aboutsummaryrefslogtreecommitdiffstats
path: root/arch/powerpc/math-emu
diff options
context:
space:
mode:
authorKumar Gala <galak@kernel.crashing.org>2006-03-28 00:43:27 -0500
committerKumar Gala <galak@kernel.crashing.org>2006-03-28 00:43:27 -0500
commit5cd272085bbc905532869f3e1fd18a7100496b56 (patch)
tree2c29f8a713e90fe55864502ed962c656014bca71 /arch/powerpc/math-emu
parentff2e6d7e27cf1f757ab0d97e1a9e46de47152a0e (diff)
powerpc: move math-emu over to arch/powerpc
Towards the goal of having arch/powerpc not build anything over in arch/ppc move math-emu over. Also, killed some references to arch/ppc/ in the arch/powerpc Makefile which should belong in drivers/ when the particular sub-arch's move over to arch/powerpc. Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Diffstat (limited to 'arch/powerpc/math-emu')
-rw-r--r--arch/powerpc/math-emu/Makefile13
-rw-r--r--arch/powerpc/math-emu/double.h129
-rw-r--r--arch/powerpc/math-emu/fabs.c18
-rw-r--r--arch/powerpc/math-emu/fadd.c38
-rw-r--r--arch/powerpc/math-emu/fadds.c39
-rw-r--r--arch/powerpc/math-emu/fcmpo.c46
-rw-r--r--arch/powerpc/math-emu/fcmpu.c42
-rw-r--r--arch/powerpc/math-emu/fctiw.c25
-rw-r--r--arch/powerpc/math-emu/fctiwz.c32
-rw-r--r--arch/powerpc/math-emu/fdiv.c53
-rw-r--r--arch/powerpc/math-emu/fdivs.c55
-rw-r--r--arch/powerpc/math-emu/fmadd.c48
-rw-r--r--arch/powerpc/math-emu/fmadds.c49
-rw-r--r--arch/powerpc/math-emu/fmr.c18
-rw-r--r--arch/powerpc/math-emu/fmsub.c51
-rw-r--r--arch/powerpc/math-emu/fmsubs.c52
-rw-r--r--arch/powerpc/math-emu/fmul.c42
-rw-r--r--arch/powerpc/math-emu/fmuls.c43
-rw-r--r--arch/powerpc/math-emu/fnabs.c18
-rw-r--r--arch/powerpc/math-emu/fneg.c18
-rw-r--r--arch/powerpc/math-emu/fnmadd.c51
-rw-r--r--arch/powerpc/math-emu/fnmadds.c52
-rw-r--r--arch/powerpc/math-emu/fnmsub.c54
-rw-r--r--arch/powerpc/math-emu/fnmsubs.c55
-rw-r--r--arch/powerpc/math-emu/fres.c12
-rw-r--r--arch/powerpc/math-emu/frsp.c25
-rw-r--r--arch/powerpc/math-emu/frsqrte.c12
-rw-r--r--arch/powerpc/math-emu/fsel.c38
-rw-r--r--arch/powerpc/math-emu/fsqrt.c37
-rw-r--r--arch/powerpc/math-emu/fsqrts.c38
-rw-r--r--arch/powerpc/math-emu/fsub.c41
-rw-r--r--arch/powerpc/math-emu/fsubs.c42
-rw-r--r--arch/powerpc/math-emu/lfd.c19
-rw-r--r--arch/powerpc/math-emu/lfs.c37
-rw-r--r--arch/powerpc/math-emu/math.c483
-rw-r--r--arch/powerpc/math-emu/mcrfs.c31
-rw-r--r--arch/powerpc/math-emu/mffs.c17
-rw-r--r--arch/powerpc/math-emu/mtfsb0.c18
-rw-r--r--arch/powerpc/math-emu/mtfsb1.c18
-rw-r--r--arch/powerpc/math-emu/mtfsf.c45
-rw-r--r--arch/powerpc/math-emu/mtfsfi.c23
-rw-r--r--arch/powerpc/math-emu/op-1.h245
-rw-r--r--arch/powerpc/math-emu/op-2.h433
-rw-r--r--arch/powerpc/math-emu/op-4.h297
-rw-r--r--arch/powerpc/math-emu/op-common.h688
-rw-r--r--arch/powerpc/math-emu/sfp-machine.h377
-rw-r--r--arch/powerpc/math-emu/single.h66
-rw-r--r--arch/powerpc/math-emu/soft-fp.h104
-rw-r--r--arch/powerpc/math-emu/stfd.c20
-rw-r--r--arch/powerpc/math-emu/stfiwx.c16
-rw-r--r--arch/powerpc/math-emu/stfs.c41
-rw-r--r--arch/powerpc/math-emu/types.c51
-rw-r--r--arch/powerpc/math-emu/udivmodti4.c191
53 files changed, 4506 insertions, 0 deletions
diff --git a/arch/powerpc/math-emu/Makefile b/arch/powerpc/math-emu/Makefile
new file mode 100644
index 000000000000..754143e8936b
--- /dev/null
+++ b/arch/powerpc/math-emu/Makefile
@@ -0,0 +1,13 @@
1
2obj-y := math.o fmr.o lfd.o stfd.o
3
4obj-$(CONFIG_MATH_EMULATION) += fabs.o fadd.o fadds.o fcmpo.o fcmpu.o \
5 fctiw.o fctiwz.o fdiv.o fdivs.o \
6 fmadd.o fmadds.o fmsub.o fmsubs.o \
7 fmul.o fmuls.o fnabs.o fneg.o types.o \
8 fnmadd.o fnmadds.o fnmsub.o fnmsubs.o \
9 fres.o frsp.o frsqrte.o fsel.o lfs.o \
10 fsqrt.o fsqrts.o fsub.o fsubs.o \
11 mcrfs.o mffs.o mtfsb0.o mtfsb1.o \
12 mtfsf.o mtfsfi.o stfiwx.o stfs.o \
13 udivmodti4.o
diff --git a/arch/powerpc/math-emu/double.h b/arch/powerpc/math-emu/double.h
new file mode 100644
index 000000000000..ffba8b67f059
--- /dev/null
+++ b/arch/powerpc/math-emu/double.h
@@ -0,0 +1,129 @@
1/*
2 * Definitions for IEEE Double Precision
3 */
4
5#if _FP_W_TYPE_SIZE < 32
6#error "Here's a nickel kid. Go buy yourself a real computer."
7#endif
8
9#if _FP_W_TYPE_SIZE < 64
10#define _FP_FRACTBITS_D (2 * _FP_W_TYPE_SIZE)
11#else
12#define _FP_FRACTBITS_D _FP_W_TYPE_SIZE
13#endif
14
15#define _FP_FRACBITS_D 53
16#define _FP_FRACXBITS_D (_FP_FRACTBITS_D - _FP_FRACBITS_D)
17#define _FP_WFRACBITS_D (_FP_WORKBITS + _FP_FRACBITS_D)
18#define _FP_WFRACXBITS_D (_FP_FRACTBITS_D - _FP_WFRACBITS_D)
19#define _FP_EXPBITS_D 11
20#define _FP_EXPBIAS_D 1023
21#define _FP_EXPMAX_D 2047
22
23#define _FP_QNANBIT_D \
24 ((_FP_W_TYPE)1 << ((_FP_FRACBITS_D-2) % _FP_W_TYPE_SIZE))
25#define _FP_IMPLBIT_D \
26 ((_FP_W_TYPE)1 << ((_FP_FRACBITS_D-1) % _FP_W_TYPE_SIZE))
27#define _FP_OVERFLOW_D \
28 ((_FP_W_TYPE)1 << (_FP_WFRACBITS_D % _FP_W_TYPE_SIZE))
29
30#if _FP_W_TYPE_SIZE < 64
31
32union _FP_UNION_D
33{
34 double flt;
35 struct {
36#if __BYTE_ORDER == __BIG_ENDIAN
37 unsigned sign : 1;
38 unsigned exp : _FP_EXPBITS_D;
39 unsigned frac1 : _FP_FRACBITS_D - (_FP_IMPLBIT_D != 0) - _FP_W_TYPE_SIZE;
40 unsigned frac0 : _FP_W_TYPE_SIZE;
41#else
42 unsigned frac0 : _FP_W_TYPE_SIZE;
43 unsigned frac1 : _FP_FRACBITS_D - (_FP_IMPLBIT_D != 0) - _FP_W_TYPE_SIZE;
44 unsigned exp : _FP_EXPBITS_D;
45 unsigned sign : 1;
46#endif
47 } bits __attribute__((packed));
48};
49
50#define FP_DECL_D(X) _FP_DECL(2,X)
51#define FP_UNPACK_RAW_D(X,val) _FP_UNPACK_RAW_2(D,X,val)
52#define FP_PACK_RAW_D(val,X) _FP_PACK_RAW_2(D,val,X)
53
54#define FP_UNPACK_D(X,val) \
55 do { \
56 _FP_UNPACK_RAW_2(D,X,val); \
57 _FP_UNPACK_CANONICAL(D,2,X); \
58 } while (0)
59
60#define FP_PACK_D(val,X) \
61 do { \
62 _FP_PACK_CANONICAL(D,2,X); \
63 _FP_PACK_RAW_2(D,val,X); \
64 } while (0)
65
66#define FP_NEG_D(R,X) _FP_NEG(D,2,R,X)
67#define FP_ADD_D(R,X,Y) _FP_ADD(D,2,R,X,Y)
68#define FP_SUB_D(R,X,Y) _FP_SUB(D,2,R,X,Y)
69#define FP_MUL_D(R,X,Y) _FP_MUL(D,2,R,X,Y)
70#define FP_DIV_D(R,X,Y) _FP_DIV(D,2,R,X,Y)
71#define FP_SQRT_D(R,X) _FP_SQRT(D,2,R,X)
72
73#define FP_CMP_D(r,X,Y,un) _FP_CMP(D,2,r,X,Y,un)
74#define FP_CMP_EQ_D(r,X,Y) _FP_CMP_EQ(D,2,r,X,Y)
75
76#define FP_TO_INT_D(r,X,rsz,rsg) _FP_TO_INT(D,2,r,X,rsz,rsg)
77#define FP_FROM_INT_D(X,r,rs,rt) _FP_FROM_INT(D,2,X,r,rs,rt)
78
79#else
80
81union _FP_UNION_D
82{
83 double flt;
84 struct {
85#if __BYTE_ORDER == __BIG_ENDIAN
86 unsigned sign : 1;
87 unsigned exp : _FP_EXPBITS_D;
88 unsigned long frac : _FP_FRACBITS_D - (_FP_IMPLBIT_D != 0);
89#else
90 unsigned long frac : _FP_FRACBITS_D - (_FP_IMPLBIT_D != 0);
91 unsigned exp : _FP_EXPBITS_D;
92 unsigned sign : 1;
93#endif
94 } bits __attribute__((packed));
95};
96
97#define FP_DECL_D(X) _FP_DECL(1,X)
98#define FP_UNPACK_RAW_D(X,val) _FP_UNPACK_RAW_1(D,X,val)
99#define FP_PACK_RAW_D(val,X) _FP_PACK_RAW_1(D,val,X)
100
101#define FP_UNPACK_D(X,val) \
102 do { \
103 _FP_UNPACK_RAW_1(D,X,val); \
104 _FP_UNPACK_CANONICAL(D,1,X); \
105 } while (0)
106
107#define FP_PACK_D(val,X) \
108 do { \
109 _FP_PACK_CANONICAL(D,1,X); \
110 _FP_PACK_RAW_1(D,val,X); \
111 } while (0)
112
113#define FP_NEG_D(R,X) _FP_NEG(D,1,R,X)
114#define FP_ADD_D(R,X,Y) _FP_ADD(D,1,R,X,Y)
115#define FP_SUB_D(R,X,Y) _FP_SUB(D,1,R,X,Y)
116#define FP_MUL_D(R,X,Y) _FP_MUL(D,1,R,X,Y)
117#define FP_DIV_D(R,X,Y) _FP_DIV(D,1,R,X,Y)
118#define FP_SQRT_D(R,X) _FP_SQRT(D,1,R,X)
119
120/* The implementation of _FP_MUL_D and _FP_DIV_D should be chosen by
121 the target machine. */
122
123#define FP_CMP_D(r,X,Y,un) _FP_CMP(D,1,r,X,Y,un)
124#define FP_CMP_EQ_D(r,X,Y) _FP_CMP_EQ(D,1,r,X,Y)
125
126#define FP_TO_INT_D(r,X,rsz,rsg) _FP_TO_INT(D,1,r,X,rsz,rsg)
127#define FP_FROM_INT_D(X,r,rs,rt) _FP_FROM_INT(D,1,X,r,rs,rt)
128
129#endif /* W_TYPE_SIZE < 64 */
diff --git a/arch/powerpc/math-emu/fabs.c b/arch/powerpc/math-emu/fabs.c
new file mode 100644
index 000000000000..41f0617f3d3a
--- /dev/null
+++ b/arch/powerpc/math-emu/fabs.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6fabs(u32 *frD, u32 *frB)
7{
8 frD[0] = frB[0] & 0x7fffffff;
9 frD[1] = frB[1];
10
11#ifdef DEBUG
12 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
13 dump_double(frD);
14 printk("\n");
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/fadd.c b/arch/powerpc/math-emu/fadd.c
new file mode 100644
index 000000000000..fc8836488b64
--- /dev/null
+++ b/arch/powerpc/math-emu/fadd.c
