; v->etype==TARRAY; v=v->link) + ; + w = ewidth[v->etype]; + if(w <= 0 || w >= SZ_VLONG) + w = SZ_VLONG; + if(packflg) + w = packflg; + break; + + case Ael2: /* width of a struct element */ + o += t->width; + break; + + case Aarg0: /* initial passbyptr argument in arg list */ + if(typesu[t->etype]) { + o = align(o, types[TIND], Aarg1); + o = align(o, types[TIND], Aarg2); + } + break; + + case Aarg1: /* initial align of parameter */ + w = ewidth[t->etype]; + if(w <= 0 || w >= SZ_VLONG) { + w = SZ_VLONG; + break; + } + w = 1; /* little endian no adjustment */ + break; + + case Aarg2: /* width of a parameter */ + o += t->width; + w = SZ_VLONG; + break; + + case Aaut3: /* total allign of automatic */ + o = align(o, t, Ael1); + o = align(o, t, Ael2); + break; + } + o = round(o, w); + if(debug['A']) + print("align %s %ld %T = %ld\n", bnames[op], i, t, o); + return o; +} + +long +maxround(long max, long v) +{ + v += SZ_VLONG-1; + if(v > max) + max = round(v, SZ_VLONG); + return max; +} diff --git a/sys/src/cmd/6c/sys.c b/sys/src/cmd/6c/sys.c new file mode 100644 index 000000000..abdc42b04 --- /dev/null +++ b/sys/src/cmd/6c/sys.c @@ -0,0 +1,106 @@ +#include <u.h> +#include <libc.h> +#include "/sys/src/libc/9syscall/sys.h" + +vlong _sysargs[6*4]; +vlong _callsys(void); + +/* + * syscalls + */ + +int +getpid(void) +{ + _sysargs[0] = -1; + return _callsys(); +} + +long +pread(int fd, void *a, long n, vlong) +{ + _sysargs[0] = PREAD; + _sysargs[1] = fd; + _sysargs[2] = (vlong)a; + _sysargs[3] = n; + return _callsys(); +} + +long +pwrite(int fd, void *a, long n, vlong) +{ + _sysargs[0] = PWRITE; + _sysargs[1] = fd; + _sysargs[2] = (vlong)a; + _sysargs[3] = n; + return _callsys(); +} + +int +close(int fd) +{ + _sysargs[0] = CLOSE; + _sysargs[1] = fd; + return _callsys(); +} + +int +open(char *name, int mode) +{ + _sysargs[0] = OPEN; + _sysargs[1] = (vlong)name; + _sysargs[2] = mode; + return _callsys(); +} + +int +create(char *f, int mode, ulong perm) +{ + _sysargs[0] = CREATE; + _sysargs[1] = (vlong)f; + _sysargs[2] = mode; + _sysargs[3] = perm; + return _callsys(); +} + +void +_exits(char *s) +{ + _sysargs[0] = EXITS; + _sysargs[1] = s!=nil? strlen(s): 0; + _callsys(); +} + +int +dup(int f, int t) +{ + _sysargs[0] = DUP; + _sysargs[1] = f; + _sysargs[2] = t; + return _callsys(); +} + +int +errstr(char *buf, uint n) +{ + _sysargs[0] = ERRSTR; + _sysargs[1] = (vlong)buf; + _sysargs[2] = n; + return _callsys(); +} + +int +brk_(void *a) +{ + _sysargs[0] = BRK_; + _sysargs[1] = (vlong)a; + return _callsys(); +} + +void* +sbrk(ulong n) +{ + _sysargs[0] = -2; + _sysargs[1] = n; + return (void*)_callsys(); +} diff --git a/sys/src/cmd/6c/txt.c b/sys/src/cmd/6c/txt.c new file mode 100644 index 000000000..24ac1db46 --- /dev/null +++ b/sys/src/cmd/6c/txt.c @@ -0,0 +1,1515 @@ +#include "gc.h" + +void +ginit(void) +{ + int i; + Type *t; + + thechar = '6'; + thestring = "amd64"; + exregoffset = REGEXT; + exfregoffset = FREGEXT; + listinit(); + nstring = 0; + mnstring = 0; + nrathole = 0; + pc = 0; + breakpc = -1; + continpc = -1; + cases = C; + firstp = P; + lastp = P; + tfield = types[TINT]; + + typeword = typechlvp; + typeswitch = typechlv; + typecmplx = typesu; + + /* TO DO */ + memmove(typechlpv, typechlp, sizeof(typechlpv)); + typechlpv[TVLONG] = 1; + typechlpv[TUVLONG] = 1; + + zprog.link = P; + zprog.as = AGOK; + zprog.from.type = D_NONE; + zprog.from.index = D_NONE; + zprog.from.scale = 0; + zprog.to = zprog.from; + + lregnode.op = OREGISTER; + lregnode.class = CEXREG; + lregnode.reg = REGTMP; + lregnode.complex = 0; + lregnode.addable = 11; + lregnode.type = types[TLONG]; + + qregnode = lregnode; + qregnode.type = types[TVLONG]; + + constnode.op = OCONST; + constnode.class = CXXX; + constnode.complex = 0; + constnode.addable = 20; + constnode.type = types[TLONG]; + + vconstnode = constnode; + vconstnode.type = types[TVLONG]; + + fconstnode.op = OCONST; + fconstnode.class = CXXX; + fconstnode.complex = 0; + fconstnode.addable = 20; + fconstnode.type = types[TDOUBLE]; + + nodsafe = new(ONAME, Z, Z); + nodsafe->sym = slookup(".safe"); + nodsafe->type = types[TINT]; + nodsafe->etype = types[TINT]->etype; + nodsafe->class = CAUTO; + complex(nodsafe); + + t = typ(TARRAY, types[TCHAR]); + symrathole = slookup(".rathole"); + symrathole->class = CGLOBL; + symrathole->type = t; + + nodrat = new(ONAME, Z, Z); + nodrat->sym = symrathole; + nodrat->type = types[TIND]; + nodrat->etype = TVOID; + nodrat->class = CGLOBL; + complex(nodrat); + nodrat->type = t; + + nodret = new(ONAME, Z, Z); + nodret->sym = slookup(".ret"); + nodret->type = types[TIND]; + nodret->etype = TIND; + nodret->class = CPARAM; + nodret = new(OIND, nodret, Z); + complex(nodret); + + if(0) + com64init(); + + for(i=0; i<nelem(reg); i++) { + reg[i] = 1; + if(i >= D_AX && i <= D_R15 && i != D_SP) + reg[i] = 0; + if(i >= D_X0 && i <= D_X7) + reg[i] = 0; + } +} + +void +gclean(void) +{ + int i; + Sym *s; + + reg[D_SP]--; + for(i=D_AX; i<=D_R15; i++) + if(reg[i]) + diag(Z, "reg %R left allocated", i); + for(i=D_X0; i<=D_X7; i++) + if(reg[i]) + diag(Z, "reg %R left allocated", i); + while(mnstring) + outstring("", 1L); + symstring->type->width = nstring; + symrathole->type->width = nrathole; + for(i=0; i<NHASH; i++) + for(s = hash[i]; s != S; s = s->link) { + if(s->type == T) + continue; + if(s->type->width == 0) + continue; + if(s->class != CGLOBL && s->class != CSTATIC) + continue; + if(s->type == types[TENUM]) + continue; + gpseudo(AGLOBL, s, nodconst(s->type->width)); + } + nextpc(); + p->as = AEND; + outcode(); +} + +void +nextpc(void) +{ + + p = alloc(sizeof(*p)); + *p = zprog; + p->lineno = nearln; + pc++; + if(firstp == P) { + firstp = p; + lastp = p; + return; + } + lastp->link = p; + lastp = p; +} + +void +gargs(Node *n, Node *tn1, Node *tn2) +{ + long regs; + Node fnxargs[20], *fnxp; + + regs = cursafe; + + fnxp = fnxargs; + garg1(n, tn1, tn2, 0, &fnxp); /* compile fns to temps */ + + curarg = 0; + fnxp = fnxargs; + garg1(n, tn1, tn2, 1, &fnxp); /* compile normal args and temps */ + + cursafe = regs; +} + +int +nareg(void) +{ + int i, n; + + n = 0; + for(i=D_AX; i<=D_R15; i++) + if(reg[i] == 0) + n++; + return n; +} + +void +garg1(Node *n, Node *tn1, Node *tn2, int f, Node **fnxp) +{ + Node nod; + + if(n == Z) + return; + if(n->op == OLIST) { + garg1(n->left, tn1, tn2, f, fnxp); + garg1(n->right, tn1, tn2, f, fnxp); + return; + } + if(f == 0) { + if(n->complex >= FNX) { + regsalloc(*fnxp, n); + nod = znode; + nod.op = OAS; + nod.left = *fnxp; + nod.right = n; + nod.type = n->type; + cgen(&nod, Z); + (*fnxp)++; + } + return; + } + if(typesu[n->type->etype]) { + regaalloc(tn2, n); + if(n->complex >= FNX) { + sugen(*fnxp, tn2, n->type->width); + (*fnxp)++; + } else + sugen(n, tn2, n->type->width); + return; + } + if(REGARG && curarg == 0 && typechlpv[n->type->etype]) { + regaalloc1(tn1, n); + if(n->complex >= FNX) { + cgen(*fnxp, tn1); + (*fnxp)++; + } else + cgen(n, tn1); + return; + } + if(vconst(n) == 0) { + regaalloc(tn2, n); + gmove(n, tn2); + return; + } + regalloc(tn1, n, Z); + if(n->complex >= FNX) { + cgen(*fnxp, tn1); + (*fnxp)++; + } else + cgen(n, tn1); + regaalloc(tn2, n); + gmove(tn1, tn2); + regfree(tn1); +} + +Node* +nodgconst(vlong v, Type *t) +{ + if(!typev[t->etype]) + return nodconst((long)v); + vconstnode.vconst = v; + return &vconstnode; +} + +Node* +nodconst(long v) +{ + constnode.vconst = v; + return &constnode; +} + +Node* +nodfconst(double d) +{ + fconstnode.fconst = d; + return &fconstnode; +} + +int +isreg(Node *n, int r) +{ + + if(n->op == OREGISTER) + if(n->reg == r) + return 1; + return 0; +} + +int +nodreg(Node *n, Node *nn, int r) +{ + int et; + + *n = qregnode; + n->reg = r; + if(nn != Z){ + et = nn->type->etype; + if(!typefd[et] && nn->type->width <= SZ_LONG && 0) + n->type = typeu[et]? types[TUINT]: types[TINT]; + else + n->type = nn->type; +//print("nodreg %s [%s]\n", tnames[et], tnames[n->type->etype]); + n->lineno = nn->lineno; + } + if(reg[r] == 0) + return 0; + if(nn != Z) { + if(nn->op == OREGISTER) + if(nn->reg == r) + return 0; + } + return 1; +} + +void +regret(Node *n, Node *nn) +{ + int r; + + r = REGRET; + if(typefd[nn->type->etype]) + r = FREGRET; + nodreg(n, nn, r); + reg[r]++; +} + +void +regalloc(Node *n, Node *tn, Node *o) +{ + int i; + + switch(tn->type->etype) { + case TCHAR: + case TUCHAR: + case TSHORT: + case TUSHORT: + case TINT: + case TUINT: + case TLONG: + case TULONG: + case TVLONG: + case TUVLONG: + case TIND: + if(o != Z && o->op == OREGISTER) { + i = o->reg; + if(i >= D_AX && i <= D_R15) + goto out; + } + for(i=D_AX; i<=D_R15; i++) + if(reg[i] == 0) + goto out; + diag(tn, "out of fixed registers"); + goto err; + + case TFLOAT: + case TDOUBLE: + if(o != Z && o->op == OREGISTER) { + i = o->reg; + if(i >= D_X0 && i <= D_X7) + goto out; + } + for(i=D_X0; i<=D_X7; i++) + if(reg[i] == 0) + goto out; + diag(tn, "out of float registers"); + goto out; + } + diag(tn, "unknown type in regalloc: %T", tn->type); +err: + i = 0; +out: + if(i) + reg[i]++; + nodreg(n, tn, i); +} + +void +regialloc(Node *n, Node *tn, Node *o) +{ + Node nod; + + nod = *tn; + nod.type = types[TIND]; + regalloc(n, &nod, o); +} + +void +regfree(Node *n) +{ + int i; + + i = 0; + if(n->op != OREGISTER && n->op != OINDREG) + goto err; + i = n->reg; + if(i < 0 || i >= sizeof(reg)) + goto err; + if(reg[i] <= 0) + goto err; + reg[i]--; + return; +err: + diag(n, "error in regfree: %R", i); +} + +void +regsalloc(Node *n, Node *nn) +{ + cursafe = align(cursafe, nn->type, Aaut3); + maxargsafe = maxround(maxargsafe, cursafe+curarg); + *n = *nodsafe; + n->xoffset = -(stkoff + cursafe); + n->type = nn->type; + n->etype = nn->type->etype; + n->lineno = nn->lineno; +} + +void +regaalloc1(Node *n, Node *nn) +{ + nodreg(n, nn, REGARG); + reg[REGARG]++; + curarg = align(curarg, nn->type, Aarg1); + curarg = align(curarg, nn->type, Aarg2); + maxargsafe = maxround(maxargsafe, cursafe+curarg); +} + +void +regaalloc(Node *n, Node *nn) +{ + curarg = align(curarg, nn->type, Aarg1); + *n = *nn; + n->op = OINDREG; + n->reg = REGSP; + n->xoffset = curarg; + n->complex = 0; + n->addable = 20; + curarg = align(curarg, nn->type, Aarg2); + maxargsafe = maxround(maxargsafe, cursafe+curarg); +} + +void +regind(Node *n, Node *nn) +{ + + if(n->op != OREGISTER) { + diag(n, "regind not OREGISTER"); + return; + } + n->op = OINDREG; + n->type = nn->type; +} + +void +naddr(Node *n, Adr *a) +{ + long v; + + a->type = D_NONE; + if(n == Z) + return; + switch(n->op) { + default: + bad: + diag(n, "bad in naddr: %O %D", n->op, a); + break; + + case OREGISTER: + a->type = n->reg; + a->sym = S; + break; + + + case OIND: + naddr(n->left, a); + if(a->type >= D_AX && a->type <= D_R15) + a->type += D_INDIR; + else + if(a->type == D_CONST) + a->type = D_NONE+D_INDIR; + else + if(a->type == D_ADDR) { + a->type = a->index; + a->index = D_NONE; + } else + goto bad; + break; + + case OINDEX: + a->type = idx.ptr; + if(n->left->op == OADDR || n->left->op == OCONST) + naddr(n->left, a); + if(a->type >= D_AX && a->type <= D_R15) + a->type += D_INDIR; + else + if(a->type == D_CONST) + a->type = D_NONE+D_INDIR; + else + if(a->type == D_ADDR) { + a->type = a->index; + a->index = D_NONE; + } else + goto bad; + a->index = idx.reg; + a->scale = n->scale; + a->offset += n->xoffset; + break; + + case OINDREG: + a->type = n->reg+D_INDIR; + a->sym = S; + a->offset = n->xoffset; + break; + + case ONAME: + a->etype = n->etype; + a->type = D_STATIC; + a->sym = n->sym; + a->offset = n->xoffset; + if(n->class == CSTATIC) + break; + if(n->class == CEXTERN || n->class == CGLOBL) { + a->type = D_EXTERN; + break; + } + if(n->class == CAUTO) { + a->type = D_AUTO; + break; + } + if(n->class == CPARAM) { + a->type = D_PARAM; + break; + } + goto bad; + + case OCONST: + if(typefd[n->type->etype]) { + a->type = D_FCONST; + a->dval = n->fconst; + break; + } + a->sym = S; + a->type = D_CONST; + if(typev[n->type->etype] || n->type->etype == TIND) + a->offset = n->vconst; + else + a->offset = convvtox(n->vconst, typeu[n->type->etype]? TULONG: TLONG); + break; + + case OADDR: + naddr(n->left, a); + if(a->type >= D_INDIR) { + a->type -= D_INDIR; + break; + } + if(a->type == D_EXTERN || a->type == D_STATIC || + a->type == D_AUTO || a->type == D_PARAM) + if(a->index == D_NONE) { + a->index = a->type; + a->type = D_ADDR; + break; + } + goto bad; + + case OADD: + if(n->right->op == OCONST) { + v = n->right->vconst; + naddr(n->left, a); + } else + if(n->left->op == OCONST) { + v = n->left->vconst; + naddr(n->right, a); + } else + goto bad; + a->offset += v; + break; + + } +} + +void +gcmp(int op, Node *n, vlong val) +{ + Node *cn, nod; + + cn = nodgconst(val, n->type); + if(!immconst(cn)){ + regalloc(&nod, n, Z); + gmove(cn, &nod); + gopcode(op, n->type, n, &nod); + regfree(&nod); + }else + gopcode(op, n->type, n, cn); +} + +#define CASE(a,b) ((a<<8)|(b<<0)) + +void +gmove(Node *f, Node *t) +{ + int ft, tt, t64, a; + Node nod, nod1, nod2, nod3; + Prog *p1, *p2; + + ft = f->type->etype; + tt = t->type->etype; + t64 = tt == TVLONG || tt == TUVLONG || tt == TIND; + if(debug['M']) + print("gop: %O %O[%s],%O[%s]\n", OAS, + f->op, tnames[ft], t->op, tnames[tt]); + if(typefd[ft] && f->op == OCONST) { + /* TO DO: pick up special constants, possibly preloaded */ + if(f->fconst == 0.0){ + regalloc(&nod, t, t); + gins(AXORPD, &nod, &nod); + gmove(&nod, t); + regfree(&nod); + return; + } + } +/* + * load + */ + if(f->op == ONAME || f->op == OINDREG || + f->op == OIND || f->op == OINDEX) + switch(ft) { + case TCHAR: + a = AMOVBLSX; + if(t64) + a = AMOVBQSX; + goto ld; + case TUCHAR: + a = AMOVBLZX; + if(t64) + a = AMOVBQZX; + goto ld; + case TSHORT: + a = AMOVWLSX; + if(t64) + a = AMOVWQSX; + goto ld; + case TUSHORT: + a = AMOVWLZX; + if(t64) + a = AMOVWQZX; + goto ld; + case TINT: + case TLONG: + if(typefd[tt]) { + regalloc(&nod, t, t); + if(tt == TDOUBLE) + a = ACVTSL2SD; + else + a = ACVTSL2SS; + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + } + a = AMOVL; + if(t64) + a = AMOVLQSX; + goto ld; + case TUINT: + case TULONG: + a = AMOVL; + if(t64) + a = AMOVLQZX; /* could probably use plain MOVL */ + goto ld; + case TVLONG: + if(typefd[tt]) { + regalloc(&nod, t, t); + if(tt == TDOUBLE) + a = ACVTSQ2SD; + else + a = ACVTSQ2SS; + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + } + case TUVLONG: + a = AMOVQ; + goto ld; + case TIND: + a = AMOVQ; + + ld: + regalloc(&nod, f, t); + nod.type = t64? types[TVLONG]: types[TINT]; + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + + case TFLOAT: + a = AMOVSS; + goto fld; + case TDOUBLE: + a = AMOVSD; + fld: + regalloc(&nod, f, t); + if(tt != TDOUBLE && tt != TFLOAT){ /* TO DO: why is this here */ + prtree(f, "odd tree"); + nod.type = t64? types[TVLONG]: types[TINT]; + } + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + } + +/* + * store + */ + if(t->op == ONAME || t->op == OINDREG || + t->op == OIND || t->op == OINDEX) + switch(tt) { + case TCHAR: + case TUCHAR: + a = AMOVB; goto st; + case TSHORT: + case TUSHORT: + a = AMOVW; goto st; + case TINT: + case TUINT: + case TLONG: + case TULONG: + a = AMOVL; goto st; + case TVLONG: + case TUVLONG: + case TIND: + a = AMOVQ; goto st; + + st: + if(f->op == OCONST) { + gins(a, f, t); + return; + } + fst: + regalloc(&nod, t, f); + gmove(f, &nod); + gins(a, &nod, t); + regfree(&nod); + return; + + case TFLOAT: + a = AMOVSS; + goto fst; + case TDOUBLE: + a = AMOVSD; + goto fst; + } + +/* + * convert + */ + switch(CASE(ft,tt)) { + default: +/* + * integer to integer + ******** + a = AGOK; break; + + case CASE( TCHAR, TCHAR): + case CASE( TUCHAR, TCHAR): + case CASE( TSHORT, TCHAR): + case CASE( TUSHORT,TCHAR): + case CASE( TINT, TCHAR): + case CASE( TUINT, TCHAR): + case CASE( TLONG, TCHAR): + case CASE( TULONG, TCHAR): + case CASE( TIND, TCHAR): + + case CASE( TCHAR, TUCHAR): + case CASE( TUCHAR, TUCHAR): + case CASE( TSHORT, TUCHAR): + case CASE( TUSHORT,TUCHAR): + case CASE( TINT, TUCHAR): + case CASE( TUINT, TUCHAR): + case CASE( TLONG, TUCHAR): + case CASE( TULONG, TUCHAR): + case CASE( TIND, TUCHAR): + + case CASE( TSHORT, TSHORT): + case CASE( TUSHORT,TSHORT): + case CASE( TINT, TSHORT): + case CASE( TUINT, TSHORT): + case CASE( TLONG, TSHORT): + case CASE( TULONG, TSHORT): + case CASE( TIND, TSHORT): + + case CASE( TSHORT, TUSHORT): + case CASE( TUSHORT,TUSHORT): + case CASE( TINT, TUSHORT): + case CASE( TUINT, TUSHORT): + case CASE( TLONG, TUSHORT): + case CASE( TULONG, TUSHORT): + case CASE( TIND, TUSHORT): + + case CASE( TINT, TINT): + case CASE( TUINT, TINT): + case CASE( TLONG, TINT): + case CASE( TULONG, TINT): + case CASE( TIND, TINT): + + case CASE( TINT, TUINT): + case CASE( TUINT, TUINT): + case CASE( TLONG, TUINT): + case CASE( TULONG, TUINT): + case CASE( TIND, TUINT): + + case CASE( TUINT, TIND): + case CASE( TVLONG, TUINT): + case CASE( TVLONG, TULONG): + case CASE( TUVLONG, TUINT): + case CASE( TUVLONG, TULONG): + *****/ + a = AMOVL; + break; + + case CASE( TVLONG, TCHAR): + case CASE( TVLONG, TSHORT): + case CASE( TVLONG, TINT): + case CASE( TVLONG, TLONG): + case CASE( TUVLONG, TCHAR): + case CASE( TUVLONG, TSHORT): + case CASE( TUVLONG, TINT): + case CASE( TUVLONG, TLONG): + case CASE( TINT, TVLONG): + case CASE( TINT, TUVLONG): + case CASE( TLONG, TVLONG): + case CASE( TINT, TIND): + case CASE( TLONG, TIND): + a = AMOVLQSX; + if(f->op == OCONST) { + f->vconst &= (uvlong)0xffffffffU; + if(f->vconst & 0x80000000) + f->vconst |= (vlong)0xffffffff << 32; + a = AMOVQ; + } + break; + + case CASE( TUINT, TIND): + case CASE( TUINT, TVLONG): + case CASE( TUINT, TUVLONG): + case CASE( TULONG, TVLONG): + case CASE( TULONG, TUVLONG): + case CASE( TULONG, TIND): + a = AMOVL; /* same effect as AMOVLQZX */ + if(f->op == OCONST) { + f->vconst &= (uvlong)0xffffffffU; + a = AMOVQ; + } + break; + + case CASE( TIND, TVLONG): + case CASE( TVLONG, TVLONG): + case CASE( TUVLONG, TVLONG): + case CASE( TVLONG, TUVLONG): + case CASE( TUVLONG, TUVLONG): + case CASE( TIND, TUVLONG): + case CASE( TVLONG, TIND): + case CASE( TUVLONG, TIND): + case CASE( TIND, TIND): + a = AMOVQ; + break; + + case CASE( TSHORT, TINT): + case CASE( TSHORT, TUINT): + case CASE( TSHORT, TLONG): + case CASE( TSHORT, TULONG): + a = AMOVWLSX; + if(f->op == OCONST) { + f->vconst &= 0xffff; + if(f->vconst & 0x8000) + f->vconst |= 0xffff0000; + a = AMOVL; + } + break; + + case CASE( TSHORT, TVLONG): + case CASE( TSHORT, TUVLONG): + case CASE( TSHORT, TIND): + a = AMOVWQSX; + if(f->op == OCONST) { + f->vconst &= 0xffff; + if(f->vconst & 0x8000){ + f->vconst |= 0xffff0000; + f->vconst |= (vlong)~0 << 32; + } + a = AMOVL; + } + break; + + case CASE( TUSHORT,TINT): + case CASE( TUSHORT,TUINT): + case CASE( TUSHORT,TLONG): + case CASE( TUSHORT,TULONG): + a = AMOVWLZX; + if(f->op == OCONST) { + f->vconst &= 0xffff; + a = AMOVL; + } + break; + + case CASE( TUSHORT,TVLONG): + case CASE( TUSHORT,TUVLONG): + case CASE( TUSHORT,TIND): + a = AMOVWQZX; + if(f->op == OCONST) { + f->vconst &= 0xffff; + a = AMOVL; /* MOVL also zero-extends to 64 bits */ + } + break; + + case CASE( TCHAR, TSHORT): + case CASE( TCHAR, TUSHORT): + case CASE( TCHAR, TINT): + case CASE( TCHAR, TUINT): + case CASE( TCHAR, TLONG): + case CASE( TCHAR, TULONG): + a = AMOVBLSX; + if(f->op == OCONST) { + f->vconst &= 0xff; + if(f->vconst & 0x80) + f->vconst |= 0xffffff00; + a = AMOVL; + } + break; + + case CASE( TCHAR, TVLONG): + case CASE( TCHAR, TUVLONG): + case CASE( TCHAR, TIND): + a = AMOVBQSX; + if(f->op == OCONST) { + f->vconst &= 0xff; + if(f->vconst & 0x80){ + f->vconst |= 0xffffff00; + f->vconst |= (vlong)~0 << 32; + } + a = AMOVQ; + } + break; + + case CASE( TUCHAR, TSHORT): + case CASE( TUCHAR, TUSHORT): + case CASE( TUCHAR, TINT): + case CASE( TUCHAR, TUINT): + case CASE( TUCHAR, TLONG): + case CASE( TUCHAR, TULONG): + a = AMOVBLZX; + if(f->op == OCONST) { + f->vconst &= 0xff; + a = AMOVL; + } + break; + + case CASE( TUCHAR, TVLONG): + case CASE( TUCHAR, TUVLONG): + case CASE( TUCHAR, TIND): + a = AMOVBQZX; + if(f->op == OCONST) { + f->vconst &= 0xff; + a = AMOVL; /* zero-extends to 64-bits */ + } + break; + +/* + * float to fix + */ + case CASE( TFLOAT, TCHAR): + case CASE( TFLOAT, TUCHAR): + case CASE( TFLOAT, TSHORT): + case CASE( TFLOAT, TUSHORT): + case CASE( TFLOAT, TINT): + case CASE( TFLOAT, TUINT): + case