ref: bc895417f804bbb78410a365412bcf8ab59a44b4
parent: 23742053f5ee70cd394c33f28f3e6547e9e8e31d
author: mischief <[email protected]>
date: Sat Aug 8 23:44:03 EDT 2015
libc: fix spim endianness
--- a/sys/src/libc/spim/mkfile
+++ b/sys/src/libc/spim/mkfile
@@ -2,7 +2,7 @@
</$objtype/mkfile
LIB=/$objtype/lib/libc.a
-SFILES=\
+MIPSSFILES=\
argv0.s\
atom.s\
getcallerpc.s\
@@ -19,23 +19,33 @@
strcmp.s\
strcpy.s\
tas.s\
- vlop.s\
-CFILES=\
+MIPSCFILES=\
cycles.c\
lock.c\
notejmp.c\
+
+SFILES=\
+ vlop.s\
+
+CFILES=\
sqrt.c\
vlrt.c\
HFILES=/sys/include/libc.h
-OFILES=${CFILES:%.c=%.$O} ${SFILES:%.s=%.$O}+OFILES=\
+ ${MIPSSFILES:%.s=%.$O}\+ ${SFILES:%.s=%.$O}\+ ${MIPSCFILES:%.c=%.$O}\+ ${CFILES:%.c=%.$O}\-%.$O: ../mips/%.c
- $CC -I../mips $CFLAGS ../mips/$stem.c
+MIPSS=`{echo $MIPSSFILES | sed 's/\.s//g; s/ /|/g'}+^($MIPSS)\.$O:R: '../mips/\1.s'
+ $AS $AFLAGS ../mips/$stem1.s
-%.$O: ../mips/%.s
- $AS -I../mips $AFLAGS ../mips/$stem.s
+MIPSC=`{echo $MIPSCFILES | sed 's/\.c//g; s/ /|/g'}+^($MIPSC)\.$O:R: '../mips/\1.c'
+ $CC $CFLAGS ../mips/$stem1.c
</sys/src/cmd/mksyslib
--- /dev/null
+++ b/sys/src/libc/spim/sqrt.c
@@ -1,0 +1,103 @@
+#include <u.h>
+#include <libc.h>
+
+static long sqtab[64] =
+{+ 0x6cdb2, 0x726d4, 0x77ea3, 0x7d52f, 0x82a85, 0x87eb1, 0x8d1c0, 0x923bd,
+ 0x974b2, 0x9c4a8, 0xa13a9, 0xa61be, 0xaaeee, 0xafb41, 0xb46bf, 0xb916e,
+ 0xbdb55, 0xc247a, 0xc6ce3, 0xcb495, 0xcfb95, 0xd41ea, 0xd8796, 0xdcca0,
+ 0xe110c, 0xe54dd, 0xe9818, 0xedac0, 0xf1cd9, 0xf5e67, 0xf9f6e, 0xfdfef,
+ 0x01fe0, 0x05ee6, 0x09cfd, 0x0da30, 0x11687, 0x1520c, 0x18cc8, 0x1c6c1,
+ 0x20000, 0x2388a, 0x27068, 0x2a79e, 0x2de32, 0x3142b, 0x3498c, 0x37e5b,
+ 0x3b29d, 0x3e655, 0x41989, 0x44c3b, 0x47e70, 0x4b02b, 0x4e16f, 0x51241,
+ 0x542a2, 0x57296, 0x5a220, 0x5d142, 0x60000, 0x62e5a, 0x65c55, 0x689f2,
+};
+
+double
+sqrt(double arg)
+{+ int e, ms;
+ double a, t;
+ union
+ {+ double d;
+ struct
+ {+ long ls;
+ long ms;
+ };
+ } u;
+
+ u.d = arg;
+ ms = u.ms;
+
+ /*
+ * sign extend the mantissa with
+ * exponent. result should be > 0 for
+ * normal case.
+ */
+ e = ms >> 20;
+ if(e <= 0) {+ if(e == 0)
+ return 0;
+ return NaN();
+ }
+
+ /*
+ * pick up arg/4 by adjusting exponent
+ */
+ u.ms = ms - (2 << 20);
+ a = u.d;
+
+ /*
+ * use 5 bits of mantissa and 1 bit
+ * of exponent to form table index.