@@ -0,0 +1,38 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fadd(void *frD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 FP_DECL_D(R);
14 int ret = 0;
15
16#ifdef DEBUG
17 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
18#endif
19
20 __FP_UNPACK_D(A, frA);
21 __FP_UNPACK_D(B, frB);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
25 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
26#endif
27
28 if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF)
29 ret |= EFLAG_VXISI;
30
31 FP_ADD_D(R, A, B);
32
33#ifdef DEBUG
34 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
35#endif
36
37 return (ret | __FP_PACK_D(frD, R));
38}
diff --git a/arch/powerpc/math-emu/fadds.c b/arch/powerpc/math-emu/fadds.c
new file mode 100644
index 000000000000..93025b6c8f3c
--- /dev/null
+++ b/arch/powerpc/math-emu/fadds.c
@@ -0,0 +1,39 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fadds(void *frD, void *frA, void *frB)
11{
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(R);
15 int ret = 0;
16
17#ifdef DEBUG
18 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
19#endif
20
21 __FP_UNPACK_D(A, frA);
22 __FP_UNPACK_D(B, frB);
23
24#ifdef DEBUG
25 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
26 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
27#endif
28
29 if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF)
30 ret |= EFLAG_VXISI;
31
32 FP_ADD_D(R, A, B);
33
34#ifdef DEBUG
35 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
36#endif
37
38 return (ret | __FP_PACK_DS(frD, R));
39}
diff --git a/arch/powerpc/math-emu/fcmpo.c b/arch/powerpc/math-emu/fcmpo.c
new file mode 100644
index 000000000000..4efac394b4cb
--- /dev/null
+++ b/arch/powerpc/math-emu/fcmpo.c
@@ -0,0 +1,46 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fcmpo(u32 *ccr, int crfD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 int code[4] = { (1 << 3), (1 << 1), (1 << 2), (1 << 0) };
14 long cmp;
15 int ret = 0;
16
17#ifdef DEBUG
18 printk("%s: %p (%08x) %d %p %p\n", __FUNCTION__, ccr, *ccr, crfD, frA, frB);
19#endif
20
21 __FP_UNPACK_D(A, frA);
22 __FP_UNPACK_D(B, frB);
23
24#ifdef DEBUG
25 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
26 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
27#endif
28
29 if (A_c == FP_CLS_NAN || B_c == FP_CLS_NAN)
30 ret |= EFLAG_VXVC;
31
32 FP_CMP_D(cmp, A, B, 2);
33 cmp = code[(cmp + 1) & 3];
34
35 __FPU_FPSCR &= ~(0x1f000);
36 __FPU_FPSCR |= (cmp << 12);
37
38 *ccr &= ~(15 << ((7 - crfD) << 2));
39 *ccr |= (cmp << ((7 - crfD) << 2));
40
41#ifdef DEBUG
42 printk("CR: %08x\n", *ccr);
43#endif
44
45 return ret;
46}
diff --git a/arch/powerpc/math-emu/fcmpu.c b/arch/powerpc/math-emu/fcmpu.c
new file mode 100644
index 000000000000..b7e33176e618
--- /dev/null
+++ b/arch/powerpc/math-emu/fcmpu.c
@@ -0,0 +1,42 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fcmpu(u32 *ccr, int crfD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 int code[4] = { (1 << 3), (1 << 1), (1 << 2), (1 << 0) };
14 long cmp;
15
16#ifdef DEBUG
17 printk("%s: %p (%08x) %d %p %p\n", __FUNCTION__, ccr, *ccr, crfD, frA, frB);
18#endif
19
20 __FP_UNPACK_D(A, frA);
21 __FP_UNPACK_D(B, frB);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
25 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
26#endif
27
28 FP_CMP_D(cmp, A, B, 2);
29 cmp = code[(cmp + 1) & 3];
30
31 __FPU_FPSCR &= ~(0x1f000);
32 __FPU_FPSCR |= (cmp << 12);
33
34 *ccr &= ~(15 << ((7 - crfD) << 2));
35 *ccr |= (cmp << ((7 - crfD) << 2));
36
37#ifdef DEBUG
38 printk("CR: %08x\n", *ccr);
39#endif
40
41 return 0;
42}
diff --git a/arch/powerpc/math-emu/fctiw.c b/arch/powerpc/math-emu/fctiw.c
new file mode 100644
index 000000000000..3b3c98b840cf
--- /dev/null
+++ b/arch/powerpc/math-emu/fctiw.c
@@ -0,0 +1,25 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fctiw(u32 *frD, void *frB)
10{
11 FP_DECL_D(B);
12 unsigned int r;
13
14 __FP_UNPACK_D(B, frB);
15 FP_TO_INT_D(r, B, 32, 1);
16 frD[1] = r;
17
18#ifdef DEBUG
19 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
20 dump_double(frD);
21 printk("\n");
22#endif
23
24 return 0;
25}
diff --git a/arch/powerpc/math-emu/fctiwz.c b/arch/powerpc/math-emu/fctiwz.c
new file mode 100644
index 000000000000..7717eb6fcfb6
--- /dev/null
+++ b/arch/powerpc/math-emu/fctiwz.c
@@ -0,0 +1,32 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fctiwz(u32 *frD, void *frB)
10{
11 FP_DECL_D(B);
12 u32 fpscr;
13 unsigned int r;
14
15 fpscr = __FPU_FPSCR;
16 __FPU_FPSCR &= ~(3);
17 __FPU_FPSCR |= FP_RND_ZERO;
18
19 __FP_UNPACK_D(B, frB);
20 FP_TO_INT_D(r, B, 32, 1);
21 frD[1] = r;
22
23 __FPU_FPSCR = fpscr;
24
25#ifdef DEBUG
26 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
27 dump_double(frD);
28 printk("\n");
29#endif
30
31 return 0;
32}
diff --git a/arch/powerpc/math-emu/fdiv.c b/arch/powerpc/math-emu/fdiv.c
new file mode 100644
index 000000000000..f2fba825b2d0
--- /dev/null
+++ b/arch/powerpc/math-emu/fdiv.c
@@ -0,0 +1,53 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fdiv(void *frD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 FP_DECL_D(R);
14 int ret = 0;
15
16#ifdef DEBUG
17 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
18#endif
19
20 __FP_UNPACK_D(A, frA);
21 __FP_UNPACK_D(B, frB);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
25 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
26#endif
27
28 if (A_c == FP_CLS_ZERO && B_c == FP_CLS_ZERO) {
29 ret |= EFLAG_VXZDZ;
30#ifdef DEBUG
31 printk("%s: FPSCR_VXZDZ raised\n", __FUNCTION__);
32#endif
33 }
34 if (A_c == FP_CLS_INF && B_c == FP_CLS_INF) {
35 ret |= EFLAG_VXIDI;
36#ifdef DEBUG
37 printk("%s: FPSCR_VXIDI raised\n", __FUNCTION__);
38#endif
39 }
40
41 if (B_c == FP_CLS_ZERO && A_c != FP_CLS_ZERO) {
42 ret |= EFLAG_DIVZERO;
43 if (__FPU_TRAP_P(EFLAG_DIVZERO))
44 return ret;
45 }
46 FP_DIV_D(R, A, B);
47
48#ifdef DEBUG
49 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
50#endif
51
52 return (ret | __FP_PACK_D(frD, R));
53}
diff --git a/arch/powerpc/math-emu/fdivs.c b/arch/powerpc/math-emu/fdivs.c
new file mode 100644
index 000000000000..b971196e3175
--- /dev/null
+++ b/arch/powerpc/math-emu/fdivs.c
@@ -0,0 +1,55 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fdivs(void *frD, void *frA, void *frB)
11{
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(R);
15 int ret = 0;
16
17#ifdef DEBUG
18 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
19#endif
20
21 __FP_UNPACK_D(A, frA);
22 __FP_UNPACK_D(B, frB);
23
24#ifdef DEBUG
25 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
26 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
27#endif
28
29 if (A_c == FP_CLS_ZERO && B_c == FP_CLS_ZERO) {
30 ret |= EFLAG_VXZDZ;
31#ifdef DEBUG
32 printk("%s: FPSCR_VXZDZ raised\n", __FUNCTION__);
33#endif
34 }
35 if (A_c == FP_CLS_INF && B_c == FP_CLS_INF) {
36 ret |= EFLAG_VXIDI;
37#ifdef DEBUG
38 printk("%s: FPSCR_VXIDI raised\n", __FUNCTION__);
39#endif
40 }
41
42 if (B_c == FP_CLS_ZERO && A_c != FP_CLS_ZERO) {
43 ret |= EFLAG_DIVZERO;
44 if (__FPU_TRAP_P(EFLAG_DIVZERO))
45 return ret;
46 }
47
48 FP_DIV_D(R, A, B);
49
50#ifdef DEBUG
51 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
52#endif
53
54 return (ret | __FP_PACK_DS(frD, R));
55}
diff --git a/arch/powerpc/math-emu/fmadd.c b/arch/powerpc/math-emu/fmadd.c
new file mode 100644
index 000000000000..0a1dbce793e9
--- /dev/null
+++ b/arch/powerpc/math-emu/fmadd.c
@@ -0,0 +1,48 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fmadd(void *frD, void *frA, void *frB, void *frC)
10{
11 FP_DECL_D(R);
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(C);
15 FP_DECL_D(T);
16 int ret = 0;
17
18#ifdef DEBUG
19 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
20#endif
21
22 __FP_UNPACK_D(A, frA);
23 __FP_UNPACK_D(B, frB);
24 __FP_UNPACK_D(C, frC);
25
26#ifdef DEBUG
27 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
28 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
29 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
30#endif
31
32 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
33 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
34 ret |= EFLAG_VXIMZ;
35
36 FP_MUL_D(T, A, C);
37
38 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
39 ret |= EFLAG_VXISI;
40
41 FP_ADD_D(R, T, B);
42
43#ifdef DEBUG
44 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
45#endif
46
47 return (ret | __FP_PACK_D(frD, R));
