aboutsummaryrefslogtreecommitdiffstats
path: root/arch
diff options
context:
space:
mode:
Diffstat (limited to 'arch')
-rw-r--r--arch/arm/vfp/vfp.h11
-rw-r--r--arch/arm/vfp/vfpdouble.c72
-rw-r--r--arch/arm/vfp/vfpsingle.c70
3 files changed, 82 insertions, 71 deletions
diff --git a/arch/arm/vfp/vfp.h b/arch/arm/vfp/vfp.h
index 96fdf30f6a3b..19ace2e37789 100644
--- a/arch/arm/vfp/vfp.h
+++ b/arch/arm/vfp/vfp.h
@@ -355,3 +355,14 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand);
355 * we check for an error. 355 * we check for an error.
356 */ 356 */
357#define VFP_EXCEPTION_ERROR ((u32)-1 & ~VFP_NAN_FLAG) 357#define VFP_EXCEPTION_ERROR ((u32)-1 & ~VFP_NAN_FLAG)
358
359/*
360 * A flag to tell vfp instruction type
361 */
362#define OP_SCALAR (1 << 0)
363#define OP_SD (1 << 1)
364
365struct op {
366 u32 (* const fn)(int dd, int dn, int dm, u32 fpscr);
367 u32 flags;
368};
diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c
index add48e36c2dc..e19a4f7620de 100644
--- a/arch/arm/vfp/vfpdouble.c
+++ b/arch/arm/vfp/vfpdouble.c
@@ -659,22 +659,22 @@ static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr)
659} 659}
660 660
661 661
662static u32 (* const fop_extfns[32])(int dd, int unused, int dm, u32 fpscr) = { 662static struct op fops_ext[32] = {
663 [FEXT_TO_IDX(FEXT_FCPY)] = vfp_double_fcpy, 663 [FEXT_TO_IDX(FEXT_FCPY)] = {vfp_double_fcpy, 0},
664 [FEXT_TO_IDX(FEXT_FABS)] = vfp_double_fabs, 664 [FEXT_TO_IDX(FEXT_FABS)] = {vfp_double_fabs, 0},
665 [FEXT_TO_IDX(FEXT_FNEG)] = vfp_double_fneg, 665 [FEXT_TO_IDX(FEXT_FNEG)] = {vfp_double_fneg, 0},
666 [FEXT_TO_IDX(FEXT_FSQRT)] = vfp_double_fsqrt, 666 [FEXT_TO_IDX(FEXT_FSQRT)] = {vfp_double_fsqrt, 0},
667 [FEXT_TO_IDX(FEXT_FCMP)] = vfp_double_fcmp, 667 [FEXT_TO_IDX(FEXT_FCMP)] = {vfp_double_fcmp, OP_SCALAR},
668 [FEXT_TO_IDX(FEXT_FCMPE)] = vfp_double_fcmpe, 668 [FEXT_TO_IDX(FEXT_FCMPE)] = {vfp_double_fcmpe, OP_SCALAR},
669 [FEXT_TO_IDX(FEXT_FCMPZ)] = vfp_double_fcmpz, 669 [FEXT_TO_IDX(FEXT_FCMPZ)] = {vfp_double_fcmpz, OP_SCALAR},
670 [FEXT_TO_IDX(FEXT_FCMPEZ)] = vfp_double_fcmpez, 670 [FEXT_TO_IDX(FEXT_FCMPEZ)] = {vfp_double_fcmpez, OP_SCALAR},
671 [FEXT_TO_IDX(FEXT_FCVT)] = vfp_double_fcvts, 671 [FEXT_TO_IDX(FEXT_FCVT)] = {vfp_double_fcvts, (OP_SD|OP_SCALAR)},
672 [FEXT_TO_IDX(FEXT_FUITO)] = vfp_double_fuito, 672 [FEXT_TO_IDX(FEXT_FUITO)] = {vfp_double_fuito, OP_SCALAR},
673 [FEXT_TO_IDX(FEXT_FSITO)] = vfp_double_fsito, 673 [FEXT_TO_IDX(FEXT_FSITO)] = {vfp_double_fsito, OP_SCALAR},
674 [FEXT_TO_IDX(FEXT_FTOUI)] = vfp_double_ftoui, 674 [FEXT_TO_IDX(FEXT_FTOUI)] = {vfp_double_ftoui, (OP_SD|OP_SCALAR)},
675 [FEXT_TO_IDX(FEXT_FTOUIZ)] = vfp_double_ftouiz, 675 [FEXT_TO_IDX(FEXT_FTOUIZ)] = {vfp_double_ftouiz, (OP_SD|OP_SCALAR)},
676 [FEXT_TO_IDX(FEXT_FTOSI)] = vfp_double_ftosi, 676 [FEXT_TO_IDX(FEXT_FTOSI)] = {vfp_double_ftosi, (OP_SD|OP_SCALAR)},
