Skip to content

L
linux
Commit baf97ce6 authored by Russell King's avatar Russell King Committed by Russell King
Browse files
Options

[ARM] Cleanups for 4cc9bd2e 

- 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: default avatarRussell King <rmk+kernel@arm.linux.org.uk>
parent 4cc9bd2e
Hide whitespace changes
Inline Side-by-side
Showing with 73 additions and 85 deletions
arch/arm/vfp/vfp.h
View file @ baf97ce6
...@@ -357,10 +357,14 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand); ...@@ -357,10 +357,14 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand);
#define VFP_EXCEPTION_ERROR ((u32)-1 & ~VFP_NAN_FLAG) #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_SCALAR (1 << 0)
#define OP_SD (1 << 1) #define OP_SD (1 << 1)
#define OP_DD (1 << 1)
struct op { struct op {
u32 (* const fn)(int dd, int dn, int dm, u32 fpscr); u32 (* const fn)(int dd, int dn, int dm, u32 fpscr);
......
arch/arm/vfp/vfpdouble.c
View file @ baf97ce6
...@@ -660,21 +660,21 @@ static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr) ...@@ -660,21 +660,21 @@ static u32 vfp_double_ftosiz(int dd, int unused, int dm, u32 fpscr)
static struct op fops_ext[32] = { static struct op fops_ext[32] = {
[FEXT_TO_IDX(FEXT_FCPY)] = {vfp_double_fcpy, 0}, [FEXT_TO_IDX(FEXT_FCPY)] = { vfp_double_fcpy, 0 },
[FEXT_TO_IDX(FEXT_FABS)] = {vfp_double_fabs, 0}, [FEXT_TO_IDX(FEXT_FABS)] = { vfp_double_fabs, 0 },
[FEXT_TO_IDX(FEXT_FNEG)] = {vfp_double_fneg, 0}, [FEXT_TO_IDX(FEXT_FNEG)] = { vfp_double_fneg, 0 },
[FEXT_TO_IDX(FEXT_FSQRT)] = {vfp_double_fsqrt, 0}, [FEXT_TO_IDX(FEXT_FSQRT)] = { vfp_double_fsqrt, 0 },
[FEXT_TO_IDX(FEXT_FCMP)] = {vfp_double_fcmp, OP_SCALAR}, [FEXT_TO_IDX(FEXT_FCMP)] = { vfp_double_fcmp, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FCMPE)] = {vfp_double_fcmpe, 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_FCMPZ)] = { vfp_double_fcmpz, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FCMPEZ)] = {vfp_double_fcmpez, 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_FCVT)] = { vfp_double_fcvts, OP_SCALAR|OP_SD },
[FEXT_TO_IDX(FEXT_FUITO)] = {vfp_double_fuito, 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_FSITO)] = { vfp_double_fsito, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FTOUI)] = {vfp_double_ftoui, (OP_SD|OP_SCALAR)}, [FEXT_TO_IDX(FEXT_FTOUI)] = { vfp_double_ftoui, OP_SCALAR|OP_SD },
[FEXT_TO_IDX(FEXT_FTOUIZ)] = {vfp_double_ftouiz, (OP_SD|OP_SCALAR)}, [FEXT_TO_IDX(FEXT_FTOUIZ)] = { vfp_double_ftouiz, OP_SCALAR|OP_SD },
[FEXT_TO_IDX(FEXT_FTOSI)] = {vfp_double_ftosi, (OP_SD|OP_SCALAR)}, [FEXT_TO_IDX(FEXT_FTOSI)] = { vfp_double_ftosi, OP_SCALAR|OP_SD },
[FEXT_TO_IDX(FEXT_FTOSIZ)] = {vfp_double_ftosiz, (OP_SD|OP_SCALAR)}, [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) ...@@ -1109,15 +1109,15 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr)
} }
static struct op fops[16] = { static struct op fops[16] = {
[FOP_TO_IDX(FOP_FMAC)] = {vfp_double_fmac, 0}, [FOP_TO_IDX(FOP_FMAC)] = { vfp_double_fmac, 0 },
[FOP_TO_IDX(FOP_FNMAC)] = {vfp_double_fnmac, 0}, [FOP_TO_IDX(FOP_FNMAC)] = { vfp_double_fnmac, 0 },
[FOP_TO_IDX(FOP_FMSC)] = {vfp_double_fmsc, 0}, [FOP_TO_IDX(FOP_FMSC)] = { vfp_double_fmsc, 0 },
[FOP_TO_IDX(FOP_FNMSC)] = {vfp_double_fnmsc, 0}, [FOP_TO_IDX(FOP_FNMSC)] = { vfp_double_fnmsc, 0 },
[FOP_TO_IDX(FOP_FMUL)] = {vfp_double_fmul, 0}, [FOP_TO_IDX(FOP_FMUL)] = { vfp_double_fmul, 0 },
[FOP_TO_IDX(FOP_FNMUL)] = {vfp_double_fnmul, 0}, [FOP_TO_IDX(FOP_FNMUL)] = { vfp_double_fnmul, 0 },
[FOP_TO_IDX(FOP_FADD)] = {vfp_double_fadd, 0}, [FOP_TO_IDX(FOP_FADD)] = { vfp_double_fadd, 0 },
[FOP_TO_IDX(FOP_FSUB)] = {vfp_double_fsub, 0}, [FOP_TO_IDX(FOP_FSUB)] = { vfp_double_fsub, 0 },
[FOP_TO_IDX(FOP_FDIV)] = {vfp_double_fdiv, 0}, [FOP_TO_IDX(FOP_FDIV)] = { vfp_double_fdiv, 0 },
}; };
#define FREG_BANK(x) ((x) & 0x0c) #define FREG_BANK(x) ((x) & 0x0c)
...@@ -1136,6 +1136,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr) ...@@ -1136,6 +1136,7 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
vecstride = (1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK)) * 2; 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)]; 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. * fcvtds takes an sN register number as destination, not dN.
