Minki/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2024-11-30 16:11:38 +00:00
Code Issues Packages Projects Releases Wiki Activity
16add41164
linux/arch/nds32/math-emu/fpuemu.c
Vincent Chen 44e92e0364 nds32: support denormalized result through FP emulator 
Currently, the nds32 FPU dose not support the arithmetic of denormalized
number. When the nds32 FPU finds the result of the instruction is a
denormlized number, the nds32 FPU considers it to be an underflow condition
and rounds the result to an appropriate number. It may causes some loss
of precision. This commit proposes a solution to re-execute the
instruction by the FPU emulator to enhance the precision. To transfer
calculations from user space to kernel space, this feature will enable
the underflow exception trap by default. Enabling this feature may cause
some side effects:
  1. Performance loss due to extra FPU exception
  2. Need another scheme to control real underflow trap
       A new parameter, UDF_trap, which is belong to FPU context is used
     to control underflow trap.

User can configure this feature via CONFIG_SUPPORT_DENORMAL_ARITHMETIC

Signed-off-by: Vincent Chen <vincentc@andestech.com>
Acked-by: Greentime Hu <greentime@andestech.com>
Signed-off-by: Greentime Hu <greentime@andestech.com>
2018-11-22 18:13:27 +08:00

358 lines
6.8 KiB
C
Raw Blame History

// SPDX-License-Identifier: GPL-2.0
// Copyright (C) 2005-2018 Andes Technology Corporation
#include <asm/bitfield.h>
#include <asm/uaccess.h>
#include <asm/sfp-machine.h>
#include <asm/fpuemu.h>
#include <asm/nds32_fpu_inst.h>
#define DPFROMREG(dp, x) (dp = (void *)((unsigned long *)fpu_reg + 2*x))
#ifdef __NDS32_EL__
#define SPFROMREG(sp, x)\
((sp) = (void *)((unsigned long *)fpu_reg + (x^1)))
#else
#define SPFROMREG(sp, x) ((sp) = (void *)((unsigned long *)fpu_reg + x))
#endif
#define DEF3OP(name, p, f1, f2) \
void fpemu_##name##p(void *ft, void *fa, void *fb) \
{ \
f1(fa, fa, fb); \
f2(ft, ft, fa); \
}
#define DEF3OPNEG(name, p, f1, f2, f3) \
void fpemu_##name##p(void *ft, void *fa, void *fb) \
{ \
f1(fa, fa, fb); \
f2(ft, ft, fa); \
f3(ft, ft); \
}
DEF3OP(fmadd, s, fmuls, fadds);
DEF3OP(fmsub, s, fmuls, fsubs);
DEF3OP(fmadd, d, fmuld, faddd);
DEF3OP(fmsub, d, fmuld, fsubd);
DEF3OPNEG(fnmadd, s, fmuls, fadds, fnegs);
DEF3OPNEG(fnmsub, s, fmuls, fsubs, fnegs);
DEF3OPNEG(fnmadd, d, fmuld, faddd, fnegd);
DEF3OPNEG(fnmsub, d, fmuld, fsubd, fnegd);
static const unsigned char cmptab[8] = {
SF_CEQ,
SF_CEQ,
SF_CLT,
SF_CLT,
SF_CLT | SF_CEQ,
SF_CLT | SF_CEQ,
SF_CUN,
SF_CUN
};
enum ARGTYPE {
S1S = 1,
S2S,
S1D,
CS,
D1D,
D2D,
D1S,
CD
};
union func_t {
void (*t)(void *ft, void *fa, void *fb);
void (*b)(void *ft, void *fa);
};
/*
* Emulate a single FPU arithmetic instruction.
*/
static int fpu_emu(struct fpu_struct *fpu_reg, unsigned long insn)
{
int rfmt; /* resulting format */
union func_t func;
int ftype = 0;
switch (rfmt = NDS32Insn_OPCODE_COP0(insn)) {
case fs1_op:{
switch (NDS32Insn_OPCODE_BIT69(insn)) {
case fadds_op:
func.t = fadds;
ftype = S2S;
break;
case fsubs_op:
func.t = fsubs;
ftype = S2S;
break;
case fmadds_op:
func.t = fpemu_fmadds;
ftype = S2S;
break;
case fmsubs_op:
func.t = fpemu_fmsubs;
ftype = S2S;
break;
case fnmadds_op:
func.t = fpemu_fnmadds;
ftype = S2S;
break;
case fnmsubs_op:
func.t = fpemu_fnmsubs;
ftype = S2S;
break;
case fmuls_op:
func.t = fmuls;
ftype = S2S;
break;
case fdivs_op:
func.t = fdivs;
ftype = S2S;
break;
case fs1_f2op_op:
switch (NDS32Insn_OPCODE_BIT1014(insn)) {
case fs2d_op:
func.b = fs2d;
ftype = S1D;
break;
case fsqrts_op:
func.b = fsqrts;
ftype = S1S;
break;
default:
return SIGILL;
}
break;
default:
return SIGILL;
}
break;
}
case fs2_op:
switch (NDS32Insn_OPCODE_BIT69(insn)) {
case fcmpeqs_op:
case fcmpeqs_e_op:
case fcmplts_op:
case fcmplts_e_op:
case fcmples_op:
case fcmples_e_op:
case fcmpuns_op:
case fcmpuns_e_op:
ftype = CS;