48}
diff --git a/arch/powerpc/math-emu/fmadds.c b/arch/powerpc/math-emu/fmadds.c
new file mode 100644
index 000000000000..0f70bba9445e
--- /dev/null
+++ b/arch/powerpc/math-emu/fmadds.c
@@ -0,0 +1,49 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fmadds(void *frD, void *frA, void *frB, void *frC)
11{
12 FP_DECL_D(R);
13 FP_DECL_D(A);
14 FP_DECL_D(B);
15 FP_DECL_D(C);
16 FP_DECL_D(T);
17 int ret = 0;
18
19#ifdef DEBUG
20 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
21#endif
22
23 __FP_UNPACK_D(A, frA);
24 __FP_UNPACK_D(B, frB);
25 __FP_UNPACK_D(C, frC);
26
27#ifdef DEBUG
28 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
29 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
30 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
31#endif
32
33 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
34 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
35 ret |= EFLAG_VXIMZ;
36
37 FP_MUL_D(T, A, C);
38
39 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
40 ret |= EFLAG_VXISI;
41
42 FP_ADD_D(R, T, B);
43
44#ifdef DEBUG
45 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
46#endif
47
48 return (ret | __FP_PACK_DS(frD, R));
49}
diff --git a/arch/powerpc/math-emu/fmr.c b/arch/powerpc/math-emu/fmr.c
new file mode 100644
index 000000000000..28df700c0c7e
--- /dev/null
+++ b/arch/powerpc/math-emu/fmr.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6fmr(u32 *frD, u32 *frB)
7{
8 frD[0] = frB[0];
9 frD[1] = frB[1];
10
11#ifdef DEBUG
12 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
13 dump_double(frD);
14 printk("\n");
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/fmsub.c b/arch/powerpc/math-emu/fmsub.c
new file mode 100644
index 000000000000..203fd48a6fec
--- /dev/null
+++ b/arch/powerpc/math-emu/fmsub.c
@@ -0,0 +1,51 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fmsub(void *frD, void *frA, void *frB, void *frC)
10{
11 FP_DECL_D(R);
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(C);
15 FP_DECL_D(T);
16 int ret = 0;
17
18#ifdef DEBUG
19 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
20#endif
21
22 __FP_UNPACK_D(A, frA);
23 __FP_UNPACK_D(B, frB);
24 __FP_UNPACK_D(C, frC);
25
26#ifdef DEBUG
27 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
28 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
29 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
30#endif
31
32 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
33 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
34 ret |= EFLAG_VXIMZ;
35
36 FP_MUL_D(T, A, C);
37
38 if (B_c != FP_CLS_NAN)
39 B_s ^= 1;
40
41 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
42 ret |= EFLAG_VXISI;
43
44 FP_ADD_D(R, T, B);
45
46#ifdef DEBUG
47 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
48#endif
49
50 return (ret | __FP_PACK_D(frD, R));
51}
diff --git a/arch/powerpc/math-emu/fmsubs.c b/arch/powerpc/math-emu/fmsubs.c
new file mode 100644
index 000000000000..8ce68624c189
--- /dev/null
+++ b/arch/powerpc/math-emu/fmsubs.c
@@ -0,0 +1,52 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fmsubs(void *frD, void *frA, void *frB, void *frC)
11{
12 FP_DECL_D(R);
13 FP_DECL_D(A);
14 FP_DECL_D(B);
15 FP_DECL_D(C);
16 FP_DECL_D(T);
17 int ret = 0;
18
19#ifdef DEBUG
20 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
21#endif
22
23 __FP_UNPACK_D(A, frA);
24 __FP_UNPACK_D(B, frB);
25 __FP_UNPACK_D(C, frC);
26
27#ifdef DEBUG
28 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
29 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
30 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
31#endif
32
33 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
34 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
35 ret |= EFLAG_VXIMZ;
36
37 FP_MUL_D(T, A, C);
38
39 if (B_c != FP_CLS_NAN)
40 B_s ^= 1;
41
42 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
43 ret |= EFLAG_VXISI;
44
45 FP_ADD_D(R, T, B);
46
47#ifdef DEBUG
48 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
49#endif
50
51 return (ret | __FP_PACK_DS(frD, R));
52}
diff --git a/arch/powerpc/math-emu/fmul.c b/arch/powerpc/math-emu/fmul.c
new file mode 100644
index 000000000000..66c7e79aae2e
--- /dev/null
+++ b/arch/powerpc/math-emu/fmul.c
@@ -0,0 +1,42 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fmul(void *frD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 FP_DECL_D(R);
14 int ret = 0;
15
16#ifdef DEBUG
17 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
18#endif
19
20 __FP_UNPACK_D(A, frA);
21 __FP_UNPACK_D(B, frB);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
25 A_s, A_f1, A_f0, A_e, A_c, A_f1, A_f0, A_e + 1023);
26 printk("B: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
27 B_s, B_f1, B_f0, B_e, B_c, B_f1, B_f0, B_e + 1023);
28#endif
29
30 if ((A_c == FP_CLS_INF && B_c == FP_CLS_ZERO) ||
31 (A_c == FP_CLS_ZERO && B_c == FP_CLS_INF))
32 ret |= EFLAG_VXIMZ;
33
34 FP_MUL_D(R, A, B);
35
36#ifdef DEBUG
37 printk("D: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
38 R_s, R_f1, R_f0, R_e, R_c, R_f1, R_f0, R_e + 1023);
39#endif
40
41 return (ret | __FP_PACK_D(frD, R));
42}
diff --git a/arch/powerpc/math-emu/fmuls.c b/arch/powerpc/math-emu/fmuls.c
new file mode 100644
index 000000000000..26bc4278271c
--- /dev/null
+++ b/arch/powerpc/math-emu/fmuls.c
@@ -0,0 +1,43 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fmuls(void *frD, void *frA, void *frB)
11{
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(R);
15 int ret = 0;
16
17#ifdef DEBUG
18 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
19#endif
20
21 __FP_UNPACK_D(A, frA);
22 __FP_UNPACK_D(B, frB);
23
24#ifdef DEBUG
25 printk("A: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
26 A_s, A_f1, A_f0, A_e, A_c, A_f1, A_f0, A_e + 1023);
27 printk("B: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
28 B_s, B_f1, B_f0, B_e, B_c, B_f1, B_f0, B_e + 1023);
29#endif
30
31 if ((A_c == FP_CLS_INF && B_c == FP_CLS_ZERO) ||
32 (A_c == FP_CLS_ZERO && B_c == FP_CLS_INF))
33 ret |= EFLAG_VXIMZ;
34
35 FP_MUL_D(R, A, B);
36
37#ifdef DEBUG
38 printk("D: %ld %lu %lu %ld (%ld) [%08lx.%08lx %lx]\n",
39 R_s, R_f1, R_f0, R_e, R_c, R_f1, R_f0, R_e + 1023);
40#endif
41
42 return (ret | __FP_PACK_DS(frD, R));
43}
diff --git a/arch/powerpc/math-emu/fnabs.c b/arch/powerpc/math-emu/fnabs.c
new file mode 100644
index 000000000000..c6b913d179e0
--- /dev/null
+++ b/arch/powerpc/math-emu/fnabs.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6fnabs(u32 *frD, u32 *frB)
7{
8 frD[0] = frB[0] | 0x80000000;
9 frD[1] = frB[1];
10
11#ifdef DEBUG
12 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
13 dump_double(frD);
14 printk("\n");
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/fneg.c b/arch/powerpc/math-emu/fneg.c
new file mode 100644
index 000000000000..fe9a98deff69
--- /dev/null
+++ b/arch/powerpc/math-emu/fneg.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6fneg(u32 *frD, u32 *frB)
7{
8 frD[0] = frB[0] ^ 0x80000000;
9 frD[1] = frB[1];
10
11#ifdef DEBUG
12 printk("%s: D %p, B %p: ", __FUNCTION__, frD, frB);
13 dump_double(frD);
14 printk("\n");
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/fnmadd.c b/arch/powerpc/math-emu/fnmadd.c
new file mode 100644
index 000000000000..7f312276d920
--- /dev/null
+++ b/arch/powerpc/math-emu/fnmadd.c
@@ -0,0 +1,51 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fnmadd(void *frD, void *frA, void *frB, void *frC)
10{
11 FP_DECL_D(R);
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(C);
15 FP_DECL_D(T);
16 int ret = 0;
17
18#ifdef DEBUG
19 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
20#endif
21
22 __FP_UNPACK_D(A, frA);
23 __FP_UNPACK_D(B, frB);
24 __FP_UNPACK_D(C, frC);
25
26#ifdef DEBUG
27 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
28 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
29 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
30#endif
31
32 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
33 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
34 ret |= EFLAG_VXIMZ;
35
36 FP_MUL_D(T, A, C);
37
38 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
39 ret |= EFLAG_VXISI;
40
41 FP_ADD_D(R, T, B);
42
43 if (R_c != FP_CLS_NAN)
44 R_s ^= 1;
45
46#ifdef DEBUG
47 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
48#endif
49
50 return (ret | __FP_PACK_D(frD, R));
51}
