diff options
Diffstat (limited to 'arch/arm/vfp/vfpsingle.c')
-rw-r--r-- | arch/arm/vfp/vfpsingle.c | 95 |
1 files changed, 44 insertions, 51 deletions
diff --git a/arch/arm/vfp/vfpsingle.c b/arch/arm/vfp/vfpsingle.c index 8f6c179cafbe..ab5e9503bae5 100644 --- a/arch/arm/vfp/vfpsingle.c +++ b/arch/arm/vfp/vfpsingle.c | |||
@@ -702,22 +702,22 @@ static u32 vfp_single_ftosiz(int sd, int unused, s32 m, u32 fpscr) | |||
702 | return vfp_single_ftosi(sd, unused, m, FPSCR_ROUND_TOZERO); | 702 | return vfp_single_ftosi(sd, unused, m, FPSCR_ROUND_TOZERO); |
703 | } | 703 | } |
704 | 704 | ||
705 | static u32 (* const fop_extfns[32])(int sd, int unused, s32 m, u32 fpscr) = { | 705 | static struct op fops_ext[32] = { |
706 | [FEXT_TO_IDX(FEXT_FCPY)] = vfp_single_fcpy, | 706 | [FEXT_TO_IDX(FEXT_FCPY)] = { vfp_single_fcpy, 0 }, |
707 | [FEXT_TO_IDX(FEXT_FABS)] = vfp_single_fabs, | 707 | [FEXT_TO_IDX(FEXT_FABS)] = { vfp_single_fabs, 0 }, |
708 | [FEXT_TO_IDX(FEXT_FNEG)] = vfp_single_fneg, | 708 | [FEXT_TO_IDX(FEXT_FNEG)] = { vfp_single_fneg, 0 }, |
709 | [FEXT_TO_IDX(FEXT_FSQRT)] = vfp_single_fsqrt, | 709 | [FEXT_TO_IDX(FEXT_FSQRT)] = { vfp_single_fsqrt, 0 }, |
710 | [FEXT_TO_IDX(FEXT_FCMP)] = vfp_single_fcmp, | 710 | [FEXT_TO_IDX(FEXT_FCMP)] = { vfp_single_fcmp, OP_SCALAR }, |
711 | [FEXT_TO_IDX(FEXT_FCMPE)] = vfp_single_fcmpe, | 711 | [FEXT_TO_IDX(FEXT_FCMPE)] = { vfp_single_fcmpe, OP_SCALAR }, |
712 | [FEXT_TO_IDX(FEXT_FCMPZ)] = vfp_single_fcmpz, | 712 | [FEXT_TO_IDX(FEXT_FCMPZ)] = { vfp_single_fcmpz, OP_SCALAR }, |
713 | [FEXT_TO_IDX(FEXT_FCMPEZ)] = vfp_single_fcmpez, | 713 | [FEXT_TO_IDX(FEXT_FCMPEZ)] = { vfp_single_fcmpez, OP_SCALAR }, |
714 | [FEXT_TO_IDX(FEXT_FCVT)] = vfp_single_fcvtd, | 714 | [FEXT_TO_IDX(FEXT_FCVT)] = { vfp_single_fcvtd, OP_SCALAR|OP_DD }, |
715 | [FEXT_TO_IDX(FEXT_FUITO)] = vfp_single_fuito, | 715 | [FEXT_TO_IDX(FEXT_FUITO)] = { vfp_single_fuito, OP_SCALAR }, |
716 | [FEXT_TO_IDX(FEXT_FSITO)] = vfp_single_fsito, | 716 | [FEXT_TO_IDX(FEXT_FSITO)] = { vfp_single_fsito, OP_SCALAR }, |
717 | [FEXT_TO_IDX(FEXT_FTOUI)] = vfp_single_ftoui, | 717 | [FEXT_TO_IDX(FEXT_FTOUI)] = { vfp_single_ftoui, OP_SCALAR }, |
718 | [FEXT_TO_IDX(FEXT_FTOUIZ)] = vfp_single_ftouiz, | 718 | [FEXT_TO_IDX(FEXT_FTOUIZ)] = { vfp_single_ftouiz, OP_SCALAR }, |
719 | [FEXT_TO_IDX(FEXT_FTOSI)] = vfp_single_ftosi, | 719 | [FEXT_TO_IDX(FEXT_FTOSI)] = { vfp_single_ftosi, OP_SCALAR }, |
720 | [FEXT_TO_IDX(FEXT_FTOSIZ)] = vfp_single_ftosiz, | 720 | [FEXT_TO_IDX(FEXT_FTOSIZ)] = { vfp_single_ftosiz, OP_SCALAR }, |
721 | }; | 721 | }; |
722 | 722 | ||
723 | 723 | ||
@@ -1151,16 +1151,16 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr) | |||
1151 | return FPSCR_IOC; | 1151 | return FPSCR_IOC; |
1152 | } | 1152 | } |
1153 | 1153 | ||
1154 | static u32 (* const fop_fns[16])(int sd, int sn, s32 m, u32 fpscr) = { | 1154 | static struct op fops[16] = { |
1155 | [FOP_TO_IDX(FOP_FMAC)] = vfp_single_fmac, | 1155 | [FOP_TO_IDX(FOP_FMAC)] = { vfp_single_fmac, 0 }, |