* It also always operates on scalars. * It also always operates on scalars.
...@@ -1162,19 +1163,17 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr) ...@@ -1162,19 +1163,17 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
u32 except; u32 except;
char type;
if (op == FOP_EXT && (fop->flags & OP_SD)) type = fop->flags & OP_SD ? 's' : 'd';
pr_debug("VFP: itr%d (s%u) = op[%u] (d%u)\n", if (op == FOP_EXT)
vecitr >> FPSCR_LENGTH_BIT, pr_debug("VFP: itr%d (%c%u) = op[%u] (d%u)\n",
dest, dn, dm);
else if (op == FOP_EXT)
pr_debug("VFP: itr%d (d%u) = op[%u] (d%u)\n",
vecitr >> FPSCR_LENGTH_BIT, vecitr >> FPSCR_LENGTH_BIT,
dest, dn, dm); type, dest, dn, dm);
else 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, 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); except = fop->fn(dest, dn, dm, fpscr);
pr_debug("VFP: itr%d: exceptions=%08x\n", pr_debug("VFP: itr%d: exceptions=%08x\n",
...@@ -1182,18 +1181,10 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr) ...@@ -1182,18 +1181,10 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr)
exceptions |= 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 * CHECK: It appears to be undefined whether we stop when
* we encounter an exception. We continue. * we encounter an exception. We continue.
*/ */
dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6); dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6);
dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6); dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6);
if (FREG_BANK(dm) != 0) if (FREG_BANK(dm) != 0)
......
arch/arm/vfp/vfpsingle.c
View file @ baf97ce6
...@@ -703,21 +703,21 @@ static u32 vfp_single_ftosiz(int sd, int unused, s32 m, u32 fpscr) ...@@ -703,21 +703,21 @@ static u32 vfp_single_ftosiz(int sd, int unused, s32 m, u32 fpscr)
} }
static struct op fops_ext[32] = { static struct op fops_ext[32] = {
[FEXT_TO_IDX(FEXT_FCPY)] = {vfp_single_fcpy, 0}, [FEXT_TO_IDX(FEXT_FCPY)] = { vfp_single_fcpy, 0 },
[FEXT_TO_IDX(FEXT_FABS)] = {vfp_single_fabs, 0}, [FEXT_TO_IDX(FEXT_FABS)] = { vfp_single_fabs, 0 },
[FEXT_TO_IDX(FEXT_FNEG)] = {vfp_single_fneg, 0}, [FEXT_TO_IDX(FEXT_FNEG)] = { vfp_single_fneg, 0 },
[FEXT_TO_IDX(FEXT_FSQRT)] = {vfp_single_fsqrt, 0}, [FEXT_TO_IDX(FEXT_FSQRT)] = { vfp_single_fsqrt, 0 },
[FEXT_TO_IDX(FEXT_FCMP)] = {vfp_single_fcmp, OP_SCALAR}, [FEXT_TO_IDX(FEXT_FCMP)] = { vfp_single_fcmp, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FCMPE)] = {vfp_single_fcmpe, 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_FCMPZ)] = { vfp_single_fcmpz, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FCMPEZ)] = {vfp_single_fcmpez, 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_FCVT)] = { vfp_single_fcvtd, OP_SCALAR|OP_DD },
[FEXT_TO_IDX(FEXT_FUITO)] = {vfp_single_fuito, 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_FSITO)] = { vfp_single_fsito, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FTOUI)] = {vfp_single_ftoui, 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_FTOUIZ)] = { vfp_single_ftouiz, OP_SCALAR },
[FEXT_TO_IDX(FEXT_FTOSI)] = {vfp_single_ftosi, 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_FTOSIZ)] = { vfp_single_ftosiz, OP_SCALAR },
}; };
...@@ -1152,15 +1152,15 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr) ...@@ -1152,15 +1152,15 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr)
} }
static struct op fops[16] = { static struct op fops[16] = {
[FOP_TO_IDX(FOP_FMAC)] = {vfp_single_fmac, 0}, [FOP_TO_IDX(FOP_FMAC)] = { vfp_single_fmac, 0 },
[FOP_TO_IDX(FOP_FNMAC)] = {vfp_single_fnmac, 0}, [FOP_TO_IDX(FOP_FNMAC)] = { vfp_single_fnmac, 0 },
