diff options
24 files changed, 8706 insertions, 104 deletions
diff --git a/sys/arch/alpha/alpha/db_instruction.h b/sys/arch/alpha/alpha/db_instruction.h index 10994cff35c..6eb6a4faf7f 100644 --- a/sys/arch/alpha/alpha/db_instruction.h +++ b/sys/arch/alpha/alpha/db_instruction.h @@ -1,5 +1,5 @@ -/* $OpenBSD: db_instruction.h,v 1.3 2001/08/12 12:03:02 heko Exp $ */ -/* $NetBSD: db_instruction.h,v 1.6 2000/03/20 02:54:45 thorpej Exp $ */ +/* $OpenBSD: db_instruction.h,v 1.4 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: db_instruction.h,v 1.7 2001/04/26 03:10:44 ross Exp $ */ /* * Copyright (c) 1999 Christopher G. Demetriou. All rights reserved. @@ -185,6 +185,16 @@ typedef union { opcode : 6; } float_format; + struct { + unsigned fc : 5, + opclass : 4, + src : 2, + rnd : 2, + trp : 3, + fb : 5, + fa : 5, + opcode : 6; + } float_detail; /* * PAL instructions just define the major opcode @@ -223,6 +233,7 @@ typedef union { #define op_bit 0x12 /* see BIT sub-table */ #define op_mul 0x13 /* see MUL sub-table */ /* reserved */ +#define op_fix_float 0x14 /* if ALPHA_AMASK_FIX */ #define op_vax_float 0x15 /* see FLOAT sub-table */ #define op_ieee_float 0x16 /* see FLOAT sub-table */ #define op_any_float 0x17 /* see FLOAT sub-table */ @@ -412,6 +423,12 @@ typedef union { * Load and store operations use opcodes op_ldf..op_stt */ + /* src encoding from function, 9..10 */ +#define op_src_sf 0 +#define op_src_xd 1 +#define op_src_tg 2 +#define op_src_qq 3 + /* any FLOAT, "function" opcodes (bits 5..11) */ #define op_cvtlq 0x010 @@ -428,7 +445,7 @@ typedef union { #define op_fcmovgt 0x02f #define op_cvtql 0x030 #define op_cvtql_v 0x130 -#define op_cvtql_sv 0x330 +#define op_cvtql_sv 0x530 /* ieee FLOAT, "function" opcodes (bits 5..11) */ @@ -521,6 +538,7 @@ typedef union { #define op_mult_ud 0x1e2 #define op_divt_ud 0x1e3 #define op_cvtts_ud 0x1ec +#define op_cvtst 0x2ac #define op_adds_suc 0x500 #define op_subs_suc 0x501 #define op_muls_suc 0x502 @@ -563,6 +581,7 @@ typedef union { #define op_mult_sud 0x5e2 #define op_divt_sud 0x5e3 #define op_cvtts_sud 0x5ec +#define op_cvtst_u 0x6ac #define op_adds_suic 0x700 #define op_subs_suic 0x701 #define op_muls_suic 0x702 diff --git a/sys/arch/alpha/alpha/fp_complete.c b/sys/arch/alpha/alpha/fp_complete.c new file mode 100644 index 00000000000..80f295050e5 --- /dev/null +++ b/sys/arch/alpha/alpha/fp_complete.c @@ -0,0 +1,698 @@ +/* $OpenBSD: fp_complete.c,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: fp_complete.c,v 1.5 2002/01/18 22:15:56 ross Exp $ */ + +/*- + * Copyright (c) 2001 Ross Harvey + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef NO_IEEE + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/proc.h> + +#include <machine/cpu.h> +#include <machine/fpu.h> +#include <machine/reg.h> +#include <machine/cpu.h> +#include <alpha/alpha/db_instruction.h> + +#include <lib/libkern/softfloat.h> + +#define TSWINSIZE 4 /* size of trap shadow window in u_int32_t units */ + +/* Set Name Opcodes AARM C.* Symbols */ + +#define CPUREG_CLASS (0xfUL << 0x10) /* INT[ALSM] */ +#define FPUREG_CLASS (0xfUL << 0x14) /* ITFP, FLT[ILV] */ +#define CHECKFUNCTIONCODE (1UL << 0x18) /* MISC */ +#define TRAPSHADOWBOUNDARY (1UL << 0x00 | /* PAL */\ + 1UL << 0x19 | /* \PAL\ */\ + 1UL << 0x1a | /* JSR */\ + 1UL << 0x1b | /* \PAL\ */\ + 1UL << 0x1d | /* \PAL\ */\ + 1UL << 0x1e | /* \PAL\ */\ + 1UL << 0x1f | /* \PAL\ */\ + 0xffffUL << 0x30 | /* branch ops */\ + CHECKFUNCTIONCODE) + +#define MAKE_FLOATXX(width, expwidth, sign, exp, msb, rest_of_frac) \ + (u_int ## width ## _t)(sign) << ((width) - 1) |\ + (u_int ## width ## _t)(exp) << ((width) - 1 - (expwidth)) |\ + (u_int ## width ## _t)(msb) << ((width) - 1 - (expwidth) - 1) |\ + (u_int ## width ## _t)(rest_of_frac) + +#define FLOAT32QNAN MAKE_FLOATXX(32, 8, 0, 0xff, 1, 0) +#define FLOAT64QNAN MAKE_FLOATXX(64, 11, 0, 0x7ff, 1, 0) + +#define IS_SUBNORMAL(v) ((v)->exp == 0 && (v)->frac != 0) + +#define PREFILTER_SUBNORMAL(p,v) if ((p)->p_md.md_flags & IEEE_MAP_DMZ \ + && IS_SUBNORMAL(v)) \ + (v)->frac = 0; else + +#define POSTFILTER_SUBNORMAL(p,v) if ((p)->p_md.md_flags & IEEE_MAP_UMZ \ + && IS_SUBNORMAL(v)) \ + (v)->frac = 0; else + + /* Alpha returns 2.0 for true, all zeroes for false. */ + +#define CMP_RESULT(flag) ((flag) ? 4UL << 60 : 0L) + + /* Move bits from sw fp_c to hw fpcr. */ + +#define CRBLIT(sw, hw, m, offs) (((sw) & ~(m)) | ((hw) >> (offs) & (m))) + +/* + * Temporary trap shadow instrumentation. The [un]resolved counters + * could be kept permanently, as they provide information on whether + * user code has met AARM trap shadow generation requirements. + */ + +struct alpha_shadow { + u_int64_t resolved; /* cases trigger pc found */ + u_int64_t unresolved; /* cases it wasn't, code problems? */ + u_int64_t scans; /* trap shadow scans */ + u_int64_t len; /* number of instructions examined */ + u_int64_t uop; /* bit mask of unexpected opcodes */ + u_int64_t sqrts; /* ev6+ square root single count */ + u_int64_t sqrtt; /* ev6+ square root double count */ + u_int32_t ufunc; /* bit mask of unexpected functions */ + u_int32_t max; /* max trap shadow scan */ + u_int32_t nilswop; /* unexpected op codes */ + u_int32_t nilswfunc; /* unexpected function codes */ + u_int32_t nilanyop; /* this "cannot happen" */ + u_int32_t vax; /* sigs from vax fp opcodes */ +} alpha_shadow, alpha_shadow_zero; + +static float64 float64_unk(float64, float64); +static float64 compare_un(float64, float64); +static float64 compare_eq(float64, float64); +static float64 compare_lt(float64, float64); +static float64 compare_le(float64, float64); +static void cvt_qs_ts_st_gf_qf(u_int32_t, struct proc *); +static void cvt_gd(u_int32_t, struct proc *); +static void cvt_qt_dg_qg(u_int32_t, struct proc *); +static void cvt_tq_gq(u_int32_t, struct proc *); + +static float32 (*swfp_s[])(float32, float32) = { + float32_add, float32_sub, float32_mul, float32_div, +}; + +static float64 (*swfp_t[])(float64, float64) = { + float64_add, float64_sub, float64_mul, float64_div, + compare_un, compare_eq, compare_lt, compare_le, + float64_unk, float64_unk, float64_unk, float64_unk +}; + +static void (*swfp_cvt[])(u_int32_t, struct proc *) = { + cvt_qs_ts_st_gf_qf, cvt_gd, cvt_qt_dg_qg, cvt_tq_gq +}; + +static void +this_cannot_happen(int what_cannot_happen, int64_t bits) +{ + static int total; + alpha_instruction inst; + static u_int64_t reported; + + inst.bits = bits; + ++alpha_shadow.nilswfunc; + if (bits != -1) + alpha_shadow.uop |= 1UL << inst.generic_format.opcode; + if (1UL << what_cannot_happen & reported) + return; + reported |= 1UL << what_cannot_happen; + if (total >= 1000) + return; /* right now, this return "cannot happen" */ + ++total; + if (bits) + printf("FP instruction %x\n", (unsigned int)bits); + printf("FP event %d/%lx/%lx\n", what_cannot_happen, reported, + alpha_shadow.uop); +} + +static __inline void +sts(unsigned int rn, s_float *v, struct proc *p) +{ + alpha_sts(rn, v); + PREFILTER_SUBNORMAL(p, v); +} + +static __inline void +stt(unsigned int rn, t_float *v, struct proc *p) +{ + alpha_stt(rn, v); + PREFILTER_SUBNORMAL(p, v); +} + +static __inline void +lds(unsigned int rn, s_float *v, struct proc *p) +{ + POSTFILTER_SUBNORMAL(p, v); + alpha_lds(rn, v); +} + +static __inline void +ldt(unsigned int rn, t_float *v, struct proc *p) +{ + POSTFILTER_SUBNORMAL(p, v); + alpha_ldt(rn, v); +} + +static float64 +compare_lt(float64 a, float64 b) +{ + return CMP_RESULT(float64_lt(a, b)); +} + +static float64 +compare_le(float64 a, float64 b) +{ + return CMP_RESULT(float64_le(a, b)); +} + +static float64 +compare_un(float64 a, float64 b) +{ + if (float64_is_nan(a) | float64_is_nan(b)) { + if (float64_is_signaling_nan(a) | float64_is_signaling_nan(b)) + float_set_invalid(); + return CMP_RESULT(1); + } + return CMP_RESULT(0); +} + +static float64 +compare_eq(float64 a, float64 b) +{ + return CMP_RESULT(float64_eq(a, b)); +} +/* + * A note regarding the VAX FP ops. + * + * The AARM gives us complete leeway to set or not set status flags on VAX + * ops, but we do any subnorm, NaN and dirty zero fixups anyway, and we set + * flags by IEEE rules. Many ops are common to d/f/g and s/t source types. + * For the purely vax ones, it's hard to imagine ever running them. + * (Generated VAX fp ops with completion flags? Hmm.) We are careful never + * to panic, assert, or print unlimited output based on a path through the + * decoder, so weird cases don't become security issues. + */ +static void +cvt_qs_ts_st_gf_qf(u_int32_t inst_bits, struct proc *p) +{ + t_float tfb, tfc; + s_float sfb, sfc; + alpha_instruction inst; + + inst.bits = inst_bits; + /* + * cvtst and cvtts have the same opcode, function, and source. The + * distinction for cvtst is hidden in the illegal modifier combinations. + * We decode even the non-/s modifier, so that the fix-up-always mode + * works on ev6 and later. The rounding bits are unused and fixed for + * cvtst, so we check those too. + */ + switch(inst.float_format.function) { + case op_cvtst: + case op_cvtst_u: + sts(inst.float_detail.fb, &sfb, p); + tfc.i = float32_to_float64(sfb.i); + ldt(inst.float_detail.fc, &tfc, p); + return; + } + if(inst.float_detail.src == 2) { + stt(inst.float_detail.fb, &tfb, p); + sfc.i = float64_to_float32(tfb.i); + lds(inst.float_detail.fc, &sfc, p); + return; + } + /* 0: S/F */ + /* 1: /D */ + /* 3: Q/Q */ + this_cannot_happen(5, inst.generic_format.opcode); + tfc.i = FLOAT64QNAN; + ldt(inst.float_detail.fc, &tfc, p); + return; +} + +static void +cvt_gd(u_int32_t inst_bits, struct proc *p) +{ + t_float tfb, tfc; + alpha_instruction inst; + + inst.bits = inst_bits; + stt(inst.float_detail.fb, &tfb, p); + (void) float64_to_float32(tfb.i); + p->p_md.md_flags &= ~OPENBSD_FLAG_TO_FP_C(FP_X_IMP); + tfc.i = float64_add(tfb.i, (float64)0); + ldt(inst.float_detail.fc, &tfc, p); +} + +static void +cvt_qt_dg_qg(u_int32_t inst_bits, struct proc *p) +{ + t_float tfb, tfc; + alpha_instruction inst; + + inst.bits = inst_bits; + switch(inst.float_detail.src) { + case 0: /* S/F */ + this_cannot_happen(3, inst.bits); + /* fall thru */ + case 1: /* D */ + /* VAX dirty 0's and reserved ops => UNPREDICTABLE */ + /* We've done what's important by just not trapping */ + tfc.i = 0; + break; + case 2: /* T/G */ + this_cannot_happen(4, inst.bits); + tfc.i = 0; + break; + case 3: /* Q/Q */ + stt(inst.float_detail.fb, &tfb, p); + tfc.i = int64_to_float64(tfb.i); + break; + } + alpha_ldt(inst.float_detail.fc, &tfc); +} +/* + * XXX: AARM and 754 seem to disagree here, also, beware of softfloat's + * unfortunate habit of always returning the nontrapping result. + * XXX: there are several apparent AARM/AAH disagreements, as well as + * the issue of trap handler pc and trapping results. + */ +static void +cvt_tq_gq(u_int32_t inst_bits, struct proc *p) +{ + t_float tfb, tfc; + alpha_instruction inst; + + inst.bits = inst_bits; + stt(inst.float_detail.fb, &tfb, p); + tfc.i = float64_to_int64(tfb.i); + alpha_ldt(inst.float_detail.fc, &tfc); /* yes, ldt */ +} + +static u_int64_t +fp_c_to_fpcr_1(u_int64_t fpcr, u_int64_t fp_c) +{ + u_int64_t disables; + + /* + * It's hard to arrange for conforming bit fields, because the FP_C + * and the FPCR are both architected, with specified (and relatively + * scrambled) bit numbers. Defining an internal unscrambled FP_C + * wouldn't help much, because every user exception requires the + * architected bit order in the sigcontext. + * + * Programs that fiddle with the fpcr exception bits (instead of fp_c) + * will lose, because those bits can be and usually are subsetted; + * the official home is in the fp_c. Furthermore, the kernel puts + * phony enables (it lies :-) in the fpcr in order to get control when + * it is necessary to initially set a sticky bit. + */ + + fpcr &= FPCR_DYN(3); + + /* + * enable traps = case where flag bit is clear OR program wants a trap + * enables = ~flags | mask + * disables = ~(~flags | mask) + * disables = flags & ~mask. Thank you, Augustus De Morgan (1806-1871) + */ + disables = FP_C_TO_OPENBSD_FLAG(fp_c) & ~FP_C_TO_OPENBSD_MASK(fp_c); + + fpcr |= (disables & (FP_X_IMP | FP_X_UFL)) << (61 - 3); + fpcr |= (disables & (FP_X_OFL | FP_X_DZ | FP_X_INV)) << (49 - 0); + +# if !(FP_X_INV == 1 && FP_X_DZ == 2 && FP_X_OFL == 4 && \ + FP_X_UFL == 8 && FP_X_IMP == 16 && FP_X_IOV == 32 && \ + FP_X_UFL << (61 - 3) == FPCR_UNFD && \ + FP_X_IMP << (61 - 3) == FPCR_INED && \ + FP_X_OFL << (49 - 0) == FPCR_OVFD) +# error "Assertion failed" + /* + * We don't care about the other built-in bit numbers because they + * have been architecturally specified. + */ +# endif + + fpcr |= fp_c & FP_C_MIRRORED << (FPCR_MIR_START - FP_C_MIR_START); + fpcr |= (fp_c & IEEE_MAP_DMZ) << 36; + if (fp_c & FP_C_MIRRORED) + fpcr |= FPCR_SUM; + if (fp_c & IEEE_MAP_UMZ) + fpcr |= FPCR_UNDZ | FPCR_UNFD; + fpcr |= (~fp_c & IEEE_TRAP_ENABLE_DNO) << 41; + return fpcr; +} + +static void +fp_c_to_fpcr(struct proc *p) +{ + alpha_write_fpcr(fp_c_to_fpcr_1(alpha_read_fpcr(), p->p_md.md_flags)); +} + +void +alpha_write_fp_c(struct proc *p, u_int64_t fp_c) +{ + u_int64_t md_flags; + + fp_c &= MDP_FP_C; + md_flags = p->p_md.md_flags; + if ((md_flags & MDP_FP_C) == fp_c) + return; + p->p_md.md_flags = (md_flags & ~MDP_FP_C) | fp_c; + alpha_enable_fp(p, 1); + fp_c_to_fpcr(p); + alpha_pal_wrfen(0); +} + +u_int64_t +alpha_read_fp_c(struct proc *p) +{ + /* + * A possibly-desireable EV6-specific optimization would deviate from + * the Alpha Architecture spec and keep some FP_C bits in the FPCR, + * but in a transparent way. Some of the code for that would need to + * go right here. + */ + return p->p_md.md_flags & MDP_FP_C; +} + +static float64 +float64_unk(float64 a, float64 b) +{ + return 0; +} + +/* + * The real function field encodings for IEEE and VAX FP instructions. + * + * Since there is only one operand type field, the cvtXX instructions + * require a variety of special cases, and these have to be analyzed as + * they don't always fit into the field descriptions in AARM section I. + * + * Lots of staring at bits in the appendix shows what's really going on. + * + * | | + * 15 14 13|12 11 10 09|08 07 06 05 + * --------======------============ + * TRAP : RND : SRC : FUNCTION : + * 0 0 0:. . .:. . . . . . . . . . . . Imprecise + * 0 0 1|. . .:. . . . . . . . . . . ./U underflow enable (if FP output) + * | /V overfloat enable (if int output) + * 0 1 0:. . .:. . . . . . . . . . . ."Unsupported", but used for CVTST + * 0 1 1|. . .:. . . . . . . . . . . . Unsupported + * 1 0 0:. . .:. . . . . . . . . . . ./S software completion (VAX only) + * 1 0 1|. . .:. . . . . . . . . . . ./SU + * | /SV + * 1 1 0:. . .:. . . . . . . . . . . ."Unsupported", but used for CVTST/S + * 1 1 1|. . .:. . . . . . . . . . . ./SUI (if FP output) (IEEE only) + * | /SVI (if int output) (IEEE only) + * S I UV: In other words: bits 15:13 are S:I:UV, except that _usually_ + * | not all combinations are valid. + * | | + * 15 14 13|12 11 10 09|08 07 06 05 + * --------======------============ + * TRAP : RND : SRC : FUNCTION : + * | 0 0 . . . . . . . . . . . ./C Chopped + * : 0 1 . . . . . . . . . . . ./M Minus Infinity + * | 1 0 . . . . . . . . . . . . Normal + * : 1 1 . . . . . . . . . . . ./D Dynamic (in FPCR: Plus Infinity) + * | | + * 15 14 13|12 11 10 09|08 07 06 05 + * --------======------============ + * TRAP : RND : SRC : FUNCTION : + * 0 0. . . . . . . . . . S/F + * 0 1. . . . . . . . . . -/D + * 1 0. . . . . . . . . . T/G + * 1 1. . . . . . . . . . Q/Q + * | | + * 15 14 13|12 11 10 09|08 07 06 05 + * --------======------============ + * TRAP : RND : SRC : FUNCTION : + * 0 0 0 0 . . . addX + * 0 0 0 1 . . . subX + * 0 0 1 0 . . . mulX + * 0 0 1 1 . . . divX + * 0 1 0 0 . . . cmpXun + * 0 1 0 1 . . . cmpXeq + * 0 1 1 0 . . . cmpXlt + * 0 1 1 1 . . . cmpXle + * 1 0 0 0 . . . reserved + * 1 0 0 1 . . . reserved + * 1 0 1 0 . . . sqrt[fg] (op_fix, not exactly "vax") + * 1 0 1 1 . . . sqrt[st] (op_fix, not exactly "ieee") + * 1 1 0 0 . . . cvtXs/f (cvt[qt]s, cvtst(!), cvt[gq]f) + * 1 1 0 1 . . . cvtXd (vax only) + * 1 1 1 0 . . . cvtXt/g (cvtqt, cvt[dq]g only) + * 1 1 1 1 . . . cvtXq/q (cvttq, cvtgq) + * | | + * 15 14 13|12 11 10 09|08 07 06 05 the twilight zone + * --------======------============ + * TRAP : RND : SRC : FUNCTION : + * /s /i /u x x 1 0 1 1 0 0 . . . cvtts, /siu only 0, 1, 5, 7 + * 0 1 0 1 0 1 0 1 1 0 0 . . . cvtst (src == T (!)) 2ac NOT /S + * 1 1 0 1 0 1 0 1 1 0 0 . . . cvtst/s (src == T (!)) 6ac + * x 0 x x x x 0 1 1 1 1 . . . cvttq/_ (src == T) + */ + +static void +alpha_fp_interpret(alpha_instruction *pc, struct proc *p, u_int64_t bits) +{ + s_float sfa, sfb, sfc; + t_float tfa, tfb, tfc; + alpha_instruction inst; + + inst.bits = bits; + switch(inst.generic_format.opcode) { + default: + /* this "cannot happen" */ + this_cannot_happen(2, inst.bits); + return; + case op_any_float: + if (inst.float_format.function == op_cvtql_sv || + inst.float_format.function == op_cvtql_v) { + alpha_stt(inst.float_detail.fb, &tfb); + sfc.i = (int64_t)tfb.i >= 0L ? INT_MAX : INT_MIN; + alpha_lds(inst.float_detail.fc, &sfc); + float_raise(FP_X_INV); + } else { + ++alpha_shadow.nilanyop; + this_cannot_happen(3, inst.bits); + } + break; + case op_vax_float: + ++alpha_shadow.vax; /* fall thru */ + case op_ieee_float: + case op_fix_float: + switch(inst.float_detail.src) { + case op_src_sf: + sts(inst.float_detail.fb, &sfb, p); + if (inst.float_detail.opclass == 10) + sfc.i = float32_sqrt(sfb.i); + else if (inst.float_detail.opclass & ~3) { + this_cannot_happen(1, inst.bits); + sfc.i = FLOAT32QNAN; + } else { + sts(inst.float_detail.fa, &sfa, p); + sfc.i = (*swfp_s[inst.float_detail.opclass])( + sfa.i, sfb.i); + } + lds(inst.float_detail.fc, &sfc, p); + break; + case op_src_xd: + case op_src_tg: + if (inst.float_detail.opclass >= 12) + (*swfp_cvt[inst.float_detail.opclass - 12])( + inst.bits, p); + else { + stt(inst.float_detail.fb, &tfb, p); + if (inst.float_detail.opclass == 10) + tfc.i = float64_sqrt(tfb.i); + else { + stt(inst.float_detail.fa, &tfa, p); + tfc.i = (*swfp_t[inst.float_detail + .opclass])(tfa.i, tfb.i); + } + ldt(inst.float_detail.fc, &tfc, p); + } + break; + case op_src_qq: + float_raise(FP_X_IMP); + break; + } + } +} + +static int +alpha_fp_complete_at(alpha_instruction *trigger_pc, struct proc *p, + u_int64_t *ucode) +{ + int needsig; + alpha_instruction inst; + u_int64_t rm, fpcr, orig_fpcr; + u_int64_t orig_flags, new_flags, changed_flags, md_flags; + + if (__predict_false(copyin(trigger_pc, &inst, sizeof inst))) { + this_cannot_happen(6, -1); + return SIGSEGV; + } + alpha_enable_fp(p, 1); + /* + * If necessary, lie about the dynamic rounding mode so emulation + * software need go to only one place for it, and so we don't have to + * lock any memory locations or pass a third parameter to every + * SoftFloat entry point. + */ + orig_fpcr = fpcr = alpha_read_fpcr(); + rm = inst.float_detail.rnd; + if (__predict_false(rm != 3 /* dynamic */ && rm != (fpcr >> 58 & 3))) { + fpcr = (fpcr & ~FPCR_DYN(3)) | FPCR_DYN(rm); + alpha_write_fpcr(fpcr); + } + orig_flags = FP_C_TO_OPENBSD_FLAG(p->p_md.md_flags); + + alpha_fp_interpret(trigger_pc, p, inst.bits); + + md_flags = p->p_md.md_flags; + + new_flags = FP_C_TO_OPENBSD_FLAG(md_flags); + changed_flags = orig_flags ^ new_flags; + KASSERT((orig_flags | changed_flags) == new_flags); /* panic on 1->0 */ + alpha_write_fpcr(fp_c_to_fpcr_1(orig_fpcr, md_flags)); + needsig = changed_flags & FP_C_TO_OPENBSD_MASK(md_flags); + alpha_pal_wrfen(0); + if (__predict_false(needsig)) { + *ucode = needsig; + return SIGFPE; + } + return 0; +} + +int +alpha_fp_complete(u_long a0, u_long a1, struct proc *p, u_int64_t *ucode) +{ + int t; + int sig; + u_int64_t op_class; + alpha_instruction inst; + /* "trigger_pc" is Compaq's term for the earliest faulting op */ + alpha_instruction *trigger_pc, *usertrap_pc; + alpha_instruction *pc, *win_begin, tsw[TSWINSIZE]; + + sig = SIGFPE; + pc = (alpha_instruction *)p->p_md.md_tf->tf_regs[FRAME_PC]; + trigger_pc = pc - 1; /* for ALPHA_AMASK_PAT case */ + if (cpu_amask & ALPHA_AMASK_PAT) { + if (a0 & 1 || alpha_fp_sync_complete) { + sig = alpha_fp_complete_at(trigger_pc, p, ucode); + goto done; + } + } + *ucode = a0; + if (!(a0 & 1)) + return sig; +/* + * At this point we are somwhere in the trap shadow of one or more instruc- + * tions that have trapped with software completion specified. We have a mask + * of the registers written by trapping instructions. + * + * Now step backwards through the trap shadow, clearing bits in the + * destination write mask until the trigger instruction is found, and + * interpret this one instruction in SW. If a SIGFPE is not required, back up + * the PC until just after this instruction and restart. This will execute all + * trap shadow instructions between the trigger pc and the trap pc twice. + * + * If a SIGFPE is generated from the OSF1 emulation, back up one more + * instruction to the trigger pc itself. Native binaries don't because it + * is non-portable and completely defeats the intended purpose of IEEE + * traps -- for example, to count the number of exponent wraps for a later + * correction. + */ + trigger_pc = 0; + win_begin = pc; + ++alpha_shadow.scans; + t = alpha_shadow.len; + for (--pc; a1; --pc) { + ++alpha_shadow.len; + if (pc < win_begin) { + win_begin = pc - TSWINSIZE + 1; + if (copyin(win_begin, tsw, sizeof tsw)) { + /* sigh, try to get just one */ + win_begin = pc; + if (copyin(win_begin, tsw, 4)) + return SIGSEGV; + } + } + assert(win_begin <= pc && !((long)pc & 3)); + inst = tsw[pc - win_begin]; + op_class = 1UL << inst.generic_format.opcode; + if (op_class & FPUREG_CLASS) { + a1 &= ~(1UL << (inst.operate_generic_format.rc + 32)); + trigger_pc = pc; + } else if (op_class & CPUREG_CLASS) { + a1 &= ~(1UL << inst.operate_generic_format.rc); + trigger_pc = pc; + } else if (op_class & TRAPSHADOWBOUNDARY) { + if (op_class & CHECKFUNCTIONCODE) { + if (inst.mem_format.displacement == op_trapb || + inst.mem_format.displacement == op_excb) + break; /* code breaks AARM rules */ + } else + break; /* code breaks AARM rules */ + } + /* Some shadow-safe op, probably load, store, or FPTI class */ + } + t = alpha_shadow.len - t; + if (t > alpha_shadow.max) + alpha_shadow.max = t; + if (__predict_true(trigger_pc != 0 && a1 == 0)) { + ++alpha_shadow.resolved; + sig = alpha_fp_complete_at(trigger_pc, p, ucode); + } else { + ++alpha_shadow.unresolved; + return sig; + } +done: + if (sig) { + usertrap_pc = trigger_pc + 1; + p->p_md.md_tf->tf_regs[FRAME_PC] = (unsigned long)usertrap_pc; + return sig; + } + return 0; +} +#endif diff --git a/sys/arch/alpha/alpha/locore.s b/sys/arch/alpha/alpha/locore.s index 8c1d9ddc315..4d38766d884 100644 --- a/sys/arch/alpha/alpha/locore.s +++ b/sys/arch/alpha/alpha/locore.s @@ -1,5 +1,5 @@ -/* $OpenBSD: locore.s,v 1.17 2001/09/30 13:08:45 art Exp $ */ -/* $NetBSD: locore.s,v 1.80 2000/09/04 00:31:59 thorpej Exp $ */ +/* $OpenBSD: locore.s,v 1.18 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: locore.s,v 1.94 2001/04/26 03:10:44 ross Exp $ */ /*- * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc. @@ -1906,6 +1906,63 @@ longjmp_botchmsg: .text END(longjmp) +/* + * void sts(int rn, u_int32_t *rval); + * void stt(int rn, u_int64_t *rval); + * void lds(int rn, u_int32_t *rval); + * void ldt(int rn, u_int64_t *rval); + */ + +#ifndef NO_IEEE +.macro make_freg_util name, op + LEAF(alpha_\name, 2) + and a0, 0x1f, a0 + s8addq a0, pv, pv + addq pv, 1f - alpha_\name, pv + jmp (pv) +1: + rn = 0 + .rept 32 + \op $f0 + rn, 0(a1) + RET + rn = rn + 1 + .endr + END(alpha_\name) +.endm +/* +LEAF(alpha_sts, 2) +LEAF(alpha_stt, 2) +LEAF(alpha_lds, 2) +LEAF(alpha_ldt, 2) + */ + make_freg_util sts, sts + make_freg_util stt, stt + make_freg_util lds, lds + make_freg_util ldt, ldt + +LEAF(alpha_read_fpcr, 0); f30save = 0; rettmp = 8; framesz = 16 + lda sp, -framesz(sp) + stt $f30, f30save(sp) + mf_fpcr $f30 + stt $f30, rettmp(sp) + ldt $f30, f30save(sp) + ldq v0, rettmp(sp) + lda sp, framesz(sp) + RET +END(alpha_read_fpcr) + +LEAF(alpha_write_fpcr, 1); f30save = 0; fpcrtmp = 8; framesz = 16 + lda sp, -framesz(sp) + stq a0, fpcrtmp(sp) + stt $f30, f30save(sp) + ldt $f30, fpcrtmp(sp) + mt_fpcr $f30 + ldt $f30, f30save(sp) + lda sp, framesz(sp) + RET +END(alpha_write_fpcr) +#endif + #if 0 NESTED(transfer_check,0,0,ra,0,0) CALL(U_need_2_run_config) diff --git a/sys/arch/alpha/alpha/machdep.c b/sys/arch/alpha/alpha/machdep.c index 9918d85d027..9b56830e56b 100644 --- a/sys/arch/alpha/alpha/machdep.c +++ b/sys/arch/alpha/alpha/machdep.c @@ -1,4 +1,4 @@ -/* $OpenBSD: machdep.c,v 1.70 2002/04/25 00:53:58 miod Exp $ */ +/* $OpenBSD: machdep.c,v 1.71 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: machdep.c,v 1.210 2000/06/01 17:12:38 thorpej Exp $ */ /*- @@ -90,6 +90,9 @@ #include <sys/core.h> #include <sys/kcore.h> #include <machine/kcore.h> +#ifndef NO_IEEE +#include <machine/fpu.h> +#endif #ifdef SYSVMSG #include <sys/msg.h> #endif @@ -113,6 +116,9 @@ #include <machine/rpb.h> #include <machine/prom.h> #include <machine/cpuconf.h> +#ifndef NO_IEEE +#include <machine/ieeefp.h> +#endif #include <dev/pci/pcivar.h> @@ -201,6 +207,9 @@ struct platform platform; int alpha_unaligned_print = 1; /* warn about unaligned accesses */ int alpha_unaligned_fix = 1; /* fix up unaligned accesses */ int alpha_unaligned_sigbus = 1; /* SIGBUS on fixed-up accesses */ +#ifndef NO_IEEE +int alpha_fp_sync_complete = 0; /* fp fixup if sync even without /s */ +#endif /* * XXX This should be dynamically sized, but we have the chicken-egg problem! @@ -1597,19 +1606,18 @@ sendsig(catcher, sig, mask, code, type, val) ksc.sc_regs[R_SP] = alpha_pal_rdusp(); /* save the floating-point state, if necessary, then copy it. */ - if (p == fpcurproc) { - alpha_pal_wrfen(1); - savefpstate(&p->p_addr->u_pcb.pcb_fp); - alpha_pal_wrfen(0); - fpcurproc = NULL; - } + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 1); ksc.sc_ownedfp = p->p_md.md_flags & MDP_FPUSED; - bcopy(&p->p_addr->u_pcb.pcb_fp, (struct fpreg *)ksc.sc_fpregs, + memcpy((struct fpreg *)ksc.sc_fpregs, &p->p_addr->u_pcb.pcb_fp, sizeof(struct fpreg)); - ksc.sc_fp_control = 0; /* XXX ? */ - bzero(ksc.sc_reserved, sizeof ksc.sc_reserved); /* XXX */ - bzero(ksc.sc_xxx, sizeof ksc.sc_xxx); /* XXX */ - +#ifndef NO_IEEE + ksc.sc_fp_control = alpha_read_fp_c(p); +#else + ksc.sc_fp_control = 0; +#endif + memset(ksc.sc_reserved, 0, sizeof ksc.sc_reserved); /* XXX */ + memset(ksc.sc_xxx, 0, sizeof ksc.sc_xxx); /* XXX */ #ifdef COMPAT_OSF1 /* @@ -1713,11 +1721,14 @@ sys_sigreturn(p, v, retval) alpha_pal_wrusp(ksc.sc_regs[R_SP]); /* XXX ksc.sc_ownedfp ? */ - if (p == fpcurproc) - fpcurproc = NULL; - bcopy((struct fpreg *)ksc.sc_fpregs, &p->p_addr->u_pcb.pcb_fp, + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 0); + memcpy(&p->p_addr->u_pcb.pcb_fp, (struct fpreg *)ksc.sc_fpregs, sizeof(struct fpreg)); - /* XXX ksc.sc_fp_control ? */ +#ifndef NO_IEEE + p->p_addr->u_pcb.pcb_fp.fpr_cr = ksc.sc_fpcr; + p->p_md.md_flags = ksc.sc_fp_control & MDP_FP_C; +#endif #ifdef DEBUG if (sigdebug & SDB_FOLLOW) @@ -1772,10 +1783,17 @@ cpu_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) case CPU_BOOTED_KERNEL: return (sysctl_rdstring(oldp, oldlenp, newp, bootinfo.booted_kernel)); - + case CPU_CHIPSET: return (alpha_sysctl_chipset(name + 1, namelen - 1, oldp, oldlenp)); + +#ifndef NO_IEEE + case CPU_FP_SYNC_COMPLETE: + return (sysctl_int(oldp, oldlenp, newp, newlen, + &alpha_fp_sync_complete)); +#endif + default: return (EOPNOTSUPP); } @@ -1812,8 +1830,6 @@ setregs(p, pack, stack, retval) bzero(tfp->tf_regs, FRAME_SIZE * sizeof tfp->tf_regs[0]); #endif bzero(&p->p_addr->u_pcb.pcb_fp, sizeof p->p_addr->u_pcb.pcb_fp); -#define FP_RN 2 /* XXX */ - p->p_addr->u_pcb.pcb_fp.fpr_cr = (long)FP_RN << 58; alpha_pal_wrusp(stack); tfp->tf_regs[FRAME_PS] = ALPHA_PSL_USERSET; tfp->tf_regs[FRAME_PC] = pack->ep_entry & ~3; @@ -1823,10 +1839,96 @@ setregs(p, pack, stack, retval) tfp->tf_regs[FRAME_T12] = tfp->tf_regs[FRAME_PC]; /* a.k.a. PV */ p->p_md.md_flags &= ~MDP_FPUSED; - if (fpcurproc == p) - fpcurproc = NULL; +#ifndef NO_IEEE + if (__predict_true((p->p_md.md_flags & IEEE_INHERIT) == 0)) { + p->p_md.md_flags &= ~MDP_FP_C; + p->p_addr->u_pcb.pcb_fp.fpr_cr = FPCR_DYN(FP_RN); + } +#endif + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 0); +} + +/* + * Release the FPU. + */ +void +fpusave_cpu(struct cpu_info *ci, int save) +{ + struct proc *p; +#if defined(MULTIPROCESSOR) + int s; +#endif + + KDASSERT(ci == curcpu()); + +#if defined(MULTIPROCESSOR) + atomic_setbits_ulong(&ci->ci_flags, CPUF_FPUSAVE); +#endif - retval[0] = retval[1] = 0; + p = ci->ci_fpcurproc; + if (p == NULL) + goto out; + + if (save) { + alpha_pal_wrfen(1); + savefpstate(&p->p_addr->u_pcb.pcb_fp); + } + + alpha_pal_wrfen(0); + + p->p_addr->u_pcb.pcb_fpcpu = NULL; + ci->ci_fpcurproc = NULL; + +out: +#if defined(MULTIPROCESSOR) + atomic_clearbits_ulong(&ci->ci_flags, CPUF_FPUSAVE); +#endif + return; +} + +/* + * Synchronize FP state for this process. + */ +void +fpusave_proc(struct proc *p, int save) +{ + struct cpu_info *ci = curcpu(); + struct cpu_info *oci; +#if defined(MULTIPROCESSOR) + u_long ipi = save ? ALPHA_IPI_SYNCH_FPU : ALPHA_IPI_DISCARD_FPU; + int s, spincount; +#endif + + KDASSERT(p->p_addr != NULL); + KDASSERT(p->p_flag & P_INMEM); + + oci = p->p_addr->u_pcb.pcb_fpcpu; + if (oci == NULL) { + return; + } + +#if defined(MULTIPROCESSOR) + if (oci == ci) { + KASSERT(ci->ci_fpcurproc == p); + fpusave_cpu(ci, save); + return; + } + + KASSERT(oci->ci_fpcurproc == p); + alpha_send_ipi(oci->ci_cpuid, ipi); + + spincount = 0; + while (p->p_addr->u_pcb.pcb_fpcpu != NULL) { + spincount++; + delay(1000); /* XXX */ + if (spincount > 10000) + panic("fpsave ipi didn't"); + } +#else + KASSERT(ci->ci_fpcurproc == p); + fpusave_cpu(ci, save); +#endif /* MULTIPROCESSOR */ } int diff --git a/sys/arch/alpha/alpha/process_machdep.c b/sys/arch/alpha/alpha/process_machdep.c index ed6817e9bcd..a99c685acfb 100644 --- a/sys/arch/alpha/alpha/process_machdep.c +++ b/sys/arch/alpha/alpha/process_machdep.c @@ -1,4 +1,4 @@ -/* $OpenBSD: process_machdep.c,v 1.8 2002/03/14 06:04:11 mickey Exp $ */ +/* $OpenBSD: process_machdep.c,v 1.9 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: process_machdep.c,v 1.7 1996/07/11 20:14:21 cgd Exp $ */ /*- @@ -154,8 +154,8 @@ process_write_fpregs(p, regs) struct fpreg *regs; { - if (p == fpcurproc) - fpcurproc = NULL; + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 1); bcopy(regs, process_fpframe(p), sizeof(struct fpreg)); return (0); @@ -333,6 +333,8 @@ process_sstep(struct proc *p, int sstep) count = 1; } + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 0); p->p_md.md_sstep[0].addr = addr[0]; error = ptrace_set_bpt(p, &p->p_md.md_sstep[0]); if (error) diff --git a/sys/arch/alpha/alpha/sys_machdep.c b/sys/arch/alpha/alpha/sys_machdep.c index 5a2bc9f6f28..587cd20986e 100644 --- a/sys/arch/alpha/alpha/sys_machdep.c +++ b/sys/arch/alpha/alpha/sys_machdep.c @@ -1,5 +1,41 @@ -/* $OpenBSD: sys_machdep.c,v 1.5 1997/01/24 19:56:44 niklas Exp $ */ -/* $NetBSD: sys_machdep.c,v 1.5 1996/11/13 22:20:57 cgd Exp $ */ +/* $OpenBSD: sys_machdep.c,v 1.6 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: sys_machdep.c,v 1.14 2002/01/14 00:53:16 thorpej Exp $ */ + +/*- + * Copyright (c) 2000 The NetBSD Foundation, Inc. + * All rights reserved. + * + * This code is derived from software contributed to The NetBSD Foundation + * by Jason R. Thorpe. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ /* * Copyright (c) 1994, 1995, 1996 Carnegie-Mellon University. @@ -30,22 +66,92 @@ #include <sys/param.h> #include <sys/systm.h> +#ifndef NO_IEEE +#include <sys/device.h> +#include <sys/proc.h> +#endif #include <sys/mount.h> #include <sys/syscallargs.h> +#ifndef NO_IEEE +#include <machine/fpu.h> +#include <machine/sysarch.h> + +#include <dev/pci/pcivar.h> + int -sys_sysarch(p, v, retval) - struct proc *p; - void *v; - register_t *retval; +sys_sysarch(struct proc *p, void *v, register_t *retval) { -#if 0 struct sys_sysarch_args /* { syscallarg(int) op; - syscallarg(char *) parms; + syscallarg(void *) parms; } */ *uap = v; -#endif + int error = 0; + + switch(SCARG(uap, op)) { + case ALPHA_FPGETMASK: + *retval = FP_C_TO_OPENBSD_MASK(p->p_md.md_flags); + break; + case ALPHA_FPGETSTICKY: + *retval = FP_C_TO_OPENBSD_FLAG(p->p_md.md_flags); + break; + case ALPHA_FPSETMASK: + case ALPHA_FPSETSTICKY: + { + fp_except m; + u_int64_t md_flags; + struct alpha_fp_except_args args; + + error = copyin(SCARG(uap, parms), &args, sizeof args); + if (error) + return error; + m = args.mask; + md_flags = p->p_md.md_flags; + switch (SCARG(uap, op)) { + case ALPHA_FPSETMASK: + *retval = FP_C_TO_OPENBSD_MASK(md_flags); + md_flags = SET_FP_C_MASK(md_flags, m); + break; + case ALPHA_FPSETSTICKY: + *retval = FP_C_TO_OPENBSD_FLAG(md_flags); + md_flags = SET_FP_C_FLAG(md_flags, m); + break; + } + alpha_write_fp_c(p, md_flags); + break; + } + case ALPHA_GET_FP_C: + { + struct alpha_fp_c_args args; + + args.fp_c = alpha_read_fp_c(p); + error = copyout(&args, SCARG(uap, parms), sizeof args); + break; + } + case ALPHA_SET_FP_C: + { + struct alpha_fp_c_args args; + + error = copyin(SCARG(uap, parms), &args, sizeof args); + if (error) + return (error); + if ((args.fp_c >> 63) != 0) + args.fp_c |= IEEE_INHERIT; + alpha_write_fp_c(p, args.fp_c); + break; + } + default: + error = EINVAL; + break; + } + + return (error); +} +#else +int sys_sysarch(struct proc *p, void *v, register_t *retval) +{ return (ENOSYS); } +#endif diff --git a/sys/arch/alpha/alpha/trap.c b/sys/arch/alpha/alpha/trap.c index d6722a9c831..d6d06f2521e 100644 --- a/sys/arch/alpha/alpha/trap.c +++ b/sys/arch/alpha/alpha/trap.c @@ -1,4 +1,4 @@ -/* $OpenBSD: trap.c,v 1.32 2002/03/16 03:21:28 art Exp $ */ +/* $OpenBSD: trap.c,v 1.33 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: trap.c,v 1.52 2000/05/24 16:48:33 thorpej Exp $ */ /*- @@ -102,6 +102,9 @@ #include <sys/user.h> #include <sys/syscall.h> #include <sys/buf.h> +#ifndef NO_IEEE +#include <sys/device.h> +#endif #ifdef KTRACE #include <sys/ktrace.h> #endif @@ -114,7 +117,7 @@ #ifdef DDB #include <machine/db_machdep.h> #endif -#include <alpha/alpha/db_instruction.h> /* for handle_opdec() */ +#include <alpha/alpha/db_instruction.h> #ifdef COMPAT_OSF1 #include <compat/osf1/osf1_syscall.h> @@ -135,6 +138,11 @@ int unaligned_fixup(unsigned long, unsigned long, unsigned long, struct proc *); int handle_opdec(struct proc *p, u_int64_t *ucodep); +#ifndef NO_IEEE +struct device fpevent_use; +struct device fpevent_reuse; +#endif + static void printtrap(const unsigned long, const unsigned long, const unsigned long, const unsigned long, struct trapframe *, int, int); @@ -331,21 +339,19 @@ trap(a0, a1, a2, entry, framep) goto dopanic; case ALPHA_KENTRY_ARITH: - /* - * If user-land, just give a SIGFPE. Should do - * software completion and IEEE handling, if the - * user has requested that. + /* + * Resolve trap shadows, interpret FP ops requiring infinities, + * NaNs, or denorms, and maintain FPCR corrections. */ if (user) { -#ifdef COMPAT_OSF1 - extern struct emul emul_osf1; - - /* just punt on OSF/1. XXX THIS IS EVIL */ - if (p->p_emul == &emul_osf1) +#ifndef NO_IEEE + i = alpha_fp_complete(a0, a1, p, &ucode); + if (i == 0) goto out; -#endif +#else i = SIGFPE; - ucode = a0; /* exception summary */ + ucode = a0; +#endif break; } @@ -401,6 +407,10 @@ trap(a0, a1, a2, entry, framep) break; case ALPHA_IF_CODE_FEN: +#ifndef NO_IEEE + alpha_enable_fp(p, 0); + alpha_pal_wrfen(0); +#else /* * on exit from the kernel, if proc == fpcurproc, * FP is enabled. @@ -410,7 +420,7 @@ trap(a0, a1, a2, entry, framep) p); goto dopanic; } - + alpha_pal_wrfen(1); if (fpcurproc) savefpstate(&fpcurproc->p_addr->u_pcb.pcb_fp); @@ -419,6 +429,7 @@ trap(a0, a1, a2, entry, framep) alpha_pal_wrfen(0); p->p_md.md_flags |= MDP_FPUSED; +#endif goto out; default: @@ -751,6 +762,45 @@ child_return(arg) #endif } +#ifndef NO_IEEE +/* + * Set the float-point enable for the current process, and return + * the FPU context to the named process. If check == 0, it is an + * error for the named process to already be fpcurproc. + */ +void +alpha_enable_fp(struct proc *p, int check) +{ + struct cpu_info *ci = curcpu(); + + if (check && ci->ci_fpcurproc == p) { + alpha_pal_wrfen(1); + return; + } + if (ci->ci_fpcurproc == p) + panic("trap: fp disabled for fpcurproc == %p", p); + + if (ci->ci_fpcurproc != NULL) + fpusave_cpu(ci, 1); + + KDASSERT(ci->ci_fpcurproc == NULL); + +#if defined(MULTIPROCESSOR) + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 1); +#else + KDASSERT(p->p_addr->u_pcb.pcb_fpcpu == NULL); +#endif + + p->p_addr->u_pcb.pcb_fpcpu = ci; + ci->ci_fpcurproc = p; + + p->p_md.md_flags |= MDP_FPUSED; + alpha_pal_wrfen(1); + restorefpstate(&p->p_addr->u_pcb.pcb_fp); +} +#endif + /* * Process an asynchronous software trap. * This is relatively easy. @@ -804,12 +854,8 @@ const static int reg_to_framereg[32] = { (&(p)->p_addr->u_pcb.pcb_fp.fpr_regs[(reg)]) #define dump_fp_regs() \ - if (p == fpcurproc) { \ - alpha_pal_wrfen(1); \ - savefpstate(&fpcurproc->p_addr->u_pcb.pcb_fp); \ - alpha_pal_wrfen(0); \ - fpcurproc = NULL; \ - } + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) \ + fpusave_proc(p, 1); #define unaligned_load(storage, ptrf, mod) \ if (copyin((caddr_t)va, &(storage), sizeof (storage)) != 0) \ @@ -957,9 +1003,6 @@ Gfloat_reg_cvt(input) } #endif /* FIX_UNALIGNED_VAX_FP */ -extern int alpha_unaligned_print, alpha_unaligned_fix; -extern int alpha_unaligned_sigbus; - struct unaligned_fixup_data { const char *type; /* opcode name */ int fixable; /* fixable, 0 if fixup not supported */ diff --git a/sys/arch/alpha/alpha/vm_machdep.c b/sys/arch/alpha/alpha/vm_machdep.c index 05d78106c69..afab58a822f 100644 --- a/sys/arch/alpha/alpha/vm_machdep.c +++ b/sys/arch/alpha/alpha/vm_machdep.c @@ -1,4 +1,4 @@ -/* $OpenBSD: vm_machdep.c,v 1.27 2001/12/08 02:24:05 art Exp $ */ +/* $OpenBSD: vm_machdep.c,v 1.28 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: vm_machdep.c,v 1.55 2000/03/29 03:49:48 simonb Exp $ */ /* @@ -68,12 +68,9 @@ cpu_coredump(p, vp, cred, chdr) cpustate.md_tf = *p->p_md.md_tf; cpustate.md_tf.tf_regs[FRAME_SP] = alpha_pal_rdusp(); /* XXX */ if (p->p_md.md_flags & MDP_FPUSED) { - if (p == fpcurproc) { - alpha_pal_wrfen(1); - savefpstate(&cpustate.md_fpstate); - alpha_pal_wrfen(0); - } else - cpustate.md_fpstate = p->p_addr->u_pcb.pcb_fp; + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 1); + cpustate.md_fpstate = p->p_addr->u_pcb.pcb_fp; } else bzero(&cpustate.md_fpstate, sizeof(cpustate.md_fpstate)); @@ -108,8 +105,8 @@ cpu_exit(p) struct proc *p; { - if (p == fpcurproc) - fpcurproc = NULL; + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 0); /* * Deactivate the exiting address space before the vmspace @@ -150,7 +147,12 @@ cpu_fork(p1, p2, stack, stacksize, func, arg) struct user *up = p2->p_addr; p2->p_md.md_tf = p1->p_md.md_tf; + +#ifndef NO_IEEE + p2->p_md.md_flags = p1->p_md.md_flags & (MDP_FPUSED | MDP_FP_C); +#else p2->p_md.md_flags = p1->p_md.md_flags & MDP_FPUSED; +#endif /* * Cache the physical address of the pcb, so we can @@ -162,11 +164,8 @@ cpu_fork(p1, p2, stack, stacksize, func, arg) * Copy floating point state from the FP chip to the PCB * if this process has state stored there. */ - if (p1 == fpcurproc) { - alpha_pal_wrfen(1); - savefpstate(&fpcurproc->p_addr->u_pcb.pcb_fp); - alpha_pal_wrfen(0); - } + if (p1->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p1, 1); /* * Copy pcb and stack from proc p1 to p2. @@ -265,13 +264,8 @@ cpu_swapout(p) struct proc *p; { - if (p != fpcurproc) - return; - - alpha_pal_wrfen(1); - savefpstate(&fpcurproc->p_addr->u_pcb.pcb_fp); - alpha_pal_wrfen(0); - fpcurproc = NULL; + if (p->p_addr->u_pcb.pcb_fpcpu != NULL) + fpusave_proc(p, 1); } /* diff --git a/sys/arch/alpha/conf/RAMDISK b/sys/arch/alpha/conf/RAMDISK index 7c681de4f26..231d1c8a513 100644 --- a/sys/arch/alpha/conf/RAMDISK +++ b/sys/arch/alpha/conf/RAMDISK @@ -1,4 +1,4 @@ -# $OpenBSD: RAMDISK,v 1.55 2002/03/30 20:21:25 deraadt Exp $ +# $OpenBSD: RAMDISK,v 1.56 2002/04/28 20:55:14 pvalchev Exp $ # $NetBSD: RAMDISK,v 1.9 1996/12/03 17:25:33 cgd Exp $ machine alpha # architecture, used by config; REQUIRED @@ -21,6 +21,8 @@ option DEC_550 # Miata: Digital Personal Workstation option RAMDISK_HOOKS option MINIROOTSIZE=5744 # 4 Megabytes! +option NO_IEEE # Disable IEEE math + # Standard system options maxusers 8 # estimated number of users diff --git a/sys/arch/alpha/conf/RAMDISKB b/sys/arch/alpha/conf/RAMDISKB index 7a7759e2b9a..d32a646f415 100644 --- a/sys/arch/alpha/conf/RAMDISKB +++ b/sys/arch/alpha/conf/RAMDISKB @@ -1,4 +1,4 @@ -# $OpenBSD: RAMDISKB,v 1.18 2002/03/30 20:21:25 deraadt Exp $ +# $OpenBSD: RAMDISKB,v 1.19 2002/04/28 20:55:14 pvalchev Exp $ # $NetBSD: RAMDISK,v 1.9 1996/12/03 17:25:33 cgd Exp $ machine alpha # architecture, used by config; REQUIRED @@ -21,6 +21,8 @@ option API_UP1000 # EV6: Alpha Processor UP1000 option RAMDISK_HOOKS option MINIROOTSIZE=5744 # 4 Megabytes! +option NO_IEEE # Disable IEEE math + # Standard system options maxusers 8 # estimated number of users diff --git a/sys/arch/alpha/conf/RAMDISKBIG b/sys/arch/alpha/conf/RAMDISKBIG index 364066a2c33..fc4d1cba8c2 100644 --- a/sys/arch/alpha/conf/RAMDISKBIG +++ b/sys/arch/alpha/conf/RAMDISKBIG @@ -1,4 +1,4 @@ -# $OpenBSD: RAMDISKBIG,v 1.24 2002/04/02 17:14:48 deraadt Exp $ +# $OpenBSD: RAMDISKBIG,v 1.25 2002/04/28 20:55:14 pvalchev Exp $ # $NetBSD: GENERIC,v 1.31 1996/12/03 17:25:29 cgd Exp $ # # Generic Alpha kernel. Enough to get booted, etc., but not much more. @@ -21,6 +21,8 @@ option API_UP1000 # EV6: Alpha Processor UP1000 option RAMDISK_HOOKS option MINIROOTSIZE=5744 # 4 Megabytes! +option NO_IEEE # Disable IEEE math + # Standard system options maxusers 8 # estimated number of users diff --git a/sys/arch/alpha/conf/files.alpha b/sys/arch/alpha/conf/files.alpha index e643643f3d4..5249e2ecadd 100644 --- a/sys/arch/alpha/conf/files.alpha +++ b/sys/arch/alpha/conf/files.alpha @@ -1,4 +1,4 @@ -# $OpenBSD: files.alpha,v 1.56 2002/03/23 14:14:25 deraadt Exp $ +# $OpenBSD: files.alpha,v 1.57 2002/04/28 20:55:14 pvalchev Exp $ # $NetBSD: files.alpha,v 1.32 1996/11/25 04:03:21 cgd Exp $ # # alpha-specific configuration info @@ -288,6 +288,7 @@ file arch/alpha/alpha/process_machdep.c file arch/alpha/alpha/prom.c file arch/alpha/alpha/sys_machdep.c file arch/alpha/alpha/trap.c +file arch/alpha/alpha/fp_complete.c file arch/alpha/alpha/vm_machdep.c file arch/alpha/alpha/disksubr.c file arch/alpha/dev/bus_dma.c diff --git a/sys/arch/alpha/include/cpu.h b/sys/arch/alpha/include/cpu.h index b599120df2e..4a4ff665401 100644 --- a/sys/arch/alpha/include/cpu.h +++ b/sys/arch/alpha/include/cpu.h @@ -1,4 +1,4 @@ -/* $OpenBSD: cpu.h,v 1.15 2001/11/06 18:41:09 art Exp $ */ +/* $OpenBSD: cpu.h,v 1.16 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: cpu.h,v 1.45 2000/08/21 02:03:12 thorpej Exp $ */ /*- @@ -83,6 +83,22 @@ #ifndef _ALPHA_CPU_H_ #define _ALPHA_CPU_H_ +#ifndef NO_IEEE +typedef union alpha_s_float { + u_int32_t i; + u_int32_t frac: 23, + exp: 8, + sign: 1; +} s_float; + +typedef union alpha_t_float { + u_int64_t i; + u_int64_t frac: 52, + exp: 11, + sign: 1; +} t_float; +#endif + /* * Exported definitions unique to Alpha cpu support. */ @@ -100,7 +116,11 @@ struct reg; struct rpb; struct trapframe; +extern u_long cpu_implver; /* from IMPLVER instruction */ +extern u_long cpu_amask; /* from AMASK instruction */ extern int bootdev_debug; +extern int alpha_fp_sync_complete; +extern int alpha_unaligned_print, alpha_unaligned_fix, alpha_unaligned_sigbus; void XentArith(u_int64_t, u_int64_t, u_int64_t); /* MAGIC */ void XentIF(u_int64_t, u_int64_t, u_int64_t); /* MAGIC */ @@ -195,6 +215,11 @@ struct cpu_info { #define CPUF_PRIMARY 0x01 /* CPU is primary CPU */ #define CPUF_PRESENT 0x02 /* CPU is present */ #define CPUF_RUNNING 0x04 /* CPU is running */ +#define CPUF_PAUSED 0x08 /* CPU is paused */ +#define CPUF_FPUSAVE 0x10 /* CPU is currently in fpusave_cpu() */ + +void fpusave_cpu(struct cpu_info *, int); +void fpusave_proc(struct proc *, int); #if defined(MULTIPROCESSOR) extern __volatile u_long cpus_running; @@ -218,9 +243,6 @@ extern struct cpu_info cpu_info_store; #define fpcurproc curcpu()->ci_fpcurproc #define curpcb curcpu()->ci_curpcb -extern u_long cpu_implver; /* from IMPLVER instruction */ -extern u_long cpu_amask; /* from AMASK instruction */ - /* * definitions of cpu-dependent requirements * referenced in generic code @@ -309,8 +331,9 @@ do { \ #define CPU_UNALIGNED_FIX 4 /* int: fix unaligned accesses */ #define CPU_UNALIGNED_SIGBUS 5 /* int: SIGBUS unaligned accesses */ #define CPU_BOOTED_KERNEL 6 /* string: booted kernel name */ -#define CPU_CHIPSET 7 /* chipset information */ -#define CPU_MAXID 8 /* 6 valid machdep IDs */ +#define CPU_FP_SYNC_COMPLETE 7 /* int: always fixup sync fp traps */ +#define CPU_MAXID 8 /* 7 valid machdep IDs */ +#define CPU_CHIPSET 9 /* chipset information */ #define CPU_CHIPSET_MEM 1 /* PCI memory address */ #define CPU_CHIPSET_BWX 2 /* PCI supports BWX */ @@ -328,6 +351,7 @@ do { \ { "unaligned_sigbus", CTLTYPE_INT }, \ { "booted_kernel", CTLTYPE_STRING }, \ { "chipset", CTLTYPE_NODE }, \ + { "fp_sync_complete", CTLTYPE_INT }, \ } #ifdef _KERNEL @@ -338,5 +362,23 @@ struct reg; struct rpb; struct trapframe; +/* IEEE and VAX FP completion */ + +#ifndef NO_IEEE +void alpha_sts(int, s_float *); /* MAGIC */ +void alpha_stt(int, t_float *); /* MAGIC */ +void alpha_lds(int, s_float *); /* MAGIC */ +void alpha_ldt(int, t_float *); /* MAGIC */ + +uint64_t alpha_read_fpcr(void); /* MAGIC */ +void alpha_write_fpcr(u_int64_t); /* MAGIC */ + +u_int64_t alpha_read_fp_c(struct proc *); +void alpha_write_fp_c(struct proc *, u_int64_t); + +void alpha_enable_fp(struct proc *, int); +int alpha_fp_complete(u_long, u_long, struct proc *, u_int64_t *); +#endif + #endif /* _KERNEL */ #endif /* _ALPHA_CPU_H_ */ diff --git a/sys/arch/alpha/include/fpu.h b/sys/arch/alpha/include/fpu.h new file mode 100644 index 00000000000..1aa71c1e765 --- /dev/null +++ b/sys/arch/alpha/include/fpu.h @@ -0,0 +1,121 @@ +/* $OpenBSD: fpu.h,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: fpu.h,v 1.4 2001/04/26 03:10:46 ross Exp $ */ + +/*- + * Copyright (c) 2001 Ross Harvey + * All rights reserved. + * + * This software was written for NetBSD. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _ALPHA_FPU_H_ +#define _ALPHA_FPU_H_ + +#define _FP_C_DEF(n) (1UL << (n)) + +/* + * Most of these next definitions were moved from <ieeefp.h>. Apparently the + * names happen to match those exported by Compaq and Linux from their fpu.h + * files. + */ + +#define FPCR_SUM _FP_C_DEF(63) +#define FPCR_INED _FP_C_DEF(62) +#define FPCR_UNFD _FP_C_DEF(61) +#define FPCR_UNDZ _FP_C_DEF(60) +#define FPCR_DYN(rm) ((unsigned long)(rm) << 58) +#define FPCR_IOV _FP_C_DEF(57) +#define FPCR_INE _FP_C_DEF(56) +#define FPCR_UNF _FP_C_DEF(55) +#define FPCR_OVF _FP_C_DEF(54) +#define FPCR_DZE _FP_C_DEF(53) +#define FPCR_INV _FP_C_DEF(52) +#define FPCR_OVFD _FP_C_DEF(51) +#define FPCR_DZED _FP_C_DEF(50) +#define FPCR_INVD _FP_C_DEF(49) +#define FPCR_DNZ _FP_C_DEF(48) +#define FPCR_DNOD _FP_C_DEF(47) + +#define FPCR_MIRRORED (FPCR_INE | FPCR_UNF | FPCR_OVF | FPCR_DZE | FPCR_INV) +#define FPCR_MIR_START 52 + +/* + * The AARM specifies the bit positions of the software word used for + * user mode interface to the control and status of the kernel completion + * routines. Although it largely just redefines the FPCR, it shuffles + * the bit order. The names of the bits are defined in the AARM, and + * the definition prefix can easily be determined from public domain + * programs written to either the Compaq or Linux interfaces, which + * appear to be identical. + */ + +#define IEEE_STATUS_DNO _FP_C_DEF(22) +#define IEEE_STATUS_INE _FP_C_DEF(21) +#define IEEE_STATUS_UNF _FP_C_DEF(20) +#define IEEE_STATUS_OVF _FP_C_DEF(19) +#define IEEE_STATUS_DZE _FP_C_DEF(18) +#define IEEE_STATUS_INV _FP_C_DEF(17) + +#define IEEE_TRAP_ENABLE_DNO _FP_C_DEF(6) +#define IEEE_TRAP_ENABLE_INE _FP_C_DEF(5) +#define IEEE_TRAP_ENABLE_UNF _FP_C_DEF(4) +#define IEEE_TRAP_ENABLE_OVF _FP_C_DEF(3) +#define IEEE_TRAP_ENABLE_DZE _FP_C_DEF(2) +#define IEEE_TRAP_ENABLE_INV _FP_C_DEF(1) + +#define IEEE_INHERIT _FP_C_DEF(14) +#define IEEE_MAP_UMZ _FP_C_DEF(13) +#define IEEE_MAP_DMZ _FP_C_DEF(12) + +#define FP_C_MIRRORED (IEEE_STATUS_INE | IEEE_STATUS_UNF | IEEE_STATUS_OVF\ + | IEEE_STATUS_DZE | IEEE_STATUS_INV) +#define FP_C_MIR_START 17 + +#ifdef _KERNEL + +#define FLD_MASK(len) ((1UL << (len)) - 1) +#define FLD_CLEAR(obj, origin, len) \ + ((obj) & ~(FLD_MASK(len) << (origin))) +#define FLD_INSERT(obj, origin, len, value) \ + (FLD_CLEAR(obj, origin, len) | (value) << origin) + +#define FP_C_TO_OPENBSD_MASK(fp_c) ((fp_c) >> 1 & 0x3f) +#define FP_C_TO_OPENBSD_FLAG(fp_c) ((fp_c) >> 17 & 0x3f) +#define OPENBSD_MASK_TO_FP_C(m) (((m) & 0x3f) << 1) +#define OPENBSD_FLAG_TO_FP_C(s) (((s) & 0x3f) << 17) +#define CLEAR_FP_C_MASK(fp_c) ((fp_c) & ~(0x3f << 1)) +#define CLEAR_FP_C_FLAG(fp_c) ((fp_c) & ~(0x3f << 17)) +#define SET_FP_C_MASK(fp_c, m) (CLEAR_FP_C_MASK(fp_c) | OPENBSD_MASK_TO_FP_C(m)) +#define SET_FP_C_FLAG(fp_c, m) (CLEAR_FP_C_FLAG(fp_c) | OPENBSD_FLAG_TO_FP_C(m)) + +#endif + +#endif diff --git a/sys/arch/alpha/include/ieeefp.h b/sys/arch/alpha/include/ieeefp.h index 4ebb20b0aa5..4cb8539a2c1 100644 --- a/sys/arch/alpha/include/ieeefp.h +++ b/sys/arch/alpha/include/ieeefp.h @@ -1,4 +1,4 @@ -/* $OpenBSD: ieeefp.h,v 1.3 1996/10/30 22:39:08 niklas Exp $ */ +/* $OpenBSD: ieeefp.h,v 1.4 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: ieeefp.h,v 1.1 1995/04/29 01:09:17 cgd Exp $ */ /* @@ -10,18 +10,45 @@ #define _ALPHA_IEEEFP_H_ typedef int fp_except; + +#ifdef _KERNEL + +#include <sys/param.h> +#include <sys/proc.h> +#include <machine/fpu.h> +#include <machine/cpu.h> + +/* FP_X_IOV is intentionally omitted from the architecture flags mask */ + +#define FP_AA_FLAGS (FP_X_INV | FP_X_DZ | FP_X_OFL | FP_X_UFL | FP_X_IMP) + +#define float_raise(f) \ + do curproc->p_md.md_flags |= OPENBSD_FLAG_TO_FP_C(f); \ + while(0) + +#define float_set_inexact() float_raise(FP_X_IMP) +#define float_set_invalid() float_raise(FP_X_INV) +#define fpgetround() (alpha_read_fpcr() >> 58 & 3) + +#endif + #define FP_X_INV 0x01 /* invalid operation exception */ #define FP_X_DZ 0x02 /* divide-by-zero exception */ #define FP_X_OFL 0x04 /* overflow exception */ #define FP_X_UFL 0x08 /* underflow exception */ #define FP_X_IMP 0x10 /* imprecise (loss of precision; "inexact") */ -#define FP_X_IOV 0x20 /* integer overflow XXX? */ +#define FP_X_IOV 0x20 /* integer overflow */ +/* + * fp_rnd bits match the fpcr, below, as well as bits 12:11 + * in fp operate instructions + */ typedef enum { - FP_RZ=0, /* round to zero (truncate) */ - FP_RM=1, /* round toward negative infinity */ - FP_RN=2, /* round to nearest representable number */ - FP_RP=3 /* round toward positive infinity */ + FP_RZ = 0, /* round to zero (truncate) */ + FP_RM = 1, /* round toward negative infinity */ + FP_RN = 2, /* round to nearest representable number */ + FP_RP = 3, /* round toward positive infinity */ + _FP_DYNAMIC=FP_RP } fp_rnd; #endif /* _ALPHA_IEEEFP_H_ */ diff --git a/sys/arch/alpha/include/pcb.h b/sys/arch/alpha/include/pcb.h index f7eb89491af..b91bdf22438 100644 --- a/sys/arch/alpha/include/pcb.h +++ b/sys/arch/alpha/include/pcb.h @@ -1,4 +1,4 @@ -/* $OpenBSD: pcb.h,v 1.5 2002/03/14 01:26:27 millert Exp $ */ +/* $OpenBSD: pcb.h,v 1.6 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: pcb.h,v 1.5 1996/11/13 22:21:00 cgd Exp $ */ /* @@ -52,6 +52,7 @@ struct pcb { struct fpreg pcb_fp; /* FP registers [SW] */ unsigned long pcb_onfault; /* for copy faults [SW] */ unsigned long pcb_accessaddr; /* for [fs]uswintr [SW] */ + struct cpu_info *__volatile pcb_fpcpu; /* CPU with our FP state[SW] */ }; /* diff --git a/sys/arch/alpha/include/proc.h b/sys/arch/alpha/include/proc.h index fd698f6f9e1..dab1d4c78c0 100644 --- a/sys/arch/alpha/include/proc.h +++ b/sys/arch/alpha/include/proc.h @@ -1,4 +1,4 @@ -/* $OpenBSD: proc.h,v 1.7 2002/03/14 01:26:27 millert Exp $ */ +/* $OpenBSD: proc.h,v 1.8 2002/04/28 20:55:14 pvalchev Exp $ */ /* $NetBSD: proc.h,v 1.2 1995/03/24 15:01:36 cgd Exp $ */ /* @@ -45,7 +45,28 @@ struct mdproc { struct mdbpt md_sstep[2]; /* two breakpoints for sstep */ }; +/* + * md_flags usage + * -------------- + * MDP_FPUSED + * A largely unused bit indicating the presence of FPU history. + * Cleared on exec. Set but not used by the fpu context switcher + * itself. + * + * MDP_FP_C + * The architected FP Control word. It should forever begin at bit 1, + * as the bits are AARM specified and this way it doesn't need to be + * shifted. + * + * Until C99 there was never an IEEE 754 API, making most of the + * standard useless. Because of overlapping AARM, OSF/1, NetBSD, and + * C99 API's, the use of the MDP_FP_C bits is defined variously in + * ieeefp.h and fpu.h. + */ #define MDP_FPUSED 0x0001 /* Process used the FPU */ +#ifndef NO_IEEE +#define MDP_FP_C 0x7ffffe /* Extended FP_C Quadword bits */ +#endif #define MDP_STEP1 0x0002 /* Single step normal */ #define MDP_STEP2 0x0003 /* Single step branch */ diff --git a/sys/arch/alpha/include/sysarch.h b/sys/arch/alpha/include/sysarch.h new file mode 100644 index 00000000000..a439c52c7e2 --- /dev/null +++ b/sys/arch/alpha/include/sysarch.h @@ -0,0 +1,69 @@ +/* $OpenBSD: sysarch.h,v 1.3 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: sysarch.h,v 1.8 2001/04/26 03:10:46 ross Exp $ */ + +/*- + * Copyright (c) 2000 The NetBSD Foundation, Inc. + * All rights reserved. + * + * This code is derived from software contributed to The NetBSD Foundation + * by Jason R. Thorpe. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef _ALPHA_SYSARCH_H_ +#define _ALPHA_SYSARCH_H_ + +#include <machine/bus.h> +#include <machine/ieeefp.h> + +/* + * Architecture specific syscalls (ALPHA) + */ + +#define ALPHA_FPGETMASK 0 +#define ALPHA_FPSETMASK 1 +#define ALPHA_FPSETSTICKY 2 +#define ALPHA_FPGETSTICKY 6 +#define ALPHA_GET_FP_C 7 +#define ALPHA_SET_FP_C 8 + +struct alpha_fp_except_args { + fp_except mask; +}; + +struct alpha_fp_c_args { + uint64_t fp_c; +}; + +#ifdef _KERNEL +int sysarch(int, void *); +#endif /* _KERNEL */ + +#endif /* !_ALPHA_SYSARCH_H_ */ diff --git a/sys/lib/libkern/arch/alpha/Makefile.inc b/sys/lib/libkern/arch/alpha/Makefile.inc index af056c8c99f..b984665e3a5 100644 --- a/sys/lib/libkern/arch/alpha/Makefile.inc +++ b/sys/lib/libkern/arch/alpha/Makefile.inc @@ -1,10 +1,10 @@ -# $OpenBSD: Makefile.inc,v 1.11 2000/12/18 18:40:45 provos Exp $ +# $OpenBSD: Makefile.inc,v 1.12 2002/04/28 20:55:14 pvalchev Exp $ # $NetBSD: Makefile.inc,v 1.9 1996/08/27 00:44:24 cgd Exp $ SRCS+= __main.c imax.c imin.c lmax.c lmin.c max.c min.c ulmax.c ulmin.c \ memchr.c memcmp.c memset.c \ - bcmp.c bzero.S ffs.S strcat.c strcmp.c strcpy.c strlcat.c strlcpy.c \ - strlen.c strncmp.c \ + bcmp.c bzero.S ffs.S softfloat.c strcat.c strcmp.c strcpy.c \ + strlcat.c strlcpy.c strlen.c strncmp.c \ strncpy.c scanc.c skpc.c htonl.S htons.S ntohl.S ntohs.S \ random.c strncasecmp.c diff --git a/sys/lib/libkern/milieu.h b/sys/lib/libkern/milieu.h new file mode 100644 index 00000000000..53538bf6e0d --- /dev/null +++ b/sys/lib/libkern/milieu.h @@ -0,0 +1,163 @@ +/* $OpenBSD: milieu.h,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: milieu.h,v 1.1 2001/04/26 03:10:47 ross Exp $ */ + +/* This is a derivative work. */ + +/*- + * Copyright (c) 2001 The NetBSD Foundation, Inc. + * All rights reserved. + * + * This code is derived from software contributed to The NetBSD Foundation + * by Ross Harvey. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +/* +=============================================================================== + +This C header file is part of TestFloat, Release 2a, a package of programs +for testing the correctness of floating-point arithmetic complying to the +IEC/IEEE Standard for Floating-Point. + +Written by John R. Hauser. More information is available through the Web +page `http://HTTP.CS.Berkeley.EDU/~jhauser/arithmetic/TestFloat.