diff --git a/arch/powerpc/math-emu/fnmadds.c b/arch/powerpc/math-emu/fnmadds.c
new file mode 100644
index 000000000000..65454c9c70bc
--- /dev/null
+++ b/arch/powerpc/math-emu/fnmadds.c
@@ -0,0 +1,52 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fnmadds(void *frD, void *frA, void *frB, void *frC)
11{
12 FP_DECL_D(R);
13 FP_DECL_D(A);
14 FP_DECL_D(B);
15 FP_DECL_D(C);
16 FP_DECL_D(T);
17 int ret = 0;
18
19#ifdef DEBUG
20 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
21#endif
22
23 __FP_UNPACK_D(A, frA);
24 __FP_UNPACK_D(B, frB);
25 __FP_UNPACK_D(C, frC);
26
27#ifdef DEBUG
28 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
29 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
30 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
31#endif
32
33 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
34 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
35 ret |= EFLAG_VXIMZ;
36
37 FP_MUL_D(T, A, C);
38
39 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
40 ret |= EFLAG_VXISI;
41
42 FP_ADD_D(R, T, B);
43
44 if (R_c != FP_CLS_NAN)
45 R_s ^= 1;
46
47#ifdef DEBUG
48 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
49#endif
50
51 return (ret | __FP_PACK_DS(frD, R));
52}
diff --git a/arch/powerpc/math-emu/fnmsub.c b/arch/powerpc/math-emu/fnmsub.c
new file mode 100644
index 000000000000..f1ca7482b5f0
--- /dev/null
+++ b/arch/powerpc/math-emu/fnmsub.c
@@ -0,0 +1,54 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fnmsub(void *frD, void *frA, void *frB, void *frC)
10{
11 FP_DECL_D(R);
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(C);
15 FP_DECL_D(T);
16 int ret = 0;
17
18#ifdef DEBUG
19 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
20#endif
21
22 __FP_UNPACK_D(A, frA);
23 __FP_UNPACK_D(B, frB);
24 __FP_UNPACK_D(C, frC);
25
26#ifdef DEBUG
27 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
28 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
29 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
30#endif
31
32 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
33 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
34 ret |= EFLAG_VXIMZ;
35
36 FP_MUL_D(T, A, C);
37
38 if (B_c != FP_CLS_NAN)
39 B_s ^= 1;
40
41 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
42 ret |= EFLAG_VXISI;
43
44 FP_ADD_D(R, T, B);
45
46 if (R_c != FP_CLS_NAN)
47 R_s ^= 1;
48
49#ifdef DEBUG
50 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
51#endif
52
53 return (ret | __FP_PACK_D(frD, R));
54}
diff --git a/arch/powerpc/math-emu/fnmsubs.c b/arch/powerpc/math-emu/fnmsubs.c
new file mode 100644
index 000000000000..5c9a09a87dc7
--- /dev/null
+++ b/arch/powerpc/math-emu/fnmsubs.c
@@ -0,0 +1,55 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fnmsubs(void *frD, void *frA, void *frB, void *frC)
11{
12 FP_DECL_D(R);
13 FP_DECL_D(A);
14 FP_DECL_D(B);
15 FP_DECL_D(C);
16 FP_DECL_D(T);
17 int ret = 0;
18
19#ifdef DEBUG
20 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
21#endif
22
23 __FP_UNPACK_D(A, frA);
24 __FP_UNPACK_D(B, frB);
25 __FP_UNPACK_D(C, frC);
26
27#ifdef DEBUG
28 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
29 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
30 printk("C: %ld %lu %lu %ld (%ld)\n", C_s, C_f1, C_f0, C_e, C_c);
31#endif
32
33 if ((A_c == FP_CLS_INF && C_c == FP_CLS_ZERO) ||
34 (A_c == FP_CLS_ZERO && C_c == FP_CLS_INF))
35 ret |= EFLAG_VXIMZ;
36
37 FP_MUL_D(T, A, C);
38
39 if (B_c != FP_CLS_NAN)
40 B_s ^= 1;
41
42 if (T_s != B_s && T_c == FP_CLS_INF && B_c == FP_CLS_INF)
43 ret |= EFLAG_VXISI;
44
45 FP_ADD_D(R, T, B);
46
47 if (R_c != FP_CLS_NAN)
48 R_s ^= 1;
49
50#ifdef DEBUG
51 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
52#endif
53
54 return (ret | __FP_PACK_DS(frD, R));
55}
diff --git a/arch/powerpc/math-emu/fres.c b/arch/powerpc/math-emu/fres.c
new file mode 100644
index 000000000000..ec11e46d20af
--- /dev/null
+++ b/arch/powerpc/math-emu/fres.c
@@ -0,0 +1,12 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6fres(void *frD, void *frB)
7{
8#ifdef DEBUG
9 printk("%s: %p %p\n", __FUNCTION__, frD, frB);
10#endif
11 return -ENOSYS;
12}
diff --git a/arch/powerpc/math-emu/frsp.c b/arch/powerpc/math-emu/frsp.c
new file mode 100644
index 000000000000..d879b2a3d0c9
--- /dev/null
+++ b/arch/powerpc/math-emu/frsp.c
@@ -0,0 +1,25 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10frsp(void *frD, void *frB)
11{
12 FP_DECL_D(B);
13
14#ifdef DEBUG
15 printk("%s: D %p, B %p\n", __FUNCTION__, frD, frB);
16#endif
17
18 __FP_UNPACK_D(B, frB);
19
20#ifdef DEBUG
21 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
22#endif
23
24 return __FP_PACK_DS(frD, B);
25}
diff --git a/arch/powerpc/math-emu/frsqrte.c b/arch/powerpc/math-emu/frsqrte.c
new file mode 100644
index 000000000000..a11ae1829850
--- /dev/null
+++ b/arch/powerpc/math-emu/frsqrte.c
@@ -0,0 +1,12 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6frsqrte(void *frD, void *frB)
7{
8#ifdef DEBUG
9 printk("%s: %p %p\n", __FUNCTION__, frD, frB);
10#endif
11 return 0;
12}
diff --git a/arch/powerpc/math-emu/fsel.c b/arch/powerpc/math-emu/fsel.c
new file mode 100644
index 000000000000..e36e6e72819a
--- /dev/null
+++ b/arch/powerpc/math-emu/fsel.c
@@ -0,0 +1,38 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fsel(u32 *frD, void *frA, u32 *frB, u32 *frC)
10{
11 FP_DECL_D(A);
12
13#ifdef DEBUG
14 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frA, frB, frC);
15#endif
16
17 __FP_UNPACK_D(A, frA);
18
19#ifdef DEBUG
20 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
21 printk("B: %08x %08x\n", frB[0], frB[1]);
22 printk("C: %08x %08x\n", frC[0], frC[1]);
23#endif
24
25 if (A_c == FP_CLS_NAN || (A_c != FP_CLS_ZERO && A_s)) {
26 frD[0] = frB[0];
27 frD[1] = frB[1];
28 } else {
29 frD[0] = frC[0];
30 frD[1] = frC[1];
31 }
32
33#ifdef DEBUG
34 printk("D: %08x.%08x\n", frD[0], frD[1]);
35#endif
36
37 return 0;
38}
diff --git a/arch/powerpc/math-emu/fsqrt.c b/arch/powerpc/math-emu/fsqrt.c
new file mode 100644
index 000000000000..6f8319f64a8a
--- /dev/null
+++ b/arch/powerpc/math-emu/fsqrt.c
@@ -0,0 +1,37 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fsqrt(void *frD, void *frB)
10{
11 FP_DECL_D(B);
12 FP_DECL_D(R);
13 int ret = 0;
14
15#ifdef DEBUG
16 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frB);
17#endif
18
19 __FP_UNPACK_D(B, frB);
20
21#ifdef DEBUG
22 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
23#endif
24
25 if (B_s && B_c != FP_CLS_ZERO)
26 ret |= EFLAG_VXSQRT;
27 if (B_c == FP_CLS_NAN)
28 ret |= EFLAG_VXSNAN;
29
30 FP_SQRT_D(R, B);
31
32#ifdef DEBUG
33 printk("R: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
34#endif
35
36 return (ret | __FP_PACK_D(frD, R));
37}
diff --git a/arch/powerpc/math-emu/fsqrts.c b/arch/powerpc/math-emu/fsqrts.c
new file mode 100644
index 000000000000..3b2b1cf55c12
--- /dev/null
+++ b/arch/powerpc/math-emu/fsqrts.c
@@ -0,0 +1,38 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fsqrts(void *frD, void *frB)
11{
12 FP_DECL_D(B);
13 FP_DECL_D(R);
14 int ret = 0;
15
16#ifdef DEBUG
17 printk("%s: %p %p %p %p\n", __FUNCTION__, frD, frB);
18#endif
19
20 __FP_UNPACK_D(B, frB);
21
22#ifdef DEBUG
23 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
24#endif
25
26 if (B_s && B_c != FP_CLS_ZERO)
27 ret |= EFLAG_VXSQRT;
28 if (B_c == FP_CLS_NAN)
29 ret |= EFLAG_VXSNAN;
30
31 FP_SQRT_D(R, B);
32
33#ifdef DEBUG
34 printk("R: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
35#endif
36
37 return (ret | __FP_PACK_DS(frD, R));
38}
diff --git a/arch/powerpc/math-emu/fsub.c b/arch/powerpc/math-emu/fsub.c
new file mode 100644
index 000000000000..956679042bb2
--- /dev/null
+++ b/arch/powerpc/math-emu/fsub.c
@@ -0,0 +1,41 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7
8int
9fsub(void *frD, void *frA, void *frB)
10{
11 FP_DECL_D(A);
12 FP_DECL_D(B);
13 FP_DECL_D(R);
14 int ret = 0;
15
16#ifdef DEBUG
17 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
18#endif
19
20 __FP_UNPACK_D(A, frA);
21 __FP_UNPACK_D(B, frB);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
25 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
26#endif
27
28 if (B_c != FP_CLS_NAN)
29 B_s ^= 1;
30
31 if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF)
32 ret |= EFLAG_VXISI;
33
34 FP_ADD_D(R, A, B);
35
36#ifdef DEBUG
37 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
38#endif
39
40 return (ret | __FP_PACK_D(frD, R));