break;
default:
return SIGILL;
}
break;
case fd1_op:{
switch (NDS32Insn_OPCODE_BIT69(insn)) {
case faddd_op:
func.t = faddd;
ftype = D2D;
break;
case fsubd_op:
func.t = fsubd;
ftype = D2D;
break;
case fmaddd_op:
func.t = fpemu_fmaddd;
ftype = D2D;
break;
case fmsubd_op:
func.t = fpemu_fmsubd;
ftype = D2D;
break;
case fnmaddd_op:
func.t = fpemu_fnmaddd;
ftype = D2D;
break;
case fnmsubd_op:
func.t = fpemu_fnmsubd;
ftype = D2D;
break;
case fmuld_op:
func.t = fmuld;
ftype = D2D;
break;
case fdivd_op:
func.t = fdivd;
ftype = D2D;
break;
case fd1_f2op_op:
switch (NDS32Insn_OPCODE_BIT1014(insn)) {
case fd2s_op:
func.b = fd2s;
ftype = D1S;
break;
case fsqrtd_op:
func.b = fsqrtd;
ftype = D1D;
break;
default:
return SIGILL;
}
break;
default:
return SIGILL;
}
break;
}
case fd2_op:
switch (NDS32Insn_OPCODE_BIT69(insn)) {
case fcmpeqd_op:
case fcmpeqd_e_op:
case fcmpltd_op:
case fcmpltd_e_op:
case fcmpled_op:
case fcmpled_e_op:
case fcmpund_op:
case fcmpund_e_op:
ftype = CD;
break;
default:
return SIGILL;
}
break;
default:
return SIGILL;
}
switch (ftype) {
case S1S:{
void *ft, *fa;
SPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
SPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
func.b(ft, fa);
break;
}
case S2S:{
void *ft, *fa, *fb;
SPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
SPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
SPFROMREG(fb, NDS32Insn_OPCODE_Rb(insn));
func.t(ft, fa, fb);
break;
}
case S1D:{
void *ft, *fa;
DPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
SPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
func.b(ft, fa);
break;
}
case CS:{
unsigned int cmpop = NDS32Insn_OPCODE_BIT69(insn);
void *ft, *fa, *fb;
SPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
SPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
SPFROMREG(fb, NDS32Insn_OPCODE_Rb(insn));
if (cmpop < 0x8) {
cmpop = cmptab[cmpop];
fcmps(ft, fa, fb, cmpop);
} else
return SIGILL;
break;
}
case D1D:{
void *ft, *fa;
DPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
DPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
func.b(ft, fa);
break;
}
case D2D:{
void *ft, *fa, *fb;
DPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
DPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
DPFROMREG(fb, NDS32Insn_OPCODE_Rb(insn));
func.t(ft, fa, fb);
break;
}
case D1S:{
void *ft, *fa;
SPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
DPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
func.b(ft, fa);
break;
}
case CD:{
unsigned int cmpop = NDS32Insn_OPCODE_BIT69(insn);
void *ft, *fa, *fb;
SPFROMREG(ft, NDS32Insn_OPCODE_Rt(insn));
DPFROMREG(fa, NDS32Insn_OPCODE_Ra(insn));
DPFROMREG(fb, NDS32Insn_OPCODE_Rb(insn));
if (cmpop < 0x8) {
cmpop = cmptab[cmpop];
fcmpd(ft, fa, fb, cmpop);
} else
return SIGILL;
break;
}
default:
return SIGILL;
}
/*
* If an exception is required, generate a tidy SIGFPE exception.
*/
#if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
if (((fpu_reg->fpcsr << 5) & fpu_reg->fpcsr & FPCSR_mskALLE_NO_UDFE) ||
((fpu_reg->fpcsr & FPCSR_mskUDF) && (fpu_reg->UDF_trap)))
#else
if ((fpu_reg->fpcsr << 5) & fpu_reg->fpcsr & FPCSR_mskALLE)
#endif
return SIGFPE;
return 0;
}
int do_fpuemu(struct pt_regs *regs, struct fpu_struct *fpu)
{
unsigned long insn = 0, addr = regs->ipc;
unsigned long emulpc, contpc;
unsigned char *pc = (void *)&insn;
char c;
int i = 0, ret;
for (i = 0; i < 4; i++) {
if (__get_user(c, (unsigned char *)addr++))
return SIGBUS;
*pc++ = c;
}
insn = be32_to_cpu(insn);
emulpc = regs->ipc;
contpc = regs->ipc + 4;
if (NDS32Insn_OPCODE(insn) != cop0_op)
return SIGILL;
switch (NDS32Insn_OPCODE_COP0(insn)) {
case fs1_op:
case fs2_op:
case fd1_op:
case fd2_op:
{
/* a real fpu computation instruction */
ret = fpu_emu(fpu, insn);
if (!ret)
regs->ipc = contpc;
}
break;
default:
return SIGILL;
}
return ret;
}
Reference in New Issue View Git Blame Copy Permalink