677 [FEXT_TO_IDX(FEXT_FTOSIZ)] = vfp_double_ftosiz, 677 [FEXT_TO_IDX(FEXT_FTOSIZ)] = {vfp_double_ftosiz, (OP_SD|OP_SCALAR)},
678}; 678};
679 679
680 680
@@ -1108,16 +1108,16 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr)
1108 return FPSCR_IOC; 1108 return FPSCR_IOC;
1109} 1109}
1110 1110
1111static u32 (* const fop_fns[16])(int dd, int dn, int dm, u32 fpscr) = { 1111static struct op fops[16] = {
1112 [FOP_TO_IDX(FOP_FMAC)] = vfp_double_fmac, 1112 [FOP_TO_IDX(FOP_FMAC)] = {vfp_double_fmac, 0},
1113 [FOP_TO_IDX(FOP_FNMAC)] = vfp_double_fnmac, 1113 [FOP_TO_IDX(FOP_FNMAC)] = {vfp_double_fnmac, 0},
1114 [FOP_TO_IDX(FOP_FMSC)] = vfp_double_fmsc, 1114 [FOP_TO_IDX(FOP_FMSC)] = {vfp_double_fmsc, 0},
1115 [FOP_TO_IDX(FOP_FNMSC)] = vfp_double_fnmsc, 1115 [FOP_TO_IDX(FOP_FNMSC)] = {vfp_double_fnmsc, 0},
1116 [FOP_TO_IDX(FOP_FMUL)] = vfp_double_fmul, 1116 [FOP_TO_IDX(FOP_FMUL)] = {vfp_double_fmul, 0},
1117 [FOP_TO_IDX(FOP_FNMUL)] = vfp_double_fnmul, 1117 [FOP_TO_IDX(FOP_FNMUL)] = {vfp_double_fnmul, 0},
1118 [FOP_TO_IDX(FOP_FADD)] = vfp_double_fadd, 1118 [FOP_TO_IDX(FOP_FADD)] = {vfp_double_fadd, 0},
1119 [FOP_TO_IDX(FOP_FSUB)] = vfp_double_fsub, 1119 [FOP_TO_IDX(FOP_FSUB)] = {vfp_double_fsub, 0},
1120 [FOP_TO_IDX(FOP_FDIV)] = vfp_double_fdiv, 1120 [FOP_TO_IDX(FOP_FDIV)] = {vfp_double_fdiv, 0},
1121}; 1121};
1122 1122
1123#define FREG_BANK(x) ((x) & 0x0c) 1123#define FREG_BANK(x) ((x) & 0x0c)
@@ -1131,39 +1131,39 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
1131 unsigned int dn = vfp_get_dn(inst); 1131 unsigned int dn = vfp_get_dn(inst);
1132 unsigned int dm = vfp_get_dm(inst); 1132 unsigned int dm = vfp_get_dm(inst);
1133 unsigned int vecitr, veclen, vecstride; 1133 unsigned int vecitr, veclen, vecstride;
1134 u32 (*fop)(int, int, s32, u32); 1134 struct op *fop;
1135 1135
1136 veclen = fpscr & FPSCR_LENGTH_MASK;
1137 vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2; 1136 vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2;
1138 1137
1138 fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
1139 /* 1139 /*
1140 * fcvtds takes an sN register number as destination, not dN. 1140 * fcvtds takes an sN register number as destination, not dN.
1141 * It also always operates on scalars. 1141 * It also always operates on scalars.
1142 */ 1142 */
1143 if ((inst & FEXT_MASK) == FEXT_FCVT) { 1143 if (fop->flags & OP_SD)
1144 veclen = 0;
1145 dest = vfp_get_sd(inst); 1144 dest = vfp_get_sd(inst);
1146 } else 1145 else
1147 dest = vfp_get_dd(inst); 1146 dest = vfp_get_dd(inst);
1148 1147
1149 /* 1148 /*
1150 * If destination bank is zero, vector length is always '1'. 1149 * If destination bank is zero, vector length is always '1'.
1151 * ARM DDI0100F C5.1.3, C5.3.2. 1150 * ARM DDI0100F C5.1.3, C5.3.2.