41}
diff --git a/arch/powerpc/math-emu/fsubs.c b/arch/powerpc/math-emu/fsubs.c
new file mode 100644
index 000000000000..3428117dfe8c
--- /dev/null
+++ b/arch/powerpc/math-emu/fsubs.c
@@ -0,0 +1,42 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10fsubs(void *frD, void *frA, void *frB)
11{
12 FP_DECL_D(A);
13 FP_DECL_D(B);
14 FP_DECL_D(R);
15 int ret = 0;
16
17#ifdef DEBUG
18 printk("%s: %p %p %p\n", __FUNCTION__, frD, frA, frB);
19#endif
20
21 __FP_UNPACK_D(A, frA);
22 __FP_UNPACK_D(B, frB);
23
24#ifdef DEBUG
25 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
26 printk("B: %ld %lu %lu %ld (%ld)\n", B_s, B_f1, B_f0, B_e, B_c);
27#endif
28
29 if (B_c != FP_CLS_NAN)
30 B_s ^= 1;
31
32 if (A_s != B_s && A_c == FP_CLS_INF && B_c == FP_CLS_INF)
33 ret |= EFLAG_VXISI;
34
35 FP_ADD_D(R, A, B);
36
37#ifdef DEBUG
38 printk("D: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
39#endif
40
41 return (ret | __FP_PACK_DS(frD, R));
42}
diff --git a/arch/powerpc/math-emu/lfd.c b/arch/powerpc/math-emu/lfd.c
new file mode 100644
index 000000000000..7d38101c329b
--- /dev/null
+++ b/arch/powerpc/math-emu/lfd.c
@@ -0,0 +1,19 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "sfp-machine.h"
6#include "double.h"
7
8int
9lfd(void *frD, void *ea)
10{
11 if (copy_from_user(frD, ea, sizeof(double)))
12 return -EFAULT;
13#ifdef DEBUG
14 printk("%s: D %p, ea %p: ", __FUNCTION__, frD, ea);
15 dump_double(frD);
16 printk("\n");
17#endif
18 return 0;
19}
diff --git a/arch/powerpc/math-emu/lfs.c b/arch/powerpc/math-emu/lfs.c
new file mode 100644
index 000000000000..c86dee3d7655
--- /dev/null
+++ b/arch/powerpc/math-emu/lfs.c
@@ -0,0 +1,37 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10lfs(void *frD, void *ea)
11{
12 FP_DECL_D(R);
13 FP_DECL_S(A);
14 float f;
15
16#ifdef DEBUG
17 printk("%s: D %p, ea %p\n", __FUNCTION__, frD, ea);
18#endif
19
20 if (copy_from_user(&f, ea, sizeof(float)))
21 return -EFAULT;
22
23 __FP_UNPACK_S(A, &f);
24
25#ifdef DEBUG
26 printk("A: %ld %lu %ld (%ld) [%08lx]\n", A_s, A_f, A_e, A_c,
27 *(unsigned long *)&f);
28#endif
29
30 FP_CONV(D, S, 2, 1, R, A);
31
32#ifdef DEBUG
33 printk("R: %ld %lu %lu %ld (%ld)\n", R_s, R_f1, R_f0, R_e, R_c);
34#endif
35
36 return __FP_PACK_D(frD, R);
37}
diff --git a/arch/powerpc/math-emu/math.c b/arch/powerpc/math-emu/math.c
new file mode 100644
index 000000000000..589153472761
--- /dev/null
+++ b/arch/powerpc/math-emu/math.c
@@ -0,0 +1,483 @@
1/*
2 * Copyright (C) 1999 Eddie C. Dost (ecd@atecom.com)
3 */
4
5#include <linux/config.h>
6#include <linux/types.h>
7#include <linux/sched.h>
8
9#include <asm/uaccess.h>
10#include <asm/reg.h>
11
12#include "sfp-machine.h"
13#include "double.h"
14
15#define FLOATFUNC(x) extern int x(void *, void *, void *, void *)
16
17FLOATFUNC(fadd);
18FLOATFUNC(fadds);
19FLOATFUNC(fdiv);
20FLOATFUNC(fdivs);
21FLOATFUNC(fmul);
22FLOATFUNC(fmuls);
23FLOATFUNC(fsub);
24FLOATFUNC(fsubs);
25
26FLOATFUNC(fmadd);
27FLOATFUNC(fmadds);
28FLOATFUNC(fmsub);
29FLOATFUNC(fmsubs);
30FLOATFUNC(fnmadd);
31FLOATFUNC(fnmadds);
32FLOATFUNC(fnmsub);
33FLOATFUNC(fnmsubs);
34
35FLOATFUNC(fctiw);
36FLOATFUNC(fctiwz);
37FLOATFUNC(frsp);
38
39FLOATFUNC(fcmpo);
40FLOATFUNC(fcmpu);
41
42FLOATFUNC(mcrfs);
43FLOATFUNC(mffs);
44FLOATFUNC(mtfsb0);
45FLOATFUNC(mtfsb1);
46FLOATFUNC(mtfsf);
47FLOATFUNC(mtfsfi);
48
49FLOATFUNC(lfd);
50FLOATFUNC(lfs);
51
52FLOATFUNC(stfd);
53FLOATFUNC(stfs);
54FLOATFUNC(stfiwx);
55
56FLOATFUNC(fabs);
57FLOATFUNC(fmr);
58FLOATFUNC(fnabs);
59FLOATFUNC(fneg);
60
61/* Optional */
62FLOATFUNC(fres);
63FLOATFUNC(frsqrte);
64FLOATFUNC(fsel);
65FLOATFUNC(fsqrt);
66FLOATFUNC(fsqrts);
67
68
69#define OP31 0x1f /* 31 */
70#define LFS 0x30 /* 48 */
71#define LFSU 0x31 /* 49 */
72#define LFD 0x32 /* 50 */
73#define LFDU 0x33 /* 51 */
74#define STFS 0x34 /* 52 */
75#define STFSU 0x35 /* 53 */
76#define STFD 0x36 /* 54 */
77#define STFDU 0x37 /* 55 */
78#define OP59 0x3b /* 59 */
79#define OP63 0x3f /* 63 */
80
81/* Opcode 31: */
82/* X-Form: */
83#define LFSX 0x217 /* 535 */
84#define LFSUX 0x237 /* 567 */
85#define LFDX 0x257 /* 599 */
86#define LFDUX 0x277 /* 631 */
87#define STFSX 0x297 /* 663 */
88#define STFSUX 0x2b7 /* 695 */
89#define STFDX 0x2d7 /* 727 */
90#define STFDUX 0x2f7 /* 759 */
91#define STFIWX 0x3d7 /* 983 */
92
93/* Opcode 59: */
94/* A-Form: */
95#define FDIVS 0x012 /* 18 */
96#define FSUBS 0x014 /* 20 */
97#define FADDS 0x015 /* 21 */
98#define FSQRTS 0x016 /* 22 */
99#define FRES 0x018 /* 24 */
100#define FMULS 0x019 /* 25 */
101#define FMSUBS 0x01c /* 28 */
102#define FMADDS 0x01d /* 29 */
103#define FNMSUBS 0x01e /* 30 */
104#define FNMADDS 0x01f /* 31 */
105
106/* Opcode 63: */
107/* A-Form: */
108#define FDIV 0x012 /* 18 */
109#define FSUB 0x014 /* 20 */
110#define FADD 0x015 /* 21 */
111#define FSQRT 0x016 /* 22 */
112#define FSEL 0x017 /* 23 */
113#define FMUL 0x019 /* 25 */
114#define FRSQRTE 0x01a /* 26 */
115#define FMSUB 0x01c /* 28 */
116#define FMADD 0x01d /* 29 */
117#define FNMSUB 0x01e /* 30 */
118#define FNMADD 0x01f /* 31 */
119
120/* X-Form: */
121#define FCMPU 0x000 /* 0 */
122#define FRSP 0x00c /* 12 */
123#define FCTIW 0x00e /* 14 */
124#define FCTIWZ 0x00f /* 15 */
125#define FCMPO 0x020 /* 32 */
126#define MTFSB1 0x026 /* 38 */
127#define FNEG 0x028 /* 40 */
128#define MCRFS 0x040 /* 64 */
129#define MTFSB0 0x046 /* 70 */
130#define FMR 0x048 /* 72 */
131#define MTFSFI 0x086 /* 134 */
132#define FNABS 0x088 /* 136 */
133#define FABS 0x108 /* 264 */
134#define MFFS 0x247 /* 583 */
135#define MTFSF 0x2c7 /* 711 */
136
137
138#define AB 2
139#define AC 3
140#define ABC 4
141#define D 5
142#define DU 6
143#define X 7
144#define XA 8
145#define XB 9
146#define XCR 11
147#define XCRB 12
148#define XCRI 13
149#define XCRL 16
150#define XE 14
151#define XEU 15
152#define XFLB 10
153
154#ifdef CONFIG_MATH_EMULATION
155static int
156record_exception(struct pt_regs *regs, int eflag)
157{
158 u32 fpscr;
159
160 fpscr = __FPU_FPSCR;
161
162 if (eflag) {
163 fpscr |= FPSCR_FX;
164 if (eflag & EFLAG_OVERFLOW)
165 fpscr |= FPSCR_OX;
166 if (eflag & EFLAG_UNDERFLOW)
167 fpscr |= FPSCR_UX;
168 if (eflag & EFLAG_DIVZERO)
169 fpscr |= FPSCR_ZX;
170 if (eflag & EFLAG_INEXACT)
171 fpscr |= FPSCR_XX;
172 if (eflag & EFLAG_VXSNAN)
173 fpscr |= FPSCR_VXSNAN;
174 if (eflag & EFLAG_VXISI)
175 fpscr |= FPSCR_VXISI;
176 if (eflag & EFLAG_VXIDI)
177 fpscr |= FPSCR_VXIDI;
178 if (eflag & EFLAG_VXZDZ)
179 fpscr |= FPSCR_VXZDZ;
180 if (eflag & EFLAG_VXIMZ)
181 fpscr |= FPSCR_VXIMZ;
182 if (eflag & EFLAG_VXVC)
183 fpscr |= FPSCR_VXVC;
184 if (eflag & EFLAG_VXSOFT)
185 fpscr |= FPSCR_VXSOFT;
186 if (eflag & EFLAG_VXSQRT)
187 fpscr |= FPSCR_VXSQRT;
188 if (eflag & EFLAG_VXCVI)
189 fpscr |= FPSCR_VXCVI;
190 }
191
192 fpscr &= ~(FPSCR_VX);
193 if (fpscr & (FPSCR_VXSNAN | FPSCR_VXISI | FPSCR_VXIDI |
194 FPSCR_VXZDZ | FPSCR_VXIMZ | FPSCR_VXVC |
195 FPSCR_VXSOFT | FPSCR_VXSQRT | FPSCR_VXCVI))
196 fpscr |= FPSCR_VX;
197
198 fpscr &= ~(FPSCR_FEX);
199 if (((fpscr & FPSCR_VX) && (fpscr & FPSCR_VE)) ||
200 ((fpscr & FPSCR_OX) && (fpscr & FPSCR_OE)) ||
201 ((fpscr & FPSCR_UX) && (fpscr & FPSCR_UE)) ||
202 ((fpscr & FPSCR_ZX) && (fpscr & FPSCR_ZE)) ||
203 ((fpscr & FPSCR_XX) && (fpscr & FPSCR_XE)))
204 fpscr |= FPSCR_FEX;
205
206 __FPU_FPSCR = fpscr;
207
208 return (fpscr & FPSCR_FEX) ? 1 : 0;
209}
210#endif /* CONFIG_MATH_EMULATION */
211
212int
213do_mathemu(struct pt_regs *regs)
214{
215 void *op0 = 0, *op1 = 0, *op2 = 0, *op3 = 0;
216 unsigned long pc = regs->nip;
217 signed short sdisp;
218 u32 insn = 0;
219 int idx = 0;
220#ifdef CONFIG_MATH_EMULATION
221 int (*func)(void *, void *, void *, void *);
222 int type = 0;
223 int eflag, trap;
224#endif
225
226 if (get_user(insn, (u32 *)pc))
227 return -EFAULT;
228
229#ifndef CONFIG_MATH_EMULATION
230 switch (insn >> 26) {
231 case LFD:
232 idx = (insn >> 16) & 0x1f;
233 sdisp = (insn & 0xffff);
234 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
235 op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
236 lfd(op0, op1, op2, op3);
237 break;
238 case LFDU:
239 idx = (insn >> 16) & 0x1f;
240 sdisp = (insn & 0xffff);
241 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
242 op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
243 lfd(op0, op1, op2, op3);
244 regs->gpr[idx] = (unsigned long)op1;
245 break;
246 case STFD:
