X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=arch%2Farm%2Fvfp%2Fvfpdouble.c;fp=arch%2Farm%2Fvfp%2Fvfpdouble.c;h=e44b9ed0f81f7e7cf885e74d222d94ad248e7f8e;hb=97bf2856c6014879bd04983a3e9dfcdac1e7fe85;hp=9b367a65cb4d65caec8afedf39b53c2cb99b0159;hpb=76828883507a47dae78837ab5dec5a5b4513c667;p=linux-2.6.git diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c index 9b367a65c..e44b9ed0f 100644 --- a/arch/arm/vfp/vfpdouble.c +++ b/arch/arm/vfp/vfpdouble.c @@ -56,7 +56,7 @@ static void vfp_double_normalise_denormal(struct vfp_double *vd) { int bits = 31 - fls(vd->significand >> 32); if (bits == 31) - bits = 62 - fls(vd->significand); + bits = 63 - fls(vd->significand); vfp_double_dump("normalise_denormal: in", vd); @@ -195,9 +195,9 @@ u32 vfp_double_normaliseround(int dd, struct vfp_double *vd, u32 fpscr, u32 exce s64 d = vfp_double_pack(vd); pr_debug("VFP: %s: d(d%d)=%016llx exceptions=%08x\n", func, dd, d, exceptions); - vfp_put_double(dd, d); + vfp_put_double(d, dd); } - return exceptions & ~VFP_NAN_FLAG; + return exceptions; } /* @@ -250,19 +250,19 @@ vfp_propagate_nan(struct vfp_double *vdd, struct vfp_double *vdn, */ static u32 vfp_double_fabs(int dd, int unused, int dm, u32 fpscr) { - vfp_put_double(dd, vfp_double_packed_abs(vfp_get_double(dm))); + vfp_put_double(vfp_double_packed_abs(vfp_get_double(dm)), dd); return 0; } static u32 vfp_double_fcpy(int dd, int unused, int dm, u32 fpscr) { - vfp_put_double(dd, vfp_get_double(dm)); + vfp_put_double(vfp_get_double(dm), dd); return 0; } static u32 vfp_double_fneg(int dd, int unused, int dm, u32 fpscr) { - vfp_put_double(dd, vfp_double_packed_negate(vfp_get_double(dm))); + vfp_put_double(vfp_double_packed_negate(vfp_get_double(dm)), dd); return 0; } @@ -287,7 +287,7 @@ static u32 vfp_double_fsqrt(int dd, int unused, int dm, u32 fpscr) vdp = &vfp_double_default_qnan; ret = FPSCR_IOC; } - vfp_put_double(dd, vfp_double_pack(vdp)); + vfp_put_double(vfp_double_pack(vdp), dd); return ret; } @@ -465,7 +465,7 @@ static u32 vfp_double_fcvts(int sd, int unused, int dm, u32 fpscr) */ if (tm & (VFP_INFINITY|VFP_NAN)) { vsd.exponent = 255; - if (tm & VFP_NAN) + if (tm == VFP_QNAN) vsd.significand |= VFP_SINGLE_SIGNIFICAND_QNAN; goto pack_nan; } else if (tm & VFP_ZERO) @@ -476,7 +476,7 @@ static u32 vfp_double_fcvts(int sd, int unused, int dm, u32 fpscr) return vfp_single_normaliseround(sd, &vsd, fpscr, exceptions, "fcvts"); pack_nan: - vfp_put_float(sd, vfp_single_pack(&vsd)); + vfp_put_float(vfp_single_pack(&vsd), sd); return exceptions; } @@ -573,7 +573,7 @@ static u32 vfp_double_ftoui(int sd, int unused, int dm, u32 fpscr) pr_debug("VFP: ftoui: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions); - vfp_put_float(sd, d); + vfp_put_float(d, sd); return exceptions; } @@ -588,6 +588,7 @@ static u32 vfp_double_ftosi(int sd, int unused, int dm, u32 fpscr) struct vfp_double vdm; u32 d, exceptions = 0; int rmode = fpscr & FPSCR_RMODE_MASK; + int tm; vfp_double_unpack(&vdm, vfp_get_double(dm)); vfp_double_dump("VDM", &vdm); @@ -595,10 +596,14 @@ static u32 vfp_double_ftosi(int sd, int unused, int dm, u32 fpscr) /* * Do we have denormalised number? */ - if (vfp_double_type(&vdm) & VFP_DENORMAL) + tm = vfp_double_type(&vdm); + if (tm & VFP_DENORMAL) exceptions |= FPSCR_IDC; - if (vdm.exponent >= 1023 + 32) { + if (tm & VFP_NAN) { + d = 0; + exceptions |= FPSCR_IOC; + } else if (vdm.exponent >= 1023 + 32) { d = 0x7fffffff; if (vdm.sign) d = ~d; @@ -643,7 +648,7 @@ static u32 vfp_double_ftosi(int sd, int unused, int dm, u32 fpscr) pr_debug("VFP: ftosi: d(s%d)=%08x exceptions=%08x\n", sd, d, exceptions); - vfp_put_float(sd, (s32)d); + vfp_put_float((s32)d, sd); return exceptions; } @@ -654,22 +659,22 @@ static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr) } -static u32 (* const fop_extfns[32])(int dd, int unused, int dm, u32 fpscr) = { - [FEXT_TO_IDX(FEXT_FCPY)] = vfp_double_fcpy, - [FEXT_TO_IDX(FEXT_FABS)] = vfp_double_fabs, - [FEXT_TO_IDX(FEXT_FNEG)] = vfp_double_fneg, - [FEXT_TO_IDX(FEXT_FSQRT)] = vfp_double_fsqrt, - [FEXT_TO_IDX(FEXT_FCMP)] = vfp_double_fcmp, - [FEXT_TO_IDX(FEXT_FCMPE)] = vfp_double_fcmpe, - [FEXT_TO_IDX(FEXT_FCMPZ)] = vfp_double_fcmpz, - [FEXT_TO_IDX(FEXT_FCMPEZ)] = vfp_double_fcmpez, - [FEXT_TO_IDX(FEXT_FCVT)] = vfp_double_fcvts, - [FEXT_TO_IDX(FEXT_FUITO)] = vfp_double_fuito, - [FEXT_TO_IDX(FEXT_FSITO)] = vfp_double_fsito, - [FEXT_TO_IDX(FEXT_FTOUI)] = vfp_double_ftoui, - [FEXT_TO_IDX(FEXT_FTOUIZ)] = vfp_double_ftouiz, - [FEXT_TO_IDX(FEXT_FTOSI)] = vfp_double_ftosi, - [FEXT_TO_IDX(FEXT_FTOSIZ)] = vfp_double_ftosiz, +static struct op fops_ext[32] = { + [FEXT_TO_IDX(FEXT_FCPY)] = { vfp_double_fcpy, 0 }, + [FEXT_TO_IDX(FEXT_FABS)] = { vfp_double_fabs, 0 }, + [FEXT_TO_IDX(FEXT_FNEG)] = { vfp_double_fneg, 0 }, + [FEXT_TO_IDX(FEXT_FSQRT)] = { vfp_double_fsqrt, 0 }, + [FEXT_TO_IDX(FEXT_FCMP)] = { vfp_double_fcmp, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FCMPE)] = { vfp_double_fcmpe, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FCMPZ)] = { vfp_double_fcmpz, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FCMPEZ)] = { vfp_double_fcmpez, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FCVT)] = { vfp_double_fcvts, OP_SCALAR|OP_SD }, + [FEXT_TO_IDX(FEXT_FUITO)] = { vfp_double_fuito, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FSITO)] = { vfp_double_fsito, OP_SCALAR }, + [FEXT_TO_IDX(FEXT_FTOUI)] = { vfp_double_ftoui, OP_SCALAR|OP_SD }, + [FEXT_TO_IDX(FEXT_FTOUIZ)] = { vfp_double_ftouiz, OP_SCALAR|OP_SD }, + [FEXT_TO_IDX(FEXT_FTOSI)] = { vfp_double_ftosi, OP_SCALAR|OP_SD }, + [FEXT_TO_IDX(FEXT_FTOSIZ)] = { vfp_double_ftosiz, OP_SCALAR|OP_SD }, }; @@ -1079,7 +1084,7 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr) vdn_nan: exceptions = vfp_propagate_nan(&vdd, &vdn, &vdm, fpscr); pack: - vfp_put_double(dd, vfp_double_pack(&vdd)); + vfp_put_double(vfp_double_pack(&vdd), dd); return exceptions; vdm_nan: @@ -1099,20 +1104,20 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr) goto pack; invalid: - vfp_put_double(dd, vfp_double_pack(&vfp_double_default_qnan)); + vfp_put_double(vfp_double_pack(&vfp_double_default_qnan), dd); return FPSCR_IOC; } -static u32 (* const fop_fns[16])(int dd, int dn, int dm, u32 fpscr) = { - [FOP_TO_IDX(FOP_FMAC)] = vfp_double_fmac, - [FOP_TO_IDX(FOP_FNMAC)] = vfp_double_fnmac, - [FOP_TO_IDX(FOP_FMSC)] = vfp_double_fmsc, - [FOP_TO_IDX(FOP_FNMSC)] = vfp_double_fnmsc, - [FOP_TO_IDX(FOP_FMUL)] = vfp_double_fmul, - [FOP_TO_IDX(FOP_FNMUL)] = vfp_double_fnmul, - [FOP_TO_IDX(FOP_FADD)] = vfp_double_fadd, - [FOP_TO_IDX(FOP_FSUB)] = vfp_double_fsub, - [FOP_TO_IDX(FOP_FDIV)] = vfp_double_fdiv, +static