1152 */ 1151 */
1153 if (FREG_BANK(dest) == 0) 1152 if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0))
1154 veclen = 0; 1153 veclen = 0;
1154 else
1155 veclen = fpscr & FPSCR_LENGTH_MASK;
1155 1156
1156 pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride, 1157 pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
1157 (veclen >> FPSCR_LENGTH_BIT) + 1); 1158 (veclen >> FPSCR_LENGTH_BIT) + 1);
1158 1159
1159 fop = (op == FOP_EXT) ? fop_extfns[FEXT_TO_IDX(inst)] : fop_fns[FOP_TO_IDX(op)]; 1160 if (!fop->fn)
1160 if (!fop)
1161 goto invalid; 1161 goto invalid;
1162 1162
1163 for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { 1163 for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
1164 u32 except; 1164 u32 except;
1165 1165
1166 if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT) 1166 if (op == FOP_EXT && (fop->flags & OP_SD))
1167 pr_debug("VFP: itr%d (s%u) = op[%u] (d%u)\n", 1167 pr_debug("VFP: itr%d (s%u) = op[%u] (d%u)\n",
1168 vecitr >> FPSCR_LENGTH_BIT, 1168 vecitr >> FPSCR_LENGTH_BIT,
1169 dest, dn, dm); 1169 dest, dn, dm);
@@ -1176,7 +1176,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
1176 vecitr >> FPSCR_LENGTH_BIT, 1176 vecitr >> FPSCR_LENGTH_BIT,
1177 dest, dn, FOP_TO_IDX(op), dm); 1177 dest, dn, FOP_TO_IDX(op), dm);
1178 1178
1179 except = fop(dest, dn, dm, fpscr); 1179 except = fop->fn(dest, dn, dm, fpscr);
1180 pr_debug("VFP: itr%d: exceptions=%08x\n", 1180 pr_debug("VFP: itr%d: exceptions=%08x\n",
1181 vecitr >> FPSCR_LENGTH_BIT, except); 1181 vecitr >> FPSCR_LENGTH_BIT, except);
1182 1182
diff --git a/arch/arm/vfp/vfpsingle.c b/arch/arm/vfp/vfpsingle.c
index 8f6c179cafbe..4f717d72eb6b 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
705static u32 (* const fop_extfns[32])(int sd, int unused, s32 m, u32 fpscr) = { 705static 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_SD|OP_SCALAR)},
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
1154static u32 (* const fop_fns[16])(int sd, int sn, s32 m, u32 fpscr) = { 1154static 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)
@@ -1174,19 +1174,18 @@ 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 * fcvtsd takes a dN register number as destination, not sN. 1183 * fcvtsd takes a dN register number as destination, not sN.
1184 * Technically, if bit 0 of dd is set, this is an invalid 1184 * Technically, if bit 0 of dd is set, this is an invalid
1185 * instruction. However, we ignore this for efficiency. 1185 * instruction. However, we ignore this for efficiency.
1186 * It also only operates on scalars. 1186 * It also only operates on scalars.
1187 */ 1187 */
1188 if ((inst & FEXT_MASK) == FEXT_FCVT) { 1188 if (fop->flags & OP_SD) {
1189 veclen = 0;
1190 dest = vfp_get_dd(inst); 1189 dest = vfp_get_dd(inst);
1191 } else 1190 } else
1192 dest = vfp_get_sd(inst); 1191 dest = vfp_get_sd(inst);
@@ -1195,21 +1194,22 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
1195 * If destination bank is zero, vector length is always '1'. 1194 * If destination bank is zero, vector length is always '1'.
1196 * ARM DDI0100F C5.1.3, C5.3.2. 1195 * ARM DDI0100F C5.1.3, C5.3.2.
1197 */ 1196 */
1198 if (FREG_BANK(dest) == 0) 1197 if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0))
1199 veclen = 0; 1198 veclen = 0;
1199 else
1200 veclen = fpscr & FPSCR_LENGTH_MASK;
1200 1201
1201 pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride, 1202 pr_debug("VFP: vecstride=%u veclen=%u\n", vecstride,
1202 (veclen >> FPSCR_LENGTH_BIT) + 1); 1203 (veclen >> FPSCR_LENGTH_BIT) + 1);
1203 1204
1204 fop = (op == FOP_EXT) ? fop_extfns[FEXT_TO_IDX(inst)] : fop_fns[FOP_TO_IDX(op)]; 1205 if (!fop->fn)
1205 if (!fop)
1206 goto invalid; 1206 goto invalid;
1207 1207
1208 for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { 1208 for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
1209 s32 m = vfp_get_float(sm); 1209 s32 m = vfp_get_float(sm);
1210 u32 except; 1210 u32 except;
1211 1211
1212 if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT) 1212 if (op == FOP_EXT && (fop->flags & OP_SD))
1213 pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n", 1213 pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n",
1214 vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m); 1214 vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
1215 else if (op == FOP_EXT) 1215 else if (op == FOP_EXT)
@@ -1220,7 +1220,7 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
1220 vecitr >> FPSCR_LENGTH_BIT, dest, sn, 1220 vecitr >> FPSCR_LENGTH_BIT, dest, sn,
1221 FOP_TO_IDX(op), sm, m); 1221 FOP_TO_IDX(op), sm, m);
1222 1222
1223 except = fop(dest, sn, m, fpscr); 1223 except = fop->fn(dest, sn, m, fpscr);
1224 pr_debug("VFP: itr%d: exceptions=%08x\n", 1224 pr_debug("VFP: itr%d: exceptions=%08x\n",
1225 vecitr >> FPSCR_LENGTH_BIT, except); 1225 vecitr >> FPSCR_LENGTH_BIT, except);
1226 1226