247 idx = (insn >> 16) & 0x1f;
248 sdisp = (insn & 0xffff);
249 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
250 op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
251 stfd(op0, op1, op2, op3);
252 break;
253 case STFDU:
254 idx = (insn >> 16) & 0x1f;
255 sdisp = (insn & 0xffff);
256 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
257 op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
258 stfd(op0, op1, op2, op3);
259 regs->gpr[idx] = (unsigned long)op1;
260 break;
261 case OP63:
262 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
263 op1 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
264 fmr(op0, op1, op2, op3);
265 break;
266 default:
267 goto illegal;
268 }
269#else /* CONFIG_MATH_EMULATION */
270 switch (insn >> 26) {
271 case LFS: func = lfs; type = D; break;
272 case LFSU: func = lfs; type = DU; break;
273 case LFD: func = lfd; type = D; break;
274 case LFDU: func = lfd; type = DU; break;
275 case STFS: func = stfs; type = D; break;
276 case STFSU: func = stfs; type = DU; break;
277 case STFD: func = stfd; type = D; break;
278 case STFDU: func = stfd; type = DU; break;
279
280 case OP31:
281 switch ((insn >> 1) & 0x3ff) {
282 case LFSX: func = lfs; type = XE; break;
283 case LFSUX: func = lfs; type = XEU; break;
284 case LFDX: func = lfd; type = XE; break;
285 case LFDUX: func = lfd; type = XEU; break;
286 case STFSX: func = stfs; type = XE; break;
287 case STFSUX: func = stfs; type = XEU; break;
288 case STFDX: func = stfd; type = XE; break;
289 case STFDUX: func = stfd; type = XEU; break;
290 case STFIWX: func = stfiwx; type = XE; break;
291 default:
292 goto illegal;
293 }
294 break;
295
296 case OP59:
297 switch ((insn >> 1) & 0x1f) {
298 case FDIVS: func = fdivs; type = AB; break;
299 case FSUBS: func = fsubs; type = AB; break;
300 case FADDS: func = fadds; type = AB; break;
301 case FSQRTS: func = fsqrts; type = AB; break;
302 case FRES: func = fres; type = AB; break;
303 case FMULS: func = fmuls; type = AC; break;
304 case FMSUBS: func = fmsubs; type = ABC; break;
305 case FMADDS: func = fmadds; type = ABC; break;
306 case FNMSUBS: func = fnmsubs; type = ABC; break;
307 case FNMADDS: func = fnmadds; type = ABC; break;
308 default:
309 goto illegal;
310 }
311 break;
312
313 case OP63:
314 if (insn & 0x20) {
315 switch ((insn >> 1) & 0x1f) {
316 case FDIV: func = fdiv; type = AB; break;
317 case FSUB: func = fsub; type = AB; break;
318 case FADD: func = fadd; type = AB; break;
319 case FSQRT: func = fsqrt; type = AB; break;
320 case FSEL: func = fsel; type = ABC; break;
321 case FMUL: func = fmul; type = AC; break;
322 case FRSQRTE: func = frsqrte; type = AB; break;
323 case FMSUB: func = fmsub; type = ABC; break;
324 case FMADD: func = fmadd; type = ABC; break;
325 case FNMSUB: func = fnmsub; type = ABC; break;
326 case FNMADD: func = fnmadd; type = ABC; break;
327 default:
328 goto illegal;
329 }
330 break;
331 }
332
333 switch ((insn >> 1) & 0x3ff) {
334 case FCMPU: func = fcmpu; type = XCR; break;
335 case FRSP: func = frsp; type = XB; break;
336 case FCTIW: func = fctiw; type = XB; break;
337 case FCTIWZ: func = fctiwz; type = XB; break;
338 case FCMPO: func = fcmpo; type = XCR; break;
339 case MTFSB1: func = mtfsb1; type = XCRB; break;
340 case FNEG: func = fneg; type = XB; break;
341 case MCRFS: func = mcrfs; type = XCRL; break;
342 case MTFSB0: func = mtfsb0; type = XCRB; break;
343 case FMR: func = fmr; type = XB; break;
344 case MTFSFI: func = mtfsfi; type = XCRI; break;
345 case FNABS: func = fnabs; type = XB; break;
346 case FABS: func = fabs; type = XB; break;
347 case MFFS: func = mffs; type = X; break;
348 case MTFSF: func = mtfsf; type = XFLB; break;
349 default:
350 goto illegal;
351 }
352 break;
353
354 default:
355 goto illegal;
356 }
357
358 switch (type) {
359 case AB:
360 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
361 op1 = (void *)&current->thread.fpr[(insn >> 16) & 0x1f];
362 op2 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
363 break;
364
365 case AC:
366 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
367 op1 = (void *)&current->thread.fpr[(insn >> 16) & 0x1f];
368 op2 = (void *)&current->thread.fpr[(insn >> 6) & 0x1f];
369 break;
370
371 case ABC:
372 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
373 op1 = (void *)&current->thread.fpr[(insn >> 16) & 0x1f];
374 op2 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
375 op3 = (void *)&current->thread.fpr[(insn >> 6) & 0x1f];
376 break;
377
378 case D:
379 idx = (insn >> 16) & 0x1f;
380 sdisp = (insn & 0xffff);
381 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
382 op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
383 break;
384
385 case DU:
386 idx = (insn >> 16) & 0x1f;
387 if (!idx)
388 goto illegal;
389
390 sdisp = (insn & 0xffff);
391 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
392 op1 = (void *)(regs->gpr[idx] + sdisp);
393 break;
394
395 case X:
396 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
397 break;
398
399 case XA:
400 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
401 op1 = (void *)&current->thread.fpr[(insn >> 16) & 0x1f];
402 break;
403
404 case XB:
405 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
406 op1 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
407 break;
408
409 case XE:
410 idx = (insn >> 16) & 0x1f;
411 if (!idx)
412 goto illegal;
413
414 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
415 op1 = (void *)(regs->gpr[idx] + regs->gpr[(insn >> 11) & 0x1f]);
416 break;
417
418 case XEU:
419 idx = (insn >> 16) & 0x1f;
420 op0 = (void *)&current->thread.fpr[(insn >> 21) & 0x1f];
421 op1 = (void *)((idx ? regs->gpr[idx] : 0)
422 + regs->gpr[(insn >> 11) & 0x1f]);
423 break;
424
425 case XCR:
426 op0 = (void *)&regs->ccr;
427 op1 = (void *)((insn >> 23) & 0x7);
428 op2 = (void *)&current->thread.fpr[(insn >> 16) & 0x1f];
429 op3 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
430 break;
431
432 case XCRL:
433 op0 = (void *)&regs->ccr;
434 op1 = (void *)((insn >> 23) & 0x7);
435 op2 = (void *)((insn >> 18) & 0x7);
436 break;
437
438 case XCRB:
439 op0 = (void *)((insn >> 21) & 0x1f);
440 break;
441
442 case XCRI:
443 op0 = (void *)((insn >> 23) & 0x7);
444 op1 = (void *)((insn >> 12) & 0xf);
445 break;
446
447 case XFLB:
448 op0 = (void *)((insn >> 17) & 0xff);
449 op1 = (void *)&current->thread.fpr[(insn >> 11) & 0x1f];
450 break;
451
452 default:
453 goto illegal;
454 }
455
456 eflag = func(op0, op1, op2, op3);
457
458 if (insn & 1) {
459 regs->ccr &= ~(0x0f000000);
460 regs->ccr |= (__FPU_FPSCR >> 4) & 0x0f000000;
461 }
462
463 trap = record_exception(regs, eflag);
464 if (trap)
465 return 1;
466
467 switch (type) {
468 case DU:
469 case XEU:
470 regs->gpr[idx] = (unsigned long)op1;
471 break;
472
473 default:
474 break;
475 }
476#endif /* CONFIG_MATH_EMULATION */
477
478 regs->nip += 4;
479 return 0;
480
481illegal:
482 return -ENOSYS;
483}
diff --git a/arch/powerpc/math-emu/mcrfs.c b/arch/powerpc/math-emu/mcrfs.c
new file mode 100644
index 000000000000..106dd912914b
--- /dev/null
+++ b/arch/powerpc/math-emu/mcrfs.c
@@ -0,0 +1,31 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mcrfs(u32 *ccr, u32 crfD, u32 crfS)
9{
10 u32 value, clear;
11
12#ifdef DEBUG
13 printk("%s: %p (%08x) %d %d\n", __FUNCTION__, ccr, *ccr, crfD, crfS);
14#endif
15
16 clear = 15 << ((7 - crfS) << 2);
17 if (!crfS)
18 clear = 0x90000000;
19
20 value = (__FPU_FPSCR >> ((7 - crfS) << 2)) & 15;
21 __FPU_FPSCR &= ~(clear);
22
23 *ccr &= ~(15 << ((7 - crfD) << 2));
24 *ccr |= (value << ((7 - crfD) << 2));
25
26#ifdef DEBUG
27 printk("CR: %08x\n", __FUNCTION__, *ccr);
28#endif
29
30 return 0;
31}
diff --git a/arch/powerpc/math-emu/mffs.c b/arch/powerpc/math-emu/mffs.c
new file mode 100644
index 000000000000..f477c9170e75
--- /dev/null
+++ b/arch/powerpc/math-emu/mffs.c
@@ -0,0 +1,17 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mffs(u32 *frD)
9{
10 frD[1] = __FPU_FPSCR;
11
12#ifdef DEBUG
13 printk("%s: frD %p: %08x.%08x\n", __FUNCTION__, frD, frD[0], frD[1]);
14#endif
15
16 return 0;
17}
diff --git a/arch/powerpc/math-emu/mtfsb0.c b/arch/powerpc/math-emu/mtfsb0.c
new file mode 100644
index 000000000000..99bfd80f4af3
--- /dev/null
+++ b/arch/powerpc/math-emu/mtfsb0.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mtfsb0(int crbD)
9{
10 if ((crbD != 1) && (crbD != 2))