CASE( TFLOAT, TLONG): + case CASE( TFLOAT, TULONG): + case CASE( TFLOAT, TVLONG): + case CASE( TFLOAT, TUVLONG): + case CASE( TFLOAT, TIND): + + case CASE( TDOUBLE,TCHAR): + case CASE( TDOUBLE,TUCHAR): + case CASE( TDOUBLE,TSHORT): + case CASE( TDOUBLE,TUSHORT): + case CASE( TDOUBLE,TINT): + case CASE( TDOUBLE,TUINT): + case CASE( TDOUBLE,TLONG): + case CASE( TDOUBLE,TULONG): + case CASE( TDOUBLE,TVLONG): + case CASE( TDOUBLE,TUVLONG): + case CASE( TDOUBLE,TIND): + regalloc(&nod, t, Z); + if(ewidth[tt] == SZ_VLONG || typeu[tt] && ewidth[tt] == SZ_INT){ + if(ft == TFLOAT) + a = ACVTTSS2SQ; + else + a = ACVTTSD2SQ; + }else{ + if(ft == TFLOAT) + a = ACVTTSS2SL; + else + a = ACVTTSD2SL; + } + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + +/* + * ulong to float + */ + case CASE( TUVLONG, TDOUBLE): + case CASE( TUVLONG, TFLOAT): + a = ACVTSQ2SS; + if(tt == TDOUBLE) + a = ACVTSQ2SD; + regalloc(&nod, f, f); + gmove(f, &nod); + regalloc(&nod1, t, t); + gins(ACMPQ, &nod, nodconst(0)); + gins(AJLT, Z, Z); + p1 = p; + gins(a, &nod, &nod1); + gins(AJMP, Z, Z); + p2 = p; + patch(p1, pc); + regalloc(&nod2, f, Z); + regalloc(&nod3, f, Z); + gmove(&nod, &nod2); + gins(ASHRQ, nodconst(1), &nod2); + gmove(&nod, &nod3); + gins(AANDL, nodconst(1), &nod3); + gins(AORQ, &nod3, &nod2); + gins(a, &nod2, &nod1); + gins(tt == TDOUBLE? AADDSD: AADDSS, &nod1, &nod1); + regfree(&nod2); + regfree(&nod3); + patch(p2, pc); + regfree(&nod); + regfree(&nod1); + return; + + case CASE( TULONG, TDOUBLE): + case CASE( TUINT, TDOUBLE): + case CASE( TULONG, TFLOAT): + case CASE( TUINT, TFLOAT): + a = ACVTSQ2SS; + if(tt == TDOUBLE) + a = ACVTSQ2SD; + regalloc(&nod, f, f); + gins(AMOVLQZX, f, &nod); + regalloc(&nod1, t, t); + gins(a, &nod, &nod1); + gmove(&nod1, t); + regfree(&nod); + regfree(&nod1); + return; + +/* + * fix to float + */ + case CASE( TCHAR, TFLOAT): + case CASE( TUCHAR, TFLOAT): + case CASE( TSHORT, TFLOAT): + case CASE( TUSHORT,TFLOAT): + case CASE( TINT, TFLOAT): + case CASE( TLONG, TFLOAT): + case CASE( TVLONG, TFLOAT): + case CASE( TIND, TFLOAT): + + case CASE( TCHAR, TDOUBLE): + case CASE( TUCHAR, TDOUBLE): + case CASE( TSHORT, TDOUBLE): + case CASE( TUSHORT,TDOUBLE): + case CASE( TINT, TDOUBLE): + case CASE( TLONG, TDOUBLE): + case CASE( TVLONG, TDOUBLE): + case CASE( TIND, TDOUBLE): + regalloc(&nod, t, t); + if(ewidth[ft] == SZ_VLONG){ + if(tt == TFLOAT) + a = ACVTSQ2SS; + else + a = ACVTSQ2SD; + }else{ + if(tt == TFLOAT) + a = ACVTSL2SS; + else + a = ACVTSL2SD; + } + gins(a, f, &nod); + gmove(&nod, t); + regfree(&nod); + return; + +/* + * float to float + */ + case CASE( TFLOAT, TFLOAT): + a = AMOVSS; + break; + case CASE( TDOUBLE,TFLOAT): + a = ACVTSD2SS; + break; + case CASE( TFLOAT, TDOUBLE): + a = ACVTSS2SD; + break; + case CASE( TDOUBLE,TDOUBLE): + a = AMOVSD; + break; + } + if(a == AMOVQ || a == AMOVSD || a == AMOVSS || a == AMOVL && ewidth[ft] == ewidth[tt]) /* TO DO: check AMOVL */ + if(samaddr(f, t)) + return; + gins(a, f, t); +} + +void +doindex(Node *n) +{ + Node nod, nod1; + long v; + +if(debug['Y']) +prtree(n, "index"); + +if(n->left->complex >= FNX) +print("botch in doindex\n"); + + regalloc(&nod, &qregnode, Z); + v = constnode.vconst; + cgen(n->right, &nod); + idx.ptr = D_NONE; + if(n->left->op == OCONST) + idx.ptr = D_CONST; + else if(n->left->op == OREGISTER) + idx.ptr = n->left->reg; + else if(n->left->op != OADDR) { + reg[D_BP]++; // cant be used as a base + regalloc(&nod1, &qregnode, Z); + cgen(n->left, &nod1); + idx.ptr = nod1.reg; + regfree(&nod1); + reg[D_BP]--; + } + idx.reg = nod.reg; + regfree(&nod); + constnode.vconst = v; +} + +void +gins(int a, Node *f, Node *t) +{ + + if(f != Z && f->op == OINDEX) + doindex(f); + if(t != Z && t->op == OINDEX) + doindex(t); + nextpc(); + p->as = a; + if(f != Z) + naddr(f, &p->from); + if(t != Z) + naddr(t, &p->to); + if(debug['g']) + print("%P\n", p); +} + +void +gopcode(int o, Type *ty, Node *f, Node *t) +{ + int a, et; + + et = TLONG; + if(ty != T) + et = ty->etype; + if(debug['M']) { + if(f != Z && f->type != T) + print("gop: %O %O[%s],", o, f->op, tnames[et]); + else + print("gop: %O Z,", o); + if(t != Z && t->type != T) + print("%O[%s]\n", t->op, tnames[t->type->etype]); + else + print("Z\n"); + } + a = AGOK; + switch(o) { + case OCOM: + a = ANOTL; + if(et == TCHAR || et == TUCHAR) + a = ANOTB; + if(et == TSHORT || et == TUSHORT) + a = ANOTW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ANOTQ; + break; + + case ONEG: + a = ANEGL; + if(et == TCHAR || et == TUCHAR) + a = ANEGB; + if(et == TSHORT || et == TUSHORT) + a = ANEGW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ANEGQ; + break; + + case OADDR: + a = ALEAQ; + break; + + case OASADD: + case OADD: + a = AADDL; + if(et == TCHAR || et == TUCHAR) + a = AADDB; + if(et == TSHORT || et == TUSHORT) + a = AADDW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AADDQ; + if(et == TFLOAT) + a = AADDSS; + if(et == TDOUBLE) + a = AADDSD; + break; + + case OASSUB: + case OSUB: + a = ASUBL; + if(et == TCHAR || et == TUCHAR) + a = ASUBB; + if(et == TSHORT || et == TUSHORT) + a = ASUBW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ASUBQ; + if(et == TFLOAT) + a = ASUBSS; + if(et == TDOUBLE) + a = ASUBSD; + break; + + case OASOR: + case OOR: + a = AORL; + if(et == TCHAR || et == TUCHAR) + a = AORB; + if(et == TSHORT || et == TUSHORT) + a = AORW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AORQ; + break; + + case OASAND: + case OAND: + a = AANDL; + if(et == TCHAR || et == TUCHAR) + a = AANDB; + if(et == TSHORT || et == TUSHORT) + a = AANDW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AANDQ; + break; + + case OASXOR: + case OXOR: + a = AXORL; + if(et == TCHAR || et == TUCHAR) + a = AXORB; + if(et == TSHORT || et == TUSHORT) + a = AXORW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AXORQ; + break; + + case OASLSHR: + case OLSHR: + a = ASHRL; + if(et == TCHAR || et == TUCHAR) + a = ASHRB; + if(et == TSHORT || et == TUSHORT) + a = ASHRW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ASHRQ; + break; + + case OASASHR: + case OASHR: + a = ASARL; + if(et == TCHAR || et == TUCHAR) + a = ASARB; + if(et == TSHORT || et == TUSHORT) + a = ASARW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ASARQ; + break; + + case OASASHL: + case OASHL: + a = ASALL; + if(et == TCHAR || et == TUCHAR) + a = ASALB; + if(et == TSHORT || et == TUSHORT) + a = ASALW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ASALQ; + break; + + case OFUNC: + a = ACALL; + break; + + case OASMUL: + case OMUL: + if(f->op == OREGISTER && t != Z && isreg(t, D_AX) && reg[D_DX] == 0) + t = Z; + a = AIMULL; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AIMULQ; + if(et == TFLOAT) + a = AMULSS; + if(et == TDOUBLE) + a = AMULSD; + break; + + case OASMOD: + case OMOD: + case OASDIV: + case ODIV: + a = AIDIVL; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AIDIVQ; + if(et == TFLOAT) + a = ADIVSS; + if(et == TDOUBLE) + a = ADIVSD; + break; + + case OASLMUL: + case OLMUL: + a = AMULL; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = AMULQ; + break; + + case OASLMOD: + case OLMOD: + case OASLDIV: + case OLDIV: + a = ADIVL; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ADIVQ; + break; + + case OEQ: + case ONE: + case OLT: + case OLE: + case OGE: + case OGT: + case OLO: + case OLS: + case OHS: + case OHI: + a = ACMPL; + if(et == TCHAR || et == TUCHAR) + a = ACMPB; + if(et == TSHORT || et == TUSHORT) + a = ACMPW; + if(et == TVLONG || et == TUVLONG || et == TIND) + a = ACMPQ; + if(et == TFLOAT) + a = AUCOMISS; + if(et == TDOUBLE) + a = AUCOMISD; + gins(a, f, t); + switch(o) { + case OEQ: a = AJEQ; break; + case ONE: a = AJNE; break; + case OLT: a = AJLT; break; + case OLE: a = AJLE; break; + case OGE: a = AJGE; break; + case OGT: a = AJGT; break; + case OLO: a = AJCS; break; + case OLS: a = AJLS; break; + case OHS: a = AJCC; break; + case OHI: a = AJHI; break; + } + gins(a, Z, Z); + return; + } + if(a == AGOK) + diag(Z, "bad in gopcode %O", o); + gins(a, f, t); +} + +int +samaddr(Node *f, Node *t) +{ + return f->op == OREGISTER && t->op == OREGISTER && f->reg == t->reg; +} + +void +gbranch(int o) +{ + int a; + + a = AGOK; + switch(o) { + case ORETURN: + a = ARET; + break; + case OGOTO: + a = AJMP; + break; + } + nextpc(); + if(a == AGOK) { + diag(Z, "bad in gbranch %O", o); + nextpc(); + } + p->as = a; +} + +void +patch(Prog *op, long pc) +{ + + op->to.offset = pc; + op->to.type = D_BRANCH; +} + +void +gpseudo(int a, Sym *s, Node *n) +{ + + nextpc(); + p->as = a; + p->from.type = D_EXTERN; + p->from.sym = s; + p->from.scale = (profileflg ? 0 : NOPROF); + if(s->class == CSTATIC) + p->from.type = D_STATIC; + naddr(n, &p->to); + if(a == ADATA || a == AGLOBL) + pc--; +} + +int +sconst(Node *n) +{ + long v; + + if(n->op == OCONST && !typefd[n->type->etype]) { + v = n->vconst; + if(v >= -32766L && v < 32766L) + return 1; + } + return 0; +} + +long +exreg(Type *t) +{ + long o; + + if(typechlpv[t->etype]) { + if(exregoffset <= REGEXT-4) + return 0; + o = exregoffset; + exregoffset--; + return o; + } + return 0; +} + +schar ewidth[NTYPE] = +{ + -1, /*[TXXX]*/ + SZ_CHAR, /*[TCHAR]*/ + SZ_CHAR, /*[TUCHAR]*/ + SZ_SHORT, /*[TSHORT]*/ + SZ_SHORT, /*[TUSHORT]*/ + SZ_INT, /*[TINT]*/ + SZ_INT, /*[TUINT]*/ + SZ_LONG, /*[TLONG]*/ + SZ_LONG, /*[TULONG]*/ + SZ_VLONG, /*[TVLONG]*/ + SZ_VLONG, /*[TUVLONG]*/ + SZ_FLOAT, /*[TFLOAT]*/ + SZ_DOUBLE, /*[TDOUBLE]*/ + SZ_IND, /*[TIND]*/ + 0, /*[TFUNC]*/ + -1, /*[TARRAY]*/ + 0, /*[TVOID]*/ + -1, /*[TSTRUCT]*/ + -1, /*[TUNION]*/ + SZ_INT, /*[TENUM]*/ +}; +long ncast[NTYPE] = +{ + 0, /*[TXXX]*/ + BCHAR|BUCHAR, /*[TCHAR]*/ + BCHAR|BUCHAR, /*[TUCHAR]*/ + BSHORT|BUSHORT, /*[TSHORT]*/ + BSHORT|BUSHORT, /*[TUSHORT]*/ + BINT|BUINT|BLONG|BULONG, /*[TINT]*/ + BINT|BUINT|BLONG|BULONG, /*[TUINT]*/ + BINT|BUINT|BLONG|BULONG, /*[TLONG]*/ + BINT|BUINT|BLONG|BULONG, /*[TULONG]*/ + BVLONG|BUVLONG|BIND, /*[TVLONG]*/ + BVLONG|BUVLONG|BIND, /*[TUVLONG]*/ + BFLOAT, /*[TFLOAT]*/ + BDOUBLE, /*[TDOUBLE]*/ + BVLONG|BUVLONG|BIND, /*[TIND]*/ + 0, /*[TFUNC]*/ + 0, /*[TARRAY]*/ + 0, /*[TVOID]*/ + BSTRUCT, /*[TSTRUCT]*/ + BUNION, /*[TUNION]*/ + 0, /*[TENUM]*/ +}; diff --git