1156 | [FOP_TO_IDX(FOP_FNMAC)] = vfp_single_fnmac, | 1156 | [FOP_TO_IDX(FOP_FNMAC)] = { vfp_single_fnmac, 0 }, |
1157 | [FOP_TO_IDX(FOP_FMSC)] = vfp_single_fmsc, | 1157 | [FOP_TO_IDX(FOP_FMSC)] = { vfp_single_fmsc, 0 }, |
1158 | [FOP_TO_IDX(FOP_FNMSC)] = vfp_single_fnmsc, | 1158 | [FOP_TO_IDX(FOP_FNMSC)] = { vfp_single_fnmsc, 0 }, |
1159 | [FOP_TO_IDX(FOP_FMUL)] = vfp_single_fmul, | 1159 | [FOP_TO_IDX(FOP_FMUL)] = { vfp_single_fmul, 0 }, |
1160 | [FOP_TO_IDX(FOP_FNMUL)] = vfp_single_fnmul, | 1160 | [FOP_TO_IDX(FOP_FNMUL)] = { vfp_single_fnmul, 0 }, |
1161 | [FOP_TO_IDX(FOP_FADD)] = vfp_single_fadd, | 1161 | [FOP_TO_IDX(FOP_FADD)] = { vfp_single_fadd, 0 }, |
1162 | [FOP_TO_IDX(FOP_FSUB)] = vfp_single_fsub, | 1162 | [FOP_TO_IDX(FOP_FSUB)] = { vfp_single_fsub, 0 }, |
1163 | [FOP_TO_IDX(FOP_FDIV)] = vfp_single_fdiv, | 1163 | [FOP_TO_IDX(FOP_FDIV)] = { vfp_single_fdiv, 0 }, |
1164 | }; | 1164 | }; |
1165 | 1165 | ||
1166 | #define FREG_BANK(x) ((x) & 0x18) | 1166 | #define FREG_BANK(x) ((x) & 0x18) |
@@ -1174,70 +1174,63 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr) | |||
1174 | unsigned int sn = vfp_get_sn(inst); | 1174 | unsigned int sn = vfp_get_sn(inst); |
1175 | unsigned int sm = vfp_get_sm(inst); | 1175 | unsigned int sm = vfp_get_sm(inst); |
1176 | unsigned int vecitr, veclen, vecstride; | 1176 | unsigned int vecitr, veclen, vecstride; |
1177 | u32 (*fop)(int, int, s32, u32); | 1177 | struct op *fop; |
1178 | 1178 | ||
1179 | veclen = fpscr & FPSCR_LENGTH_MASK; | ||
1180 | vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK); | 1179 | vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK); |
1181 | 1180 | ||
1181 | fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)]; | ||
1182 | |||
1182 | /* | 1183 | /* |
1183 | * fcvtsd takes a dN register number as destination, not sN. | 1184 | * fcvtsd takes a dN register number as destination, not sN. |
1184 | * Technically, if bit 0 of dd is set, this is an invalid | 1185 | * Technically, if bit 0 of dd is set, this is an invalid |
1185 | * instruction. However, we ignore this for efficiency. | 1186 | * instruction. However, we ignore this for efficiency. |
1186 | * It also only operates on scalars. | 1187 | * It also only operates on scalars. |
1187 | */ | 1188 | */ |
1188 | if ((inst & FEXT_MASK) == FEXT_FCVT) { | 1189 | if (fop->flags & OP_DD) |
1189 | veclen = 0; | ||
1190 | dest = vfp_get_dd(inst); | 1190 | dest = vfp_get_dd(inst); |
1191 | } else | 1191 | else |
1192 | dest = vfp_get_sd(inst); | 1192 | dest = vfp_get_sd(inst); |
1193 | 1193 | ||
1194 | /* | 1194 | /* |
1195 | * If destination bank is zero, vector length is always '1'. | 1195 | * If destination bank is zero, vector length is always '1'. |
1196 | * ARM DDI0100F C5.1.3, C5.3.2. | 1196 | * ARM DDI0100F C5.1.3, C5.3.2. |
1197 | */ | 1197 | */ |
1198 | if (FREG_BANK(dest) == 0) | 1198 | if ((fop->flags & OP_SCALAR) || FREG_BANK(dest) == 0) |
1199 | veclen = 0; | 1199 | veclen = 0; |
1200 | else | ||
