diff options
author | Steven Whitehouse <swhiteho@redhat.com> | 2006-10-02 16:45:08 +0400 |
---|---|---|
committer | Steven Whitehouse <swhiteho@redhat.com> | 2006-10-02 16:45:08 +0400 |
commit | 59458f40e25915a355d8b1d701425fe9f4f9ea23 (patch) | |
tree | f1c9a2934df686e36d75f759ab7313b6f0e0e5f9 /arch/arm/vfp | |
parent | 825f9075d74028d11d7f5932f04e1b5db3022b51 (diff) | |
parent | d834c16516d1ebec4766fc58c059bf01311e6045 (diff) | |
download | linux-59458f40e25915a355d8b1d701425fe9f4f9ea23.tar.xz |
Merge branch 'master' into gfs2
Diffstat (limited to 'arch/arm/vfp')
-rw-r--r-- | arch/arm/vfp/vfp.h | 15 | ||||
-rw-r--r-- | arch/arm/vfp/vfpdouble.c | 95 | ||||
-rw-r--r-- | arch/arm/vfp/vfpinstr.h | 8 | ||||
-rw-r--r-- | arch/arm/vfp/vfpmodule.c | 40 | ||||
-rw-r--r-- | arch/arm/vfp/vfpsingle.c | 95 |
5 files changed, 122 insertions, 131 deletions
diff --git a/arch/arm/vfp/vfp.h b/arch/arm/vfp/vfp.h index 96fdf30f6a3b..f2797896e6d5 100644 --- a/arch/arm/vfp/vfp.h +++ b/arch/arm/vfp/vfp.h @@ -355,3 +355,18 @@ u32 vfp_estimate_sqrt_significand(u32 exponent, u32 significand); * we check for an error. */ #define VFP_EXCEPTION_ERROR ((u32)-1 & ~VFP_NAN_FLAG) + +/* + * 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); + u32 flags; +}; diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c index add48e36c2dc..4fc05ee0a2ef 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) } -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 }, }; @@ -1108,16 +1108,16 @@ static u32 vfp_double_fdiv(int dd, int dn, int dm, u32 fpscr) 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) @@ -1131,69 +1131,60 @@ u32 vfp_double_cpdo(u32 inst, u32 fpscr) 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 ((inst & FEXT_MASK) == FEXT_FCVT) { - veclen = 0; + if (fop->flags & OP_SD) dest = vfp_get_sd(inst); - } else + 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(dest) == 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[FEXT_TO_IDX(inst)] : 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; - if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT) - 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(dest, 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. */ - dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 6); dn = FREG_BANK(dn) + ((FREG_IDX(dn) + vecstride) & 6); if (FREG_BANK(dm) != 0) diff --git a/arch/arm/vfp/vfpinstr.h b/arch/arm/vfp/vfpinstr.h index 6c819aeae006..7f343a4beca0 100644 --- a/arch/arm/vfp/vfpinstr.h +++ b/arch/arm/vfp/vfpinstr.h @@ -73,14 +73,14 @@ #define fmrx(_vfp_) ({ \ u32 __v; \ - asm("mrc%? p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmrx %0, " #_vfp_ \ - : "=r" (__v)); \ + asm("mrc p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmrx %0, " #_vfp_ \ + : "=r" (__v) : : "cc"); \ __v; \ }) #define fmxr(_vfp_,_var_) \ - asm("mcr%? p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmxr " #_vfp_ ", %0" \ - : : "r" (_var_)) + asm("mcr p10, 7, %0, " vfpreg(_vfp_) ", cr0, 0 @ fmxr " #_vfp_ ", %0" \ + : : "r" (_var_) : "cc") u32 vfp_single_cpdo(u32 inst, u32 fpscr); u32 vfp_single_cprt(u32 inst, u32 fpscr, struct pt_regs *regs); diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c index 4178f6cc3d37..dedbb449632e 100644 --- a/arch/arm/vfp/vfpmodule.c +++ b/arch/arm/vfp/vfpmodule.c @@ -40,10 +40,19 @@ unsigned int VFP_arch; static int vfp_notifier(struct notifier_block *self, unsigned long cmd, void *v) { struct thread_info *thread = v; - union vfp_state *vfp = &thread->vfpstate; + union vfp_state *vfp; - switch (cmd) { - case THREAD_NOTIFY_FLUSH: + if (likely(cmd == THREAD_NOTIFY_SWITCH)) { + /* + * Always disable VFP so we can lazily save/restore the + * old state. + */ + fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_ENABLE); + return NOTIFY_DONE; + } + + vfp = &thread->vfpstate; + if (cmd == THREAD_NOTIFY_FLUSH) { /* * Per-thread VFP initialisation. */ @@ -56,29 +65,12 @@ static int vfp_notifier(struct notifier_block *self, unsigned long cmd, void *v) * Disable VFP to ensure we initialise it first. */ fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_ENABLE); - - /* - * FALLTHROUGH: Ensure we don't try to overwrite our newly - * initialised state information on the first fault. - */ - - case THREAD_NOTIFY_RELEASE: - /* - * Per-thread VFP cleanup. - */ - if (last_VFP_context == vfp) - last_VFP_context = NULL; - break; - - case THREAD_NOTIFY_SWITCH: - /* - * Always disable VFP so we can lazily save/restore the - * old state. - */ - fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_ENABLE); - break; } + /* flush and release case: Per-thread VFP cleanup. */ + if (last_VFP_context == vfp) + last_VFP_context = NULL; + return NOTIFY_DONE; } 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) return vfp_single_ftosi(sd, unused, m, FPSCR_ROUND_TOZERO); } -static u32 (* const fop_extfns[32])(int sd, int unused, s32 m, u32 fpscr) = { - [FEXT_TO_IDX(FEXT_FCPY)] = vfp_single_fcpy, - [FEXT_TO_IDX(FEXT_FABS)] = vfp_single_fabs, - [FEXT_TO_IDX(FEXT_FNEG)] = vfp_single_fneg, - [FEXT_TO_IDX(FEXT_FSQRT)] = vfp_single_fsqrt, - [FEXT_TO_IDX(FEXT_FCMP)] = vfp_single_fcmp, - [FEXT_TO_IDX(FEXT_FCMPE)] = vfp_single_fcmpe, - [FEXT_TO_IDX(FEXT_FCMPZ)] = vfp_single_fcmpz, - [FEXT_TO_IDX(FEXT_FCMPEZ)] = vfp_single_fcmpez, - [FEXT_TO_IDX(FEXT_FCVT)] = vfp_single_fcvtd, - [FEXT_TO_IDX(FEXT_FUITO)] = vfp_single_fuito, - [FEXT_TO_IDX(FEXT_FSITO)] = vfp_single_fsito, - [FEXT_TO_IDX(FEXT_FTOUI)] = vfp_single_ftoui, - [FEXT_TO_IDX(FEXT_FTOUIZ)] = vfp_single_ftouiz, - [FEXT_TO_IDX(FEXT_FTOSI)] = vfp_single_ftosi, - [FEXT_TO_IDX(FEXT_FTOSIZ)] = vfp_single_ftosiz, +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_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 }, }; @@ -1151,16 +1151,16 @@ static u32 vfp_single_fdiv(int sd, int sn, s32 m, u32 fpscr) return FPSCR_IOC; } -static u32 (* const fop_fns[16])(int sd, int sn, s32 m, u32 fpscr) = { - [FOP_TO_IDX(FOP_FMAC)] = vfp_single_fmac, - [FOP_TO_IDX(FOP_FNMAC)] = vfp_single_fnmac, - [FOP_TO_IDX(FOP_FMSC)] = vfp_single_fmsc, - [FOP_TO_IDX(FOP_FNMSC)] = vfp_single_fnmsc, - [FOP_TO_IDX(FOP_FMUL)] = vfp_single_fmul, - [FOP_TO_IDX(FOP_FNMUL)] = vfp_single_fnmul, - [FOP_TO_IDX(FOP_FADD)] = vfp_single_fadd, - [FOP_TO_IDX(FOP_FSUB)] = vfp_single_fsub, - [FOP_TO_IDX(FOP_FDIV)] = vfp_single_fdiv, +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 }, }; #define FREG_BANK(x) ((x) & 0x18) @@ -1174,70 +1174,63 @@ u32 vfp_single_cpdo(u32 inst, u32 fpscr) unsigned int sn = vfp_get_sn(inst); unsigned int sm = vfp_get_sm(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); + 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 ((inst & FEXT_MASK) == FEXT_FCVT) { - veclen = 0; + 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 (FREG_BANK(dest) == 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[FEXT_TO_IDX(inst)] : fop_fns[FOP_TO_IDX(op)]; - if (!fop) + if (!fop->fn) goto invalid; for (vecitr = 0; vecitr <= veclen; vecitr += 1 << FPSCR_LENGTH_BIT) { s32 m = vfp_get_float(sm); u32 except; + char type; - if (op == FOP_EXT && (inst & FEXT_MASK) == FEXT_FCVT) - 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(dest, sn, m, fpscr); + except = fop->fn(dest, sn, m, 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. */ - dest = FREG_BANK(dest) + ((FREG_IDX(dest) + vecstride) & 7); sn = FREG_BANK(sn) + ((FREG_IDX(sn) + vecstride) & 7); if (FREG_BANK(sm) != 0) |