[FOP_TO_IDX(FOP_FMSC)] = {vfp_single_fmsc, 0}, [FOP_TO_IDX(FOP_FMSC)] = { vfp_single_fmsc, 0 },
[FOP_TO_IDX(FOP_FNMSC)] = {vfp_single_fnmsc, 0}, [FOP_TO_IDX(FOP_FNMSC)] = { vfp_single_fnmsc, 0 },
[FOP_TO_IDX(FOP_FMUL)] = {vfp_single_fmul, 0}, [FOP_TO_IDX(FOP_FMUL)] = { vfp_single_fmul, 0 },
[FOP_TO_IDX(FOP_FNMUL)] = {vfp_single_fnmul, 0}, [FOP_TO_IDX(FOP_FNMUL)] = { vfp_single_fnmul, 0 },
[FOP_TO_IDX(FOP_FADD)] = {vfp_single_fadd, 0}, [FOP_TO_IDX(FOP_FADD)] = { vfp_single_fadd, 0 },
[FOP_TO_IDX(FOP_FSUB)] = {vfp_single_fsub, 0}, [FOP_TO_IDX(FOP_FSUB)] = { vfp_single_fsub, 0 },
[FOP_TO_IDX(FOP_FDIV)] = {vfp_single_fdiv, 0}, [FOP_TO_IDX(FOP_FDIV)] = { vfp_single_fdiv, 0 },
}; };
#define FREG_BANK(x) ((x) & 0x18) #define FREG_BANK(x) ((x) & 0x18)
...@@ -1179,22 +1179,23 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr) ...@@ -1179,22 +1179,23 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK); vecstride = 1 + ((fpscr & FPSCR_STRIDE_MASK) == FPSCR_STRIDE_MASK);
fop = (op == FOP_EXT) ? &fops_ext[FEXT_TO_IDX(inst)] : &fops[FOP_TO_IDX(op)]; 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. * fcvtsd takes a dN register number as destination, not sN.
* Technically, if bit 0 of dd is set, this is an invalid * Technically, if bit 0 of dd is set, this is an invalid
* instruction. However, we ignore this for efficiency. * instruction. However, we ignore this for efficiency.
* It also only operates on scalars. * It also only operates on scalars.
*/ */
if (fop->flags & OP_SD) { if (fop->flags & OP_DD)
dest = vfp_get_dd(inst); dest = vfp_get_dd(inst);
} else else
dest = vfp_get_sd(inst); dest = vfp_get_sd(inst);
/* /*
* If destination bank is zero, vector length is always '1'. * If destination bank is zero, vector length is always '1'.
* ARM DDI0100F C5.1.3, C5.3.2. * 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; veclen = 0;
else else
veclen = fpscr & FPSCR_LENGTH_MASK; veclen = fpscr & FPSCR_LENGTH_MASK;
...@@ -1208,16 +1209,16 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr) ...@@ -1208,16 +1209,16 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) {
s32 m = vfp_get_float(sm); s32 m = vfp_get_float(sm);
u32 except; u32 except;
char type;
if (op == FOP_EXT && (fop->flags & OP_SD)) type = fop->flags & OP_DD ? 'd' : 's';
pr_debug("VFP: itr%d (d%u) = op[%u] (s%u=%08x)\n", if (op == FOP_EXT)
vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m); pr_debug("VFP: itr%d (%c%u) = op[%u] (s%u=%08x)\n",
else if (op == FOP_EXT) vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
pr_debug("VFP: itr%d (s%u) = op[%u] (s%u=%08x)\n", sm, m);
vecitr >> FPSCR_LENGTH_BIT, dest, sn, sm, m);
else else
pr_debug("VFP: itr%d (s%u) = (s%u) op[%u] (s%u=%08x)\n", pr_debug("VFP: itr%d (%c%u) = (s%u) op[%u] (s%u=%08x)\n",
vecitr >> FPSCR_LENGTH_BIT, dest, sn, vecitr >> FPSCR_LENGTH_BIT, type, dest, sn,
FOP_TO_IDX(op), sm, m); FOP_TO_IDX(op), sm, m);
except = fop->fn(dest, sn, m, fpscr); except = fop->fn(dest, sn, m, fpscr);
...@@ -1226,18 +1227,10 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr) ...@@ -1226,18 +1227,10 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr)
exceptions |= 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 * CHECK: It appears to be undefined whether we stop when
* we encounter an exception. We continue. * we encounter an exception. We continue.
*/ */
dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7); dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7);
sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7); sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7);
if (FREG_BANK(sm) != 0) if (FREG_BANK(sm) != 0)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7