struct op fops[16] = { + [FOP_TO_IDX(FOP_FMAC)] = { vfp_double_fmac, 0 }, + [FOP_TO_IDX(FOP_FNMAC)] = { vfp_double_fnmac, 0 }, + [FOP_TO_IDX(FOP_FMSC)] = { vfp_double_fmsc, 0 }, + [FOP_TO_IDX(FOP_FNMSC)] = { vfp_double_fnmsc, 0 }, + [FOP_TO_IDX(FOP_FMUL)] = { vfp_double_fmul, 0 }, + [FOP_TO_IDX(FOP_FNMUL)] = { vfp_double_fnmul, 0 }, + [FOP_TO_IDX(FOP_FADD)] = { vfp_double_fadd, 0 }, + [FOP_TO_IDX(FOP_FSUB)] = { vfp_double_fsub, 0 }, + [FOP_TO_IDX(FOP_FDIV)] = { vfp_double_fdiv, 0 }, }; #define FREG_BANK(x) ((x) & 0x0c) @@ -1122,64 +1127,65 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr) { u32 op = inst & FOP_MASK; u32 exceptions = 0; - unsigned int dd = vfp_get_sd(inst); - unsigned int dn = vfp_get_sn(inst); - unsigned int dm = vfp_get_sm(inst); + unsigned int dest; + unsigned int dn = vfp_get_dn(inst); + unsigned int dm = vfp_get_dm(inst); unsigned int vecitr, veclen, vecstride; - u32 (*fop)(int, int, s32, u32); + struct op *fop; - veclen = fpscr & FPSCR_LENGTH_MASK; vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2; + fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)]; + + /* + * fcvtds takes an sN register number as destination, not dN. + * It also always operates on scalars. + */ + if (fop->flags & OP_SD) + dest = vfp_get_sd(inst); + else + dest = vfp_get_dd(inst); + /* * If destination bank is zero, vector length is always '1'. * ARM DDI0100F C5.1.3, C5.3.2. */ - if (FREG_BANK(dd) == 0) + if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0)) veclen = 0; + else + veclen = fpscr & FPSCR_LENGTH_MASK; pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride, (veclen >> FPSCR_LENGTH_BIT) + 1); - fop = (op == FOP_EXT) ? fop_extfns[dn] : fop_fns[FOP_TO_IDX(op)]; - if (!fop) + if (!fop->fn) goto invalid; for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { u32 except; + char type; + type = fop->flags & OP_SD ? 's' : 'd'; if (op == FOP_EXT) - pr_debug("VFP: itr%d (d%u.%u) = op[%u] (d%u.%u)\n", + pr_debug("VFP: itr%d (%c%u) = op[%u] (d%u)\n", vecitr >> FPSCR_LENGTH_BIT, - dd >> 1, dd & 1, dn, - dm >> 1, dm & 1); + type, dest, dn, dm); else - pr_debug("VFP: itr%d (d%u.%u) = (d%u.%u) op[%u] (d%u.%u)\n", + pr_debug("VFP: itr%d (%c%u) = (d%u) op[%u] (d%u)\n", vecitr >> FPSCR_LENGTH_BIT, - dd >> 1, dd & 1, - dn >> 1, dn & 1, - FOP_TO_IDX(op), - dm >> 1, dm & 1); + type, dest, dn, FOP_TO_IDX(op), dm); - except = fop(dd, dn, dm, fpscr); + except = fop->fn(dest, dn, dm, fpscr); pr_debug("VFP: itr%d: exceptions=%08x\n", vecitr >> FPSCR_LENGTH_BIT, except); exceptions |= except; - /* - * This ensures that comparisons only operate on scalars; - * comparisons always return with one FPSCR status bit set. - */ - if (except & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V)) - break; - /* * CHECK: It appears to be undefined whether we stop when * we encounter an exception. We continue. */ - - dd = FREG_BANK(dd) + ((FREG_IDX(dd) + vecstride) & 6); + dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6); dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6); if (FREG_BANK(dm) != 0) dm = FREG_BANK(dm) + ((FREG_IDX(dm) + vecstride) & 6);