[ARM] Cleanups for 4cc9bd2eaa1063c68341c1c00e66660adcfdf254
authorRussell King <rmk@dyn-67.arm.linux.org.uk>
Thu, 21 Sep 2006 16:00:08 +0000 (17:00 +0100)
committerRussell King <rmk+kernel@arm.linux.org.uk>
Mon, 25 Sep 2006 09:34:59 +0000 (10:34 +0100)
- Document the meaning for OP_SCALAR, OP_SD and add OP_DD.
- Formatting cleanups
- Remove now redundant code for making compare instructions
  operate on scalar values.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
arch/arm/vfp/vfp.h
arch/arm/vfp/vfpdouble.c
arch/arm/vfp/vfpsingle.c

index 19ace2e..f279789 100644 (file)
@@ -357,10 +357,14 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand);
 #define VFP_EXCEPTION_ERROR    ((u32)-1 & ~VFP_NAN_FLAG)
 
 /*
- * A flag to tell vfp instruction type
+ * A flag to tell vfp instruction type.
+ *  OP_SCALAR - this operation always operates in scalar mode
+ *  OP_SD - the instruction exceptionally writes to a single precision result.
+ *  OP_DD - the instruction exceptionally writes to a double precision result.
  */
 #define OP_SCALAR      (1 << 0)
 #define OP_SD          (1 << 1)
+#define OP_DD          (1 << 1)
 
 struct op {
        u32 (* const fn)(int dd, int dn, int dm, u32 fpscr);
index e19a4f7..4fc05ee 100644 (file)
@@ -660,21 +660,21 @@ static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr)
 
 
 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_SD|OP_SCALAR)},
-       [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_SD|OP_SCALAR)},
-       [FEXT_TO_IDX(FEXT_FTOUIZ)]      = {vfp_double_ftouiz, (OP_SD|OP_SCALAR)},
-       [FEXT_TO_IDX(FEXT_FTOSI)]       = {vfp_double_ftosi, (OP_SD|OP_SCALAR)},
-       [FEXT_TO_IDX(FEXT_FTOSIZ)]      = {vfp_double_ftosiz, (OP_SD|OP_SCALAR)},
+       [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 },
 };
 
 
@@ -1109,15 +1109,15 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr)
 }
 
 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},
+       [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)
@@ -1136,6 +1136,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
        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.
@@ -1162,19 +1163,17 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
 
        for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
                u32 except;
+               char type;
 
-               if (op == FOP_EXT && (fop->flags & OP_SD))
-                       pr_debug("VFP: itr%d (s%u) = op[%u] (d%u)\n",
-                                vecitr >> FPSCR_LENGTH_BIT,
-                                dest, dn, dm);
-               else if (op == FOP_EXT)
-                       pr_debug("VFP: itr%d (d%u) = op[%u] (d%u)\n",
+               type = fop->flags & OP_SD ? 's' : 'd';
+               if (op == FOP_EXT)
+                       pr_debug("VFP: itr%d (%c%u) = op[%u] (d%u)\n",
                                 vecitr >> FPSCR_LENGTH_BIT,
-                                dest, dn, dm);
+                                type, dest, dn, dm);
                else
-                       pr_debug("VFP: itr%d (d%u) = (d%u) op[%u] (d%u)\n",
+                       pr_debug("VFP: itr%d (%c%u) = (d%u) op[%u] (d%u)\n",
                                 vecitr >> FPSCR_LENGTH_BIT,
-                                dest, dn, FOP_TO_IDX(op), dm);
+                                type, dest, dn, FOP_TO_IDX(op), dm);
 
                except = fop->fn(dest, dn, dm, fpscr);
                pr_debug("VFP: itr%d: exceptions=%08x\n",
@@ -1183,17 +1182,9 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
                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.
                 */
-
                dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6);
                dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6);
                if (FREG_BANK(dm) != 0)
index 4f717d7..ab5e950 100644 (file)
@@ -703,21 +703,21 @@ static u32 vfp_single_ftosiz(int sd, int unused, s32 m, u32 fpscr)
 }
 
 static struct op fops_ext[32] = {
-       [FEXT_TO_IDX(FEXT_FCPY)]        = {vfp_single_fcpy, 0},
-       [FEXT_TO_IDX(FEXT_FABS)]        = {vfp_single_fabs, 0},
-       [FEXT_TO_IDX(FEXT_FNEG)]        = {vfp_single_fneg, 0},
-       [FEXT_TO_IDX(FEXT_FSQRT)]       = {vfp_single_fsqrt, 0},
-       [FEXT_TO_IDX(FEXT_FCMP)]        = {vfp_single_fcmp, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FCMPE)]       = {vfp_single_fcmpe, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FCMPZ)]       = {vfp_single_fcmpz, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FCMPEZ)]      = {vfp_single_fcmpez, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FCVT)]        = {vfp_single_fcvtd, (OP_SD|OP_SCALAR)},
-       [FEXT_TO_IDX(FEXT_FUITO)]       = {vfp_single_fuito, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FSITO)]       = {vfp_single_fsito, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FTOUI)]       = {vfp_single_ftoui, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FTOUIZ)]      = {vfp_single_ftouiz, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FTOSI)]       = {vfp_single_ftosi, OP_SCALAR},
-       [FEXT_TO_IDX(FEXT_FTOSIZ)]      = {vfp_single_ftosiz, OP_SCALAR},
+       [FEXT_TO_IDX(FEXT_FCPY)]        = { vfp_single_fcpy,   0 },
+       [FEXT_TO_IDX(FEXT_FABS)]        = { vfp_single_fabs,   0 },
+       [FEXT_TO_IDX(FEXT_FNEG)]        = { vfp_single_fneg,   0 },
+       [FEXT_TO_IDX(FEXT_FSQRT)]       = { vfp_single_fsqrt,  0 },
+       [FEXT_TO_IDX(FEXT_FCMP)]        = { vfp_single_fcmp,   OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FCMPE)]       = { vfp_single_fcmpe,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FCMPZ)]       = { vfp_single_fcmpz,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FCMPEZ)]      = { vfp_single_fcmpez, OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FCVT)]        = { vfp_single_fcvtd,  OP_SCALAR|OP_DD },
+       [FEXT_TO_IDX(FEXT_FUITO)]       = { vfp_single_fuito,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FSITO)]       = { vfp_single_fsito,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FTOUI)]       = { vfp_single_ftoui,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FTOUIZ)]      = { vfp_single_ftouiz, OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FTOSI)]       = { vfp_single_ftosi,  OP_SCALAR },
+       [FEXT_TO_IDX(FEXT_FTOSIZ)]      = { vfp_single_ftosiz, OP_SCALAR },
 };
 