a/sys/src/cmd/6c/vlrt.c b/sys/src/cmd/6c/vlrt.c new file mode 100644 index 000000000..7c9f1a07b --- /dev/null +++ b/sys/src/cmd/6c/vlrt.c @@ -0,0 +1,746 @@ +typedef unsigned long ulong; +typedef unsigned int uint; +typedef unsigned short ushort; +typedef unsigned char uchar; +typedef signed char schar; + +#define SIGN(n) (1UL<<(n-1)) + +typedef struct Vlong Vlong; +struct Vlong +{ + union + { + struct + { + uint lo; + uint hi; + }; + struct + { + ushort lols; + ushort loms; + ushort hils; + ushort hims; + }; + }; +}; + +void abort(void); + +void _subv(Vlong*, Vlong, Vlong); + +void +_d2v(Vlong *y, double d) +{ + union { double d; struct Vlong; } x; + uint xhi, xlo, ylo, yhi; + int sh; + + x.d = d; + + xhi = (x.hi & 0xfffff) | 0x100000; + xlo = x.lo; + sh = 1075 - ((x.hi >> 20) & 0x7ff); + + ylo = 0; + yhi = 0; + if(sh >= 0) { + /* v = (hi||lo) >> sh */ + if(sh < 32) { + if(sh == 0) { + ylo = xlo; + yhi = xhi; + } else { + ylo = (xlo >> sh) | (xhi << (32-sh)); + yhi = xhi >> sh; + } + } else { + if(sh == 32) { + ylo = xhi; + } else + if(sh < 64) { + ylo = xhi >> (sh-32); + } + } + } else { + /* v = (hi||lo) << -sh */ + sh = -sh; + if(sh <= 10) { + ylo = xlo << sh; + yhi = (xhi << sh) | (xlo >> (32-sh)); + } else { + /* overflow */ + yhi = d; /* causes something awful */ + } + } + if(x.hi & SIGN(32)) { + if(ylo != 0) { + ylo = -ylo; + yhi = ~yhi; + } else + yhi = -yhi; + } + + y->hi = yhi; + y->lo = ylo; +} + +void +_f2v(Vlong *y, float f) +{ + + _d2v(y, f); +} + +double +_v2d(Vlong x) +{ + if(x.hi & SIGN(32)) { + if(x.lo) { + x.lo = -x.lo; + x.hi = ~x.hi; + } else + x.hi = -x.hi; + return -((long)x.hi*4294967296. + x.lo); + } + return (long)x.hi*4294967296. + x.lo; +} + +float +_v2f(Vlong x) +{ + return _v2d(x); +} + +uint _div64by32(Vlong, uint, uint*); +void _mul64by32(Vlong*, Vlong, uint); + +static void +slowdodiv(Vlong num, Vlong den, Vlong *q, Vlong *r) +{ + uint numlo, numhi, denhi, denlo, quohi, quolo, t; + int i; + + numhi = num.hi; + numlo = num.lo; + denhi = den.hi; + denlo = den.lo; + + /* + * get a divide by zero + */ + if(denlo==0 && denhi==0) { + numlo = numlo / denlo; + } + + /* + * set up the divisor and find the number of iterations needed + */ + if(numhi >= SIGN(32)) { + quohi = SIGN(32); + quolo = 0; + } else { + quohi = numhi; + quolo = numlo; + } + i = 0; + while(denhi < quohi || (denhi == quohi && denlo < quolo)) { + denhi = (denhi<<1) | (denlo>>31); + denlo <<= 1; + i++; + } + + quohi = 0; + quolo = 0; + for(; i >= 0; i--) { + quohi = (quohi<<1) | (quolo>>31); + quolo <<= 1; + if(numhi > denhi || (numhi == denhi && numlo >= denlo)) { + t = numlo; + numlo -= denlo; + if(numlo > t) + numhi--; + numhi -= denhi; + quolo |= 1; + } + denlo = (denlo>>1) | (denhi<<31); + denhi >>= 1; + } + + if(q) { + q->lo = quolo; + q->hi = quohi; + } + if(r) { + r->lo = numlo; + r->hi = numhi; + } +} + +static void +dodiv(Vlong num, Vlong den, Vlong *qp, Vlong *rp) +{ + uint n; + Vlong x, q, r; + + if(den.hi > num.hi || (den.hi == num.hi && den.lo > num.lo)){ + if(qp) { + qp->hi = 0; + qp->lo = 0; + } + if(rp) { + rp->hi = num.hi; + rp->lo = num.lo; + } + return; + } + + if(den.hi != 0){ + q.hi = 0; + n = num.hi/den.hi; + _mul64by32(&x, den, n); + if(x.hi > num.hi || (x.hi == num.hi && x.lo > num.lo)) + slowdodiv(num, den, &q, &r); + else { + q.lo = n; + _subv(&r, num, x); + } + } else { + if(num.hi >= den.lo){ + q.hi = n = num.hi/den.lo; + num.hi -= den.lo*n; + } else { + q.hi = 0; + } + q.lo = _div64by32(num, den.lo, &r.lo); + r.hi = 0; + } + if(qp) { + qp->lo = q.lo; + qp->hi = q.hi; + } + if(rp) { + rp->lo = r.lo; + rp->hi = r.hi; + } +} + +void +_divvu(Vlong *q, Vlong n, Vlong d) +{ + + if(n.hi == 0 && d.hi == 0) { + q->hi = 0; + q->lo = n.lo / d.lo; + return; + } + dodiv(n, d, q, 0); +} + +void +_modvu(Vlong *r, Vlong n, Vlong d) +{ + + if(n.hi == 0 && d.hi == 0) { + r->hi = 0; + r->lo = n.lo % d.lo; + return; + } + dodiv(n, d, 0, r); +} + +static void +vneg(Vlong *v) +{ + + if(v->lo == 0) { + v->hi = -v->hi; + return; + } + v->lo = -v->lo; + v->hi = ~v->hi; +} + +void +_divv(Vlong *q, Vlong n, Vlong d) +{ + long nneg, dneg; + + if(n.hi == (((long)n.lo)>>31) && d.hi == (((long)d.lo)>>31)) { + q->lo = (long)n.lo / (long)d.lo; + q->hi = ((long)q->lo) >> 31; + return; + } + nneg = n.hi >> 31; + if(nneg) + vneg(&n); + dneg = d.hi >> 31; + if(dneg) + vneg(&d); + dodiv(n, d, q, 0); + if(nneg != dneg) + vneg(q); +} + +void +_modv(Vlong *r, Vlong n, Vlong d) +{ + long nneg, dneg; + + if(n.hi == (((long)n.lo)>>31) && d.hi == (((long)d.lo)>>31)) { + r->lo = (long)n.lo % (long)d.lo; + r->hi = ((long)r->lo) >> 31; + return; + } + nneg = n.hi >> 31; + if(nneg) + vneg(&n); + dneg = d.hi >> 31; + if(dneg) + vneg(&d); + dodiv(n, d, 0, r); + if(nneg) + vneg(r); +} + +void +_rshav(Vlong *r, Vlong a, int b) +{ + long t; + + t = a.hi; + if(b >= 32) { + r->hi = t>>31; + if(b >= 64) { + /* this is illegal re C standard */ + r->lo = t>>31; + return; + } + r->lo = t >> (b-32); + return; + } + if(b <= 0) { + r->hi = t; + r->lo = a.lo; + return; + } + r->hi = t >> b; + r->lo = (t << (32-b)) | (a.lo >> b); +} + +void +_rshlv(Vlong *r, Vlong a, int b) +{ + uint t; + + t = a.hi; + if(b >= 32) { + r->hi = 0; + if(b >= 64) { + /* this is illegal re C standard */ + r->lo = 0; + return; + } + r->lo = t >> (b-32); + return; + } + if(b <= 0) { + r->hi = t; + r->lo = a.lo; + return; + } + r->hi = t >> b; + r->lo = (t << (32-b)) | (a.lo >> b); +} + +void +_lshv(Vlong *r, Vlong a, int b) +{ + uint t; + + t = a.lo; + if(b >= 32) { + r->lo = 0; + if(b >= 64) { + /* this is illegal re C standard */ + r->hi = 0; + return; + } + r->hi = t << (b-32); + return; + } + if(b <= 0) { + r->lo = t; + r->hi = a.hi; + return; + } + r->lo = t << b; + r->hi = (t >> (32-b)) | (a.hi << b); +} + +void +_andv(Vlong *r, Vlong a, Vlong b) +{ + r->hi = a.hi & b.hi; + r->lo = a.lo & b.lo; +} + +void +_orv(Vlong *r, Vlong a, Vlong b) +{ + r->hi = a.hi | b.hi; + r->lo = a.lo | b.lo; +} + +void +_xorv(Vlong *r, Vlong a, Vlong b) +{ + r->hi = a.hi ^ b.hi; + r->lo = a.lo ^ b.lo; +} + +void +_vpp(Vlong *l, Vlong *r) +{ + + l->hi = r->hi; + l->lo = r->lo; + r->lo++; + if(r->lo == 0) + r->hi++; +} + +void +_vmm(Vlong *l, Vlong *r) +{ + + l->hi = r->hi; + l->lo = r->lo; + if(r->lo == 0) + r->hi--; + r->lo--; +} + +void +_ppv(Vlong *l, Vlong *r) +{ + + r->lo++; + if(r->lo == 0) + r->hi++; + l->hi = r->hi; + l->lo = r->lo; +} + +void +_mmv(Vlong *l, Vlong *r) +{ + + if(r->lo == 0) + r->hi--; + r->lo--; + l->hi = r->hi; + l->lo = r->lo; +} + +void +_vasop(Vlong *ret, void *lv, void fn(Vlong*, Vlong, Vlong), int type, Vlong rv) +{ + Vlong t, u; + + u.lo = 0; + u.hi = 0; + switch(type) { + default: + abort(); + break; + + case 1: /* schar */ + t.lo = *(schar*)lv; + t.hi = t.lo >> 31; + fn(&u, t, rv); + *(schar*)lv = u.lo; + break; + + case 2: /* uchar */ + t.lo = *(uchar*)lv; + t.hi = 0; + fn(&u, t, rv); + *(uchar*)lv = u.lo; + break; + + case 3: /* short */ + t.lo = *(short*)lv; + t.hi = t.lo >> 31; + fn(&u, t, rv); + *(short*)lv = u.lo; + break; + + case 4: /* ushort */ + t.lo = *(ushort*)lv; + t.hi = 0; + fn(&u, t, rv); + *(ushort*)lv = u.lo; + break; + + case 9: /* int */ + t.lo = *(int*)lv; + t.hi = t.lo >> 31; + fn(&u, t, rv); + *(int*)lv = u.lo; + break; + + case 10: /* uint */ + t.lo = *(uint*)lv; + t.hi = 0; + fn(&u, t, rv); + *(uint*)lv = u.lo; + break; + + case 5: /* long */ + t.lo = *(long*)lv; + t.hi = t.lo >> 31; + fn(&u, t, rv); + *(long*)lv = u.lo; + break; + + case 6: /* uint */ + t.lo = *(uint*)lv; + t.hi = 0; + fn(&u, t, rv); + *(uint*)lv = u.lo; + break; + + case 7: /* vlong */ + case 8: /* uvlong */ + fn(&u, *(Vlong*)lv, rv); + *(Vlong*)lv = u; + break; + } + *ret = u; +} + +void +_p2v(Vlong *ret, void *p) +{ + long t; + + t = (uint)p; + ret->lo = t; + ret->hi = 0; +} + +void +_sl2v(Vlong *ret, long sl) +{ + long t; + + t = sl; + ret->lo = t; + ret->hi = t >> 31; +} + +void +_ul2v(Vlong *ret, uint ul) +{ + long t; + + t = ul; + ret->lo = t; + ret->hi = 0; +} + +void +_si2v(Vlong *ret, int si) +{ + long t; + + t = si; + ret->lo = t; + ret->hi = t >> 31; +} + +void +_ui2v(Vlong *ret, uint ui) +{ + long t; + + t = ui; + ret->lo = t; + ret->hi = 0; +} + +void +_sh2v(Vlong *ret, long sh) +{ + long t; + + t = (sh << 16) >> 16; + ret->lo = t; + ret->hi = t >> 31; +} + +void +_uh2v(Vlong *ret, uint ul) +{ + long t; + + t = ul & 0xffff; + ret->lo = t; + ret->hi = 0; +} + +void +_sc2v(Vlong *ret, long uc) +{ + long t; + + t = (uc << 24) >> 24; + ret->lo = t; + ret->hi = t >> 31; +} + +void +_uc2v(Vlong *ret, uint ul) +{ + long t; + + t = ul & 0xff; + ret->lo = t; + ret->hi = 0; +} + +long +_v2sc(Vlong rv) +{ + long t; + + t = rv.lo & 0xff; + return (t << 24) >> 24; +} + +long +_v2uc(Vlong rv) +{ + + return rv.lo & 0xff; +} + +long +_v2sh(Vlong rv) +{ + long t; + + t = rv.lo & 0xffff; + return (t << 16) >> 16; +} + +long +_v2uh(Vlong rv) +{ + + return rv.lo & 0xffff; +} + +long +_v2sl(Vlong rv) +{ + + return rv.lo; +} + +long +_v2ul(Vlong rv) +{ + + return rv.lo; +} + +long +_v2si(Vlong rv) +{ + + return rv.lo; +} + +long +_v2ui(Vlong rv) +{ + + return rv.lo; +} + +int +_testv(Vlong rv) +{ + return rv.lo || rv.hi; +} + +int +_eqv(Vlong lv, Vlong rv) +{ + return lv.lo == rv.lo && lv.hi == rv.hi; +} + +int +_nev(Vlong lv, Vlong rv) +{ + return lv.lo != rv.lo || lv.hi != rv.hi; +} + +int +_ltv(Vlong lv, Vlong rv) +{ + return (long)lv.hi < (long)rv.hi || + (lv.hi == rv.hi && lv.lo < rv.lo); +} + +int +_lev(Vlong lv, Vlong rv) +{ + return (long)lv.hi < (long)rv.hi || + (lv.hi == rv.hi && lv.lo <= rv.lo); +} + +int +_gtv(Vlong lv, Vlong rv) +{ + return (long)lv.hi > (long)rv.hi || + (lv.hi == rv.hi && lv.lo > rv.lo); +} + +int +_gev(Vlong lv, Vlong rv) +{ + return (long)lv.hi > (long)rv.hi || + (lv.hi == rv.hi && lv.lo >= rv.lo); +} + +int +_lov(Vlong lv, Vlong rv) +{ + return lv.hi < rv.hi || + (lv.hi == rv.hi && lv.lo < rv.lo); +} + +int +_lsv(Vlong lv, Vlong rv) +{ + return lv.hi < rv.hi || + (lv.hi == rv.hi && lv.lo <= rv.lo); +} + +int +_hiv(Vlong lv, Vlong rv) +{ + return lv.hi > rv.hi || + (lv.hi == rv.hi && lv.lo > rv.lo); +} + +int +_hsv(Vlong lv, Vlong rv) +{ + return lv.hi > rv.hi || + (lv.hi == rv.hi && lv.lo >= rv.lo); +} |