1201 | veclen = fpscr & FPSCR_LENGTH_MASK; | ||
1200 | 1202 | ||
1201 | pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride, | 1203 | pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride, |
1202 | (veclen >> FPSCR_LENGTH_BIT) + 1); | 1204 | (veclen >> FPSCR_LENGTH_BIT) + 1); |
1203 | 1205 | ||
1204 | fop = (op == FOP_EXT) ? fop_extfns[FEXT_TO_IDX(inst)] : fop_fns[FOP_TO_IDX(op)]; | 1206 | if (!fop->fn) |
1205 | if (!fop) | ||
1206 | goto invalid; | 1207 | goto invalid; |
1207 | 1208 | ||
1208 | for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { | 1209 | for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { |
1209 | s32 m = vfp_get_float(sm); | 1210 | s32 m = vfp_get_float(sm); |
1210 | u32 except; | 1211 | u32 except; |
1212 | char type; | ||
1211 | 1213 | ||
1212 | if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT) | 1214 | type = fop->flags & OP_DD ? 'd' : 's'; |
1213 | pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n", | 1215 | if (op == FOP_EXT) |
1214 | vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m); | 1216 | pr_debug("VFP: itr%d (%c%u) = op[%u] (s%u=%08x)\n", |
1215 | else if (op == FOP_EXT) | 1217 | vecitr >> FPSCR_LENGTH_BIT, type, dest, sn, |
1216 | pr_debug("VFP: itr%d (s%u) = op[%u] (s%u=%08x)\n", | 1218 | sm, m); |
1217 | vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m); | ||
1218 | else | 1219 | else |
1219 | pr_debug("VFP: itr%d (s%u) = (s%u) op[%u] (s%u=%08x)\n", | 1220 | pr_debug("VFP: itr%d (%c%u) = (s%u) op[%u] (s%u=%08x)\n", |
1220 | vecitr >> FPSCR_LENGTH_BIT, dest, sn, | 1221 | vecitr >> FPSCR_LENGTH_BIT, type, dest, sn, |
1221 | FOP_TO_IDX(op), sm, m); | 1222 | FOP_TO_IDX(op), sm, m); |
1222 | 1223 | ||
1223 | except = fop(dest, sn, m, fpscr); | 1224 | except = fop->fn(dest, sn, m, fpscr); |
1224 | pr_debug("VFP: itr%d: exceptions=%08x\n", | 1225 | pr_debug("VFP: itr%d: exceptions=%08x\n", |
1225 | vecitr >> FPSCR_LENGTH_BIT, except); | 1226 | vecitr >> FPSCR_LENGTH_BIT, except); |
1226 | 1227 | ||
1227 | exceptions |= except; | 1228 | exceptions |= except; |
1228 | 1229 | ||
1229 | /* | 1230 | /* |
1230 | * This ensures that comparisons only operate on scalars; | ||
1231 | * comparisons always return with one FPSCR status bit set. | ||
1232 | */ | ||
1233 | if (except & (FPSCR_N|FPSCR_Z|FPSCR_C|FPSCR_V)) | ||
1234 | break; | ||
1235 | |||
1236 | /* | ||
1237 | * CHECK: It appears to be undefined whether we stop when | 1231 | * CHECK: It appears to be undefined whether we stop when |
1238 | * we encounter an exception. We continue. | 1232 | * we encounter an exception. We continue. |
1239 | */ | 1233 | */ |
1240 | |||
1241 | dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7); | 1234 | dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7); |
1242 | sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7); | 1235 | sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7); |
1243 | if (FREG_BANK(sm) != 0) | 1236 | if (FREG_BANK(sm) != 0) |