11 __FPU_FPSCR &= ~(1 << (31 - crbD));
12
13#ifdef DEBUG
14 printk("%s: %d %08lx\n", __FUNCTION__, crbD, __FPU_FPSCR);
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/mtfsb1.c b/arch/powerpc/math-emu/mtfsb1.c
new file mode 100644
index 000000000000..3d9e7ed92d2b
--- /dev/null
+++ b/arch/powerpc/math-emu/mtfsb1.c
@@ -0,0 +1,18 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mtfsb1(int crbD)
9{
10 if ((crbD != 1) && (crbD != 2))
11 __FPU_FPSCR |= (1 << (31 - crbD));
12
13#ifdef DEBUG
14 printk("%s: %d %08lx\n", __FUNCTION__, crbD, __FPU_FPSCR);
15#endif
16
17 return 0;
18}
diff --git a/arch/powerpc/math-emu/mtfsf.c b/arch/powerpc/math-emu/mtfsf.c
new file mode 100644
index 000000000000..d70cf714994c
--- /dev/null
+++ b/arch/powerpc/math-emu/mtfsf.c
@@ -0,0 +1,45 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mtfsf(unsigned int FM, u32 *frB)
9{
10 u32 mask;
11
12 if (FM == 0)
13 return 0;
14
15 if (FM == 0xff)
16 mask = 0x9fffffff;
17 else {
18 mask = 0;
19 if (FM & (1 << 0))
20 mask |= 0x90000000;
21 if (FM & (1 << 1))
22 mask |= 0x0f000000;
23 if (FM & (1 << 2))
24 mask |= 0x00f00000;
25 if (FM & (1 << 3))
26 mask |= 0x000f0000;
27 if (FM & (1 << 4))
28 mask |= 0x0000f000;
29 if (FM & (1 << 5))
30 mask |= 0x00000f00;
31 if (FM & (1 << 6))
32 mask |= 0x000000f0;
33 if (FM & (1 << 7))
34 mask |= 0x0000000f;
35 }
36
37 __FPU_FPSCR &= ~(mask);
38 __FPU_FPSCR |= (frB[1] & mask);
39
40#ifdef DEBUG
41 printk("%s: %02x %p: %08lx\n", __FUNCTION__, FM, frB, __FPU_FPSCR);
42#endif
43
44 return 0;
45}
diff --git a/arch/powerpc/math-emu/mtfsfi.c b/arch/powerpc/math-emu/mtfsfi.c
new file mode 100644
index 000000000000..71df854baa7e
--- /dev/null
+++ b/arch/powerpc/math-emu/mtfsfi.c
@@ -0,0 +1,23 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6
7int
8mtfsfi(unsigned int crfD, unsigned int IMM)
9{
10 u32 mask = 0xf;
11
12 if (!crfD)
13 mask = 9;
14
15 __FPU_FPSCR &= ~(mask << ((7 - crfD) << 2));
16 __FPU_FPSCR |= (IMM & 0xf) << ((7 - crfD) << 2);
17
18#ifdef DEBUG
19 printk("%s: %d %x: %08lx\n", __FUNCTION__, crfD, IMM, __FPU_FPSCR);
20#endif
21
22 return 0;
23}
diff --git a/arch/powerpc/math-emu/op-1.h b/arch/powerpc/math-emu/op-1.h
new file mode 100644
index 000000000000..c92fa95f562e
--- /dev/null
+++ b/arch/powerpc/math-emu/op-1.h
@@ -0,0 +1,245 @@
1/*
2 * Basic one-word fraction declaration and manipulation.
3 */
4
5#define _FP_FRAC_DECL_1(X) _FP_W_TYPE X##_f
6#define _FP_FRAC_COPY_1(D,S) (D##_f = S##_f)
7#define _FP_FRAC_SET_1(X,I) (X##_f = I)
8#define _FP_FRAC_HIGH_1(X) (X##_f)
9#define _FP_FRAC_LOW_1(X) (X##_f)
10#define _FP_FRAC_WORD_1(X,w) (X##_f)
11
12#define _FP_FRAC_ADDI_1(X,I) (X##_f += I)
13#define _FP_FRAC_SLL_1(X,N) \
14 do { \
15 if (__builtin_constant_p(N) && (N) == 1) \
16 X##_f += X##_f; \
17 else \
18 X##_f <<= (N); \
19 } while (0)
20#define _FP_FRAC_SRL_1(X,N) (X##_f >>= N)
21
22/* Right shift with sticky-lsb. */
23#define _FP_FRAC_SRS_1(X,N,sz) __FP_FRAC_SRS_1(X##_f, N, sz)
24
25#define __FP_FRAC_SRS_1(X,N,sz) \
26 (X = (X >> (N) | (__builtin_constant_p(N) && (N) == 1 \
27 ? X & 1 : (X << (_FP_W_TYPE_SIZE - (N))) != 0)))
28
29#define _FP_FRAC_ADD_1(R,X,Y) (R##_f = X##_f + Y##_f)
30#define _FP_FRAC_SUB_1(R,X,Y) (R##_f = X##_f - Y##_f)
31#define _FP_FRAC_CLZ_1(z, X) __FP_CLZ(z, X##_f)
32
33/* Predicates */
34#define _FP_FRAC_NEGP_1(X) ((_FP_WS_TYPE)X##_f < 0)
35#define _FP_FRAC_ZEROP_1(X) (X##_f == 0)
36#define _FP_FRAC_OVERP_1(fs,X) (X##_f & _FP_OVERFLOW_##fs)
37#define _FP_FRAC_EQ_1(X, Y) (X##_f == Y##_f)
38#define _FP_FRAC_GE_1(X, Y) (X##_f >= Y##_f)
39#define _FP_FRAC_GT_1(X, Y) (X##_f > Y##_f)
40
41#define _FP_ZEROFRAC_1 0
42#define _FP_MINFRAC_1 1
43
44/*
45 * Unpack the raw bits of a native fp value. Do not classify or
46 * normalize the data.
47 */
48
49#define _FP_UNPACK_RAW_1(fs, X, val) \
50 do { \
51 union _FP_UNION_##fs _flo; _flo.flt = (val); \
52 \
53 X##_f = _flo.bits.frac; \
54 X##_e = _flo.bits.exp; \
55 X##_s = _flo.bits.sign; \
56 } while (0)
57
58
59/*
60 * Repack the raw bits of a native fp value.
61 */
62
63#define _FP_PACK_RAW_1(fs, val, X) \
64 do { \
65 union _FP_UNION_##fs _flo; \
66 \
67 _flo.bits.frac = X##_f; \
68 _flo.bits.exp = X##_e; \
69 _flo.bits.sign = X##_s; \
70 \
71 (val) = _flo.flt; \
72 } while (0)
73
74
75/*
76 * Multiplication algorithms:
77 */
78
79/* Basic. Assuming the host word size is >= 2*FRACBITS, we can do the
80 multiplication immediately. */
81
82#define _FP_MUL_MEAT_1_imm(fs, R, X, Y) \
83 do { \
84 R##_f = X##_f * Y##_f; \
85 /* Normalize since we know where the msb of the multiplicands \
86 were (bit B), we know that the msb of the of the product is \
87 at either 2B or 2B-1. */ \
88 _FP_FRAC_SRS_1(R, _FP_WFRACBITS_##fs-1, 2*_FP_WFRACBITS_##fs); \
89 } while (0)
90
91/* Given a 1W * 1W => 2W primitive, do the extended multiplication. */
92
93#define _FP_MUL_MEAT_1_wide(fs, R, X, Y, doit) \
94 do { \
95 _FP_W_TYPE _Z_f0, _Z_f1; \
96 doit(_Z_f1, _Z_f0, X##_f, Y##_f); \
97 /* Normalize since we know where the msb of the multiplicands \
98 were (bit B), we know that the msb of the of the product is \
99 at either 2B or 2B-1. */ \
100 _FP_FRAC_SRS_2(_Z, _FP_WFRACBITS_##fs-1, 2*_FP_WFRACBITS_##fs); \
101 R##_f = _Z_f0; \
102 } while (0)
103
104/* Finally, a simple widening multiply algorithm. What fun! */
105
106#define _FP_MUL_MEAT_1_hard(fs, R, X, Y) \
107 do { \
108 _FP_W_TYPE _xh, _xl, _yh, _yl, _z_f0, _z_f1, _a_f0, _a_f1; \
109 \
110 /* split the words in half */ \
111 _xh = X##_f >> (_FP_W_TYPE_SIZE/2); \
112 _xl = X##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1); \
113 _yh = Y##_f >> (_FP_W_TYPE_SIZE/2); \
114 _yl = Y##_f & (((_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2)) - 1); \
115 \
116 /* multiply the pieces */ \
117 _z_f0 = _xl * _yl; \
118 _a_f0 = _xh * _yl; \
119 _a_f1 = _xl * _yh; \
120 _z_f1 = _xh * _yh; \
121 \
122 /* reassemble into two full words */ \
123 if ((_a_f0 += _a_f1) < _a_f1) \
124 _z_f1 += (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE/2); \
125 _a_f1 = _a_f0 >> (_FP_W_TYPE_SIZE/2); \
126 _a_f0 = _a_f0 << (_FP_W_TYPE_SIZE/2); \
127 _FP_FRAC_ADD_2(_z, _z, _a); \
128 \
129 /* normalize */ \
130 _FP_FRAC_SRS_2(_z, _FP_WFRACBITS_##fs - 1, 2*_FP_WFRACBITS_##fs); \
131 R##_f = _z_f0; \
132 } while (0)
133
134
135/*
136 * Division algorithms:
137 */
138
139/* Basic. Assuming the host word size is >= 2*FRACBITS, we can do the
140 division immediately. Give this macro either _FP_DIV_HELP_imm for
141 C primitives or _FP_DIV_HELP_ldiv for the ISO function. Which you
142 choose will depend on what the compiler does with divrem4. */
143
144#define _FP_DIV_MEAT_1_imm(fs, R, X, Y, doit) \
145 do { \
146 _FP_W_TYPE _q, _r; \
147 X##_f <<= (X##_f < Y##_f \
148 ? R##_e--, _FP_WFRACBITS_##fs \
149 : _FP_WFRACBITS_##fs - 1); \
150 doit(_q, _r, X##_f, Y##_f); \
151 R##_f = _q | (_r != 0); \
152 } while (0)
153
154/* GCC's longlong.h defines a 2W / 1W => (1W,1W) primitive udiv_qrnnd
155 that may be useful in this situation. This first is for a primitive
156 that requires normalization, the second for one that does not. Look
157 for UDIV_NEEDS_NORMALIZATION to tell which your machine needs. */
158
159#define _FP_DIV_MEAT_1_udiv_norm(fs, R, X, Y) \
160 do { \
161 _FP_W_TYPE _nh, _nl, _q, _r; \
162 \
163 /* Normalize Y -- i.e. make the most significant bit set. */ \
164 Y##_f <<= _FP_WFRACXBITS_##fs - 1; \
165 \
166 /* Shift X op correspondingly high, that is, up one full word. */ \
167 if (X##_f <= Y##_f) \
168 { \
169 _nl = 0; \
170 _nh = X##_f; \
171 } \
172 else \
173 { \
174 R##_e++; \
175 _nl = X##_f << (_FP_W_TYPE_SIZE-1); \
176 _nh = X##_f >> 1; \
177 } \
178 \
179 udiv_qrnnd(_q, _r, _nh, _nl, Y##_f); \
180 R##_f = _q | (_r != 0); \
181 } while (0)
182
183#define _FP_DIV_MEAT_1_udiv(fs, R, X, Y) \
184 do { \
185 _FP_W_TYPE _nh, _nl, _q, _r; \
186 if (X##_f < Y##_f) \
187 { \
188 R##_e--; \
189 _nl = X##_f << _FP_WFRACBITS_##fs; \
190 _nh = X##_f >> _FP_WFRACXBITS_##fs; \
191 } \
192 else \
193 { \
194 _nl = X##_f << (_FP_WFRACBITS_##fs - 1); \
195 _nh = X##_f >> (_FP_WFRACXBITS_##fs + 1); \
196 } \
197 udiv_qrnnd(_q, _r, _nh, _nl, Y##_f); \
198 R##_f = _q | (_r != 0); \
199 } while (0)
200
201
202/*
203 * Square root algorithms:
204 * We have just one right now, maybe Newton approximation
205 * should be added for those machines where division is fast.