 
@@ -1152,15 +1152,15 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr)
 }
 
 static struct op fops[16] = {
-       [FOP_TO_IDX(FOP_FMAC)]  = {vfp_single_fmac, 0},
-       [FOP_TO_IDX(FOP_FNMAC)] = {vfp_single_fnmac, 0},
-       [FOP_TO_IDX(FOP_FMSC)]  = {vfp_single_fmsc, 0},
-       [FOP_TO_IDX(FOP_FNMSC)] = {vfp_single_fnmsc, 0},
-       [FOP_TO_IDX(FOP_FMUL)]  = {vfp_single_fmul, 0},
-       [FOP_TO_IDX(FOP_FNMUL)] = {vfp_single_fnmul, 0},
-       [FOP_TO_IDX(FOP_FADD)]  = {vfp_single_fadd, 0},
-       [FOP_TO_IDX(FOP_FSUB)]  = {vfp_single_fsub, 0},
-       [FOP_TO_IDX(FOP_FDIV)]  = {vfp_single_fdiv, 0},
+       [FOP_TO_IDX(FOP_FMAC)]  = { vfp_single_fmac,  0 },
+       [FOP_TO_IDX(FOP_FNMAC)] = { vfp_single_fnmac, 0 },
+       [FOP_TO_IDX(FOP_FMSC)]  = { vfp_single_fmsc,  0 },
+       [FOP_TO_IDX(FOP_FNMSC)] = { vfp_single_fnmsc, 0 },
+       [FOP_TO_IDX(FOP_FMUL)]  = { vfp_single_fmul,  0 },
+       [FOP_TO_IDX(FOP_FNMUL)] = { vfp_single_fnmul, 0 },
+       [FOP_TO_IDX(FOP_FADD)]  = { vfp_single_fadd,  0 },
+       [FOP_TO_IDX(FOP_FSUB)]  = { vfp_single_fsub,  0 },
+       [FOP_TO_IDX(FOP_FDIV)]  = { vfp_single_fdiv,  0 },
 };
 
 #define FREG_BANK(x)   ((x) & 0x18)
@@ -1179,22 +1179,23 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
        vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
 
        fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)];
+
        /*
         * fcvtsd takes a dN register number as destination, not sN.
         * Technically, if bit 0 of dd is set, this is an invalid
         * instruction.  However, we ignore this for efficiency.
         * It also only operates on scalars.
         */
-       if (fop->flags & OP_SD) {
+       if (fop->flags & OP_DD)
                dest = vfp_get_dd(inst);
-       else
+       else
                dest = vfp_get_sd(inst);
 
        /*
         * If destination bank is zero, vector length is always '1'.
         * ARM DDI0100F C5.1.3, C5.3.2.
         */
-       if ((fop->flags & OP_SCALAR) || (FREG_BANK(dest) == 0))
+       if ((fop->flags & OP_SCALAR) || FREG_BANK(dest) == 0)
                veclen = 0;
        else
                veclen = fpscr & FPSCR_LENGTH_MASK;
@@ -1208,16 +1209,16 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
        for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
                s32 m = vfp_get_float(sm);
                u32 except;
+               char type;
 
-               if (op == FOP_EXT && (fop->flags & OP_SD))
-                       pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n",
-                                vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
-               else if (op == FOP_EXT)
-                       pr_debug("VFP: itr%d (s%u) = op[%u] (s%u=%08x)\n",
-                                vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
+               type = fop->flags & OP_DD ? 'd' : 's';
+               if (op == FOP_EXT)
+                       pr_debug("VFP: itr%d (%c%u) = op[%u] (s%u=%08x)\n",
+                                vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
+                                sm, m);
                else
-                       pr_debug("VFP: itr%d (s%u) = (s%u) op[%u] (s%u=%08x)\n",
-                                vecitr >> FPSCR_LENGTH_BIT, dest, sn,
+                       pr_debug("VFP: itr%d (%c%u) = (s%u) op[%u] (s%u=%08x)\n",
+                                vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
                                 FOP_TO_IDX(op), sm, m);
 
                except = fop->fn(dest, sn, m, fpscr);
@@ -1227,17 +1228,9 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
                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.
                 */
-
                dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7);
                sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
                if (FREG_BANK(sm) != 0)