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 ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM +THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY +(possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL +COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS +ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS. + +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 four paragraphs for those parts of +this code that are retained. + +=============================================================================== +*/ + +#ifndef NO_IEEE + +#ifndef MILIEU_H +#define MILIEU_H + +#include <sys/types.h> +#include <sys/endian.h> + +enum { + FALSE = 0, + TRUE = 1 +}; + + +/* +------------------------------------------------------------------------------- +One of the macros `BIGENDIAN' or `LITTLEENDIAN' must be defined. +------------------------------------------------------------------------------- +*/ + +#if _BYTE_ORDER == _LITTLE_ENDIAN +#define LITTLEENDIAN +#else +#define BIGENDIAN +#endif + +#define BITS64 + +/* +------------------------------------------------------------------------------- +Each of the following `typedef's defines the most convenient type that holds +integers of at least as many bits as specified. For example, `uint8' should +be the most convenient type that can hold unsigned integers of as many as +8 bits. The `flag' type must be able to hold either a 0 or 1. For most +implementations of C, `flag', `uint8', and `int8' should all be `typedef'ed +to the same as `int'. +------------------------------------------------------------------------------- +*/ +typedef int flag; +typedef unsigned int uint8; +typedef signed int int8; +typedef unsigned int uint16; +typedef int int16; +typedef unsigned int uint32; +typedef signed int int32; +#ifdef BITS64 +typedef uint64_t uint64; +typedef int64_t int64; +#endif + +/* +------------------------------------------------------------------------------- +Each of the following `typedef's defines a type that holds integers +of _exactly_ the number of bits specified. For instance, for most +implementation of C, `bits16' and `sbits16' should be `typedef'ed to +`unsigned short int' and `signed short int' (or `short int'), respectively. +------------------------------------------------------------------------------- +*/ +typedef uint8_t bits8; +typedef int8_t sbits8; +typedef uint16_t bits16; +typedef int16_t sbits16; +typedef uint32_t bits32; +typedef int32_t sbits32; +#ifdef BITS64 +typedef uint64_t bits64; +typedef int64_t sbits64; +#endif + +#ifdef BITS64 +/* +------------------------------------------------------------------------------- +The `LIT64' macro takes as its argument a textual integer literal and +if necessary ``marks'' the literal as having a 64-bit integer type. +For example, the GNU C Compiler (`gcc') requires that 64-bit literals be +appended with the letters `LL' standing for `long long', which is `gcc's +name for the 64-bit integer type. Some compilers may allow `LIT64' to be +defined as the identity macro: `#define LIT64( a ) a'. +------------------------------------------------------------------------------- +*/ +#define LIT64( a ) a##LL +#endif + +/* +------------------------------------------------------------------------------- +The macro `INLINE' can be used before functions that should be inlined. If +a compiler does not support explicit inlining, this macro should be defined +to be `static'. +------------------------------------------------------------------------------- +*/ +#define INLINE static inline + +#endif +#endif /* !NO_IEEE */ diff --git a/sys/lib/libkern/softfloat-macros.h b/sys/lib/libkern/softfloat-macros.h new file mode 100644 index 00000000000..b6dedb99f29 --- /dev/null +++ b/sys/lib/libkern/softfloat-macros.h @@ -0,0 +1,753 @@ +/* $OpenBSD: softfloat-macros.h,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: softfloat-macros.h,v 1.1 2001/04/26 03:10:47 ross Exp $ */ + +/* +=============================================================================== + +This C source fragment is part of the SoftFloat IEC/IEEE Floating-point +Arithmetic Package, Release 2a. + +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 ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM +THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY +(possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL +COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS +ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS. + +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 four paragraphs for those parts of +this code that are retained. + +=============================================================================== +*/ + +#ifndef NO_IEEE + +/* +------------------------------------------------------------------------------- +Shifts `a' right by the number of bits given in `count'. If any nonzero +bits are shifted off, they are ``jammed'' into the least significant bit of +the result by setting the least significant bit to 1. The value of `count' +can be arbitrarily large; in particular, if `count' is greater than 32, the +result will be either 0 or 1, depending on whether `a' is zero or nonzero. +The result is stored in the location pointed to by `zPtr'. +------------------------------------------------------------------------------- +*/ +INLINE void shift32RightJamming( bits32 a, int16 count, bits32 *zPtr ) +{ + bits32 z; + + if ( count == 0 ) { + z = a; + } + else if ( count < 32 ) { + z = ( a>>count ) | ( ( a<<( ( - count ) & 31 ) ) != 0 ); + } + else { + z = ( a != 0 ); + } + *zPtr = z; + +} + +/* +------------------------------------------------------------------------------- +Shifts `a' right by the number of bits given in `count'. If any nonzero +bits are shifted off, they are ``jammed'' into the least significant bit of +the result by setting the least significant bit to 1. The value of `count' +can be arbitrarily large; in particular, if `count' is greater than 64, the +result will be either 0 or 1, depending on whether `a' is zero or nonzero. +The result is stored in the location pointed to by `zPtr'. +------------------------------------------------------------------------------- +*/ +INLINE void shift64RightJamming( bits64 a, int16 count, bits64 *zPtr ) +{ + bits64 z; + + if ( count == 0 ) { + z = a; + } + else if ( count < 64 ) { + z = ( a>>count ) | ( ( a<<( ( - count ) & 63 ) ) != 0 ); + } + else { + z = ( a != 0 ); + } + *zPtr = z; + +} + +/* +------------------------------------------------------------------------------- +Shifts the 128-bit value formed by concatenating `a0' and `a1' right by 64 +_plus_ the number of bits given in `count'. The shifted result is at most +64 nonzero bits; this is stored at the location pointed to by `z0Ptr'. The +bits shifted off form a second 64-bit result as follows: The _last_ bit +shifted off is the most-significant bit of the extra result, and the other +63 bits of the extra result are all zero if and only if _all_but_the_last_ +bits shifted off were all zero. This extra result is stored in the location +pointed to by `z1Ptr'. The value of `count' can be arbitrarily large. + (This routine makes more sense if `a0' and `a1' are considered to form a +fixed-point value with binary point between `a0' and `a1'. This fixed-point +value is shifted right by the number of bits given in `count', and the +integer part of the result is returned at the location pointed to by +`z0Ptr'. The fractional part of the result may be slightly corrupted as +described above, and is returned at the location pointed to by `z1Ptr'.) +------------------------------------------------------------------------------- +*/ +INLINE void + shift64ExtraRightJamming( + bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + bits64 z0, z1; + int8 negCount = ( - count ) & 63; + + if ( count == 0 ) { + z1 = a1; + z0 = a0; + } + else if ( count < 64 ) { + z1 = ( a0<<negCount ) | ( a1 != 0 ); + z0 = a0>>count; + } + else { + if ( count == 64 ) { + z1 = a0 | ( a1 != 0 ); + } + else { + z1 = ( ( a0 | a1 ) != 0 ); + } + z0 = 0; + } + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the +number of bits given in `count'. Any bits shifted off are lost. The value +of `count' can be arbitrarily large; in particular, if `count' is greater +than 128, the result will be 0. The result is broken into two 64-bit pieces +which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + shift128Right( + bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + bits64 z0, z1; + int8 negCount = ( - count ) & 63; + + if ( count == 0 ) { + z1 = a1; + z0 = a0; + } + else if ( count < 64 ) { + z1 = ( a0<<negCount ) | ( a1>>count ); + z0 = a0>>count; + } + else { + z1 = ( count < 64 ) ? ( a0>>( count & 63 ) ) : 0; + z0 = 0; + } + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the +number of bits given in `count'. If any nonzero bits are shifted off, they +are ``jammed'' into the least significant bit of the result by setting the +least significant bit to 1. The value of `count' can be arbitrarily large; +in particular, if `count' is greater than 128, the result will be either +0 or 1, depending on whether the concatenation of `a0' and `a1' is zero or +nonzero. The result is broken into two 64-bit pieces which are stored at +the locations pointed to by `z0Ptr' and `z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + shift128RightJamming( + bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + bits64 z0, z1; + int8 negCount = ( - count ) & 63; + + if ( count == 0 ) { + z1 = a1; + z0 = a0; + } + else if ( count < 64 ) { + z1 = ( a0<<negCount ) | ( a1>>count ) | ( ( a1<<negCount ) != 0 ); + z0 = a0>>count; + } + else { + if ( count == 64 ) { + z1 = a0 | ( a1 != 0 ); + } + else if ( count < 128 ) { + z1 = ( a0>>( count & 63 ) ) | ( ( ( a0<<negCount ) | a1 ) != 0 ); + } + else { + z1 = ( ( a0 | a1 ) != 0 ); + } + z0 = 0; + } + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' right +by 64 _plus_ the number of bits given in `count'. The shifted result is +at most 128 nonzero bits; these are broken into two 64-bit pieces which are +stored at the locations pointed to by `z0Ptr' and `z1Ptr'. The bits shifted +off form a third 64-bit result as follows: The _last_ bit shifted off is +the most-significant bit of the extra result, and the other 63 bits of the +extra result are all zero if and only if _all_but_the_last_ bits shifted off +were all zero. This extra result is stored in the location pointed to by +`z2Ptr'. The value of `count' can be arbitrarily large. + (This routine makes more sense if `a0', `a1', and `a2' are considered +to form a fixed-point value with binary point between `a1' and `a2'. This +fixed-point value is shifted right by the number of bits given in `count', +and the integer part of the result is returned at the locations pointed to +by `z0Ptr' and `z1Ptr'. The fractional part of the result may be slightly +corrupted as described above, and is returned at the location pointed to by +`z2Ptr'.) +------------------------------------------------------------------------------- +*/ +INLINE void + shift128ExtraRightJamming( + bits64 a0, + bits64 a1, + bits64 a2, + int16 count, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr + ) +{ + bits64 z0, z1, z2; + int8 negCount = ( - count ) & 63; + + if ( count == 0 ) { + z2 = a2; + z1 = a1; + z0 = a0; + } + else { + if ( count < 64 ) { + z2 = a1<<negCount; + z1 = ( a0<<negCount ) | ( a1>>count ); + z0 = a0>>count; + } + else { + if ( count == 64 ) { + z2 = a1; + z1 = a0; + } + else { + a2 |= a1; + if ( count < 128 ) { + z2 = a0<<negCount; + z1 = a0>>( count & 63 ); + } + else { + z2 = ( count == 128 ) ? a0 : ( a0 != 0 ); + z1 = 0; + } + } + z0 = 0; + } + z2 |= ( a2 != 0 ); + } + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Shifts the 128-bit value formed by concatenating `a0' and `a1' left by the +number of bits given in `count'. Any bits shifted off are lost. The value +of `count' must be less than 64. The result is broken into two 64-bit +pieces which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + shortShift128Left( + bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + + *z1Ptr = a1<<count; + *z0Ptr = + ( count == 0 ) ? a0 : ( a0<<count ) | ( a1>>( ( - count ) & 63 ) ); + +} + +/* +------------------------------------------------------------------------------- +Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' left +by the number of bits given in `count'. Any bits shifted off are lost. +The value of `count' must be less than 64. The result is broken into three +64-bit pieces which are stored at the locations pointed to by `z0Ptr', +`z1Ptr', and `z2Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + shortShift192Left( + bits64 a0, + bits64 a1, + bits64 a2, + int16 count, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr + ) +{ + bits64 z0, z1, z2; + int8 negCount; + + z2 = a2<<count; + z1 = a1<<count; + z0 = a0<<count; + if ( 0 < count ) { + negCount = ( ( - count ) & 63 ); + z1 |= a2>>negCount; + z0 |= a1>>negCount; + } + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Adds the 128-bit value formed by concatenating `a0' and `a1' to the 128-bit +value formed by concatenating `b0' and `b1'. Addition is modulo 2^128, so +any carry out is lost. The result is broken into two 64-bit pieces which +are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + add128( + bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + bits64 z1; + + z1 = a1 + b1; + *z1Ptr = z1; + *z0Ptr = a0 + b0 + ( z1 < a1 ); + +} + +/* +------------------------------------------------------------------------------- +Adds the 192-bit value formed by concatenating `a0', `a1', and `a2' to the +192-bit value formed by concatenating `b0', `b1', and `b2'. Addition is +modulo 2^192, so any carry out is lost. The result is broken into three +64-bit pieces which are stored at the locations pointed to by `z0Ptr', +`z1Ptr', and `z2Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + add192( + bits64 a0, + bits64 a1, + bits64 a2, + bits64 b0, + bits64 b1, + bits64 b2, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr + ) +{ + bits64 z0, z1, z2; + int8 carry0, carry1; + + z2 = a2 + b2; + carry1 = ( z2 < a2 ); + z1 = a1 + b1; + carry0 = ( z1 < a1 ); + z0 = a0 + b0; + z1 += carry1; + z0 += ( z1 < carry1 ); + z0 += carry0; + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Subtracts the 128-bit value formed by concatenating `b0' and `b1' from the +128-bit value formed by concatenating `a0' and `a1'. Subtraction is modulo +2^128, so any borrow out (carry out) is lost. The result is broken into two +64-bit pieces which are stored at the locations pointed to by `z0Ptr' and +`z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + sub128( + bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + + *z1Ptr = a1 - b1; + *z0Ptr = a0 - b0 - ( a1 < b1 ); + +} + +/* +------------------------------------------------------------------------------- +Subtracts the 192-bit value formed by concatenating `b0', `b1', and `b2' +from the 192-bit value formed by concatenating `a0', `a1', and `a2'. +Subtraction is modulo 2^192, so any borrow out (carry out) is lost. The +result is broken into three 64-bit pieces which are stored at the locations +pointed to by `z0Ptr', `z1Ptr', and `z2Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + sub192( + bits64 a0, + bits64 a1, + bits64 a2, + bits64 b0, + bits64 b1, + bits64 b2, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr + ) +{ + bits64 z0, z1, z2; + int8 borrow0, borrow1; + + z2 = a2 - b2; + borrow1 = ( a2 < b2 ); + z1 = a1 - b1; + borrow0 = ( a1 < b1 ); + z0 = a0 - b0; + z0 -= ( z1 < borrow1 ); + z1 -= borrow1; + z0 -= borrow0; + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Multiplies `a' by `b' to obtain a 128-bit product. The product is broken +into two 64-bit pieces which are stored at the locations pointed to by +`z0Ptr' and `z1Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void mul64To128( bits64 a, bits64 b, bits64 *z0Ptr, bits64 *z1Ptr ) +{ + bits32 aHigh, aLow, bHigh, bLow; + bits64 z0, zMiddleA, zMiddleB, z1; + + aLow = a; + aHigh = a>>32; + bLow = b; + bHigh = b>>32; + z1 = ( (bits64) aLow ) * bLow; + zMiddleA = ( (bits64) aLow ) * bHigh; + zMiddleB = ( (bits64) aHigh ) * bLow; + z0 = ( (bits64) aHigh ) * bHigh; + zMiddleA += zMiddleB; + z0 += ( ( (bits64) ( zMiddleA < zMiddleB ) )<<32 ) + ( zMiddleA>>32 ); + zMiddleA <<= 32; + z1 += zMiddleA; + z0 += ( z1 < zMiddleA ); + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Multiplies the 128-bit value formed by concatenating `a0' and `a1' by +`b' to obtain a 192-bit product. The product is broken into three 64-bit +pieces which are stored at the locations pointed to by `z0Ptr', `z1Ptr', and +`z2Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + mul128By64To192( + bits64 a0, + bits64 a1, + bits64 b, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr + ) +{ + bits64 z0, z1, z2, more1; + + mul64To128( a1, b, &z1, &z2 ); + mul64To128( a0, b, &z0, &more1 ); + add128( z0, more1, 0, z1, &z0, &z1 ); + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Multiplies the 128-bit value formed by concatenating `a0' and `a1' to the +128-bit value formed by concatenating `b0' and `b1' to obtain a 256-bit +product. The product is broken into four 64-bit pieces which are stored at +the locations pointed to by `z0Ptr', `z1Ptr', `z2Ptr', and `z3Ptr'. +------------------------------------------------------------------------------- +*/ +INLINE void + mul128To256( + bits64 a0, + bits64 a1, + bits64 b0, + bits64 b1, + bits64 *z0Ptr, + bits64 *z1Ptr, + bits64 *z2Ptr, + bits64 *z3Ptr + ) +{ + bits64 z0, z1, z2, z3; + bits64 more1, more2; + + mul64To128( a1, b1, &z2, &z3 ); + mul64To128( a1, b0, &z1, &more2 ); + add128( z1, more2, 0, z2, &z1, &z2 ); + mul64To128( a0, b0, &z0, &more1 ); + add128( z0, more1, 0, z1, &z0, &z1 ); + mul64To128( a0, b1, &more1, &more2 ); + add128( more1, more2, 0, z2, &more1, &z2 ); + add128( z0, z1, 0, more1, &z0, &z1 ); + *z3Ptr = z3; + *z2Ptr = z2; + *z1Ptr = z1; + *z0Ptr = z0; + +} + +/* +------------------------------------------------------------------------------- +Returns an approximation to the 64-bit integer quotient obtained by dividing +`b' into the 128-bit value formed by concatenating `a0' and `a1'. The +divisor `b' must be at least 2^63. If q is the exact quotient truncated +toward zero, the approximation returned lies between q and q + 2 inclusive. +If the exact quotient q is larger than 64 bits, the maximum positive 64-bit +unsigned integer is returned. +------------------------------------------------------------------------------- +*/ +static bits64 estimateDiv128To64( bits64 a0, bits64 a1, bits64 b ) +{ + bits64 b0, b1; + bits64 rem0, rem1, term0, term1; + bits64 z; + + if ( b <= a0 ) return LIT64( 0xFFFFFFFFFFFFFFFF ); + b0 = b>>32; + z = ( b0<<32 <= a0 ) ? LIT64( 0xFFFFFFFF00000000 ) : ( a0 / b0 )<<32; + mul64To128( b, z, &term0, &term1 ); + sub128( a0, a1, term0, term1, &rem0, &rem1 ); + while ( ( (sbits64) rem0 ) < 0 ) { + z -= LIT64( 0x100000000 ); + b1 = b<<32; + add128( rem0, rem1, b0, b1, &rem0, &rem1 ); + } + rem0 = ( rem0<<32 ) | ( rem1>>32 ); + z |= ( b0<<32 <= rem0 ) ? 0xFFFFFFFF : rem0 / b0; + return z; + +} + +#ifndef SOFTFLOAT_FOR_GCC /* Not used */ +/* +------------------------------------------------------------------------------- +Returns an approximation to the square root of the 32-bit significand given +by `a'. Considered as an integer, `a' must be at least 2^31. If bit 0 of +`aExp' (the least significant bit) is 1, the integer returned approximates +2^31*sqrt(`a'/2^31), where `a' is considered an integer. If bit 0 of `aExp' +is 0, the integer returned approximates 2^31*sqrt(`a'/2^30). In either +case, the approximation returned lies strictly within +/-2 of the exact +value. +------------------------------------------------------------------------------- +*/ +static bits32 estimateSqrt32( int16 aExp, bits32 a ) +{ + static const bits16 sqrtOddAdjustments[] = { + 0x0004, 0x0022, 0x005D, 0x00B1, 0x011D, 0x019F, 0x0236, 0x02E0, + 0x039C, 0x0468, 0x0545, 0x0631, 0x072B, 0x0832, 0x0946, 0x0A67 + }; + static const bits16 sqrtEvenAdjustments[] = { + 0x0A2D, 0x08AF, 0x075A, 0x0629, 0x051A, 0x0429, 0x0356, 0x029E, + 0x0200, 0x0179, 0x0109, 0x00AF, 0x0068, 0x0034, 0x0012, 0x0002 + }; + int8 index; + bits32 z; + + index = ( a>>27 ) & 15; + if ( aExp & 1 ) { + z = 0x4000 + ( a>>17 ) - sqrtOddAdjustments[ index ]; + z = ( ( a / z )<<14 ) + ( z<<15 ); + a >>= 1; + } + else { + z = 0x8000 + ( a>>17 ) - sqrtEvenAdjustments[ index ]; + z = a / z + z; + z = ( 0x20000 <= z ) ? 0xFFFF8000 : ( z<<15 ); + if ( z <= a ) return (bits32) ( ( (sbits32) a )>>1 ); + } + return ( (bits32) ( ( ( (bits64) a )<<31 ) / z ) ) + ( z>>1 ); + +} +#endif + +/* +------------------------------------------------------------------------------- +Returns the number of leading 0 bits before the most-significant 1 bit of +`a'. If `a' is zero, 32 is returned. +------------------------------------------------------------------------------- +*/ +static int8 countLeadingZeros32( bits32 a ) +{ + static const int8 countLeadingZerosHigh[] = { + 8, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 + }; + int8 shiftCount; + + shiftCount = 0; + if ( a < 0x10000 ) { + shiftCount += 16; + a <<= 16; + } + if ( a < 0x1000000 ) { + shiftCount += 8; + a <<= 8; + } + shiftCount += countLeadingZerosHigh[ a>>24 ]; + return shiftCount; + +} + +/* +------------------------------------------------------------------------------- +Returns the number of leading 0 bits before the most-significant 1 bit of +`a'. If `a' is zero, 64 is returned. +------------------------------------------------------------------------------- +*/ +static int8 countLeadingZeros64( bits64 a ) +{ + int8 shiftCount; + + shiftCount = 0; + if ( a < ( (bits64) 1 )<<32 ) { + shiftCount += 32; + } + else { + a >>= 32; + } + shiftCount += countLeadingZeros32( a ); + return shiftCount; + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' +is equal to the 128-bit value formed by concatenating `b0' and `b1'. +Otherwise, returns 0. +------------------------------------------------------------------------------- +*/ +INLINE flag eq128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) +{ + + return ( a0 == b0 ) && ( a1 == b1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less +than or equal to the 128-bit value formed by concatenating `b0' and `b1'. +Otherwise, returns 0. +------------------------------------------------------------------------------- +*/ +INLINE flag le128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) +{ + + return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 <= b1 ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less +than the 128-bit value formed by concatenating `b0' and `b1'. Otherwise, +returns 0. +------------------------------------------------------------------------------- +*/ +INLINE flag lt128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) +{ + + return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 < b1 ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is +not equal to the 128-bit value formed by concatenating `b0' and `b1'. +Otherwise, returns 0. +------------------------------------------------------------------------------- +*/ +INLINE flag ne128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) +{ + + return ( a0 != b0 ) || ( a1 != b1 ); + +} + +#endif /* !NO_IEEE */ diff --git a/sys/lib/libkern/softfloat-specialize.h b/sys/lib/libkern/softfloat-specialize.h new file mode 100644 index 00000000000..db9fb08b4ed --- /dev/null +++ b/sys/lib/libkern/softfloat-specialize.h @@ -0,0 +1,495 @@ +/* $OpenBSD: softfloat-specialize.h,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: softfloat-specialize.h,v 1.1 2001/04/26 03:10:47 ross Exp $ */ + +/* This is a derivative work. */ + +/*- + * Copyright (c) 2001 The NetBSD Foundation, Inc. + * All rights reserved. + * + * This code is derived from software contributed to The NetBSD Foundation + * by Ross Harvey. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +/* +=============================================================================== + +This C source fragment is part of the SoftFloat IEC/IEEE Floating-point +Arithmetic Package, Release 2a. + +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 ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM +THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY +(possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL +COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS +ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS. + +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 four paragraphs for those parts of +this code that are retained. + +=============================================================================== +*/ + +/* +------------------------------------------------------------------------------- +Underflow tininess-detection mode, statically initialized to default value. +------------------------------------------------------------------------------- +*/ + +#ifndef NO_IEEE + +/* [ MP safe, does not change dynamically ] */ +int float_detect_tininess = float_tininess_after_rounding; + +/* +------------------------------------------------------------------------------- +Internal canonical NaN format. +------------------------------------------------------------------------------- +*/ +typedef struct { + flag sign; + bits64 high, low; +} commonNaNT; + +/* +------------------------------------------------------------------------------- +The pattern for a default generated single-precision NaN. +------------------------------------------------------------------------------- +*/ +#define float32_default_nan 0xFFC00000 + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is a NaN; +otherwise returns 0. +------------------------------------------------------------------------------- +*/ +static flag float32_is_nan( float32 a ) +{ + + return ( 0xFF000000 < (bits32) ( a<<1 ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is a signaling +NaN; otherwise returns 0. +------------------------------------------------------------------------------- +*/ +flag float32_is_signaling_nan( float32 a ) +{ + + return ( ( ( a>>22 ) & 0x1FF ) == 0x1FE ) && ( a & 0x003FFFFF ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point NaN +`a' to the canonical NaN format. If `a' is a signaling NaN, the invalid +exception is raised. +------------------------------------------------------------------------------- +*/ +static commonNaNT float32ToCommonNaN( float32 a ) +{ + commonNaNT z; + + if ( float32_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); + z.sign = a>>31; + z.low = 0; + z.high = ( (bits64) a )<<41; + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the canonical NaN `a' to the single- +precision floating-point format. +------------------------------------------------------------------------------- +*/ +static float32 commonNaNToFloat32( commonNaNT a ) +{ + + return ( ( (bits32) a.sign )<<31 ) | 0x7FC00000 | ( a.high>>41 ); + +} + +/* +------------------------------------------------------------------------------- +Takes two single-precision floating-point values `a' and `b', one of which +is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a +signaling NaN, the invalid exception is raised. +------------------------------------------------------------------------------- +*/ +static float32 propagateFloat32NaN( float32 a, float32 b ) +{ + flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; + + aIsNaN = float32_is_nan( a ); + aIsSignalingNaN = float32_is_signaling_nan( a ); + bIsNaN = float32_is_nan( b ); + bIsSignalingNaN = float32_is_signaling_nan( b ); + a |= 0x00400000; + b |= 0x00400000; + if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); + if ( aIsSignalingNaN ) { + if ( bIsSignalingNaN ) goto returnLargerSignificand; + return bIsNaN ? b : a; + } + else if ( aIsNaN ) { + if ( bIsSignalingNaN | ! bIsNaN ) return a; + returnLargerSignificand: + if ( (bits32) ( a<<1 ) < (bits32) ( b<<1 ) ) return b; + if ( (bits32) ( b<<1 ) < (bits32) ( a<<1 ) ) return a; + return ( a < b ) ? a : b; + } + else { + return b; + } + +} + + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point NaN +`a' to the canonical NaN format. If `a' is a signaling NaN, the invalid +exception is raised. +------------------------------------------------------------------------------- +*/ +static commonNaNT float64ToCommonNaN( float64 a ) +{ + commonNaNT z; + + if ( float64_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); + z.sign = a>>63; + z.low = 0; + z.high = a<<12; + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the canonical NaN `a' to the double- +precision floating-point format. +------------------------------------------------------------------------------- +*/ +static float64 commonNaNToFloat64( commonNaNT a ) +{ + + return + ( ( (bits64) a.sign )<<63 ) + | LIT64( 0x7FF8000000000000 ) + | ( a.high>>12 ); + +} + +/* +------------------------------------------------------------------------------- +Takes two double-precision floating-point values `a' and `b', one of which +is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a +signaling NaN, the invalid exception is raised. +------------------------------------------------------------------------------- +*/ +static float64 propagateFloat64NaN( float64 a, float64 b ) +{ + flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; + + aIsNaN = float64_is_nan( a ); + aIsSignalingNaN = float64_is_signaling_nan( a ); + bIsNaN = float64_is_nan( b ); + bIsSignalingNaN = float64_is_signaling_nan( b ); + a |= LIT64( 0x0008000000000000 ); + b |= LIT64( 0x0008000000000000 ); + if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); + if ( aIsSignalingNaN ) { + if ( bIsSignalingNaN ) goto returnLargerSignificand; + return bIsNaN ? b : a; + } + else if ( aIsNaN ) { + if ( bIsSignalingNaN | ! bIsNaN ) return a; + returnLargerSignificand: + if ( (bits64) ( a<<1 ) < (bits64) ( b<<1 ) ) return b; + if ( (bits64) ( b<<1 ) < (bits64) ( a<<1 ) ) return a; + return ( a < b ) ? a : b; + } + else { + return b; + } + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +The pattern for a default generated extended double-precision NaN. The +`high' and `low' values hold the most- and least-significant bits, +respectively. +------------------------------------------------------------------------------- +*/ +#define floatx80_default_nan_high 0xFFFF +#define floatx80_default_nan_low LIT64( 0xC000000000000000 ) + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is a +NaN; otherwise returns 0. +------------------------------------------------------------------------------- +*/ +static flag floatx80_is_nan( floatx80 a ) +{ + + return ( ( a.high & 0x7FFF ) == 0x7FFF ) && (bits64) ( a.low<<1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is a +signaling NaN; otherwise returns 0. +------------------------------------------------------------------------------- +*/ +flag floatx80_is_signaling_nan( floatx80 a ) +{ + bits64 aLow; + + aLow = a.low & ~ LIT64( 0x4000000000000000 ); + return + ( ( a.high & 0x7FFF ) == 0x7FFF ) + && (bits64) ( aLow<<1 ) + && ( a.low == aLow ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point NaN `a' to the canonical NaN format. If `a' is a signaling NaN, the +invalid exception is raised. +------------------------------------------------------------------------------- +*/ +static commonNaNT floatx80ToCommonNaN( floatx80 a ) +{ + commonNaNT z; + + if ( floatx80_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); + z.sign = a.high>>15; + z.low = 0; + z.high = a.low<<1; + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the canonical NaN `a' to the extended +double-precision floating-point format. +------------------------------------------------------------------------------- +*/ +static floatx80 commonNaNToFloatx80( commonNaNT a ) +{ + floatx80 z; + + z.low = LIT64( 0xC000000000000000 ) | ( a.high>>1 ); + z.high = ( ( (bits16) a.sign )<<15 ) | 0x7FFF; + return z; + +} + +/* +------------------------------------------------------------------------------- +Takes two extended double-precision floating-point values `a' and `b', one +of which is a NaN, and returns the appropriate NaN result. If either `a' or +`b' is a signaling NaN, the invalid exception is raised. +------------------------------------------------------------------------------- +*/ +static floatx80 propagateFloatx80NaN( floatx80 a, floatx80 b ) +{ + flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; + + aIsNaN = floatx80_is_nan( a ); + aIsSignalingNaN = floatx80_is_signaling_nan( a ); + bIsNaN = floatx80_is_nan( b ); + bIsSignalingNaN = floatx80_is_signaling_nan( b ); + a.low |= LIT64( 0xC000000000000000 ); + b.low |= LIT64( 0xC000000000000000 ); + if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); + if ( aIsSignalingNaN ) { + if ( bIsSignalingNaN ) goto returnLargerSignificand; + return bIsNaN ? b : a; + } + else if ( aIsNaN ) { + if ( bIsSignalingNaN | ! bIsNaN ) return a; + returnLargerSignificand: + if ( a.low < b.low ) return b; + if ( b.low < a.low ) return a; + return ( a.high < b.high ) ? a : b; + } + else { + return b; + } + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +The pattern for a default generated quadruple-precision NaN. The `high' and +`low' values hold the most- and least-significant bits, respectively. +------------------------------------------------------------------------------- +*/ +#define float128_default_nan_high LIT64( 0xFFFF800000000000 ) +#define float128_default_nan_low LIT64( 0x0000000000000000 ) + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is a NaN; +otherwise returns 0. +------------------------------------------------------------------------------- +*/ +flag float128_is_nan( float128 a ) +{ + + return + ( LIT64( 0xFFFE000000000000 ) <= (bits64) ( a.high<<1 ) ) + && ( a.low || ( a.high & LIT64( 0x0000FFFFFFFFFFFF ) ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is a +signaling NaN; otherwise returns 0. +------------------------------------------------------------------------------- +*/ +flag float128_is_signaling_nan( float128 a ) +{ + + return + ( ( ( a.high>>47 ) & 0xFFFF ) == 0xFFFE ) + && ( a.low || ( a.high & LIT64( 0x00007FFFFFFFFFFF ) ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point NaN +`a' to the canonical NaN format. If `a' is a signaling NaN, the invalid +exception is raised. +------------------------------------------------------------------------------- +*/ +static commonNaNT float128ToCommonNaN( float128 a ) +{ + commonNaNT z; + + if ( float128_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); + z.sign = a.high>>63; + shortShift128Left( a.high, a.low, 16, &z.high, &z.low ); + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the canonical NaN `a' to the quadruple- +precision floating-point format. +------------------------------------------------------------------------------- +*/ +static float128 commonNaNToFloat128( commonNaNT a ) +{ + float128 z; + + shift128Right( a.high, a.low, 16, &z.high, &z.low ); + z.high |= ( ( (bits64) a.sign )<<63 ) | LIT64( 0x7FFF800000000000 ); + return z; + +} + +/* +------------------------------------------------------------------------------- +Takes two quadruple-precision floating-point values `a' and `b', one of +which is a NaN, and returns the appropriate NaN result. If either `a' or +`b' is a signaling NaN, the invalid exception is raised. +------------------------------------------------------------------------------- +*/ +static float128 propagateFloat128NaN( float128 a, float128 b ) +{ + flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; + + aIsNaN = float128_is_nan( a ); + aIsSignalingNaN = float128_is_signaling_nan( a ); + bIsNaN = float128_is_nan( b ); + bIsSignalingNaN = float128_is_signaling_nan( b ); + a.high |= LIT64( 0x0000800000000000 ); + b.high |= LIT64( 0x0000800000000000 ); + if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); + if ( aIsSignalingNaN ) { + if ( bIsSignalingNaN ) goto returnLargerSignificand; + return bIsNaN ? b : a; + } + else if ( aIsNaN ) { + if ( bIsSignalingNaN | ! bIsNaN ) return a; + returnLargerSignificand: + if ( lt128( a.high<<1, a.low, b.high<<1, b.low ) ) return b; + if ( lt128( b.high<<1, b.low, a.high<<1, a.low ) ) return a; + return ( a.high < b.high ) ? a : b; + } + else { + return b; + } + +} + +#endif + +#endif /* !NO_IEEE */ diff --git a/sys/lib/libkern/softfloat.c b/sys/lib/libkern/softfloat.c new file mode 100644 index 00000000000..853f9fb4972 --- /dev/null +++ b/sys/lib/libkern/softfloat.c @@ -0,0 +1,5506 @@ +/* $OpenBSD: softfloat.c,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: softfloat.c,v 1.1 2001/04/26 03:10:47 ross Exp $ */ + +/* + * This version hacked for use with gcc -msoft-float by bjh21. + * (Mostly a case of #ifdefing out things GCC doesn't need or provides + * itself). + */ + +/* + * Things you may want to define: + * + * SOFTFLOAT_FOR_GCC - build only those functions necessary for GCC (with + * -msoft-float) to work. Include "softfloat-for-gcc.h" to get them + * properly renamed. + */ + +/* +=============================================================================== + +This C source file is part of the SoftFloat IEC/IEEE Floating-point +Arithmetic Package, Release 2a. + +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 ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM +THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY +(possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL +COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS +ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS. + +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 four paragraphs for those parts of +this code that are retained. + +=============================================================================== +*/ + +#ifndef NO_IEEE + +#include <sys/cdefs.h> +#if defined(LIBC_SCCS) && !defined(lint) +__RCSID("$NetBSD: softfloat.c,v 1.1 2001/04/26 03:10:47 ross Exp $"); +#endif /* LIBC_SCCS and not lint */ + +#ifdef SOFTFLOAT_FOR_GCC +#include "softfloat-for-gcc.h" +#endif + +#include "milieu.h" +#include "softfloat.h" + +/* + * Conversions between floats as stored in memory and floats as + * SoftFloat uses them + */ +#ifndef FLOAT64_DEMANGLE +#define FLOAT64_DEMANGLE(a) (a) +#endif +#ifndef FLOAT64_MANGLE +#define FLOAT64_MANGLE(a) (a) +#endif + +/* +------------------------------------------------------------------------------- +Floating-point rounding mode, extended double-precision rounding precision, +and exception flags. +------------------------------------------------------------------------------- +*/ + +/* + * XXX: This may cause options-MULTIPROCESSOR or thread problems someday. + * Right now, it does not. I've removed all other dynamic global + * variables. [ross] + */ +#ifdef FLOATX80 +int8 floatx80_rounding_precision = 80; +#endif + +/* +------------------------------------------------------------------------------- +Primitive arithmetic functions, including multi-word arithmetic, and +division and square root approximations. (Can be specialized to target if +desired.) +------------------------------------------------------------------------------- +*/ +#include "softfloat-macros.h" + +/* +------------------------------------------------------------------------------- +Functions and definitions to determine: (1) whether tininess for underflow +is detected before or after rounding by default, (2) what (if anything) +happens when exceptions are raised, (3) how signaling NaNs are distinguished +from quiet NaNs, (4) the default generated quiet NaNs, and (5) how NaNs +are propagated from function inputs to output. These details are target- +specific. +------------------------------------------------------------------------------- +*/ +#include "softfloat-specialize.h" + +#ifndef SOFTFLOAT_FOR_GCC /* Not used */ +/* +------------------------------------------------------------------------------- +Takes a 64-bit fixed-point value `absZ' with binary point between bits 6 +and 7, and returns the properly rounded 32-bit integer corresponding to the +input. If `zSign' is 1, the input is negated before being converted to an +integer. Bit 63 of `absZ' must be zero. Ordinarily, the fixed-point input +is simply rounded to an integer, with the inexact exception raised if the +input cannot be represented exactly as an integer. However, if the fixed- +point input is too large, the invalid exception is raised and the largest +positive or negative integer is returned. +------------------------------------------------------------------------------- +*/ +static int32 roundAndPackInt32( flag zSign, bits64 absZ ) +{ + int8 roundingMode; + flag roundNearestEven; + int8 roundIncrement, roundBits; + int32 z; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + roundIncrement = 0x40; + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + roundIncrement = 0; + } + else { + roundIncrement = 0x7F; + if ( zSign ) { + if ( roundingMode == float_round_up ) roundIncrement = 0; + } + else { + if ( roundingMode == float_round_down ) roundIncrement = 0; + } + } + } + roundBits = absZ & 0x7F; + absZ = ( absZ + roundIncrement )>>7; + absZ &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); + z = absZ; + if ( zSign ) z = - z; + if ( ( absZ>>32 ) || ( z && ( ( z < 0 ) ^ zSign ) ) ) { + float_raise( float_flag_invalid ); + return zSign ? (sbits32) 0x80000000 : 0x7FFFFFFF; + } + if ( roundBits ) float_set_inexact(); + return z; + +} + +/* +------------------------------------------------------------------------------- +Takes the 128-bit fixed-point value formed by concatenating `absZ0' and +`absZ1', with binary point between bits 63 and 64 (between the input words), +and returns the properly rounded 64-bit integer corresponding to the input. +If `zSign' is 1, the input is negated before being converted to an integer. +Ordinarily, the fixed-point input is simply rounded to an integer, with +the inexact exception raised if the input cannot be represented exactly as +an integer. However, if the fixed-point input is too large, the invalid +exception is raised and the largest positive or negative integer is +returned. +------------------------------------------------------------------------------- +*/ +static int64 roundAndPackInt64( flag zSign, bits64 absZ0, bits64 absZ1 ) +{ + int8 roundingMode; + flag roundNearestEven, increment; + int64 z; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + increment = ( (sbits64) absZ1 < 0 ); + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + increment = 0; + } + else { + if ( zSign ) { + increment = ( roundingMode == float_round_down ) && absZ1; + } + else { + increment = ( roundingMode == float_round_up ) && absZ1; + } + } + } + if ( increment ) { + ++absZ0; + if ( absZ0 == 0 ) goto overflow; + absZ0 &= ~ ( ( (bits64) ( absZ1<<1 ) == 0 ) & roundNearestEven ); + } + z = absZ0; + if ( zSign ) z = - z; + if ( z && ( ( z < 0 ) ^ zSign ) ) { + overflow: + float_raise( float_flag_invalid ); + return + zSign ? (sbits64) LIT64( 0x8000000000000000 ) + : LIT64( 0x7FFFFFFFFFFFFFFF ); + } + if ( absZ1 ) float_set_inexact(); + return z; + +} +#endif + +/* +------------------------------------------------------------------------------- +Returns the fraction bits of the single-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE bits32 extractFloat32Frac( float32 a ) +{ + + return a & 0x007FFFFF; + +} + +/* +------------------------------------------------------------------------------- +Returns the exponent bits of the single-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE int16 extractFloat32Exp( float32 a ) +{ + + return ( a>>23 ) & 0xFF; + +} + +/* +------------------------------------------------------------------------------- +Returns the sign bit of the single-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE flag extractFloat32Sign( float32 a ) +{ + + return a>>31; + +} + +/* +------------------------------------------------------------------------------- +Normalizes the subnormal single-precision floating-point value represented +by the denormalized significand `aSig'. The normalized exponent and +significand are stored at the locations pointed to by `zExpPtr' and +`zSigPtr', respectively. +------------------------------------------------------------------------------- +*/ +static void + normalizeFloat32Subnormal( bits32 aSig, int16 *zExpPtr, bits32 *zSigPtr ) +{ + int8 shiftCount; + + shiftCount = countLeadingZeros32( aSig ) - 8; + *zSigPtr = aSig<<shiftCount; + *zExpPtr = 1 - shiftCount; + +} + +/* +------------------------------------------------------------------------------- +Packs the sign `zSign', exponent `zExp', and significand `zSig' into a +single-precision floating-point value, returning the result. After being +shifted into the proper positions, the three fields are simply added +together to form the result. This means that any integer portion of `zSig' +will be added into the exponent. Since a properly normalized significand +will have an integer portion equal to 1, the `zExp' input should be 1 less +than the desired result exponent whenever `zSig' is a complete, normalized +significand. +------------------------------------------------------------------------------- +*/ +INLINE float32 packFloat32( flag zSign, int16 zExp, bits32 zSig ) +{ + + return ( ( (bits32) zSign )<<31 ) + ( ( (bits32) zExp )<<23 ) + zSig; + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and significand `zSig', and returns the proper single-precision floating- +point value corresponding to the abstract input. Ordinarily, the abstract +value is simply rounded and packed into the single-precision format, with +the inexact exception raised if the abstract input cannot be represented +exactly. However, if the abstract value is too large, the overflow and +inexact exceptions are raised and an infinity or maximal finite value is +returned. If the abstract value is too small, the input value is rounded to +a subnormal number, and the underflow and inexact exceptions are raised if +the abstract input cannot be represented exactly as a subnormal single- +precision floating-point number. + The input significand `zSig' has its binary point between bits 30 +and 29, which is 7 bits to the left of the usual location. This shifted +significand must be normalized or smaller. If `zSig' is not normalized, +`zExp' must be 0; in that case, the result returned is a subnormal number, +and it must not require rounding. In the usual case that `zSig' is +normalized, `zExp' must be 1 less than the ``true'' floating-point exponent. +The handling of underflow and overflow follows the IEC/IEEE Standard for +Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float32 roundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) +{ + int8 roundingMode; + flag roundNearestEven; + int8 roundIncrement, roundBits; + flag isTiny; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + roundIncrement = 0x40; + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + roundIncrement = 0; + } + else { + roundIncrement = 0x7F; + if ( zSign ) { + if ( roundingMode == float_round_up ) roundIncrement = 0; + } + else { + if ( roundingMode == float_round_down ) roundIncrement = 0; + } + } + } + roundBits = zSig & 0x7F; + if ( 0xFD <= (bits16) zExp ) { + if ( ( 0xFD < zExp ) + || ( ( zExp == 0xFD ) + && ( (sbits32) ( zSig + roundIncrement ) < 0 ) ) + ) { + float_raise( float_flag_overflow | float_flag_inexact ); + return packFloat32( zSign, 0xFF, 0 ) - ( roundIncrement == 0 ); + } + if ( zExp < 0 ) { + isTiny = + ( float_detect_tininess == float_tininess_before_rounding ) + || ( zExp < -1 ) + || ( zSig + roundIncrement < 0x80000000 ); + shift32RightJamming( zSig, - zExp, &zSig ); + zExp = 0; + roundBits = zSig & 0x7F; + if ( isTiny && roundBits ) float_raise( float_flag_underflow ); + } + } + if ( roundBits ) float_set_inexact(); + zSig = ( zSig + roundIncrement )>>7; + zSig &= ~ ( ( ( roundBits ^ 0x40 ) == 0 ) & roundNearestEven ); + if ( zSig == 0 ) zExp = 0; + return packFloat32( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and significand `zSig', and returns the proper single-precision floating- +point value corresponding to the abstract input. This routine is just like +`roundAndPackFloat32' except that `zSig' does not have to be normalized. +Bit 31 of `zSig' must be zero, and `zExp' must be 1 less than the ``true'' +floating-point exponent. +------------------------------------------------------------------------------- +*/ +static float32 + normalizeRoundAndPackFloat32( flag zSign, int16 zExp, bits32 zSig ) +{ + int8 shiftCount; + + shiftCount = countLeadingZeros32( zSig ) - 1; + return roundAndPackFloat32( zSign, zExp - shiftCount, zSig<<shiftCount ); + +} + +/* +------------------------------------------------------------------------------- +Returns the fraction bits of the double-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE bits64 extractFloat64Frac( float64 a ) +{ + + return FLOAT64_DEMANGLE(a) & LIT64( 0x000FFFFFFFFFFFFF ); + +} + +/* +------------------------------------------------------------------------------- +Returns the exponent bits of the double-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE int16 extractFloat64Exp( float64 a ) +{ + + return ( FLOAT64_DEMANGLE(a)>>52 ) & 0x7FF; + +} + +/* +------------------------------------------------------------------------------- +Returns the sign bit of the double-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE flag extractFloat64Sign( float64 a ) +{ + + return FLOAT64_DEMANGLE(a)>>63; + +} + +/* +------------------------------------------------------------------------------- +Normalizes the subnormal double-precision floating-point value represented +by the denormalized significand `aSig'. The normalized exponent and +significand are stored at the locations pointed to by `zExpPtr' and +`zSigPtr', respectively. +------------------------------------------------------------------------------- +*/ +static void + normalizeFloat64Subnormal( bits64 aSig, int16 *zExpPtr, bits64 *zSigPtr ) +{ + int8 shiftCount; + + shiftCount = countLeadingZeros64( aSig ) - 11; + *zSigPtr = aSig<<shiftCount; + *zExpPtr = 1 - shiftCount; + +} + +/* +------------------------------------------------------------------------------- +Packs the sign `zSign', exponent `zExp', and significand `zSig' into a +double-precision floating-point value, returning the result. After being +shifted into the proper positions, the three fields are simply added +together to form the result. This means that any integer portion of `zSig' +will be added into the exponent. Since a properly normalized significand +will have an integer portion equal to 1, the `zExp' input should be 1 less +than the desired result exponent whenever `zSig' is a complete, normalized +significand. +------------------------------------------------------------------------------- +*/ +INLINE float64 packFloat64( flag zSign, int16 zExp, bits64 zSig ) +{ + + return FLOAT64_MANGLE( ( ( (bits64) zSign )<<63 ) + + ( ( (bits64) zExp )<<52 ) + zSig ); + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and significand `zSig', and returns the proper double-precision floating- +point value corresponding to the abstract input. Ordinarily, the abstract +value is simply rounded and packed into the double-precision format, with +the inexact exception raised if the abstract input cannot be represented +exactly. However, if the abstract value is too large, the overflow and +inexact exceptions are raised and an infinity or maximal finite value is +returned. If the abstract value is too small, the input value is rounded to +a subnormal number, and the underflow and inexact exceptions are raised if +the abstract input cannot be represented exactly as a subnormal double- +precision floating-point number. + The input significand `zSig' has its binary point between bits 62 +and 61, which is 10 bits to the left of the usual location. This shifted +significand must be normalized or smaller. If `zSig' is not normalized, +`zExp' must be 0; in that case, the result returned is a subnormal number, +and it must not require rounding. In the usual case that `zSig' is +normalized, `zExp' must be 1 less than the ``true'' floating-point exponent. +The handling of underflow and overflow follows the IEC/IEEE Standard for +Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float64 roundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) +{ + int8 roundingMode; + flag roundNearestEven; + int16 roundIncrement, roundBits; + flag isTiny; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + roundIncrement = 0x200; + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + roundIncrement = 0; + } + else { + roundIncrement = 0x3FF; + if ( zSign ) { + if ( roundingMode == float_round_up ) roundIncrement = 0; + } + else { + if ( roundingMode == float_round_down ) roundIncrement = 0; + } + } + } + roundBits = zSig & 0x3FF; + if ( 0x7FD <= (bits16) zExp ) { + if ( ( 0x7FD < zExp ) + || ( ( zExp == 0x7FD ) + && ( (sbits64) ( zSig + roundIncrement ) < 0 ) ) + ) { + float_raise( float_flag_overflow | float_flag_inexact ); + return FLOAT64_MANGLE( + FLOAT64_DEMANGLE(packFloat64( zSign, 0x7FF, 0 )) - + ( roundIncrement == 0 )); + } + if ( zExp < 0 ) { + isTiny = + ( float_detect_tininess == float_tininess_before_rounding ) + || ( zExp < -1 ) + || ( zSig + roundIncrement < LIT64( 0x8000000000000000 ) ); + shift64RightJamming( zSig, - zExp, &zSig ); + zExp = 0; + roundBits = zSig & 0x3FF; + if ( isTiny && roundBits ) float_raise( float_flag_underflow ); + } + } + if ( roundBits ) float_set_inexact(); + zSig = ( zSig + roundIncrement )>>10; + zSig &= ~ ( ( ( roundBits ^ 0x200 ) == 0 ) & roundNearestEven ); + if ( zSig == 0 ) zExp = 0; + return packFloat64( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and significand `zSig', and returns the proper double-precision floating- +point value corresponding to the abstract input. This routine is just like +`roundAndPackFloat64' except that `zSig' does not have to be normalized. +Bit 63 of `zSig' must be zero, and `zExp' must be 1 less than the ``true'' +floating-point exponent. +------------------------------------------------------------------------------- +*/ +static float64 + normalizeRoundAndPackFloat64( flag zSign, int16 zExp, bits64 zSig ) +{ + int8 shiftCount; + + shiftCount = countLeadingZeros64( zSig ) - 1; + return roundAndPackFloat64( zSign, zExp - shiftCount, zSig<<shiftCount ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the fraction bits of the extended double-precision floating-point +value `a'. +------------------------------------------------------------------------------- +*/ +INLINE bits64 extractFloatx80Frac( floatx80 a ) +{ + + return a.low; + +} + +/* +------------------------------------------------------------------------------- +Returns the exponent bits of the extended double-precision floating-point +value `a'. +------------------------------------------------------------------------------- +*/ +INLINE int32 extractFloatx80Exp( floatx80 a ) +{ + + return a.high & 0x7FFF; + +} + +/* +------------------------------------------------------------------------------- +Returns the sign bit of the extended double-precision floating-point value +`a'. +------------------------------------------------------------------------------- +*/ +INLINE flag extractFloatx80Sign( floatx80 a ) +{ + + return a.high>>15; + +} + +/* +------------------------------------------------------------------------------- +Normalizes the subnormal extended double-precision floating-point value +represented by the denormalized significand `aSig'. The normalized exponent +and significand are stored at the locations pointed to by `zExpPtr' and +`zSigPtr', respectively. +------------------------------------------------------------------------------- +*/ +static void + normalizeFloatx80Subnormal( bits64 aSig, int32 *zExpPtr, bits64 *zSigPtr ) +{ + int8 shiftCount; + + shiftCount = countLeadingZeros64( aSig ); + *zSigPtr = aSig<<shiftCount; + *zExpPtr = 1 - shiftCount; + +} + +/* +------------------------------------------------------------------------------- +Packs the sign `zSign', exponent `zExp', and significand `zSig' into an +extended double-precision floating-point value, returning the result. +------------------------------------------------------------------------------- +*/ +INLINE floatx80 packFloatx80( flag zSign, int32 zExp, bits64 zSig ) +{ + floatx80 z; + + z.low = zSig; + z.high = ( ( (bits16) zSign )<<15 ) + zExp; + return z; + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and extended significand formed by the concatenation of `zSig0' and `zSig1', +and returns the proper extended double-precision floating-point value +corresponding to the abstract input. Ordinarily, the abstract value is +rounded and packed into the extended double-precision format, with the +inexact exception raised if the abstract input cannot be represented +exactly. However, if the abstract value is too large, the overflow and +inexact exceptions are raised and an infinity or maximal finite value is +returned. If the abstract value is too small, the input value is rounded to +a subnormal number, and the underflow and inexact exceptions are raised if +the abstract input cannot be represented exactly as a subnormal extended +double-precision floating-point number. + If `roundingPrecision' is 32 or 64, the result is rounded to the same +number of bits as single or double precision, respectively. Otherwise, the +result is rounded to the full precision of the extended double-precision +format. + The input significand must be normalized or smaller. If the input +significand is not normalized, `zExp' must be 0; in that case, the result +returned is a subnormal number, and it must not require rounding. The +handling of underflow and overflow follows the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static floatx80 + roundAndPackFloatx80( + int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 + ) +{ + int8 roundingMode; + flag roundNearestEven, increment, isTiny; + int64 roundIncrement, roundMask, roundBits; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + if ( roundingPrecision == 80 ) goto precision80; + if ( roundingPrecision == 64 ) { + roundIncrement = LIT64( 0x0000000000000400 ); + roundMask = LIT64( 0x00000000000007FF ); + } + else if ( roundingPrecision == 32 ) { + roundIncrement = LIT64( 0x0000008000000000 ); + roundMask = LIT64( 0x000000FFFFFFFFFF ); + } + else { + goto precision80; + } + zSig0 |= ( zSig1 != 0 ); + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + roundIncrement = 0; + } + else { + roundIncrement = roundMask; + if ( zSign ) { + if ( roundingMode == float_round_up ) roundIncrement = 0; + } + else { + if ( roundingMode == float_round_down ) roundIncrement = 0; + } + } + } + roundBits = zSig0 & roundMask; + if ( 0x7FFD <= (bits32) ( zExp - 1 ) ) { + if ( ( 0x7FFE < zExp ) + || ( ( zExp == 0x7FFE ) && ( zSig0 + roundIncrement < zSig0 ) ) + ) { + goto overflow; + } + if ( zExp <= 0 ) { + isTiny = + ( float_detect_tininess == float_tininess_before_rounding ) + || ( zExp < 0 ) + || ( zSig0 <= zSig0 + roundIncrement ); + shift64RightJamming( zSig0, 1 - zExp, &zSig0 ); + zExp = 0; + roundBits = zSig0 & roundMask; + if ( isTiny && roundBits ) float_raise( float_flag_underflow ); + if ( roundBits ) float_set_inexact(); + zSig0 += roundIncrement; + if ( (sbits64) zSig0 < 0 ) zExp = 1; + roundIncrement = roundMask + 1; + if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) { + roundMask |= roundIncrement; + } + zSig0 &= ~ roundMask; + return packFloatx80( zSign, zExp, zSig0 ); + } + } + if ( roundBits ) float_set_inexact(); + zSig0 += roundIncrement; + if ( zSig0 < roundIncrement ) { + ++zExp; + zSig0 = LIT64( 0x8000000000000000 ); + } + roundIncrement = roundMask + 1; + if ( roundNearestEven && ( roundBits<<1 == roundIncrement ) ) { + roundMask |= roundIncrement; + } + zSig0 &= ~ roundMask; + if ( zSig0 == 0 ) zExp = 0; + return packFloatx80( zSign, zExp, zSig0 ); + precision80: + increment = ( (sbits64) zSig1 < 0 ); + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + increment = 0; + } + else { + if ( zSign ) { + increment = ( roundingMode == float_round_down ) && zSig1; + } + else { + increment = ( roundingMode == float_round_up ) && zSig1; + } + } + } + if ( 0x7FFD <= (bits32) ( zExp - 1 ) ) { + if ( ( 0x7FFE < zExp ) + || ( ( zExp == 0x7FFE ) + && ( zSig0 == LIT64( 0xFFFFFFFFFFFFFFFF ) ) + && increment + ) + ) { + roundMask = 0; + overflow: + float_raise( float_flag_overflow | float_flag_inexact ); + if ( ( roundingMode == float_round_to_zero ) + || ( zSign && ( roundingMode == float_round_up ) ) + || ( ! zSign && ( roundingMode == float_round_down ) ) + ) { + return packFloatx80( zSign, 0x7FFE, ~ roundMask ); + } + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( zExp <= 0 ) { + isTiny = + ( float_detect_tininess == float_tininess_before_rounding ) + || ( zExp < 0 ) + || ! increment + || ( zSig0 < LIT64( 0xFFFFFFFFFFFFFFFF ) ); + shift64ExtraRightJamming( zSig0, zSig1, 1 - zExp, &zSig0, &zSig1 ); + zExp = 0; + if ( isTiny && zSig1 ) float_raise( float_flag_underflow ); + if ( zSig1 ) float_set_inexact(); + if ( roundNearestEven ) { + increment = ( (sbits64) zSig1 < 0 ); + } + else { + if ( zSign ) { + increment = ( roundingMode == float_round_down ) && zSig1; + } + else { + increment = ( roundingMode == float_round_up ) && zSig1; + } + } + if ( increment ) { + ++zSig0; + zSig0 &= + ~ ( ( (bits64) ( zSig1<<1 ) == 0 ) & roundNearestEven ); + if ( (sbits64) zSig0 < 0 ) zExp = 1; + } + return packFloatx80( zSign, zExp, zSig0 ); + } + } + if ( zSig1 ) float_set_inexact(); + if ( increment ) { + ++zSig0; + if ( zSig0 == 0 ) { + ++zExp; + zSig0 = LIT64( 0x8000000000000000 ); + } + else { + zSig0 &= ~ ( ( (bits64) ( zSig1<<1 ) == 0 ) & roundNearestEven ); + } + } + else { + if ( zSig0 == 0 ) zExp = 0; + } + return packFloatx80( zSign, zExp, zSig0 ); + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent +`zExp', and significand formed by the concatenation of `zSig0' and `zSig1', +and returns the proper extended double-precision floating-point value +corresponding to the abstract input. This routine is just like +`roundAndPackFloatx80' except that the input significand does not have to be +normalized. +------------------------------------------------------------------------------- +*/ +static floatx80 + normalizeRoundAndPackFloatx80( + int8 roundingPrecision, flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 + ) +{ + int8 shiftCount; + + if ( zSig0 == 0 ) { + zSig0 = zSig1; + zSig1 = 0; + zExp -= 64; + } + shiftCount = countLeadingZeros64( zSig0 ); + shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); + zExp -= shiftCount; + return + roundAndPackFloatx80( roundingPrecision, zSign, zExp, zSig0, zSig1 ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the least-significant 64 fraction bits of the quadruple-precision +floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE bits64 extractFloat128Frac1( float128 a ) +{ + + return a.low; + +} + +/* +------------------------------------------------------------------------------- +Returns the most-significant 48 fraction bits of the quadruple-precision +floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE bits64 extractFloat128Frac0( float128 a ) +{ + + return a.high & LIT64( 0x0000FFFFFFFFFFFF ); + +} + +/* +------------------------------------------------------------------------------- +Returns the exponent bits of the quadruple-precision floating-point value +`a'. +------------------------------------------------------------------------------- +*/ +INLINE int32 extractFloat128Exp( float128 a ) +{ + + return ( a.high>>48 ) & 0x7FFF; + +} + +/* +------------------------------------------------------------------------------- +Returns the sign bit of the quadruple-precision floating-point value `a'. +------------------------------------------------------------------------------- +*/ +INLINE flag extractFloat128Sign( float128 a ) +{ + + return a.high>>63; + +} + +/* +------------------------------------------------------------------------------- +Normalizes the subnormal quadruple-precision floating-point value +represented by the denormalized significand formed by the concatenation of +`aSig0' and `aSig1'. The normalized exponent is stored at the location +pointed to by `zExpPtr'. The most significant 49 bits of the normalized +significand are stored at the location pointed to by `zSig0Ptr', and the +least significant 64 bits of the normalized significand are stored at the +location pointed to by `zSig1Ptr'. +------------------------------------------------------------------------------- +*/ +static void + normalizeFloat128Subnormal( + bits64 aSig0, + bits64 aSig1, + int32 *zExpPtr, + bits64 *zSig0Ptr, + bits64 *zSig1Ptr + ) +{ + int8 shiftCount; + + if ( aSig0 == 0 ) { + shiftCount = countLeadingZeros64( aSig1 ) - 15; + if ( shiftCount < 0 ) { + *zSig0Ptr = aSig1>>( - shiftCount ); + *zSig1Ptr = aSig1<<( shiftCount & 63 ); + } + else { + *zSig0Ptr = aSig1<<shiftCount; + *zSig1Ptr = 0; + } + *zExpPtr = - shiftCount - 63; + } + else { + shiftCount = countLeadingZeros64( aSig0 ) - 15; + shortShift128Left( aSig0, aSig1, shiftCount, zSig0Ptr, zSig1Ptr ); + *zExpPtr = 1 - shiftCount; + } + +} + +/* +------------------------------------------------------------------------------- +Packs the sign `zSign', the exponent `zExp', and the significand formed +by the concatenation of `zSig0' and `zSig1' into a quadruple-precision +floating-point value, returning the result. After being shifted into the +proper positions, the three fields `zSign', `zExp', and `zSig0' are simply +added together to form the most significant 32 bits of the result. This +means that any integer portion of `zSig0' will be added into the exponent. +Since a properly normalized significand will have an integer portion equal +to 1, the `zExp' input should be 1 less than the desired result exponent +whenever `zSig0' and `zSig1' concatenated form a complete, normalized +significand. +------------------------------------------------------------------------------- +*/ +INLINE float128 + packFloat128( flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 ) +{ + float128 z; + + z.low = zSig1; + z.high = ( ( (bits64) zSign )<<63 ) + ( ( (bits64) zExp )<<48 ) + zSig0; + return z; + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and extended significand formed by the concatenation of `zSig0', `zSig1', +and `zSig2', and returns the proper quadruple-precision floating-point value +corresponding to the abstract input. Ordinarily, the abstract value is +simply rounded and packed into the quadruple-precision format, with the +inexact exception raised if the abstract input cannot be represented +exactly. However, if the abstract value is too large, the overflow and +inexact exceptions are raised and an infinity or maximal finite value is +returned. If the abstract value is too small, the input value is rounded to +a subnormal number, and the underflow and inexact exceptions are raised if +the abstract input cannot be represented exactly as a subnormal quadruple- +precision floating-point number. + The input significand must be normalized or smaller. If the input +significand is not normalized, `zExp' must be 0; in that case, the result +returned is a subnormal number, and it must not require rounding. In the +usual case that the input significand is normalized, `zExp' must be 1 less +than the ``true'' floating-point exponent. The handling of underflow and +overflow follows the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float128 + roundAndPackFloat128( + flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1, bits64 zSig2 ) +{ + int8 roundingMode; + flag roundNearestEven, increment, isTiny; + + roundingMode = float_rounding_mode(); + roundNearestEven = ( roundingMode == float_round_nearest_even ); + increment = ( (sbits64) zSig2 < 0 ); + if ( ! roundNearestEven ) { + if ( roundingMode == float_round_to_zero ) { + increment = 0; + } + else { + if ( zSign ) { + increment = ( roundingMode == float_round_down ) && zSig2; + } + else { + increment = ( roundingMode == float_round_up ) && zSig2; + } + } + } + if ( 0x7FFD <= (bits32) zExp ) { + if ( ( 0x7FFD < zExp ) + || ( ( zExp == 0x7FFD ) + && eq128( + LIT64( 0x0001FFFFFFFFFFFF ), + LIT64( 0xFFFFFFFFFFFFFFFF ), + zSig0, + zSig1 + ) + && increment + ) + ) { + float_raise( float_flag_overflow | float_flag_inexact ); + if ( ( roundingMode == float_round_to_zero ) + || ( zSign && ( roundingMode == float_round_up ) ) + || ( ! zSign && ( roundingMode == float_round_down ) ) + ) { + return + packFloat128( + zSign, + 0x7FFE, + LIT64( 0x0000FFFFFFFFFFFF ), + LIT64( 0xFFFFFFFFFFFFFFFF ) + ); + } + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + if ( zExp < 0 ) { + isTiny = + ( float_detect_tininess == float_tininess_before_rounding ) + || ( zExp < -1 ) + || ! increment + || lt128( + zSig0, + zSig1, + LIT64( 0x0001FFFFFFFFFFFF ), + LIT64( 0xFFFFFFFFFFFFFFFF ) + ); + shift128ExtraRightJamming( + zSig0, zSig1, zSig2, - zExp, &zSig0, &zSig1, &zSig2 ); + zExp = 0; + if ( isTiny && zSig2 ) float_raise( float_flag_underflow ); + if ( roundNearestEven ) { + increment = ( (sbits64) zSig2 < 0 ); + } + else { + if ( zSign ) { + increment = ( roundingMode == float_round_down ) && zSig2; + } + else { + increment = ( roundingMode == float_round_up ) && zSig2; + } + } + } + } + if ( zSig2 ) float_set_inexact(); + if ( increment ) { + add128( zSig0, zSig1, 0, 1, &zSig0, &zSig1 ); + zSig1 &= ~ ( ( zSig2 + zSig2 == 0 ) & roundNearestEven ); + } + else { + if ( ( zSig0 | zSig1 ) == 0 ) zExp = 0; + } + return packFloat128( zSign, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Takes an abstract floating-point value having sign `zSign', exponent `zExp', +and significand formed by the concatenation of `zSig0' and `zSig1', and +returns the proper quadruple-precision floating-point value corresponding +to the abstract input. This routine is just like `roundAndPackFloat128' +except that the input significand has fewer bits and does not have to be +normalized. In all cases, `zExp' must be 1 less than the ``true'' floating- +point exponent. +------------------------------------------------------------------------------- +*/ +static float128 + normalizeRoundAndPackFloat128( + flag zSign, int32 zExp, bits64 zSig0, bits64 zSig1 ) +{ + int8 shiftCount; + bits64 zSig2; + + if ( zSig0 == 0 ) { + zSig0 = zSig1; + zSig1 = 0; + zExp -= 64; + } + shiftCount = countLeadingZeros64( zSig0 ) - 15; + if ( 0 <= shiftCount ) { + zSig2 = 0; + shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); + } + else { + shift128ExtraRightJamming( + zSig0, zSig1, 0, - shiftCount, &zSig0, &zSig1, &zSig2 ); + } + zExp -= shiftCount; + return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 ); + +} + +#endif + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 32-bit two's complement integer `a' +to the single-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 int32_to_float32( int32 a ) +{ + flag zSign; + + if ( a == 0 ) return 0; + if ( a == (sbits32) 0x80000000 ) return packFloat32( 1, 0x9E, 0 ); + zSign = ( a < 0 ); + return normalizeRoundAndPackFloat32( zSign, 0x9C, zSign ? - a : a ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 32-bit two's complement integer `a' +to the double-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 int32_to_float64( int32 a ) +{ + flag zSign; + uint32 absA; + int8 shiftCount; + bits64 zSig; + + if ( a == 0 ) return 0; + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros32( absA ) + 21; + zSig = absA; + return packFloat64( zSign, 0x432 - shiftCount, zSig<<shiftCount ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 32-bit two's complement integer `a' +to the extended double-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 int32_to_floatx80( int32 a ) +{ + flag zSign; + uint32 absA; + int8 shiftCount; + bits64 zSig; + + if ( a == 0 ) return packFloatx80( 0, 0, 0 ); + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros32( absA ) + 32; + zSig = absA; + return packFloatx80( zSign, 0x403E - shiftCount, zSig<<shiftCount ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 32-bit two's complement integer `a' to +the quadruple-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 int32_to_float128( int32 a ) +{ + flag zSign; + uint32 absA; + int8 shiftCount; + bits64 zSig0; + + if ( a == 0 ) return packFloat128( 0, 0, 0, 0 ); + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros32( absA ) + 17; + zSig0 = absA; + return packFloat128( zSign, 0x402E - shiftCount, zSig0<<shiftCount, 0 ); + +} + +#endif + +#ifndef SOFTFLOAT_FOR_GCC /* __floatdi?f is in libgcc2.c */ +/* +------------------------------------------------------------------------------- +Returns the result of converting the 64-bit two's complement integer `a' +to the single-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 int64_to_float32( int64 a ) +{ + flag zSign; + uint64 absA; + int8 shiftCount; + + if ( a == 0 ) return 0; + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros64( absA ) - 40; + if ( 0 <= shiftCount ) { + return packFloat32( zSign, 0x95 - shiftCount, absA<<shiftCount ); + } + else { + shiftCount += 7; + if ( shiftCount < 0 ) { + shift64RightJamming( absA, - shiftCount, &absA ); + } + else { + absA <<= shiftCount; + } + return roundAndPackFloat32( zSign, 0x9C - shiftCount, absA ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 64-bit two's complement integer `a' +to the double-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 int64_to_float64( int64 a ) +{ + flag zSign; + + if ( a == 0 ) return 0; + if ( a == (sbits64) LIT64( 0x8000000000000000 ) ) { + return packFloat64( 1, 0x43E, 0 ); + } + zSign = ( a < 0 ); + return normalizeRoundAndPackFloat64( zSign, 0x43C, zSign ? - a : a ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 64-bit two's complement integer `a' +to the extended double-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 int64_to_floatx80( int64 a ) +{ + flag zSign; + uint64 absA; + int8 shiftCount; + + if ( a == 0 ) return packFloatx80( 0, 0, 0 ); + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros64( absA ); + return packFloatx80( zSign, 0x403E - shiftCount, absA<<shiftCount ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the 64-bit two's complement integer `a' to +the quadruple-precision floating-point format. The conversion is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 int64_to_float128( int64 a ) +{ + flag zSign; + uint64 absA; + int8 shiftCount; + int32 zExp; + bits64 zSig0, zSig1; + + if ( a == 0 ) return packFloat128( 0, 0, 0, 0 ); + zSign = ( a < 0 ); + absA = zSign ? - a : a; + shiftCount = countLeadingZeros64( absA ) + 49; + zExp = 0x406E - shiftCount; + if ( 64 <= shiftCount ) { + zSig1 = 0; + zSig0 = absA; + shiftCount -= 64; + } + else { + zSig1 = absA; + zSig0 = 0; + } + shortShift128Left( zSig0, zSig1, shiftCount, &zSig0, &zSig1 ); + return packFloat128( zSign, zExp, zSig0, zSig1 ); + +} + +#endif +#endif /* !SOFTFLOAT_FOR_GCC */ + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the 32-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int32 float32_to_int32( float32 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits32 aSig; + bits64 aSig64; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + if ( ( aExp == 0xFF ) && aSig ) aSign = 0; + if ( aExp ) aSig |= 0x00800000; + shiftCount = 0xAF - aExp; + aSig64 = aSig; + aSig64 <<= 32; + if ( 0 < shiftCount ) shift64RightJamming( aSig64, shiftCount, &aSig64 ); + return roundAndPackInt32( aSign, aSig64 ); + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the 32-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. +If `a' is a NaN, the largest positive integer is returned. Otherwise, if +the conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int32 float32_to_int32_round_to_zero( float32 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits32 aSig; + int32 z; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + shiftCount = aExp - 0x9E; + if ( 0 <= shiftCount ) { + if ( a != 0xCF000000 ) { + float_raise( float_flag_invalid ); + if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) return 0x7FFFFFFF; + } + return (sbits32) 0x80000000; + } + else if ( aExp <= 0x7E ) { + if ( aExp | aSig ) float_set_inexact(); + return 0; + } + aSig = ( aSig | 0x00800000 )<<8; + z = aSig>>( - shiftCount ); + if ( (bits32) ( aSig<<( shiftCount & 31 ) ) ) { + float_set_inexact(); + } + if ( aSign ) z = - z; + return z; + +} + +#ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */ +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the 64-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int64 float32_to_int64( float32 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits32 aSig; + bits64 aSig64, aSigExtra; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + shiftCount = 0xBE - aExp; + if ( shiftCount < 0 ) { + float_raise( float_flag_invalid ); + if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + if ( aExp ) aSig |= 0x00800000; + aSig64 = aSig; + aSig64 <<= 40; + shift64ExtraRightJamming( aSig64, 0, shiftCount, &aSig64, &aSigExtra ); + return roundAndPackInt64( aSign, aSig64, aSigExtra ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the 64-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. If +`a' is a NaN, the largest positive integer is returned. Otherwise, if the +conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int64 float32_to_int64_round_to_zero( float32 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits32 aSig; + bits64 aSig64; + int64 z; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + shiftCount = aExp - 0xBE; + if ( 0 <= shiftCount ) { + if ( a != 0xDF000000 ) { + float_raise( float_flag_invalid ); + if ( ! aSign || ( ( aExp == 0xFF ) && aSig ) ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + else if ( aExp <= 0x7E ) { + if ( aExp | aSig ) float_set_inexact(); + return 0; + } + aSig64 = aSig | 0x00800000; + aSig64 <<= 40; + z = aSig64>>( - shiftCount ); + if ( (bits64) ( aSig64<<( shiftCount & 63 ) ) ) { + float_set_inexact(); + } + if ( aSign ) z = - z; + return z; + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the double-precision floating-point format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float32_to_float64( float32 a ) +{ + flag aSign; + int16 aExp; + bits32 aSig; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + if ( aExp == 0xFF ) { + if ( aSig ) return commonNaNToFloat64( float32ToCommonNaN( a ) ); + return packFloat64( aSign, 0x7FF, 0 ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat64( aSign, 0, 0 ); + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + --aExp; + } + return packFloat64( aSign, aExp + 0x380, ( (bits64) aSig )<<29 ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the extended double-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 float32_to_floatx80( float32 a ) +{ + flag aSign; + int16 aExp; + bits32 aSig; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + if ( aExp == 0xFF ) { + if ( aSig ) return commonNaNToFloatx80( float32ToCommonNaN( a ) ); + return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloatx80( aSign, 0, 0 ); + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + } + aSig |= 0x00800000; + return packFloatx80( aSign, aExp + 0x3F80, ( (bits64) aSig )<<40 ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the double-precision floating-point format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float32_to_float128( float32 a ) +{ + flag aSign; + int16 aExp; + bits32 aSig; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + if ( aExp == 0xFF ) { + if ( aSig ) return commonNaNToFloat128( float32ToCommonNaN( a ) ); + return packFloat128( aSign, 0x7FFF, 0, 0 ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat128( aSign, 0, 0, 0 ); + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + --aExp; + } + return packFloat128( aSign, aExp + 0x3F80, ( (bits64) aSig )<<25, 0 ); + +} + +#endif + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Rounds the single-precision floating-point value `a' to an integer, and +returns the result as a single-precision floating-point value. The +operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_round_to_int( float32 a ) +{ + flag aSign; + int16 aExp; + bits32 lastBitMask, roundBitsMask; + int8 roundingMode; + float32 z; + + aExp = extractFloat32Exp( a ); + if ( 0x96 <= aExp ) { + if ( ( aExp == 0xFF ) && extractFloat32Frac( a ) ) { + return propagateFloat32NaN( a, a ); + } + return a; + } + if ( aExp <= 0x7E ) { + if ( (bits32) ( a<<1 ) == 0 ) return a; + float_set_inexact(); + aSign = extractFloat32Sign( a ); + switch ( float_rounding_mode() ) { + case float_round_nearest_even: + if ( ( aExp == 0x7E ) && extractFloat32Frac( a ) ) { + return packFloat32( aSign, 0x7F, 0 ); + } + break; + case float_round_down: + return aSign ? 0xBF800000 : 0; + case float_round_up: + return aSign ? 0x80000000 : 0x3F800000; + } + return packFloat32( aSign, 0, 0 ); + } + lastBitMask = 1; + lastBitMask <<= 0x96 - aExp; + roundBitsMask = lastBitMask - 1; + z = a; + roundingMode = float_rounding_mode(); + if ( roundingMode == float_round_nearest_even ) { + z += lastBitMask>>1; + if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; + } + else if ( roundingMode != float_round_to_zero ) { + if ( extractFloat32Sign( z ) ^ ( roundingMode == float_round_up ) ) { + z += roundBitsMask; + } + } + z &= ~ roundBitsMask; + if ( z != a ) float_set_inexact(); + return z; + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns the result of adding the absolute values of the single-precision +floating-point values `a' and `b'. If `zSign' is 1, the sum is negated +before being returned. `zSign' is ignored if the result is a NaN. +The addition is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float32 addFloat32Sigs( float32 a, float32 b, flag zSign ) +{ + int16 aExp, bExp, zExp; + bits32 aSig, bSig, zSig; + int16 expDiff; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + bSig = extractFloat32Frac( b ); + bExp = extractFloat32Exp( b ); + expDiff = aExp - bExp; + aSig <<= 6; + bSig <<= 6; + if ( 0 < expDiff ) { + if ( aExp == 0xFF ) { + if ( aSig ) return propagateFloat32NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig |= 0x20000000; + } + shift32RightJamming( bSig, expDiff, &bSig ); + zExp = aExp; + } + else if ( expDiff < 0 ) { + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + return packFloat32( zSign, 0xFF, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig |= 0x20000000; + } + shift32RightJamming( aSig, - expDiff, &aSig ); + zExp = bExp; + } + else { + if ( aExp == 0xFF ) { + if ( aSig | bSig ) return propagateFloat32NaN( a, b ); + return a; + } + if ( aExp == 0 ) return packFloat32( zSign, 0, ( aSig + bSig )>>6 ); + zSig = 0x40000000 + aSig + bSig; + zExp = aExp; + goto roundAndPack; + } + aSig |= 0x20000000; + zSig = ( aSig + bSig )<<1; + --zExp; + if ( (sbits32) zSig < 0 ) { + zSig = aSig + bSig; + ++zExp; + } + roundAndPack: + return roundAndPackFloat32( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the absolute values of the single- +precision floating-point values `a' and `b'. If `zSign' is 1, the +difference is negated before being returned. `zSign' is ignored if the +result is a NaN. The subtraction is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float32 subFloat32Sigs( float32 a, float32 b, flag zSign ) +{ + int16 aExp, bExp, zExp; + bits32 aSig, bSig, zSig; + int16 expDiff; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + bSig = extractFloat32Frac( b ); + bExp = extractFloat32Exp( b ); + expDiff = aExp - bExp; + aSig <<= 7; + bSig <<= 7; + if ( 0 < expDiff ) goto aExpBigger; + if ( expDiff < 0 ) goto bExpBigger; + if ( aExp == 0xFF ) { + if ( aSig | bSig ) return propagateFloat32NaN( a, b ); + float_raise( float_flag_invalid ); + return float32_default_nan; + } + if ( aExp == 0 ) { + aExp = 1; + bExp = 1; + } + if ( bSig < aSig ) goto aBigger; + if ( aSig < bSig ) goto bBigger; + return packFloat32( float_rounding_mode() == float_round_down, 0, 0 ); + bExpBigger: + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + return packFloat32( zSign ^ 1, 0xFF, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig |= 0x40000000; + } + shift32RightJamming( aSig, - expDiff, &aSig ); + bSig |= 0x40000000; + bBigger: + zSig = bSig - aSig; + zExp = bExp; + zSign ^= 1; + goto normalizeRoundAndPack; + aExpBigger: + if ( aExp == 0xFF ) { + if ( aSig ) return propagateFloat32NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig |= 0x40000000; + } + shift32RightJamming( bSig, expDiff, &bSig ); + aSig |= 0x40000000; + aBigger: + zSig = aSig - bSig; + zExp = aExp; + normalizeRoundAndPack: + --zExp; + return normalizeRoundAndPackFloat32( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the single-precision floating-point values `a' +and `b'. The operation is performed according to the IEC/IEEE Standard for +Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_add( float32 a, float32 b ) +{ + flag aSign, bSign; + + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign == bSign ) { + return addFloat32Sigs( a, b, aSign ); + } + else { + return subFloat32Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the single-precision floating-point values +`a' and `b'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_sub( float32 a, float32 b ) +{ + flag aSign, bSign; + + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign == bSign ) { + return subFloat32Sigs( a, b, aSign ); + } + else { + return addFloat32Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of multiplying the single-precision floating-point values +`a' and `b'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_mul( float32 a, float32 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, zExp; + bits32 aSig, bSig; + bits64 zSig64; + bits32 zSig; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + bSig = extractFloat32Frac( b ); + bExp = extractFloat32Exp( b ); + bSign = extractFloat32Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0xFF ) { + if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { + return propagateFloat32NaN( a, b ); + } + if ( ( bExp | bSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float32_default_nan; + } + return packFloat32( zSign, 0xFF, 0 ); + } + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + if ( ( aExp | aSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float32_default_nan; + } + return packFloat32( zSign, 0xFF, 0 ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat32( zSign, 0, 0 ); + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) return packFloat32( zSign, 0, 0 ); + normalizeFloat32Subnormal( bSig, &bExp, &bSig ); + } + zExp = aExp + bExp - 0x7F; + aSig = ( aSig | 0x00800000 )<<7; + bSig = ( bSig | 0x00800000 )<<8; + shift64RightJamming( ( (bits64) aSig ) * bSig, 32, &zSig64 ); + zSig = zSig64; + if ( 0 <= (sbits32) ( zSig<<1 ) ) { + zSig <<= 1; + --zExp; + } + return roundAndPackFloat32( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of dividing the single-precision floating-point value `a' +by the corresponding value `b'. The operation is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_div( float32 a, float32 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, zExp; + bits32 aSig, bSig, zSig; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + bSig = extractFloat32Frac( b ); + bExp = extractFloat32Exp( b ); + bSign = extractFloat32Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0xFF ) { + if ( aSig ) return propagateFloat32NaN( a, b ); + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + float_raise( float_flag_invalid ); + return float32_default_nan; + } + return packFloat32( zSign, 0xFF, 0 ); + } + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + return packFloat32( zSign, 0, 0 ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + if ( ( aExp | aSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float32_default_nan; + } + float_raise( float_flag_divbyzero ); + return packFloat32( zSign, 0xFF, 0 ); + } + normalizeFloat32Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat32( zSign, 0, 0 ); + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + } + zExp = aExp - bExp + 0x7D; + aSig = ( aSig | 0x00800000 )<<7; + bSig = ( bSig | 0x00800000 )<<8; + if ( bSig <= ( aSig + aSig ) ) { + aSig >>= 1; + ++zExp; + } + zSig = ( ( (bits64) aSig )<<32 ) / bSig; + if ( ( zSig & 0x3F ) == 0 ) { + zSig |= ( (bits64) bSig * zSig != ( (bits64) aSig )<<32 ); + } + return roundAndPackFloat32( zSign, zExp, zSig ); + +} + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns the remainder of the single-precision floating-point value `a' +with respect to the corresponding value `b'. The operation is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_rem( float32 a, float32 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, expDiff; + bits32 aSig, bSig; + bits32 q; + bits64 aSig64, bSig64, q64; + bits32 alternateASig; + sbits32 sigMean; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + bSig = extractFloat32Frac( b ); + bExp = extractFloat32Exp( b ); + bSign = extractFloat32Sign( b ); + if ( aExp == 0xFF ) { + if ( aSig || ( ( bExp == 0xFF ) && bSig ) ) { + return propagateFloat32NaN( a, b ); + } + float_raise( float_flag_invalid ); + return float32_default_nan; + } + if ( bExp == 0xFF ) { + if ( bSig ) return propagateFloat32NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + float_raise( float_flag_invalid ); + return float32_default_nan; + } + normalizeFloat32Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return a; + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + } + expDiff = aExp - bExp; + aSig |= 0x00800000; + bSig |= 0x00800000; + if ( expDiff < 32 ) { + aSig <<= 8; + bSig <<= 8; + if ( expDiff < 0 ) { + if ( expDiff < -1 ) return a; + aSig >>= 1; + } + q = ( bSig <= aSig ); + if ( q ) aSig -= bSig; + if ( 0 < expDiff ) { + q = ( ( (bits64) aSig )<<32 ) / bSig; + q >>= 32 - expDiff; + bSig >>= 2; + aSig = ( ( aSig>>1 )<<( expDiff - 1 ) ) - bSig * q; + } + else { + aSig >>= 2; + bSig >>= 2; + } + } + else { + if ( bSig <= aSig ) aSig -= bSig; + aSig64 = ( (bits64) aSig )<<40; + bSig64 = ( (bits64) bSig )<<40; + expDiff -= 64; + while ( 0 < expDiff ) { + q64 = estimateDiv128To64( aSig64, 0, bSig64 ); + q64 = ( 2 < q64 ) ? q64 - 2 : 0; + aSig64 = - ( ( bSig * q64 )<<38 ); + expDiff -= 62; + } + expDiff += 64; + q64 = estimateDiv128To64( aSig64, 0, bSig64 ); + q64 = ( 2 < q64 ) ? q64 - 2 : 0; + q = q64>>( 64 - expDiff ); + bSig <<= 6; + aSig = ( ( aSig64>>33 )<<( expDiff - 1 ) ) - bSig * q; + } + do { + alternateASig = aSig; + ++q; + aSig -= bSig; + } while ( 0 <= (sbits32) aSig ); + sigMean = aSig + alternateASig; + if ( ( sigMean < 0 ) || ( ( sigMean == 0 ) && ( q & 1 ) ) ) { + aSig = alternateASig; + } + zSign = ( (sbits32) aSig < 0 ); + if ( zSign ) aSig = - aSig; + return normalizeRoundAndPackFloat32( aSign ^ zSign, bExp, aSig ); + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns the square root of the single-precision floating-point value `a'. +The operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float32_sqrt( float32 a ) +{ + flag aSign; + int16 aExp, zExp; + bits32 aSig, zSig; + bits64 rem, term; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + if ( aExp == 0xFF ) { + if ( aSig ) return propagateFloat32NaN( a, 0 ); + if ( ! aSign ) return a; + float_raise( float_flag_invalid ); + return float32_default_nan; + } + if ( aSign ) { + if ( ( aExp | aSig ) == 0 ) return