+ * insert exponent/2 - 1.
+ */
+ e = (((e - 1023) >> 1) + 1022) << 20;
+ u.ms = *(long*)((char*)sqtab + ((ms >> 13) & 0xfc)) | e;
+ u.ls = 0;
+
+ /*
+ * three laps of newton
+ */
+ e = 1 << 20;
+ t = u.d;
+ u.d = t + a/t;
+ u.ms -= e; /* u.d /= 2; */
+ t = u.d;
+ u.d = t + a/t;
+ u.ms -= e; /* u.d /= 2; */
+ t = u.d;
+
+ return t + a/t;
+}
+
+/*
+ * this is the program that generated the table.
+ * it calls sqrt by some other means.
+ *
+ * void
+ * main(void)
+ * {+ * int i;
+ * union U
+ * {+ * double d;
+ * struct
+ * {+ * long ms;
+ * long ls;
+ * };
+ * } u;
+ *
+ * for(i=0; i<64; i++) {+ * u.ms = (i<<15) | 0x3fe04000;
+ * u.ls = 0;
+ * u.d = sqrt(u.d);
+ * print(" 0x%.5lux,", u.ms & 0xfffff);+ * }
+ * print("\n");+ * exits(0);
+ * }
+ */
--- /dev/null
+++ b/sys/src/libc/spim/vlop.s
@@ -1,0 +1,20 @@
+/*
+ * from https://bitbucket.org/cherry9/plan9-loongson
+ */
+TEXT _mulv(SB), $0
+ MOVW 8(FP), R2 /* hi1 */
+ MOVW 4(FP), R3 /* lo1 */
+ MOVW 16(FP), R4 /* hi2 */
+ MOVW 12(FP), R5 /* lo2 */
+ MULU R5, R3 /* lo1*lo2 -> hi:lo*/
+ MOVW LO, R6
+ MOVW HI, R7
+ MULU R3, R4 /* lo1*hi2 -> _:hi */
+ MOVW LO, R8
+ ADDU R8, R7
+ MULU R2, R5 /* hi1*lo2 -> _:hi */
+ MOVW LO, R8
+ ADDU R8, R7
+ MOVW R6, 0(R1) /* lo */
+ MOVW R7, 4(R1) /* hi */
+ RET
--- /dev/null
+++ b/sys/src/libc/spim/vlrt.c
@@ -1,0 +1,723 @@
+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
+ {+ ulong lo;
+ ulong hi;
+ };
+ struct
+ {+ ushort loms;
+ ushort lols;
+ ushort hims;
+ ushort hils;
+ };
+ };
+};
+
+void abort(void);
+
+/* needed by profiler; can't be profiled. */
+#pragma profile off
+void
+_addv(Vlong *r, Vlong a, Vlong b)
+{+ ulong lo, hi;
+
+ lo = a.lo + b.lo;
+ hi = a.hi + b.hi;
+ if(lo < a.lo)
+ hi++;
+ r->lo = lo;
+ r->hi = hi;
+}
+
+void
+_subv(Vlong *r, Vlong a, Vlong b)
+{+ ulong lo, hi;
+
+ lo = a.lo - b.lo;
+ hi = a.hi - b.hi;
+ if(lo > a.lo)
+ hi--;
+ r->lo = lo;
+ r->hi = hi;
+}
+
+#pragma profile on
+
+void
+_d2v(Vlong *y, double d)
+{+ union { double d; struct Vlong; } x;+ ulong 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);
+}
+
+static void
+dodiv(Vlong num, Vlong den, Vlong *qp, Vlong *rp)
+{+ ulong 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(qp) {+ qp->lo = quolo;
+ qp->hi = quohi;
+ }
+ if(rp) {+ rp->lo = numlo;
+ rp->hi = numhi;
+ }
+}
+
+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)
+{+ ulong 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)
+{+ ulong 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: /* ulong */
+ t.lo = *(ulong*)lv;
+ t.hi = 0;
+ fn(&u, t, rv);
+ *(ulong*)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 = (ulong)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, ulong 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, ulong 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, ulong 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);
+}