code review 81370046: runtime: get rid of most uses of REP for copying/zeroing.

src/liblink/asm8.c

 // Inferno utils/8l/span.c
 // http://code.google.com/p/inferno-os/source/browse/utils/8l/span.c
 //
 //      Copyright © 1994-1999 Lucent Technologies Inc.  All rights reserved.
 //      Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
 //      Portions Copyright © 1997-1999 Vita Nuova Limited
 //      Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com)
 //      Portions Copyright © 2004,2006 Bruce Ellis
 //      Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
 //      Revisions Copyright © 2000-2007 Lucent Technologies Inc. and others
@@ skipping 400 matching lines @@
         Ynone,  Ybr,    Zloop,  1,
         0
 };
 static uchar    ycall[] =
 {
         Ynone,  Yml,    Zo_m,   0,
         Yrx,    Yrx,    Zo_m,   2,
         Ynone,  Ycol,   Zcallind,       2,
         Ynone,  Ybr,    Zcall,  0,
         Ynone,  Yi32,   Zcallcon,       1,
+        0
+};
+static uchar    yduff[] =
+{
+        Ynone,  Yi32,   Zcall,  1,
         0
 };
 static uchar    yjmp[] =
 {
         Ynone,  Yml,    Zo_m,   2,
         Ynone,  Ybr,    Zjmp,   0,
         Ynone,  Yi32,   Zjmpcon,        1,
         0
 };
 
@@ skipping 709 matching lines @@
         { AAESENC,      yaes,   Pq, 0x38,0xdc,(0) },
         { APINSRD,      yinsrd, Pq, 0x3a, 0x22, (00) },
         { APSHUFB,      ymshufb,Pq, 0x38, 0x00 },
 
         { AUSEFIELD,    ynop,   Px, 0,0 },
         { ATYPE },
         { AFUNCDATA,    yfuncdata,      Px, 0,0 },
         { APCDATA,      ypcdata,        Px, 0,0 },
         { ACHECKNIL },
         { AVARDEF },
+        { AVARKILL },
+        { ADUFFCOPY,    yduff,  Px, 0xe8 },
+        { ADUFFZERO,    yduff,  Px, 0xe8 },
 
         0
 };
 
 static int32    vaddr(Link*, Addr*, Reloc*);
 
 // single-instruction no-ops of various lengths.
 // constructed by hand and disassembled with gdb to verify.
 // see http://www.agner.org/optimize/optimizing_assembly.pdf for discussion.
 static uchar nop[][16] = {
@@ skipping 1210 matching lines @@
                 if(p->to.sym == nil) {
                         ctxt->diag("call without target");
                         sysfatal("bad code");
                 }
                 *ctxt->andptr++ = op;
                 r = addrel(ctxt->cursym);
                 r->off = p->pc + ctxt->andptr - ctxt->and;
                 r->type = D_PCREL;
                 r->siz = 4;
                 r->sym = p->to.sym;
+                r->add = p->to.offset;
                 put4(ctxt, 0);
                 break;
 
         case Zbr:
         case Zjmp:
         case Zloop:
                 if(p->to.sym != nil) {
                         if(t[2] != Zjmp) {
                                 ctxt->diag("branch to ATEXT");
                                 sysfatal("bad code");
@@ skipping 285 matching lines @@
                         return;
                 }
         }
 
         doasm(ctxt, p);
         if(ctxt->andptr > ctxt->and+sizeof ctxt->and) {
                 print("and[] is too short - %ld byte instruction\n", ctxt->andptr - ctxt->and);
                 sysfatal("bad code");
         }
 }