a; + float_raise( float_flag_invalid ); + return float32_default_nan; + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return 0; + normalizeFloat32Subnormal( aSig, &aExp, &aSig ); + } + zExp = ( ( aExp - 0x7F )>>1 ) + 0x7E; + aSig = ( aSig | 0x00800000 )<<8; + zSig = estimateSqrt32( aExp, aSig ) + 2; + if ( ( zSig & 0x7F ) <= 5 ) { + if ( zSig < 2 ) { + zSig = 0x7FFFFFFF; + goto roundAndPack; + } + aSig >>= aExp & 1; + term = ( (bits64) zSig ) * zSig; + rem = ( ( (bits64) aSig )<<32 ) - term; + while ( (sbits64) rem < 0 ) { + --zSig; + rem += ( ( (bits64) zSig )<<1 ) | 1; + } + zSig |= ( rem != 0 ); + } + shift32RightJamming( zSig, 1, &zSig ); + roundAndPack: + return roundAndPackFloat32( 0, zExp, zSig ); + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is equal to +the corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_eq( float32 a, float32 b ) +{ + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + return ( a == b ) || ( (bits32) ( ( a | b )<<1 ) == 0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is less than +or equal to the corresponding value `b', and 0 otherwise. The comparison +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_le( float32 a, float32 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign != bSign ) return aSign || ( (bits32) ( ( a | b )<<1 ) == 0 ); + return ( a == b ) || ( aSign ^ ( a < b ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_lt( float32 a, float32 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign != bSign ) return aSign && ( (bits32) ( ( a | b )<<1 ) != 0 ); + return ( a != b ) && ( aSign ^ ( a < b ) ); + +} + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is equal to +the corresponding value `b', and 0 otherwise. The invalid exception is +raised if either operand is a NaN. Otherwise, the comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_eq_signaling( float32 a, float32 b ) +{ + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + return ( a == b ) || ( (bits32) ( ( a | b )<<1 ) == 0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is less than or +equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not +cause an exception. Otherwise, the comparison is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_le_quiet( float32 a, float32 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign != bSign ) return aSign || ( (bits32) ( ( a | b )<<1 ) == 0 ); + return ( a == b ) || ( aSign ^ ( a < b ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the single-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an +exception. Otherwise, the comparison is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float32_lt_quiet( float32 a, float32 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat32Exp( a ) == 0xFF ) && extractFloat32Frac( a ) ) + || ( ( extractFloat32Exp( b ) == 0xFF ) && extractFloat32Frac( b ) ) + ) { + if ( float32_is_signaling_nan( a ) || float32_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat32Sign( a ); + bSign = extractFloat32Sign( b ); + if ( aSign != bSign ) return aSign && ( (bits32) ( ( a | b )<<1 ) != 0 ); + return ( a != b ) && ( aSign ^ ( a < b ) ); + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the 32-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int32 float64_to_int32( float64 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits64 aSig; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( ( aExp == 0x7FF ) && aSig ) aSign = 0; + if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); + shiftCount = 0x42C - aExp; + if ( 0 < shiftCount ) shift64RightJamming( aSig, shiftCount, &aSig ); + return roundAndPackInt32( aSign, aSig ); + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the 32-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. +If `a' is a NaN, the largest positive integer is returned. Otherwise, if +the conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int32 float64_to_int32_round_to_zero( float64 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits64 aSig, savedASig; + int32 z; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( 0x41E < aExp ) { + if ( ( aExp == 0x7FF ) && aSig ) aSign = 0; + goto invalid; + } + else if ( aExp < 0x3FF ) { + if ( aExp || aSig ) float_set_inexact(); + return 0; + } + aSig |= LIT64( 0x0010000000000000 ); + shiftCount = 0x433 - aExp; + savedASig = aSig; + aSig >>= shiftCount; + z = aSig; + if ( aSign ) z = - z; + if ( ( z < 0 ) ^ aSign ) { + invalid: + float_raise( float_flag_invalid ); + return aSign ? (sbits32) 0x80000000 : 0x7FFFFFFF; + } + if ( ( aSig<<shiftCount ) != savedASig ) { + float_set_inexact(); + } + return z; + +} + +#ifndef SOFTFLOAT_FOR_GCC /* Not needed */ +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the 64-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int64 float64_to_int64( float64 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits64 aSig, aSigExtra; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); + shiftCount = 0x433 - aExp; + if ( shiftCount <= 0 ) { + if ( 0x43E < aExp ) { + float_raise( float_flag_invalid ); + if ( ! aSign + || ( ( aExp == 0x7FF ) + && ( aSig != LIT64( 0x0010000000000000 ) ) ) + ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + aSigExtra = 0; + aSig <<= - shiftCount; + } + else { + shift64ExtraRightJamming( aSig, 0, shiftCount, &aSig, &aSigExtra ); + } + return roundAndPackInt64( aSign, aSig, aSigExtra ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the 64-bit two's complement integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. +If `a' is a NaN, the largest positive integer is returned. Otherwise, if +the conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int64 float64_to_int64_round_to_zero( float64 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits64 aSig; + int64 z; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp ) aSig |= LIT64( 0x0010000000000000 ); + shiftCount = aExp - 0x433; + if ( 0 <= shiftCount ) { + if ( 0x43E <= aExp ) { + if ( a != LIT64( 0xC3E0000000000000 ) ) { + float_raise( float_flag_invalid ); + if ( ! aSign + || ( ( aExp == 0x7FF ) + && ( aSig != LIT64( 0x0010000000000000 ) ) ) + ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + z = aSig<<shiftCount; + } + else { + if ( aExp < 0x3FE ) { + if ( aExp | aSig ) float_set_inexact(); + return 0; + } + z = aSig>>( - shiftCount ); + if ( (bits64) ( aSig<<( shiftCount & 63 ) ) ) { + float_set_inexact(); + } + } + if ( aSign ) z = - z; + return z; + +} +#endif /* !SOFTFLOAT_FOR_GCC */ + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the single-precision floating-point format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float64_to_float32( float64 a ) +{ + flag aSign; + int16 aExp; + bits64 aSig; + bits32 zSig; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp == 0x7FF ) { + if ( aSig ) return commonNaNToFloat32( float64ToCommonNaN( a ) ); + return packFloat32( aSign, 0xFF, 0 ); + } + shift64RightJamming( aSig, 22, &aSig ); + zSig = aSig; + if ( aExp || zSig ) { + zSig |= 0x40000000; + aExp -= 0x381; + } + return roundAndPackFloat32( aSign, aExp, zSig ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the extended double-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 float64_to_floatx80( float64 a ) +{ + flag aSign; + int16 aExp; + bits64 aSig; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp == 0x7FF ) { + if ( aSig ) return commonNaNToFloatx80( float64ToCommonNaN( a ) ); + return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloatx80( aSign, 0, 0 ); + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + } + return + packFloatx80( + aSign, aExp + 0x3C00, ( aSig | LIT64( 0x0010000000000000 ) )<<11 ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the quadruple-precision floating-point format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float64_to_float128( float64 a ) +{ + flag aSign; + int16 aExp; + bits64 aSig, zSig0, zSig1; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp == 0x7FF ) { + if ( aSig ) return commonNaNToFloat128( float64ToCommonNaN( a ) ); + return packFloat128( aSign, 0x7FFF, 0, 0 ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat128( aSign, 0, 0, 0 ); + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + --aExp; + } + shift128Right( aSig, 0, 4, &zSig0, &zSig1 ); + return packFloat128( aSign, aExp + 0x3C00, zSig0, zSig1 ); + +} + +#endif + +#ifndef SOFTFLOAT_FOR_GCC +/* +------------------------------------------------------------------------------- +Rounds the double-precision floating-point value `a' to an integer, and +returns the result as a double-precision floating-point value. The +operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_round_to_int( float64 a ) +{ + flag aSign; + int16 aExp; + bits64 lastBitMask, roundBitsMask; + int8 roundingMode; + float64 z; + + aExp = extractFloat64Exp( a ); + if ( 0x433 <= aExp ) { + if ( ( aExp == 0x7FF ) && extractFloat64Frac( a ) ) { + return propagateFloat64NaN( a, a ); + } + return a; + } + if ( aExp < 0x3FF ) { + if ( (bits64) ( a<<1 ) == 0 ) return a; + float_set_inexact(); + aSign = extractFloat64Sign( a ); + switch ( float_rounding_mode() ) { + case float_round_nearest_even: + if ( ( aExp == 0x3FE ) && extractFloat64Frac( a ) ) { + return packFloat64( aSign, 0x3FF, 0 ); + } + break; + case float_round_down: + return aSign ? LIT64( 0xBFF0000000000000 ) : 0; + case float_round_up: + return + aSign ? LIT64( 0x8000000000000000 ) : LIT64( 0x3FF0000000000000 ); + } + return packFloat64( aSign, 0, 0 ); + } + lastBitMask = 1; + lastBitMask <<= 0x433 - aExp; + roundBitsMask = lastBitMask - 1; + z = a; + roundingMode = float_rounding_mode(); + if ( roundingMode == float_round_nearest_even ) { + z += lastBitMask>>1; + if ( ( z & roundBitsMask ) == 0 ) z &= ~ lastBitMask; + } + else if ( roundingMode != float_round_to_zero ) { + if ( extractFloat64Sign( z ) ^ ( roundingMode == float_round_up ) ) { + z += roundBitsMask; + } + } + z &= ~ roundBitsMask; + if ( z != a ) float_set_inexact(); + return z; + +} +#endif + +/* +------------------------------------------------------------------------------- +Returns the result of adding the absolute values of the double-precision +floating-point values `a' and `b'. If `zSign' is 1, the sum is negated +before being returned. `zSign' is ignored if the result is a NaN. +The addition is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float64 addFloat64Sigs( float64 a, float64 b, flag zSign ) +{ + int16 aExp, bExp, zExp; + bits64 aSig, bSig, zSig; + int16 expDiff; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + bSig = extractFloat64Frac( b ); + bExp = extractFloat64Exp( b ); + expDiff = aExp - bExp; + aSig <<= 9; + bSig <<= 9; + if ( 0 < expDiff ) { + if ( aExp == 0x7FF ) { + if ( aSig ) return propagateFloat64NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig |= LIT64( 0x2000000000000000 ); + } + shift64RightJamming( bSig, expDiff, &bSig ); + zExp = aExp; + } + else if ( expDiff < 0 ) { + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + return packFloat64( zSign, 0x7FF, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig |= LIT64( 0x2000000000000000 ); + } + shift64RightJamming( aSig, - expDiff, &aSig ); + zExp = bExp; + } + else { + if ( aExp == 0x7FF ) { + if ( aSig | bSig ) return propagateFloat64NaN( a, b ); + return a; + } + if ( aExp == 0 ) return packFloat64( zSign, 0, ( aSig + bSig )>>9 ); + zSig = LIT64( 0x4000000000000000 ) + aSig + bSig; + zExp = aExp; + goto roundAndPack; + } + aSig |= LIT64( 0x2000000000000000 ); + zSig = ( aSig + bSig )<<1; + --zExp; + if ( (sbits64) zSig < 0 ) { + zSig = aSig + bSig; + ++zExp; + } + roundAndPack: + return roundAndPackFloat64( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the absolute values of the double- +precision floating-point values `a' and `b'. If `zSign' is 1, the +difference is negated before being returned. `zSign' is ignored if the +result is a NaN. The subtraction is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float64 subFloat64Sigs( float64 a, float64 b, flag zSign ) +{ + int16 aExp, bExp, zExp; + bits64 aSig, bSig, zSig; + int16 expDiff; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + bSig = extractFloat64Frac( b ); + bExp = extractFloat64Exp( b ); + expDiff = aExp - bExp; + aSig <<= 10; + bSig <<= 10; + if ( 0 < expDiff ) goto aExpBigger; + if ( expDiff < 0 ) goto bExpBigger; + if ( aExp == 0x7FF ) { + if ( aSig | bSig ) return propagateFloat64NaN( a, b ); + float_raise( float_flag_invalid ); + return float64_default_nan; + } + if ( aExp == 0 ) { + aExp = 1; + bExp = 1; + } + if ( bSig < aSig ) goto aBigger; + if ( aSig < bSig ) goto bBigger; + return packFloat64( float_rounding_mode() == float_round_down, 0, 0 ); + bExpBigger: + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + return packFloat64( zSign ^ 1, 0x7FF, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig |= LIT64( 0x4000000000000000 ); + } + shift64RightJamming( aSig, - expDiff, &aSig ); + bSig |= LIT64( 0x4000000000000000 ); + bBigger: + zSig = bSig - aSig; + zExp = bExp; + zSign ^= 1; + goto normalizeRoundAndPack; + aExpBigger: + if ( aExp == 0x7FF ) { + if ( aSig ) return propagateFloat64NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig |= LIT64( 0x4000000000000000 ); + } + shift64RightJamming( bSig, expDiff, &bSig ); + aSig |= LIT64( 0x4000000000000000 ); + aBigger: + zSig = aSig - bSig; + zExp = aExp; + normalizeRoundAndPack: + --zExp; + return normalizeRoundAndPackFloat64( zSign, zExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the double-precision floating-point values `a' +and `b'. The operation is performed according to the IEC/IEEE Standard for +Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_add( float64 a, float64 b ) +{ + flag aSign, bSign; + + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign == bSign ) { + return addFloat64Sigs( a, b, aSign ); + } + else { + return subFloat64Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the double-precision floating-point values +`a' and `b'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_sub( float64 a, float64 b ) +{ + flag aSign, bSign; + + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign == bSign ) { + return subFloat64Sigs( a, b, aSign ); + } + else { + return addFloat64Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of multiplying the double-precision floating-point values +`a' and `b'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_mul( float64 a, float64 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, zExp; + bits64 aSig, bSig, zSig0, zSig1; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + bSig = extractFloat64Frac( b ); + bExp = extractFloat64Exp( b ); + bSign = extractFloat64Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FF ) { + if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { + return propagateFloat64NaN( a, b ); + } + if ( ( bExp | bSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float64_default_nan; + } + return packFloat64( zSign, 0x7FF, 0 ); + } + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + if ( ( aExp | aSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float64_default_nan; + } + return packFloat64( zSign, 0x7FF, 0 ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat64( zSign, 0, 0 ); + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) return packFloat64( zSign, 0, 0 ); + normalizeFloat64Subnormal( bSig, &bExp, &bSig ); + } + zExp = aExp + bExp - 0x3FF; + aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<10; + bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; + mul64To128( aSig, bSig, &zSig0, &zSig1 ); + zSig0 |= ( zSig1 != 0 ); + if ( 0 <= (sbits64) ( zSig0<<1 ) ) { + zSig0 <<= 1; + --zExp; + } + return roundAndPackFloat64( zSign, zExp, zSig0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of dividing the double-precision floating-point value `a' +by the corresponding value `b'. The operation is performed according to +the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_div( float64 a, float64 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, zExp; + bits64 aSig, bSig, zSig; + bits64 rem0, rem1; + bits64 term0, term1; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + bSig = extractFloat64Frac( b ); + bExp = extractFloat64Exp( b ); + bSign = extractFloat64Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FF ) { + if ( aSig ) return propagateFloat64NaN( a, b ); + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + float_raise( float_flag_invalid ); + return float64_default_nan; + } + return packFloat64( zSign, 0x7FF, 0 ); + } + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + return packFloat64( zSign, 0, 0 ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + if ( ( aExp | aSig ) == 0 ) { + float_raise( float_flag_invalid ); + return float64_default_nan; + } + float_raise( float_flag_divbyzero ); + return packFloat64( zSign, 0x7FF, 0 ); + } + normalizeFloat64Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloat64( zSign, 0, 0 ); + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + } + zExp = aExp - bExp + 0x3FD; + aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<10; + bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; + if ( bSig <= ( aSig + aSig ) ) { + aSig >>= 1; + ++zExp; + } + zSig = estimateDiv128To64( aSig, 0, bSig ); + if ( ( zSig & 0x1FF ) <= 2 ) { + mul64To128( bSig, zSig, &term0, &term1 ); + sub128( aSig, 0, term0, term1, &rem0, &rem1 ); + while ( (sbits64) rem0 < 0 ) { + --zSig; + add128( rem0, rem1, 0, bSig, &rem0, &rem1 ); + } + zSig |= ( rem1 != 0 ); + } + return roundAndPackFloat64( zSign, zExp, zSig ); + +} + +#ifndef SOFTFLOAT_FOR_GCC +/* +------------------------------------------------------------------------------- +Returns the remainder of the double-precision floating-point value `a' +with respect to the corresponding value `b'. The operation is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_rem( float64 a, float64 b ) +{ + flag aSign, bSign, zSign; + int16 aExp, bExp, expDiff; + bits64 aSig, bSig; + bits64 q, alternateASig; + sbits64 sigMean; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + bSig = extractFloat64Frac( b ); + bExp = extractFloat64Exp( b ); + bSign = extractFloat64Sign( b ); + if ( aExp == 0x7FF ) { + if ( aSig || ( ( bExp == 0x7FF ) && bSig ) ) { + return propagateFloat64NaN( a, b ); + } + float_raise( float_flag_invalid ); + return float64_default_nan; + } + if ( bExp == 0x7FF ) { + if ( bSig ) return propagateFloat64NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + float_raise( float_flag_invalid ); + return float64_default_nan; + } + normalizeFloat64Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return a; + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + } + expDiff = aExp - bExp; + aSig = ( aSig | LIT64( 0x0010000000000000 ) )<<11; + bSig = ( bSig | LIT64( 0x0010000000000000 ) )<<11; + if ( expDiff < 0 ) { + if ( expDiff < -1 ) return a; + aSig >>= 1; + } + q = ( bSig <= aSig ); + if ( q ) aSig -= bSig; + expDiff -= 64; + while ( 0 < expDiff ) { + q = estimateDiv128To64( aSig, 0, bSig ); + q = ( 2 < q ) ? q - 2 : 0; + aSig = - ( ( bSig>>2 ) * q ); + expDiff -= 62; + } + expDiff += 64; + if ( 0 < expDiff ) { + q = estimateDiv128To64( aSig, 0, bSig ); + q = ( 2 < q ) ? q - 2 : 0; + q >>= 64 - expDiff; + bSig >>= 2; + aSig = ( ( aSig>>1 )<<( expDiff - 1 ) ) - bSig * q; + } + else { + aSig >>= 2; + bSig >>= 2; + } + do { + alternateASig = aSig; + ++q; + aSig -= bSig; + } while ( 0 <= (sbits64) aSig ); + sigMean = aSig + alternateASig; + if ( ( sigMean < 0 ) || ( ( sigMean == 0 ) && ( q & 1 ) ) ) { + aSig = alternateASig; + } + zSign = ( (sbits64) aSig < 0 ); + if ( zSign ) aSig = - aSig; + return normalizeRoundAndPackFloat64( aSign ^ zSign, bExp, aSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the square root of the double-precision floating-point value `a'. +The operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float64_sqrt( float64 a ) +{ + flag aSign; + int16 aExp, zExp; + bits64 aSig, zSig, doubleZSig; + bits64 rem0, rem1, term0, term1; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + if ( aExp == 0x7FF ) { + if ( aSig ) return propagateFloat64NaN( a, a ); + if ( ! aSign ) return a; + float_raise( float_flag_invalid ); + return float64_default_nan; + } + if ( aSign ) { + if ( ( aExp | aSig ) == 0 ) return a; + float_raise( float_flag_invalid ); + return float64_default_nan; + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return 0; + normalizeFloat64Subnormal( aSig, &aExp, &aSig ); + } + zExp = ( ( aExp - 0x3FF )>>1 ) + 0x3FE; + aSig |= LIT64( 0x0010000000000000 ); + zSig = estimateSqrt32( aExp, aSig>>21 ); + aSig <<= 9 - ( aExp & 1 ); + zSig = estimateDiv128To64( aSig, 0, zSig<<32 ) + ( zSig<<30 ); + if ( ( zSig & 0x1FF ) <= 5 ) { + doubleZSig = zSig<<1; + mul64To128( zSig, zSig, &term0, &term1 ); + sub128( aSig, 0, term0, term1, &rem0, &rem1 ); + while ( (sbits64) rem0 < 0 ) { + --zSig; + doubleZSig -= 2; + add128( rem0, rem1, zSig>>63, doubleZSig | 1, &rem0, &rem1 ); + } + zSig |= ( ( rem0 | rem1 ) != 0 ); + } + return roundAndPackFloat64( 0, zExp, zSig ); + +} +#endif + +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is equal to the +corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_eq( float64 a, float64 b ) +{ + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + return ( a == b ) || + ( (bits64) ( ( FLOAT64_DEMANGLE(a) | FLOAT64_DEMANGLE(b) )<<1 ) == 0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is less than or +equal to the corresponding value `b', and 0 otherwise. The comparison is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_le( float64 a, float64 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign != bSign ) + return aSign || + ( (bits64) ( ( FLOAT64_DEMANGLE(a) | FLOAT64_DEMANGLE(b) )<<1 ) == + 0 ); + return ( a == b ) || + ( aSign ^ ( FLOAT64_DEMANGLE(a) < FLOAT64_DEMANGLE(b) ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_lt( float64 a, float64 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign != bSign ) + return aSign && + ( (bits64) ( ( FLOAT64_DEMANGLE(a) | FLOAT64_DEMANGLE(b) )<<1 ) != + 0 ); + return ( a != b ) && + ( aSign ^ ( FLOAT64_DEMANGLE(a) < FLOAT64_DEMANGLE(b) ) ); + +} + +#ifndef SOFTFLOAT_FOR_GCC +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is equal to the +corresponding value `b', and 0 otherwise. The invalid exception is raised +if either operand is a NaN. Otherwise, the comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_eq_signaling( float64 a, float64 b ) +{ + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + return ( a == b ) || ( (bits64) ( ( a | b )<<1 ) == 0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is less than or +equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not +cause an exception. Otherwise, the comparison is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_le_quiet( float64 a, float64 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign != bSign ) return aSign || ( (bits64) ( ( a | b )<<1 ) == 0 ); + return ( a == b ) || ( aSign ^ ( a < b ) ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the double-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an +exception. Otherwise, the comparison is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float64_lt_quiet( float64 a, float64 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat64Exp( a ) == 0x7FF ) && extractFloat64Frac( a ) ) + || ( ( extractFloat64Exp( b ) == 0x7FF ) && extractFloat64Frac( b ) ) + ) { + if ( float64_is_signaling_nan( a ) || float64_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat64Sign( a ); + bSign = extractFloat64Sign( b ); + if ( aSign != bSign ) return aSign && ( (bits64) ( ( a | b )<<1 ) != 0 ); + return ( a != b ) && ( aSign ^ ( a < b ) ); + +} +#endif + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the 32-bit two's complement integer format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic---which means in particular that the conversion +is rounded according to the current rounding mode. If `a' is a NaN, the +largest positive integer is returned. Otherwise, if the conversion +overflows, the largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int32 floatx80_to_int32( floatx80 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( ( aExp == 0x7FFF ) && (bits64) ( aSig<<1 ) ) aSign = 0; + shiftCount = 0x4037 - aExp; + if ( shiftCount <= 0 ) shiftCount = 1; + shift64RightJamming( aSig, shiftCount, &aSig ); + return roundAndPackInt32( aSign, aSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the 32-bit two's complement integer format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic, except that the conversion is always rounded +toward zero. If `a' is a NaN, the largest positive integer is returned. +Otherwise, if the conversion overflows, the largest integer with the same +sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int32 floatx80_to_int32_round_to_zero( floatx80 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig, savedASig; + int32 z; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( 0x401E < aExp ) { + if ( ( aExp == 0x7FFF ) && (bits64) ( aSig<<1 ) ) aSign = 0; + goto invalid; + } + else if ( aExp < 0x3FFF ) { + if ( aExp || aSig ) float_set_inexact(); + return 0; + } + shiftCount = 0x403E - aExp; + savedASig = aSig; + aSig >>= shiftCount; + z = aSig; + if ( aSign ) z = - z; + if ( ( z < 0 ) ^ aSign ) { + invalid: + float_raise( float_flag_invalid ); + return aSign ? (sbits32) 0x80000000 : 0x7FFFFFFF; + } + if ( ( aSig<<shiftCount ) != savedASig ) { + float_set_inexact(); + } + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the 64-bit two's complement integer format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic---which means in particular that the conversion +is rounded according to the current rounding mode. If `a' is a NaN, +the largest positive integer is returned. Otherwise, if the conversion +overflows, the largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int64 floatx80_to_int64( floatx80 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig, aSigExtra; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + shiftCount = 0x403E - aExp; + if ( shiftCount <= 0 ) { + if ( shiftCount ) { + float_raise( float_flag_invalid ); + if ( ! aSign + || ( ( aExp == 0x7FFF ) + && ( aSig != LIT64( 0x8000000000000000 ) ) ) + ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + aSigExtra = 0; + } + else { + shift64ExtraRightJamming( aSig, 0, shiftCount, &aSig, &aSigExtra ); + } + return roundAndPackInt64( aSign, aSig, aSigExtra ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the 64-bit two's complement integer format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic, except that the conversion is always rounded +toward zero. If `a' is a NaN, the largest positive integer is returned. +Otherwise, if the conversion overflows, the largest integer with the same +sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int64 floatx80_to_int64_round_to_zero( floatx80 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig; + int64 z; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + shiftCount = aExp - 0x403E; + if ( 0 <= shiftCount ) { + aSig &= LIT64( 0x7FFFFFFFFFFFFFFF ); + if ( ( a.high != 0xC03E ) || aSig ) { + float_raise( float_flag_invalid ); + if ( ! aSign || ( ( aExp == 0x7FFF ) && aSig ) ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + else if ( aExp < 0x3FFF ) { + if ( aExp | aSig ) float_set_inexact(); + return 0; + } + z = aSig>>( - shiftCount ); + if ( (bits64) ( aSig<<( shiftCount & 63 ) ) ) { + float_set_inexact(); + } + if ( aSign ) z = - z; + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the single-precision floating-point format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 floatx80_to_float32( floatx80 a ) +{ + flag aSign; + int32 aExp; + bits64 aSig; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) ) { + return commonNaNToFloat32( floatx80ToCommonNaN( a ) ); + } + return packFloat32( aSign, 0xFF, 0 ); + } + shift64RightJamming( aSig, 33, &aSig ); + if ( aExp || aSig ) aExp -= 0x3F81; + return roundAndPackFloat32( aSign, aExp, aSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the double-precision floating-point format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 floatx80_to_float64( floatx80 a ) +{ + flag aSign; + int32 aExp; + bits64 aSig, zSig; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) ) { + return commonNaNToFloat64( floatx80ToCommonNaN( a ) ); + } + return packFloat64( aSign, 0x7FF, 0 ); + } + shift64RightJamming( aSig, 1, &zSig ); + if ( aExp || aSig ) aExp -= 0x3C01; + return roundAndPackFloat64( aSign, aExp, zSig ); + +} + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the extended double-precision floating- +point value `a' to the quadruple-precision floating-point format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 floatx80_to_float128( floatx80 a ) +{ + flag aSign; + int16 aExp; + bits64 aSig, zSig0, zSig1; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( ( aExp == 0x7FFF ) && (bits64) ( aSig<<1 ) ) { + return commonNaNToFloat128( floatx80ToCommonNaN( a ) ); + } + shift128Right( aSig<<1, 0, 16, &zSig0, &zSig1 ); + return packFloat128( aSign, aExp, zSig0, zSig1 ); + +} + +#endif + +/* +------------------------------------------------------------------------------- +Rounds the extended double-precision floating-point value `a' to an integer, +and returns the result as an extended quadruple-precision floating-point +value. The operation is performed according to the IEC/IEEE Standard for +Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_round_to_int( floatx80 a ) +{ + flag aSign; + int32 aExp; + bits64 lastBitMask, roundBitsMask; + int8 roundingMode; + floatx80 z; + + aExp = extractFloatx80Exp( a ); + if ( 0x403E <= aExp ) { + if ( ( aExp == 0x7FFF ) && (bits64) ( extractFloatx80Frac( a )<<1 ) ) { + return propagateFloatx80NaN( a, a ); + } + return a; + } + if ( aExp < 0x3FFF ) { + if ( ( aExp == 0 ) + && ( (bits64) ( extractFloatx80Frac( a )<<1 ) == 0 ) ) { + return a; + } + float_set_inexact(); + aSign = extractFloatx80Sign( a ); + switch ( float_rounding_mode() ) { + case float_round_nearest_even: + if ( ( aExp == 0x3FFE ) && (bits64) ( extractFloatx80Frac( a )<<1 ) + ) { + return + packFloatx80( aSign, 0x3FFF, LIT64( 0x8000000000000000 ) ); + } + break; + case float_round_down: + return + aSign ? + packFloatx80( 1, 0x3FFF, LIT64( 0x8000000000000000 ) ) + : packFloatx80( 0, 0, 0 ); + case float_round_up: + return + aSign ? packFloatx80( 1, 0, 0 ) + : packFloatx80( 0, 0x3FFF, LIT64( 0x8000000000000000 ) ); + } + return packFloatx80( aSign, 0, 0 ); + } + lastBitMask = 1; + lastBitMask <<= 0x403E - aExp; + roundBitsMask = lastBitMask - 1; + z = a; + roundingMode = float_rounding_mode(); + if ( roundingMode == float_round_nearest_even ) { + z.low += lastBitMask>>1; + if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; + } + else if ( roundingMode != float_round_to_zero ) { + if ( extractFloatx80Sign( z ) ^ ( roundingMode == float_round_up ) ) { + z.low += roundBitsMask; + } + } + z.low &= ~ roundBitsMask; + if ( z.low == 0 ) { + ++z.high; + z.low = LIT64( 0x8000000000000000 ); + } + if ( z.low != a.low ) float_set_inexact(); + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the absolute values of the extended double- +precision floating-point values `a' and `b'. If `zSign' is 1, the sum is +negated before being returned. `zSign' is ignored if the result is a NaN. +The addition is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static floatx80 addFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) +{ + int32 aExp, bExp, zExp; + bits64 aSig, bSig, zSig0, zSig1; + int32 expDiff; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + bSig = extractFloatx80Frac( b ); + bExp = extractFloatx80Exp( b ); + expDiff = aExp - bExp; + if ( 0 < expDiff ) { + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return a; + } + if ( bExp == 0 ) --expDiff; + shift64ExtraRightJamming( bSig, 0, expDiff, &bSig, &zSig1 ); + zExp = aExp; + } + else if ( expDiff < 0 ) { + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) ++expDiff; + shift64ExtraRightJamming( aSig, 0, - expDiff, &aSig, &zSig1 ); + zExp = bExp; + } + else { + if ( aExp == 0x7FFF ) { + if ( (bits64) ( ( aSig | bSig )<<1 ) ) { + return propagateFloatx80NaN( a, b ); + } + return a; + } + zSig1 = 0; + zSig0 = aSig + bSig; + if ( aExp == 0 ) { + normalizeFloatx80Subnormal( zSig0, &zExp, &zSig0 ); + goto roundAndPack; + } + zExp = aExp; + goto shiftRight1; + } + zSig0 = aSig + bSig; + if ( (sbits64) zSig0 < 0 ) goto roundAndPack; + shiftRight1: + shift64ExtraRightJamming( zSig0, zSig1, 1, &zSig0, &zSig1 ); + zSig0 |= LIT64( 0x8000000000000000 ); + ++zExp; + roundAndPack: + return + roundAndPackFloatx80( + floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the absolute values of the extended +double-precision floating-point values `a' and `b'. If `zSign' is 1, the +difference is negated before being returned. `zSign' is ignored if the +result is a NaN. The subtraction is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static floatx80 subFloatx80Sigs( floatx80 a, floatx80 b, flag zSign ) +{ + int32 aExp, bExp, zExp; + bits64 aSig, bSig, zSig0, zSig1; + int32 expDiff; + floatx80 z; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + bSig = extractFloatx80Frac( b ); + bExp = extractFloatx80Exp( b ); + expDiff = aExp - bExp; + if ( 0 < expDiff ) goto aExpBigger; + if ( expDiff < 0 ) goto bExpBigger; + if ( aExp == 0x7FFF ) { + if ( (bits64) ( ( aSig | bSig )<<1 ) ) { + return propagateFloatx80NaN( a, b ); + } + float_raise( float_flag_invalid ); + z.low = floatx80_default_nan_low; + z.high = floatx80_default_nan_high; + return z; + } + if ( aExp == 0 ) { + aExp = 1; + bExp = 1; + } + zSig1 = 0; + if ( bSig < aSig ) goto aBigger; + if ( aSig < bSig ) goto bBigger; + return packFloatx80( float_rounding_mode() == float_round_down, 0, 0 ); + bExpBigger: + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return packFloatx80( zSign ^ 1, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) ++expDiff; + shift128RightJamming( aSig, 0, - expDiff, &aSig, &zSig1 ); + bBigger: + sub128( bSig, 0, aSig, zSig1, &zSig0, &zSig1 ); + zExp = bExp; + zSign ^= 1; + goto normalizeRoundAndPack; + aExpBigger: + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return a; + } + if ( bExp == 0 ) --expDiff; + shift128RightJamming( bSig, 0, expDiff, &bSig, &zSig1 ); + aBigger: + sub128( aSig, 0, bSig, zSig1, &zSig0, &zSig1 ); + zExp = aExp; + normalizeRoundAndPack: + return + normalizeRoundAndPackFloatx80( + floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the extended double-precision floating-point +values `a' and `b'. The operation is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_add( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign == bSign ) { + return addFloatx80Sigs( a, b, aSign ); + } + else { + return subFloatx80Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the extended double-precision floating- +point values `a' and `b'. The operation is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_sub( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign == bSign ) { + return subFloatx80Sigs( a, b, aSign ); + } + else { + return addFloatx80Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of multiplying the extended double-precision floating- +point values `a' and `b'. The operation is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_mul( floatx80 a, floatx80 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, zExp; + bits64 aSig, bSig, zSig0, zSig1; + floatx80 z; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + bSig = extractFloatx80Frac( b ); + bExp = extractFloatx80Exp( b ); + bSign = extractFloatx80Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) + || ( ( bExp == 0x7FFF ) && (bits64) ( bSig<<1 ) ) ) { + return propagateFloatx80NaN( a, b ); + } + if ( ( bExp | bSig ) == 0 ) goto invalid; + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + if ( ( aExp | aSig ) == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = floatx80_default_nan_low; + z.high = floatx80_default_nan_high; + return z; + } + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloatx80( zSign, 0, 0 ); + normalizeFloatx80Subnormal( aSig, &aExp, &aSig ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) return packFloatx80( zSign, 0, 0 ); + normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); + } + zExp = aExp + bExp - 0x3FFE; + mul64To128( aSig, bSig, &zSig0, &zSig1 ); + if ( 0 < (sbits64) zSig0 ) { + shortShift128Left( zSig0, zSig1, 1, &zSig0, &zSig1 ); + --zExp; + } + return + roundAndPackFloatx80( + floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of dividing the extended double-precision floating-point +value `a' by the corresponding value `b'. The operation is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_div( floatx80 a, floatx80 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, zExp; + bits64 aSig, bSig, zSig0, zSig1; + bits64 rem0, rem1, rem2, term0, term1, term2; + floatx80 z; + + aSig = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + bSig = extractFloatx80Frac( b ); + bExp = extractFloatx80Exp( b ); + bSign = extractFloatx80Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig<<1 ) ) return propagateFloatx80NaN( a, b ); + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + goto invalid; + } + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return packFloatx80( zSign, 0, 0 ); + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + if ( ( aExp | aSig ) == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = floatx80_default_nan_low; + z.high = floatx80_default_nan_high; + return z; + } + float_raise( float_flag_divbyzero ); + return packFloatx80( zSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( aSig == 0 ) return packFloatx80( zSign, 0, 0 ); + normalizeFloatx80Subnormal( aSig, &aExp, &aSig ); + } + zExp = aExp - bExp + 0x3FFE; + rem1 = 0; + if ( bSig <= aSig ) { + shift128Right( aSig, 0, 1, &aSig, &rem1 ); + ++zExp; + } + zSig0 = estimateDiv128To64( aSig, rem1, bSig ); + mul64To128( bSig, zSig0, &term0, &term1 ); + sub128( aSig, rem1, term0, term1, &rem0, &rem1 ); + while ( (sbits64) rem0 < 0 ) { + --zSig0; + add128( rem0, rem1, 0, bSig, &rem0, &rem1 ); + } + zSig1 = estimateDiv128To64( rem1, 0, bSig ); + if ( (bits64) ( zSig1<<1 ) <= 8 ) { + mul64To128( bSig, zSig1, &term1, &term2 ); + sub128( rem1, 0, term1, term2, &rem1, &rem2 ); + while ( (sbits64) rem1 < 0 ) { + --zSig1; + add128( rem1, rem2, 0, bSig, &rem1, &rem2 ); + } + zSig1 |= ( ( rem1 | rem2 ) != 0 ); + } + return + roundAndPackFloatx80( + floatx80_rounding_precision, zSign, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the remainder of the extended double-precision floating-point value +`a' with respect to the corresponding value `b'. The operation is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_rem( floatx80 a, floatx80 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, expDiff; + bits64 aSig0, aSig1, bSig; + bits64 q, term0, term1, alternateASig0, alternateASig1; + floatx80 z; + + aSig0 = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + bSig = extractFloatx80Frac( b ); + bExp = extractFloatx80Exp( b ); + bSign = extractFloatx80Sign( b ); + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig0<<1 ) + || ( ( bExp == 0x7FFF ) && (bits64) ( bSig<<1 ) ) ) { + return propagateFloatx80NaN( a, b ); + } + goto invalid; + } + if ( bExp == 0x7FFF ) { + if ( (bits64) ( bSig<<1 ) ) return propagateFloatx80NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + if ( bSig == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = floatx80_default_nan_low; + z.high = floatx80_default_nan_high; + return z; + } + normalizeFloatx80Subnormal( bSig, &bExp, &bSig ); + } + if ( aExp == 0 ) { + if ( (bits64) ( aSig0<<1 ) == 0 ) return a; + normalizeFloatx80Subnormal( aSig0, &aExp, &aSig0 ); + } + bSig |= LIT64( 0x8000000000000000 ); + zSign = aSign; + expDiff = aExp - bExp; + aSig1 = 0; + if ( expDiff < 0 ) { + if ( expDiff < -1 ) return a; + shift128Right( aSig0, 0, 1, &aSig0, &aSig1 ); + expDiff = 0; + } + q = ( bSig <= aSig0 ); + if ( q ) aSig0 -= bSig; + expDiff -= 64; + while ( 0 < expDiff ) { + q = estimateDiv128To64( aSig0, aSig1, bSig ); + q = ( 2 < q ) ? q - 2 : 0; + mul64To128( bSig, q, &term0, &term1 ); + sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); + shortShift128Left( aSig0, aSig1, 62, &aSig0, &aSig1 ); + expDiff -= 62; + } + expDiff += 64; + if ( 0 < expDiff ) { + q = estimateDiv128To64( aSig0, aSig1, bSig ); + q = ( 2 < q ) ? q - 2 : 0; + q >>= 64 - expDiff; + mul64To128( bSig, q<<( 64 - expDiff ), &term0, &term1 ); + sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); + shortShift128Left( 0, bSig, 64 - expDiff, &term0, &term1 ); + while ( le128( term0, term1, aSig0, aSig1 ) ) { + ++q; + sub128( aSig0, aSig1, term0, term1, &aSig0, &aSig1 ); + } + } + else { + term1 = 0; + term0 = bSig; + } + sub128( term0, term1, aSig0, aSig1, &alternateASig0, &alternateASig1 ); + if ( lt128( alternateASig0, alternateASig1, aSig0, aSig1 ) + || ( eq128( alternateASig0, alternateASig1, aSig0, aSig1 ) + && ( q & 1 ) ) + ) { + aSig0 = alternateASig0; + aSig1 = alternateASig1; + zSign = ! zSign; + } + return + normalizeRoundAndPackFloatx80( + 80, zSign, bExp + expDiff, aSig0, aSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the square root of the extended double-precision floating-point +value `a'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_sqrt( floatx80 a ) +{ + flag aSign; + int32 aExp, zExp; + bits64 aSig0, aSig1, zSig0, zSig1, doubleZSig0; + bits64 rem0, rem1, rem2, rem3, term0, term1, term2, term3; + floatx80 z; + + aSig0 = extractFloatx80Frac( a ); + aExp = extractFloatx80Exp( a ); + aSign = extractFloatx80Sign( a ); + if ( aExp == 0x7FFF ) { + if ( (bits64) ( aSig0<<1 ) ) return propagateFloatx80NaN( a, a ); + if ( ! aSign ) return a; + goto invalid; + } + if ( aSign ) { + if ( ( aExp | aSig0 ) == 0 ) return a; + invalid: + float_raise( float_flag_invalid ); + z.low = floatx80_default_nan_low; + z.high = floatx80_default_nan_high; + return z; + } + if ( aExp == 0 ) { + if ( aSig0 == 0 ) return packFloatx80( 0, 0, 0 ); + normalizeFloatx80Subnormal( aSig0, &aExp, &aSig0 ); + } + zExp = ( ( aExp - 0x3FFF )>>1 ) + 0x3FFF; + zSig0 = estimateSqrt32( aExp, aSig0>>32 ); + shift128Right( aSig0, 0, 2 + ( aExp & 1 ), &aSig0, &aSig1 ); + zSig0 = estimateDiv128To64( aSig0, aSig1, zSig0<<32 ) + ( zSig0<<30 ); + doubleZSig0 = zSig0<<1; + mul64To128( zSig0, zSig0, &term0, &term1 ); + sub128( aSig0, aSig1, term0, term1, &rem0, &rem1 ); + while ( (sbits64) rem0 < 0 ) { + --zSig0; + doubleZSig0 -= 2; + add128( rem0, rem1, zSig0>>63, doubleZSig0 | 1, &rem0, &rem1 ); + } + zSig1 = estimateDiv128To64( rem1, 0, doubleZSig0 ); + if ( ( zSig1 & LIT64( 0x3FFFFFFFFFFFFFFF ) ) <= 5 ) { + if ( zSig1 == 0 ) zSig1 = 1; + mul64To128( doubleZSig0, zSig1, &term1, &term2 ); + sub128( rem1, 0, term1, term2, &rem1, &rem2 ); + mul64To128( zSig1, zSig1, &term2, &term3 ); + sub192( rem1, rem2, 0, 0, term2, term3, &rem1, &rem2, &rem3 ); + while ( (sbits64) rem1 < 0 ) { + --zSig1; + shortShift128Left( 0, zSig1, 1, &term2, &term3 ); + term3 |= 1; + term2 |= doubleZSig0; + add192( rem1, rem2, rem3, 0, term2, term3, &rem1, &rem2, &rem3 ); + } + zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 ); + } + shortShift128Left( 0, zSig1, 1, &zSig0, &zSig1 ); + zSig0 |= doubleZSig0; + return + roundAndPackFloatx80( + floatx80_rounding_precision, 0, zExp, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is +equal to the corresponding value `b', and 0 otherwise. The comparison is +performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_eq( floatx80 a, floatx80 b ) +{ + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + if ( floatx80_is_signaling_nan( a ) + || floatx80_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + return + ( a.low == b.low ) + && ( ( a.high == b.high ) + || ( ( a.low == 0 ) + && ( (bits16) ( ( a.high | b.high )<<1 ) == 0 ) ) + ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is +less than or equal to the corresponding value `b', and 0 otherwise. The +comparison is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_le( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign != bSign ) { + return + aSign + || ( ( ( (bits16) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + == 0 ); + } + return + aSign ? le128( b.high, b.low, a.high, a.low ) + : le128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is +less than the corresponding value `b', and 0 otherwise. The comparison +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_lt( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign != bSign ) { + return + aSign + && ( ( ( (bits16) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + != 0 ); + } + return + aSign ? lt128( b.high, b.low, a.high, a.low ) + : lt128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is equal +to the corresponding value `b', and 0 otherwise. The invalid exception is +raised if either operand is a NaN. Otherwise, the comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_eq_signaling( floatx80 a, floatx80 b ) +{ + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + return + ( a.low == b.low ) + && ( ( a.high == b.high ) + || ( ( a.low == 0 ) + && ( (bits16) ( ( a.high | b.high )<<1 ) == 0 ) ) + ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is less +than or equal to the corresponding value `b', and 0 otherwise. Quiet NaNs +do not cause an exception. Otherwise, the comparison is performed according +to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_le_quiet( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + if ( floatx80_is_signaling_nan( a ) + || floatx80_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign != bSign ) { + return + aSign + || ( ( ( (bits16) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + == 0 ); + } + return + aSign ? le128( b.high, b.low, a.high, a.low ) + : le128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the extended double-precision floating-point value `a' is less +than the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause +an exception. Otherwise, the comparison is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag floatx80_lt_quiet( floatx80 a, floatx80 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloatx80Exp( a ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( a )<<1 ) ) + || ( ( extractFloatx80Exp( b ) == 0x7FFF ) + && (bits64) ( extractFloatx80Frac( b )<<1 ) ) + ) { + if ( floatx80_is_signaling_nan( a ) + || floatx80_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloatx80Sign( a ); + bSign = extractFloatx80Sign( b ); + if ( aSign != bSign ) { + return + aSign + && ( ( ( (bits16) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + != 0 ); + } + return + aSign ? lt128( b.high, b.low, a.high, a.low ) + : lt128( a.high, a.low, b.high, b.low ); + +} + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the 32-bit two's complement integer format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int32 float128_to_int32( float128 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig0, aSig1; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( ( aExp == 0x7FFF ) && ( aSig0 | aSig1 ) ) aSign = 0; + if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); + aSig0 |= ( aSig1 != 0 ); + shiftCount = 0x4028 - aExp; + if ( 0 < shiftCount ) shift64RightJamming( aSig0, shiftCount, &aSig0 ); + return roundAndPackInt32( aSign, aSig0 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the 32-bit two's complement integer format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. If +`a' is a NaN, the largest positive integer is returned. Otherwise, if the +conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int32 float128_to_int32_round_to_zero( float128 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig0, aSig1, savedASig; + int32 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + aSig0 |= ( aSig1 != 0 ); + if ( 0x401E < aExp ) { + if ( ( aExp == 0x7FFF ) && aSig0 ) aSign = 0; + goto invalid; + } + else if ( aExp < 0x3FFF ) { + if ( aExp || aSig0 ) float_set_inexact(); + return 0; + } + aSig0 |= LIT64( 0x0001000000000000 ); + shiftCount = 0x402F - aExp; + savedASig = aSig0; + aSig0 >>= shiftCount; + z = aSig0; + if ( aSign ) z = - z; + if ( ( z < 0 ) ^ aSign ) { + invalid: + float_raise( float_flag_invalid ); + return aSign ? (sbits32) 0x80000000 : 0x7FFFFFFF; + } + if ( ( aSig0<<shiftCount ) != savedASig ) { + float_set_inexact(); + } + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the 64-bit two's complement integer format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic---which means in particular that the conversion is rounded +according to the current rounding mode. If `a' is a NaN, the largest +positive integer is returned. Otherwise, if the conversion overflows, the +largest integer with the same sign as `a' is returned. +------------------------------------------------------------------------------- +*/ +int64 float128_to_int64( float128 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig0, aSig1; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); + shiftCount = 0x402F - aExp; + if ( shiftCount <= 0 ) { + if ( 0x403E < aExp ) { + float_raise( float_flag_invalid ); + if ( ! aSign + || ( ( aExp == 0x7FFF ) + && ( aSig1 || ( aSig0 != LIT64( 0x0001000000000000 ) ) ) + ) + ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + shortShift128Left( aSig0, aSig1, - shiftCount, &aSig0, &aSig1 ); + } + else { + shift64ExtraRightJamming( aSig0, aSig1, shiftCount, &aSig0, &aSig1 ); + } + return roundAndPackInt64( aSign, aSig0, aSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the 64-bit two's complement integer format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic, except that the conversion is always rounded toward zero. +If `a' is a NaN, the largest positive integer is returned. Otherwise, if +the conversion overflows, the largest integer with the same sign as `a' is +returned. +------------------------------------------------------------------------------- +*/ +int64 float128_to_int64_round_to_zero( float128 a ) +{ + flag aSign; + int32 aExp, shiftCount; + bits64 aSig0, aSig1; + int64 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp ) aSig0 |= LIT64( 0x0001000000000000 ); + shiftCount = aExp - 0x402F; + if ( 0 < shiftCount ) { + if ( 0x403E <= aExp ) { + aSig0 &= LIT64( 0x0000FFFFFFFFFFFF ); + if ( ( a.high == LIT64( 0xC03E000000000000 ) ) + && ( aSig1 < LIT64( 0x0002000000000000 ) ) ) { + if ( aSig1 ) float_set_inexact(); + } + else { + float_raise( float_flag_invalid ); + if ( ! aSign || ( ( aExp == 0x7FFF ) && ( aSig0 | aSig1 ) ) ) { + return LIT64( 0x7FFFFFFFFFFFFFFF ); + } + } + return (sbits64) LIT64( 0x8000000000000000 ); + } + z = ( aSig0<<shiftCount ) | ( aSig1>>( ( - shiftCount ) & 63 ) ); + if ( (bits64) ( aSig1<<shiftCount ) ) { + float_set_inexact(); + } + } + else { + if ( aExp < 0x3FFF ) { + if ( aExp | aSig0 | aSig1 ) { + float_set_inexact(); + } + return 0; + } + z = aSig0>>( - shiftCount ); + if ( aSig1 + || ( shiftCount && (bits64) ( aSig0<<( shiftCount & 63 ) ) ) ) { + float_set_inexact(); + } + } + if ( aSign ) z = - z; + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the single-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float32 float128_to_float32( float128 a ) +{ + flag aSign; + int32 aExp; + bits64 aSig0, aSig1; + bits32 zSig; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) { + return commonNaNToFloat32( float128ToCommonNaN( a ) ); + } + return packFloat32( aSign, 0xFF, 0 ); + } + aSig0 |= ( aSig1 != 0 ); + shift64RightJamming( aSig0, 18, &aSig0 ); + zSig = aSig0; + if ( aExp || zSig ) { + zSig |= 0x40000000; + aExp -= 0x3F81; + } + return roundAndPackFloat32( aSign, aExp, zSig ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the double-precision floating-point format. The conversion +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +float64 float128_to_float64( float128 a ) +{ + flag aSign; + int32 aExp; + bits64 aSig0, aSig1; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) { + return commonNaNToFloat64( float128ToCommonNaN( a ) ); + } + return packFloat64( aSign, 0x7FF, 0 ); + } + shortShift128Left( aSig0, aSig1, 14, &aSig0, &aSig1 ); + aSig0 |= ( aSig1 != 0 ); + if ( aExp || aSig0 ) { + aSig0 |= LIT64( 0x4000000000000000 ); + aExp -= 0x3C01; + } + return roundAndPackFloat64( aSign, aExp, aSig0 ); + +} + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Returns the result of converting the quadruple-precision floating-point +value `a' to the extended double-precision floating-point format. The +conversion is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +floatx80 float128_to_floatx80( float128 a ) +{ + flag aSign; + int32 aExp; + bits64 aSig0, aSig1; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) { + return commonNaNToFloatx80( float128ToCommonNaN( a ) ); + } + return packFloatx80( aSign, 0x7FFF, LIT64( 0x8000000000000000 ) ); + } + if ( aExp == 0 ) { + if ( ( aSig0 | aSig1 ) == 0 ) return packFloatx80( aSign, 0, 0 ); + normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); + } + else { + aSig0 |= LIT64( 0x0001000000000000 ); + } + shortShift128Left( aSig0, aSig1, 15, &aSig0, &aSig1 ); + return roundAndPackFloatx80( 80, aSign, aExp, aSig0, aSig1 ); + +} + +#endif + +/* +------------------------------------------------------------------------------- +Rounds the quadruple-precision floating-point value `a' to an integer, and +returns the result as a quadruple-precision floating-point value. The +operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_round_to_int( float128 a ) +{ + flag aSign; + int32 aExp; + bits64 lastBitMask, roundBitsMask; + int8 roundingMode; + float128 z; + + aExp = extractFloat128Exp( a ); + if ( 0x402F <= aExp ) { + if ( 0x406F <= aExp ) { + if ( ( aExp == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) + ) { + return propagateFloat128NaN( a, a ); + } + return a; + } + lastBitMask = 1; + lastBitMask = ( lastBitMask<<( 0x406E - aExp ) )<<1; + roundBitsMask = lastBitMask - 1; + z = a; + roundingMode = float_rounding_mode(); + if ( roundingMode == float_round_nearest_even ) { + if ( lastBitMask ) { + add128( z.high, z.low, 0, lastBitMask>>1, &z.high, &z.low ); + if ( ( z.low & roundBitsMask ) == 0 ) z.low &= ~ lastBitMask; + } + else { + if ( (sbits64) z.low < 0 ) { + ++z.high; + if ( (bits64) ( z.low<<1 ) == 0 ) z.high &= ~1; + } + } + } + else if ( roundingMode != float_round_to_zero ) { + if ( extractFloat128Sign( z ) + ^ ( roundingMode == float_round_up ) ) { + add128( z.high, z.low, 0, roundBitsMask, &z.high, &z.low ); + } + } + z.low &= ~ roundBitsMask; + } + else { + if ( aExp < 0x3FFF ) { + if ( ( ( (bits64) ( a.high<<1 ) ) | a.low ) == 0 ) return a; + float_set_inexact(); + aSign = extractFloat128Sign( a ); + switch ( float_rounding_mode() ) { + case float_round_nearest_even: + if ( ( aExp == 0x3FFE ) + && ( extractFloat128Frac0( a ) + | extractFloat128Frac1( a ) ) + ) { + return packFloat128( aSign, 0x3FFF, 0, 0 ); + } + break; + case float_round_down: + return + aSign ? packFloat128( 1, 0x3FFF, 0, 0 ) + : packFloat128( 0, 0, 0, 0 ); + case float_round_up: + return + aSign ? packFloat128( 1, 0, 0, 0 ) + : packFloat128( 0, 0x3FFF, 0, 0 ); + } + return packFloat128( aSign, 0, 0, 0 ); + } + lastBitMask = 1; + lastBitMask <<= 0x402F - aExp; + roundBitsMask = lastBitMask - 1; + z.low = 0; + z.high = a.high; + roundingMode = float_rounding_mode(); + if ( roundingMode == float_round_nearest_even ) { + z.high += lastBitMask>>1; + if ( ( ( z.high & roundBitsMask ) | a.low ) == 0 ) { + z.high &= ~ lastBitMask; + } + } + else if ( roundingMode != float_round_to_zero ) { + if ( extractFloat128Sign( z ) + ^ ( roundingMode == float_round_up ) ) { + z.high |= ( a.low != 0 ); + z.high += roundBitsMask; + } + } + z.high &= ~ roundBitsMask; + } + if ( ( z.low != a.low ) || ( z.high != a.high ) ) { + float_set_inexact(); + } + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the absolute values of the quadruple-precision +floating-point values `a' and `b'. If `zSign' is 1, the sum is negated +before being returned. `zSign' is ignored if the result is a NaN. +The addition is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float128 addFloat128Sigs( float128 a, float128 b, flag zSign ) +{ + int32 aExp, bExp, zExp; + bits64 aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2; + int32 expDiff; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + bSig1 = extractFloat128Frac1( b ); + bSig0 = extractFloat128Frac0( b ); + bExp = extractFloat128Exp( b ); + expDiff = aExp - bExp; + if ( 0 < expDiff ) { + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig0 |= LIT64( 0x0001000000000000 ); + } + shift128ExtraRightJamming( + bSig0, bSig1, 0, expDiff, &bSig0, &bSig1, &zSig2 ); + zExp = aExp; + } + else if ( expDiff < 0 ) { + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig0 |= LIT64( 0x0001000000000000 ); + } + shift128ExtraRightJamming( + aSig0, aSig1, 0, - expDiff, &aSig0, &aSig1, &zSig2 ); + zExp = bExp; + } + else { + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 | bSig0 | bSig1 ) { + return propagateFloat128NaN( a, b ); + } + return a; + } + add128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); + if ( aExp == 0 ) return packFloat128( zSign, 0, zSig0, zSig1 ); + zSig2 = 0; + zSig0 |= LIT64( 0x0002000000000000 ); + zExp = aExp; + goto shiftRight1; + } + aSig0 |= LIT64( 0x0001000000000000 ); + add128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); + --zExp; + if ( zSig0 < LIT64( 0x0002000000000000 ) ) goto roundAndPack; + ++zExp; + shiftRight1: + shift128ExtraRightJamming( + zSig0, zSig1, zSig2, 1, &zSig0, &zSig1, &zSig2 ); + roundAndPack: + return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the absolute values of the quadruple- +precision floating-point values `a' and `b'. If `zSign' is 1, the +difference is negated before being returned. `zSign' is ignored if the +result is a NaN. The subtraction is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +static float128 subFloat128Sigs( float128 a, float128 b, flag zSign ) +{ + int32 aExp, bExp, zExp; + bits64 aSig0, aSig1, bSig0, bSig1, zSig0, zSig1; + int32 expDiff; + float128 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + bSig1 = extractFloat128Frac1( b ); + bSig0 = extractFloat128Frac0( b ); + bExp = extractFloat128Exp( b ); + expDiff = aExp - bExp; + shortShift128Left( aSig0, aSig1, 14, &aSig0, &aSig1 ); + shortShift128Left( bSig0, bSig1, 14, &bSig0, &bSig1 ); + if ( 0 < expDiff ) goto aExpBigger; + if ( expDiff < 0 ) goto bExpBigger; + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 | bSig0 | bSig1 ) { + return propagateFloat128NaN( a, b ); + } + float_raise( float_flag_invalid ); + z.low = float128_default_nan_low; + z.high = float128_default_nan_high; + return z; + } + if ( aExp == 0 ) { + aExp = 1; + bExp = 1; + } + if ( bSig0 < aSig0 ) goto aBigger; + if ( aSig0 < bSig0 ) goto bBigger; + if ( bSig1 < aSig1 ) goto aBigger; + if ( aSig1 < bSig1 ) goto bBigger; + return packFloat128( float_rounding_mode() == float_round_down, 0, 0, 0 ); + bExpBigger: + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + return packFloat128( zSign ^ 1, 0x7FFF, 0, 0 ); + } + if ( aExp == 0 ) { + ++expDiff; + } + else { + aSig0 |= LIT64( 0x4000000000000000 ); + } + shift128RightJamming( aSig0, aSig1, - expDiff, &aSig0, &aSig1 ); + bSig0 |= LIT64( 0x4000000000000000 ); + bBigger: + sub128( bSig0, bSig1, aSig0, aSig1, &zSig0, &zSig1 ); + zExp = bExp; + zSign ^= 1; + goto normalizeRoundAndPack; + aExpBigger: + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + --expDiff; + } + else { + bSig0 |= LIT64( 0x4000000000000000 ); + } + shift128RightJamming( bSig0, bSig1, expDiff, &bSig0, &bSig1 ); + aSig0 |= LIT64( 0x4000000000000000 ); + aBigger: + sub128( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1 ); + zExp = aExp; + normalizeRoundAndPack: + --zExp; + return normalizeRoundAndPackFloat128( zSign, zExp - 14, zSig0, zSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of adding the quadruple-precision floating-point values +`a' and `b'. The operation is performed according to the IEC/IEEE Standard +for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_add( float128 a, float128 b ) +{ + flag aSign, bSign; + + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign == bSign ) { + return addFloat128Sigs( a, b, aSign ); + } + else { + return subFloat128Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of subtracting the quadruple-precision floating-point +values `a' and `b'. The operation is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_sub( float128 a, float128 b ) +{ + flag aSign, bSign; + + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign == bSign ) { + return subFloat128Sigs( a, b, aSign ); + } + else { + return addFloat128Sigs( a, b, aSign ); + } + +} + +/* +------------------------------------------------------------------------------- +Returns the result of multiplying the quadruple-precision floating-point +values `a' and `b'. The operation is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_mul( float128 a, float128 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, zExp; + bits64 aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2, zSig3; + float128 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + bSig1 = extractFloat128Frac1( b ); + bSig0 = extractFloat128Frac0( b ); + bExp = extractFloat128Exp( b ); + bSign = extractFloat128Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FFF ) { + if ( ( aSig0 | aSig1 ) + || ( ( bExp == 0x7FFF ) && ( bSig0 | bSig1 ) ) ) { + return propagateFloat128NaN( a, b ); + } + if ( ( bExp | bSig0 | bSig1 ) == 0 ) goto invalid; + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + if ( ( aExp | aSig0 | aSig1 ) == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = float128_default_nan_low; + z.high = float128_default_nan_high; + return z; + } + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + if ( aExp == 0 ) { + if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); + normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); + } + if ( bExp == 0 ) { + if ( ( bSig0 | bSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); + normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); + } + zExp = aExp + bExp - 0x4000; + aSig0 |= LIT64( 0x0001000000000000 ); + shortShift128Left( bSig0, bSig1, 16, &bSig0, &bSig1 ); + mul128To256( aSig0, aSig1, bSig0, bSig1, &zSig0, &zSig1, &zSig2, &zSig3 ); + add128( zSig0, zSig1, aSig0, aSig1, &zSig0, &zSig1 ); + zSig2 |= ( zSig3 != 0 ); + if ( LIT64( 0x0002000000000000 ) <= zSig0 ) { + shift128ExtraRightJamming( + zSig0, zSig1, zSig2, 1, &zSig0, &zSig1, &zSig2 ); + ++zExp; + } + return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the result of dividing the quadruple-precision floating-point value +`a' by the corresponding value `b'. The operation is performed according to +the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_div( float128 a, float128 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, zExp; + bits64 aSig0, aSig1, bSig0, bSig1, zSig0, zSig1, zSig2; + bits64 rem0, rem1, rem2, rem3, term0, term1, term2, term3; + float128 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + bSig1 = extractFloat128Frac1( b ); + bSig0 = extractFloat128Frac0( b ); + bExp = extractFloat128Exp( b ); + bSign = extractFloat128Sign( b ); + zSign = aSign ^ bSign; + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, b ); + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + goto invalid; + } + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + return packFloat128( zSign, 0, 0, 0 ); + } + if ( bExp == 0 ) { + if ( ( bSig0 | bSig1 ) == 0 ) { + if ( ( aExp | aSig0 | aSig1 ) == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = float128_default_nan_low; + z.high = float128_default_nan_high; + return z; + } + float_raise( float_flag_divbyzero ); + return packFloat128( zSign, 0x7FFF, 0, 0 ); + } + normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); + } + if ( aExp == 0 ) { + if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( zSign, 0, 0, 0 ); + normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); + } + zExp = aExp - bExp + 0x3FFD; + shortShift128Left( + aSig0 | LIT64( 0x0001000000000000 ), aSig1, 15, &aSig0, &aSig1 ); + shortShift128Left( + bSig0 | LIT64( 0x0001000000000000 ), bSig1, 15, &bSig0, &bSig1 ); + if ( le128( bSig0, bSig1, aSig0, aSig1 ) ) { + shift128Right( aSig0, aSig1, 1, &aSig0, &aSig1 ); + ++zExp; + } + zSig0 = estimateDiv128To64( aSig0, aSig1, bSig0 ); + mul128By64To192( bSig0, bSig1, zSig0, &term0, &term1, &term2 ); + sub192( aSig0, aSig1, 0, term0, term1, term2, &rem0, &rem1, &rem2 ); + while ( (sbits64) rem0 < 0 ) { + --zSig0; + add192( rem0, rem1, rem2, 0, bSig0, bSig1, &rem0, &rem1, &rem2 ); + } + zSig1 = estimateDiv128To64( rem1, rem2, bSig0 ); + if ( ( zSig1 & 0x3FFF ) <= 4 ) { + mul128By64To192( bSig0, bSig1, zSig1, &term1, &term2, &term3 ); + sub192( rem1, rem2, 0, term1, term2, term3, &rem1, &rem2, &rem3 ); + while ( (sbits64) rem1 < 0 ) { + --zSig1; + add192( rem1, rem2, rem3, 0, bSig0, bSig1, &rem1, &rem2, &rem3 ); + } + zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 ); + } + shift128ExtraRightJamming( zSig0, zSig1, 0, 15, &zSig0, &zSig1, &zSig2 ); + return roundAndPackFloat128( zSign, zExp, zSig0, zSig1, zSig2 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the remainder of the quadruple-precision floating-point value `a' +with respect to the corresponding value `b'. The operation is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_rem( float128 a, float128 b ) +{ + flag aSign, bSign, zSign; + int32 aExp, bExp, expDiff; + bits64 aSig0, aSig1, bSig0, bSig1, q, term0, term1, term2; + bits64 allZero, alternateASig0, alternateASig1, sigMean1; + sbits64 sigMean0; + float128 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + bSig1 = extractFloat128Frac1( b ); + bSig0 = extractFloat128Frac0( b ); + bExp = extractFloat128Exp( b ); + bSign = extractFloat128Sign( b ); + if ( aExp == 0x7FFF ) { + if ( ( aSig0 | aSig1 ) + || ( ( bExp == 0x7FFF ) && ( bSig0 | bSig1 ) ) ) { + return propagateFloat128NaN( a, b ); + } + goto invalid; + } + if ( bExp == 0x7FFF ) { + if ( bSig0 | bSig1 ) return propagateFloat128NaN( a, b ); + return a; + } + if ( bExp == 0 ) { + if ( ( bSig0 | bSig1 ) == 0 ) { + invalid: + float_raise( float_flag_invalid ); + z.low = float128_default_nan_low; + z.high = float128_default_nan_high; + return z; + } + normalizeFloat128Subnormal( bSig0, bSig1, &bExp, &bSig0, &bSig1 ); + } + if ( aExp == 0 ) { + if ( ( aSig0 | aSig1 ) == 0 ) return a; + normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); + } + expDiff = aExp - bExp; + if ( expDiff < -1 ) return a; + shortShift128Left( + aSig0 | LIT64( 0x0001000000000000 ), + aSig1, + 15 - ( expDiff < 0 ), + &aSig0, + &aSig1 + ); + shortShift128Left( + bSig0 | LIT64( 0x0001000000000000 ), bSig1, 15, &bSig0, &bSig1 ); + q = le128( bSig0, bSig1, aSig0, aSig1 ); + if ( q ) sub128( aSig0, aSig1, bSig0, bSig1, &aSig0, &aSig1 ); + expDiff -= 64; + while ( 0 < expDiff ) { + q = estimateDiv128To64( aSig0, aSig1, bSig0 ); + q = ( 4 < q ) ? q - 4 : 0; + mul128By64To192( bSig0, bSig1, q, &term0, &term1, &term2 ); + shortShift192Left( term0, term1, term2, 61, &term1, &term2, &allZero ); + shortShift128Left( aSig0, aSig1, 61, &aSig0, &allZero ); + sub128( aSig0, 0, term1, term2, &aSig0, &aSig1 ); + expDiff -= 61; + } + if ( -64 < expDiff ) { + q = estimateDiv128To64( aSig0, aSig1, bSig0 ); + q = ( 4 < q ) ? q - 4 : 0; + q >>= - expDiff; + shift128Right( bSig0, bSig1, 12, &bSig0, &bSig1 ); + expDiff += 52; + if ( expDiff < 0 ) { + shift128Right( aSig0, aSig1, - expDiff, &aSig0, &aSig1 ); + } + else { + shortShift128Left( aSig0, aSig1, expDiff, &aSig0, &aSig1 ); + } + mul128By64To192( bSig0, bSig1, q, &term0, &term1, &term2 ); + sub128( aSig0, aSig1, term1, term2, &aSig0, &aSig1 ); + } + else { + shift128Right( aSig0, aSig1, 12, &aSig0, &aSig1 ); + shift128Right( bSig0, bSig1, 12, &bSig0, &bSig1 ); + } + do { + alternateASig0 = aSig0; + alternateASig1 = aSig1; + ++q; + sub128( aSig0, aSig1, bSig0, bSig1, &aSig0, &aSig1 ); + } while ( 0 <= (sbits64) aSig0 ); + add128( + aSig0, aSig1, alternateASig0, alternateASig1, &sigMean0, &sigMean1 ); + if ( ( sigMean0 < 0 ) + || ( ( ( sigMean0 | sigMean1 ) == 0 ) && ( q & 1 ) ) ) { + aSig0 = alternateASig0; + aSig1 = alternateASig1; + } + zSign = ( (sbits64) aSig0 < 0 ); + if ( zSign ) sub128( 0, 0, aSig0, aSig1, &aSig0, &aSig1 ); + return + normalizeRoundAndPackFloat128( aSign ^ zSign, bExp - 4, aSig0, aSig1 ); + +} + +/* +------------------------------------------------------------------------------- +Returns the square root of the quadruple-precision floating-point value `a'. +The operation is performed according to the IEC/IEEE Standard for Binary +Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +float128 float128_sqrt( float128 a ) +{ + flag aSign; + int32 aExp, zExp; + bits64 aSig0, aSig1, zSig0, zSig1, zSig2, doubleZSig0; + bits64 rem0, rem1, rem2, rem3, term0, term1, term2, term3; + float128 z; + + aSig1 = extractFloat128Frac1( a ); + aSig0 = extractFloat128Frac0( a ); + aExp = extractFloat128Exp( a ); + aSign = extractFloat128Sign( a ); + if ( aExp == 0x7FFF ) { + if ( aSig0 | aSig1 ) return propagateFloat128NaN( a, a ); + if ( ! aSign ) return a; + goto invalid; + } + if ( aSign ) { + if ( ( aExp | aSig0 | aSig1 ) == 0 ) return a; + invalid: + float_raise( float_flag_invalid ); + z.low = float128_default_nan_low; + z.high = float128_default_nan_high; + return z; + } + if ( aExp == 0 ) { + if ( ( aSig0 | aSig1 ) == 0 ) return packFloat128( 0, 0, 0, 0 ); + normalizeFloat128Subnormal( aSig0, aSig1, &aExp, &aSig0, &aSig1 ); + } + zExp = ( ( aExp - 0x3FFF )>>1 ) + 0x3FFE; + aSig0 |= LIT64( 0x0001000000000000 ); + zSig0 = estimateSqrt32( aExp, aSig0>>17 ); + shortShift128Left( aSig0, aSig1, 13 - ( aExp & 1 ), &aSig0, &aSig1 ); + zSig0 = estimateDiv128To64( aSig0, aSig1, zSig0<<32 ) + ( zSig0<<30 ); + doubleZSig0 = zSig0<<1; + mul64To128( zSig0, zSig0, &term0, &term1 ); + sub128( aSig0, aSig1, term0, term1, &rem0, &rem1 ); + while ( (sbits64) rem0 < 0 ) { + --zSig0; + doubleZSig0 -= 2; + add128( rem0, rem1, zSig0>>63, doubleZSig0 | 1, &rem0, &rem1 ); + } + zSig1 = estimateDiv128To64( rem1, 0, doubleZSig0 ); + if ( ( zSig1 & 0x1FFF ) <= 5 ) { + if ( zSig1 == 0 ) zSig1 = 1; + mul64To128( doubleZSig0, zSig1, &term1, &term2 ); + sub128( rem1, 0, term1, term2, &rem1, &rem2 ); + mul64To128( zSig1, zSig1, &term2, &term3 ); + sub192( rem1, rem2, 0, 0, term2, term3, &rem1, &rem2, &rem3 ); + while ( (sbits64) rem1 < 0 ) { + --zSig1; + shortShift128Left( 0, zSig1, 1, &term2, &term3 ); + term3 |= 1; + term2 |= doubleZSig0; + add192( rem1, rem2, rem3, 0, term2, term3, &rem1, &rem2, &rem3 ); + } + zSig1 |= ( ( rem1 | rem2 | rem3 ) != 0 ); + } + shift128ExtraRightJamming( zSig0, zSig1, 0, 14, &zSig0, &zSig1, &zSig2 ); + return roundAndPackFloat128( 0, zExp, zSig0, zSig1, zSig2 ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is equal to +the corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_eq( float128 a, float128 b ) +{ + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + if ( float128_is_signaling_nan( a ) + || float128_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + return + ( a.low == b.low ) + && ( ( a.high == b.high ) + || ( ( a.low == 0 ) + && ( (bits64) ( ( a.high | b.high )<<1 ) == 0 ) ) + ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is less than +or equal to the corresponding value `b', and 0 otherwise. The comparison +is performed according to the IEC/IEEE Standard for Binary Floating-Point +Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_le( float128 a, float128 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign != bSign ) { + return + aSign + || ( ( ( (bits64) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + == 0 ); + } + return + aSign ? le128( b.high, b.low, a.high, a.low ) + : le128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. The comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_lt( float128 a, float128 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign != bSign ) { + return + aSign + && ( ( ( (bits64) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + != 0 ); + } + return + aSign ? lt128( b.high, b.low, a.high, a.low ) + : lt128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is equal to +the corresponding value `b', and 0 otherwise. The invalid exception is +raised if either operand is a NaN. Otherwise, the comparison is performed +according to the IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_eq_signaling( float128 a, float128 b ) +{ + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + float_raise( float_flag_invalid ); + return 0; + } + return + ( a.low == b.low ) + && ( ( a.high == b.high ) + || ( ( a.low == 0 ) + && ( (bits64) ( ( a.high | b.high )<<1 ) == 0 ) ) + ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is less than +or equal to the corresponding value `b', and 0 otherwise. Quiet NaNs do not +cause an exception. Otherwise, the comparison is performed according to the +IEC/IEEE Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_le_quiet( float128 a, float128 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + if ( float128_is_signaling_nan( a ) + || float128_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign != bSign ) { + return + aSign + || ( ( ( (bits64) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + == 0 ); + } + return + aSign ? le128( b.high, b.low, a.high, a.low ) + : le128( a.high, a.low, b.high, b.low ); + +} + +/* +------------------------------------------------------------------------------- +Returns 1 if the quadruple-precision floating-point value `a' is less than +the corresponding value `b', and 0 otherwise. Quiet NaNs do not cause an +exception. Otherwise, the comparison is performed according to the IEC/IEEE +Standard for Binary Floating-Point Arithmetic. +------------------------------------------------------------------------------- +*/ +flag float128_lt_quiet( float128 a, float128 b ) +{ + flag aSign, bSign; + + if ( ( ( extractFloat128Exp( a ) == 0x7FFF ) + && ( extractFloat128Frac0( a ) | extractFloat128Frac1( a ) ) ) + || ( ( extractFloat128Exp( b ) == 0x7FFF ) + && ( extractFloat128Frac0( b ) | extractFloat128Frac1( b ) ) ) + ) { + if ( float128_is_signaling_nan( a ) + || float128_is_signaling_nan( b ) ) { + float_raise( float_flag_invalid ); + } + return 0; + } + aSign = extractFloat128Sign( a ); + bSign = extractFloat128Sign( b ); + if ( aSign != bSign ) { + return + aSign + && ( ( ( (bits64) ( ( a.high | b.high )<<1 ) ) | a.low | b.low ) + != 0 ); + } + return + aSign ? lt128( b.high, b.low, a.high, a.low ) + : lt128( a.high, a.low, b.high, b.low ); + +} + +#endif + + +#if defined(SOFTFLOAT_FOR_GCC) && defined(SOFTFLOAT_NEED_FIXUNS) + +/* + * These two routines are not part of the original softfloat distribution. + * + * They are based on the corresponding conversions to integer but return + * unsigned numbers instead since these functions are required by GCC. + * + * Added by Mark Brinicombe <mark@netbsd.org> 27/09/97 + * + * float64 version overhauled for SoftFloat 2a [bjh21 2000-07-15] + */ + +/* +------------------------------------------------------------------------------- +Returns the result of converting the double-precision floating-point value +`a' to the 32-bit unsigned integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-point +Arithmetic, except that the conversion is always rounded toward zero. If +`a' is a NaN, the largest positive integer is returned. If the conversion +overflows, the largest integer positive is returned. +------------------------------------------------------------------------------- +*/ +uint32 float64_to_uint32_round_to_zero( float64 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits64 aSig, savedASig; + uint32 z; + + aSig = extractFloat64Frac( a ); + aExp = extractFloat64Exp( a ); + aSign = extractFloat64Sign( a ); + + if (aSign) { + float_raise( float_flag_invalid ); + return(0); + } + + if ( 0x41E < aExp ) { + float_raise( float_flag_invalid ); + return 0xffffffff; + } + else if ( aExp < 0x3FF ) { + if ( aExp || aSig ) float_set_inexact(); + return 0; + } + aSig |= LIT64( 0x0010000000000000 ); + shiftCount = 0x433 - aExp; + savedASig = aSig; + aSig >>= shiftCount; + z = aSig; + if ( ( aSig<<shiftCount ) != savedASig ) { + float_set_inexact(); + } + return z; + +} + +/* +------------------------------------------------------------------------------- +Returns the result of converting the single-precision floating-point value +`a' to the 32-bit unsigned integer format. The conversion is +performed according to the IEC/IEEE Standard for Binary Floating-point +Arithmetic, except that the conversion is always rounded toward zero. If +`a' is a NaN, the largest positive integer is returned. If the conversion +overflows, the largest positive integer is returned. +------------------------------------------------------------------------------- +*/ +uint32 float32_to_uint32_round_to_zero( float32 a ) +{ + flag aSign; + int16 aExp, shiftCount; + bits32 aSig; + uint32 z; + + aSig = extractFloat32Frac( a ); + aExp = extractFloat32Exp( a ); + aSign = extractFloat32Sign( a ); + shiftCount = aExp - 0x9E; + + if (aSign) { + float_raise( float_flag_invalid ); + return(0); + } + if ( 0 < shiftCount ) { + float_raise( float_flag_invalid ); + return 0xFFFFFFFF; + } + else if ( aExp <= 0x7E ) { + if ( aExp | aSig ) float_set_inexact(); + return 0; + } + aSig = ( aSig | 0x800000 )<<8; + z = aSig>>( - shiftCount ); + if ( aSig<<( shiftCount & 31 ) ) { + float_set_inexact(); + } + return z; + +} + +#endif + +#endif /* !NO_IEEE */ diff --git a/sys/lib/libkern/softfloat.h b/sys/lib/libkern/softfloat.h new file mode 100644 index 00000000000..032408f40c7 --- /dev/null +++ b/sys/lib/libkern/softfloat.h @@ -0,0 +1,376 @@ +/* $OpenBSD: softfloat.h,v 1.1 2002/04/28 20:55:14 pvalchev Exp $ */ +/* $NetBSD: softfloat.h,v 1.1 2001/04/26 03:10:48 ross Exp $ */ + +/* This is a derivative work. */ + +/*- + * Copyright (c) 2001 The NetBSD Foundation, Inc. + * All rights reserved. + * + * This code is derived from software contributed to The NetBSD Foundation + * by Ross Harvey. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * 3. All advertising materials mentioning features or use of this software + * must display the following acknowledgement: + * This product includes software developed by the NetBSD + * Foundation, Inc. and its contributors. + * 4. Neither the name of The NetBSD Foundation nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED + * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + * POSSIBILITY OF SUCH DAMAGE. + */ + +/* +=============================================================================== + +This C header file is part of the SoftFloat IEC/IEEE Floating-point +Arithmetic Package, Release 2a. + +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 ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM +THEIR OWN USE OF THE SOFTWARE, AND WHO ALSO EFFECTIVELY INDEMNIFY +(possibly via similar legal warning) JOHN HAUSER AND THE INTERNATIONAL +COMPUTER SCIENCE INSTITUTE AGAINST ALL LOSSES, COSTS, OR OTHER PROBLEMS +ARISING FROM THE USE OF THE SOFTWARE BY THEIR CUSTOMERS AND CLIENTS. + +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 four paragraphs for those parts of +this code that are retained. + +=============================================================================== +*/ + +#ifndef NO_IEEE + +#include <sys/types.h> + +#if !defined(_KERNEL) && !defined(_STANDALONE) +#include <ieeefp.h> +#else +#include "machine/ieeefp.h" +#endif +#include <sys/endian.h> + +/* +------------------------------------------------------------------------------- +The macro `FLOATX80' must be defined to enable the extended double-precision +floating-point format `floatx80'. If this macro is not defined, the +`floatx80' type will not be defined, and none of the functions that either +input or output the `floatx80' type will be defined. The same applies to +the `FLOAT128' macro and the quadruple-precision format `float128'. +------------------------------------------------------------------------------- +*/ +/* #define FLOATX80 */ +/* #define FLOAT128 */ + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE floating-point types. +------------------------------------------------------------------------------- +*/ +typedef u_int32_t float32; +typedef u_int64_t float64; +#ifdef FLOATX80 +typedef struct { +#if BYTE_ORDER == BIG_ENDIAN + u_int16_t high; + u_int64_t low; +#else + u_int64_t low; + u_int16_t high; +#endif +} floatx80; +#endif +#ifdef FLOAT128 +typedef struct { + u_int64_t high, low; +} float128; +#endif + +/* + * Some of the global variables that used to be here have been removed for + * fairly obvious (defopt-MULTIPROCESSOR) reasons. The rest (which don't + * change dynamically) will be removed later. [ross] + */ + +#define float_rounding_mode() fpgetround() + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE floating-point underflow tininess-detection mode. +------------------------------------------------------------------------------- +*/ + +extern int float_detect_tininess; +enum { + float_tininess_after_rounding = 1, + float_tininess_before_rounding = 0 +}; + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE floating-point rounding mode. +------------------------------------------------------------------------------- +*/ + +enum { + float_round_nearest_even = FP_RN, + float_round_to_zero = FP_RZ, + float_round_down = FP_RM, + float_round_up = FP_RP +}; + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE floating-point exception flags. +------------------------------------------------------------------------------- +*/ + +enum { + float_flag_inexact = FP_X_IMP, + float_flag_underflow = FP_X_UFL, + float_flag_overflow = FP_X_OFL, + float_flag_divbyzero = FP_X_DZ, + float_flag_invalid = FP_X_INV +}; + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE integer-to-floating-point conversion routines. +------------------------------------------------------------------------------- +*/ +float32 int32_to_float32( int ); +float64 int32_to_float64( int ); +#ifdef FLOATX80 +floatx80 int32_to_floatx80( int ); +#endif +#ifdef FLOAT128 +float128 int32_to_float128( int ); +#endif +#ifndef SOFTFLOAT_FOR_GCC /* __floatdi?f is in libgcc2.c */ +float32 int64_to_float32( int64_t ); +float64 int64_to_float64( int64_t ); +#ifdef FLOATX80 +floatx80 int64_to_floatx80( int64_t ); +#endif +#ifdef FLOAT128 +float128 int64_to_float128( int64_t ); +#endif +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE single-precision conversion routines. +------------------------------------------------------------------------------- +*/ +int float32_to_int32( float32 ); +int float32_to_int32_round_to_zero( float32 ); +#ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */ +int64_t float32_to_int64( float32 ); +int64_t float32_to_int64_round_to_zero( float32 ); +#endif +float64 float32_to_float64( float32 ); +#ifdef FLOATX80 +floatx80 float32_to_floatx80( float32 ); +#endif +#ifdef FLOAT128 +float128 float32_to_float128( float32 ); +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE single-precision operations. +------------------------------------------------------------------------------- +*/ +float32 float32_round_to_int( float32 ); +float32 float32_add( float32, float32 ); +float32 float32_sub( float32, float32 ); +float32 float32_mul( float32, float32 ); +float32 float32_div( float32, float32 ); +float32 float32_rem( float32, float32 ); +float32 float32_sqrt( float32 ); +int float32_eq( float32, float32 ); +int float32_le( float32, float32 ); +int float32_lt( float32, float32 ); +int float32_eq_signaling( float32, float32 ); +int float32_le_quiet( float32, float32 ); +int float32_lt_quiet( float32, float32 ); +#ifndef SOFTFLOAT_FOR_GCC +int float32_is_signaling_nan( float32 ); +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE double-precision conversion routines. +------------------------------------------------------------------------------- +*/ +int float64_to_int32( float64 ); +int float64_to_int32_round_to_zero( float64 ); +#ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */ +int64_t float64_to_int64( float64 ); +int64_t float64_to_int64_round_to_zero( float64 ); +#endif +float32 float64_to_float32( float64 ); +#ifdef FLOATX80 +floatx80 float64_to_floatx80( float64 ); +#endif +#ifdef FLOAT128 +float128 float64_to_float128( float64 ); +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE double-precision operations. +------------------------------------------------------------------------------- +*/ +#define float64_default_nan 0xFFF8000000000000LL + +static __inline int +float64_is_nan(float64 a) +{ + return 0xFFE0000000000000LL < a << 1; +} + +static __inline int +float64_is_signaling_nan(float64 a) +{ + return (a >> 51 & 0xFFF) == 0xFFE && (a & 0x0007FFFFFFFFFFFFLL); +} + +float64 float64_round_to_int( float64 ); +float64 float64_add( float64, float64 ); +float64 float64_sub( float64, float64 ); +float64 float64_mul( float64, float64 ); +float64 float64_div( float64, float64 ); +float64 float64_rem( float64, float64 ); +float64 float64_sqrt( float64 ); +int float64_eq( float64, float64 ); +int float64_le( float64, float64 ); +int float64_lt( float64, float64 ); +int float64_eq_signaling( float64, float64 ); +int float64_le_quiet( float64, float64 ); +int float64_lt_quiet( float64, float64 ); +#ifndef SOFTFLOAT_FOR_GCC +int float64_is_signaling_nan( float64 ); +#endif + +#ifdef FLOATX80 + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE extended double-precision conversion routines. +------------------------------------------------------------------------------- +*/ +int floatx80_to_int32( floatx80 ); +int floatx80_to_int32_round_to_zero( floatx80 ); +int64_t floatx80_to_int64( floatx80 ); +int64_t floatx80_to_int64_round_to_zero( floatx80 ); +float32 floatx80_to_float32( floatx80 ); +float64 floatx80_to_float64( floatx80 ); +#ifdef FLOAT128 +float128 floatx80_to_float128( floatx80 ); +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE extended double-precision rounding precision. Valid +values are 32, 64, and 80. +------------------------------------------------------------------------------- +*/ +extern int floatx80_rounding_precision; + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE extended double-precision operations. +------------------------------------------------------------------------------- +*/ +floatx80 floatx80_round_to_int( floatx80 ); +floatx80 floatx80_add( floatx80, floatx80 ); +floatx80 floatx80_sub( floatx80, floatx80 ); +floatx80 floatx80_mul( floatx80, floatx80 ); +floatx80 floatx80_div( floatx80, floatx80 ); +floatx80 floatx80_rem( floatx80, floatx80 ); +floatx80 floatx80_sqrt( floatx80 ); +int floatx80_eq( floatx80, floatx80 ); +int floatx80_le( floatx80, floatx80 ); +int floatx80_lt( floatx80, floatx80 ); +int floatx80_eq_signaling( floatx80, floatx80 ); +int floatx80_le_quiet( floatx80, floatx80 ); +int floatx80_lt_quiet( floatx80, floatx80 ); +int floatx80_is_signaling_nan( floatx80 ); + +#endif + +#ifdef FLOAT128 + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE quadruple-precision conversion routines. +------------------------------------------------------------------------------- +*/ +int float128_to_int32( float128 ); +int float128_to_int32_round_to_zero( float128 ); +int64_t float128_to_int64( float128 ); +int64_t float128_to_int64_round_to_zero( float128 ); +float32 float128_to_float32( float128 ); +float64 float128_to_float64( float128 ); +#ifdef FLOATX80 +floatx80 float128_to_floatx80( float128 ); +#endif + +/* +------------------------------------------------------------------------------- +Software IEC/IEEE quadruple-precision operations. +------------------------------------------------------------------------------- +*/ +float128 float128_round_to_int( float128 ); +float128 float128_add( float128, float128 ); +float128 float128_sub( float128, float128 ); +float128 float128_mul( float128, float128 ); +float128 float128_div( float128, float128 ); +float128 float128_rem( float128, float128 ); +float128 float128_sqrt( float128 ); +int float128_eq( float128, float128 ); +int float128_le( float128, float128 ); +int float128_lt( float128, float128 ); +int float128_eq_signaling( float128, float128 ); +int float128_le_quiet( float128, float128 ); +int float128_lt_quiet( float128, float128 ); +int float128_is_signaling_nan( float128 ); + +#endif + +#endif /* !NO_IEEE */ |