123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256 |
- /*
- * linux/arch/arm/vfp/vfpdouble.c
- *
- * This code is derived in part from John R. Housers softfloat library, which
- * carries the following notice:
- *
- * ===========================================================================
- * This C source file is part of the SoftFloat IEC/IEEE Floating-point
- * Arithmetic Package, Release 2.
- *
- * Written by John R. Hauser. This work was made possible in part by the
- * International Computer Science Institute, located at Suite 600, 1947 Center
- * Street, Berkeley, California 94704. Funding was partially provided by the
- * National Science Foundation under grant MIP-9311980. The original version
- * of this code was written as part of a project to build a fixed-point vector
- * processor in collaboration with the University of California at Berkeley,
- * overseen by Profs. Nelson Morgan and John Wawrzynek. More information
- * is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
- * arithmetic/softfloat.html'.
- *
- * THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
- * has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
- * TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
- * PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
- * AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
- *
- * Derivative works are acceptable, even for commercial purposes, so long as
- * (1) they include prominent notice that the work is derivative, and (2) they
- * include prominent notice akin to these three paragraphs for those parts of
- * this code that are retained.
- * ===========================================================================
- */
- #include <linux/kernel.h>
- #include <linux/bitops.h>
- #include <asm/div64.h>
- #include <asm/vfp.h>
- #include "vfpinstr.h"
- #include "vfp.h"
- static struct vfp_double vfp_double_default_qnan = {
- .exponent = 2047,
- .sign = 0,
- .significand = VFP_DOUBLE_SIGNIFICAND_QNAN,
- };
- static void vfp_double_dump(const char *str, struct vfp_double *d)
- {
- pr_debug("VFP: %s: sign=%d exponent=%d significand=%016llx\n",
- str, d->sign != 0, d->exponent, d->significand);
- }
- static void vfp_double_normalise_denormal(struct vfp_double *vd)
- {
- int bits = 31 - fls(vd->significand >> 32);
- if (bits == 31)
- bits = 63 - fls(vd->significand);
- vfp_double_dump("normalise_denormal: in", vd);
- if (bits) {
- vd->exponent -= bits - 1;
- vd->significand <<= bits;
- }
- vfp_double_dump("normalise_denormal: out", vd);
- }
- u32 vfp_double_normaliseround(int dd, struct vfp_double *vd, u32 fpscr, u32 exceptions, const char *func)
- {
- u64 significand, incr;
- int exponent, shift, underflow;
- u32 rmode;
- vfp_double_dump("pack: in", vd);
- /*
- * Infinities and NaNs are a special case.
- */
- if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
- goto pack;
- /*
- * Special-case zero.
- */
- if (vd->significand == 0) {
- vd->exponent = 0;
- goto pack;
- }
- exponent = vd->exponent;
- significand = vd->significand;
- shift = 32 - fls(significand >> 32);
- if (shift == 32)
- shift = 64 - fls(significand);
- if (shift) {
- exponent -= shift;
- significand <<= shift;
- }
- #ifdef DEBUG
- vd->exponent = exponent;
- vd->significand = significand;
- vfp_double_dump("pack: normalised", vd);
- #endif
- /*
- * Tiny number?
- */
- underflow = exponent < 0;
- if (underflow) {
- significand = vfp_shiftright64jamming(significand, -exponent);
- exponent = 0;
- #ifdef DEBUG
- vd->exponent = exponent;
- vd->significand = significand;
- vfp_double_dump("pack: tiny number", vd);
- #endif
- if (!(significand & ((1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1)))
- underflow = 0;
- }
- /*
- * Select rounding increment.
- */
- incr = 0;
- rmode = fpscr & FPSCR_RMODE_MASK;
- if (rmode == FPSCR_ROUND_NEAREST) {
- incr = 1ULL << VFP_DOUBLE_LOW_BITS;
- if ((significand & (1ULL << (VFP_DOUBLE_LOW_BITS + 1))) == 0)
- incr -= 1;
- } else if (rmode == FPSCR_ROUND_TOZERO) {
- incr = 0;
- } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
- incr = (1ULL << (VFP_DOUBLE_LOW_BITS + 1)) - 1;
- pr_debug("VFP: rounding increment = 0x%08llx\n", incr);
- /*
- * Is our rounding going to overflow?
- */
- if ((significand + incr) < significand) {
- exponent += 1;
- significand = (significand >> 1) | (significand & 1);
- incr >>= 1;
- #ifdef DEBUG
- vd->exponent = exponent;
- vd->significand = significand;
- vfp_double_dump("pack: overflow", vd);
- #endif
- }
- /*
- * If any of the low bits (which will be shifted out of the
- * number) are non-zero, the result is inexact.
- */
- if (significand & ((1 << (VFP_DOUBLE_LOW_BITS + 1)) - 1))
- exceptions |= FPSCR_IXC;
- /*
- * Do our rounding.
- */
- significand += incr;
- /*
- * Infinity?
- */
- if (exponent >= 2046) {
- exceptions |= FPSCR_OFC | FPSCR_IXC;
- if (incr == 0) {
- vd->exponent = 2045;
- vd->significand = 0x7fffffffffffffffULL;
- } else {
- vd->exponent = 2047; /* infinity */
- vd->significand = 0;
- }
- } else {
- if (significand >> (VFP_DOUBLE_LOW_BITS + 1) == 0)
- exponent = 0;
- if (exponent || significand > 0x8000000000000000ULL)
- underflow = 0;
- if (underflow)
- exceptions |= FPSCR_UFC;
- vd->exponent = exponent;
- vd->significand = significand >> 1;
- }
- pack:
- vfp_double_dump("pack: final", vd);
- {
- s64 d = vfp_double_pack(vd);
- pr_debug("VFP: %s: d(d%d)=%016llx exceptions=%08x\n", func,
- dd, d, exceptions);
- vfp_put_double(d, dd);
- }
- return exceptions;
- }
- /*
- * Propagate the NaN, setting exceptions if it is signalling.
- * 'n' is always a NaN. 'm' may be a number, NaN or infinity.
- */
- static u32
- vfp_propagate_nan(struct vfp_double *vdd, struct vfp_double *vdn,
- struct vfp_double *vdm, u32 fpscr)
- {
- struct vfp_double *nan;
- int tn, tm = 0;
- tn = vfp_double_type(vdn);
- if (vdm)
- tm = vfp_double_type(vdm);
- if (fpscr & FPSCR_DEFAULT_NAN)
- /*
- * Default NaN mode - always returns a quiet NaN
- */
- nan = &vfp_double_default_qnan;
- else {
- /*
- * Contemporary mode - select the first signalling
- * NAN, or if neither are signalling, the first
- * quiet NAN.
- */
- if (tn == VFP_SNAN || (tm != VFP_SNAN && tn == VFP_QNAN))
- nan = vdn;
- else
- nan = vdm;
- /*
- * Make the NaN quiet.
- */
- nan->significand |= VFP_DOUBLE_SIGNIFICAND_QNAN;
- }
- *vdd = *nan;
- /*
- * If one was a signalling NAN, raise invalid operation.
- */
- return tn == VFP_SNAN || tm == VFP_SNAN ? FPSCR_IOC : VFP_NAN_FLAG;
- }
- /*
- * Extended operations
- */
- static u32 vfp_double_fabs(int dd, int unused, int dm, u32 fpscr)
- {
- vfp_put_double(vfp_double_packed_abs(vfp_get_double(dm)), dd);
- return 0;
- }
- static u32 vfp_double_fcpy(int dd, int unused, int dm, u32 fpscr)
|