206 */
207
208#define _FP_SQRT_MEAT_1(R, S, T, X, q) \
209 do { \
210 while (q) \
211 { \
212 T##_f = S##_f + q; \
213 if (T##_f <= X##_f) \
214 { \
215 S##_f = T##_f + q; \
216 X##_f -= T##_f; \
217 R##_f += q; \
218 } \
219 _FP_FRAC_SLL_1(X, 1); \
220 q >>= 1; \
221 } \
222 } while (0)
223
224/*
225 * Assembly/disassembly for converting to/from integral types.
226 * No shifting or overflow handled here.
227 */
228
229#define _FP_FRAC_ASSEMBLE_1(r, X, rsize) (r = X##_f)
230#define _FP_FRAC_DISASSEMBLE_1(X, r, rsize) (X##_f = r)
231
232
233/*
234 * Convert FP values between word sizes
235 */
236
237#define _FP_FRAC_CONV_1_1(dfs, sfs, D, S) \
238 do { \
239 D##_f = S##_f; \
240 if (_FP_WFRACBITS_##sfs > _FP_WFRACBITS_##dfs) \
241 _FP_FRAC_SRS_1(D, (_FP_WFRACBITS_##sfs-_FP_WFRACBITS_##dfs), \
242 _FP_WFRACBITS_##sfs); \
243 else \
244 D##_f <<= _FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs; \
245 } while (0)
diff --git a/arch/powerpc/math-emu/op-2.h b/arch/powerpc/math-emu/op-2.h
new file mode 100644
index 000000000000..b9b06b4c6ea1
--- /dev/null
+++ b/arch/powerpc/math-emu/op-2.h
@@ -0,0 +1,433 @@
1/*
2 * Basic two-word fraction declaration and manipulation.
3 */
4
5#define _FP_FRAC_DECL_2(X) _FP_W_TYPE X##_f0, X##_f1
6#define _FP_FRAC_COPY_2(D,S) (D##_f0 = S##_f0, D##_f1 = S##_f1)
7#define _FP_FRAC_SET_2(X,I) __FP_FRAC_SET_2(X, I)
8#define _FP_FRAC_HIGH_2(X) (X##_f1)
9#define _FP_FRAC_LOW_2(X) (X##_f0)
10#define _FP_FRAC_WORD_2(X,w) (X##_f##w)
11
12#define _FP_FRAC_SLL_2(X,N) \
13 do { \
14 if ((N) < _FP_W_TYPE_SIZE) \
15 { \
16 if (__builtin_constant_p(N) && (N) == 1) \
17 { \
18 X##_f1 = X##_f1 + X##_f1 + (((_FP_WS_TYPE)(X##_f0)) < 0); \
19 X##_f0 += X##_f0; \
20 } \
21 else \
22 { \
23 X##_f1 = X##_f1 << (N) | X##_f0 >> (_FP_W_TYPE_SIZE - (N)); \
24 X##_f0 <<= (N); \
25 } \
26 } \
27 else \
28 { \
29 X##_f1 = X##_f0 << ((N) - _FP_W_TYPE_SIZE); \
30 X##_f0 = 0; \
31 } \
32 } while (0)
33
34#define _FP_FRAC_SRL_2(X,N) \
35 do { \
36 if ((N) < _FP_W_TYPE_SIZE) \
37 { \
38 X##_f0 = X##_f0 >> (N) | X##_f1 << (_FP_W_TYPE_SIZE - (N)); \
39 X##_f1 >>= (N); \
40 } \
41 else \
42 { \
43 X##_f0 = X##_f1 >> ((N) - _FP_W_TYPE_SIZE); \
44 X##_f1 = 0; \
45 } \
46 } while (0)
47
48/* Right shift with sticky-lsb. */
49#define _FP_FRAC_SRS_2(X,N,sz) \
50 do { \
51 if ((N) < _FP_W_TYPE_SIZE) \
52 { \
53 X##_f0 = (X##_f1 << (_FP_W_TYPE_SIZE - (N)) | X##_f0 >> (N) | \
54 (__builtin_constant_p(N) && (N) == 1 \
55 ? X##_f0 & 1 \
56 : (X##_f0 << (_FP_W_TYPE_SIZE - (N))) != 0)); \
57 X##_f1 >>= (N); \
58 } \
59 else \
60 { \
61 X##_f0 = (X##_f1 >> ((N) - _FP_W_TYPE_SIZE) | \
62 (((X##_f1 << (sz - (N))) | X##_f0) != 0)); \
63 X##_f1 = 0; \
64 } \
65 } while (0)
66
67#define _FP_FRAC_ADDI_2(X,I) \
68 __FP_FRAC_ADDI_2(X##_f1, X##_f0, I)
69
70#define _FP_FRAC_ADD_2(R,X,Y) \
71 __FP_FRAC_ADD_2(R##_f1, R##_f0, X##_f1, X##_f0, Y##_f1, Y##_f0)
72
73#define _FP_FRAC_SUB_2(R,X,Y) \
74 __FP_FRAC_SUB_2(R##_f1, R##_f0, X##_f1, X##_f0, Y##_f1, Y##_f0)
75
76#define _FP_FRAC_CLZ_2(R,X) \
77 do { \
78 if (X##_f1) \
79 __FP_CLZ(R,X##_f1); \
80 else \
81 { \
82 __FP_CLZ(R,X##_f0); \
83 R += _FP_W_TYPE_SIZE; \
84 } \
85 } while(0)
86
87/* Predicates */
88#define _FP_FRAC_NEGP_2(X) ((_FP_WS_TYPE)X##_f1 < 0)
89#define _FP_FRAC_ZEROP_2(X) ((X##_f1 | X##_f0) == 0)
90#define _FP_FRAC_OVERP_2(fs,X) (X##_f1 & _FP_OVERFLOW_##fs)
91#define _FP_FRAC_EQ_2(X, Y) (X##_f1 == Y##_f1 && X##_f0 == Y##_f0)
92#define _FP_FRAC_GT_2(X, Y) \
93 ((X##_f1 > Y##_f1) || (X##_f1 == Y##_f1 && X##_f0 > Y##_f0))
94#define _FP_FRAC_GE_2(X, Y) \
95 ((X##_f1 > Y##_f1) || (X##_f1 == Y##_f1 && X##_f0 >= Y##_f0))
96
97#define _FP_ZEROFRAC_2 0, 0
98#define _FP_MINFRAC_2 0, 1
99
100/*
101 * Internals
102 */
103
104#define __FP_FRAC_SET_2(X,I1,I0) (X##_f0 = I0, X##_f1 = I1)
105
106#define __FP_CLZ_2(R, xh, xl) \
107 do { \
108 if (xh) \
109 __FP_CLZ(R,xl); \
110 else \
111 { \
112 __FP_CLZ(R,xl); \
113 R += _FP_W_TYPE_SIZE; \
114 } \
115 } while(0)
116
117#if 0
118
119#ifndef __FP_FRAC_ADDI_2
120#define __FP_FRAC_ADDI_2(xh, xl, i) \
121 (xh += ((xl += i) < i))
122#endif
123#ifndef __FP_FRAC_ADD_2
124#define __FP_FRAC_ADD_2(rh, rl, xh, xl, yh, yl) \
125 (rh = xh + yh + ((rl = xl + yl) < xl))
126#endif
127#ifndef __FP_FRAC_SUB_2
128#define __FP_FRAC_SUB_2(rh, rl, xh, xl, yh, yl) \
129 (rh = xh - yh - ((rl = xl - yl) > xl))
130#endif
131
132#else
133
134#undef __FP_FRAC_ADDI_2
135#define __FP_FRAC_ADDI_2(xh, xl, i) add_ssaaaa(xh, xl, xh, xl, 0, i)
136#undef __FP_FRAC_ADD_2
137#define __FP_FRAC_ADD_2 add_ssaaaa
138#undef __FP_FRAC_SUB_2
139#define __FP_FRAC_SUB_2 sub_ddmmss
140
141#endif
142
143/*
144 * Unpack the raw bits of a native fp value. Do not classify or
145 * normalize the data.
146 */
147
148#define _FP_UNPACK_RAW_2(fs, X, val) \
149 do { \
150 union _FP_UNION_##fs _flo; _flo.flt = (val); \
151 \
152 X##_f0 = _flo.bits.frac0; \
153 X##_f1 = _flo.bits.frac1; \
154 X##_e = _flo.bits.exp; \
155 X##_s = _flo.bits.sign; \
156 } while (0)
157
158
159/*
160 * Repack the raw bits of a native fp value.
161 */
162
163#define _FP_PACK_RAW_2(fs, val, X) \
164 do { \
165 union _FP_UNION_##fs _flo; \
166 \
167 _flo.bits.frac0 = X##_f0; \
168 _flo.bits.frac1 = X##_f1; \
169 _flo.bits.exp = X##_e; \
170 _flo.bits.sign = X##_s; \
171 \
172 (val) = _flo.flt; \
173 } while (0)
174
175
176/*
177 * Multiplication algorithms:
178 */
179
180/* Given a 1W * 1W => 2W primitive, do the extended multiplication. */
181
182#define _FP_MUL_MEAT_2_wide(fs, R, X, Y, doit) \
183 do { \
184 _FP_FRAC_DECL_4(_z); _FP_FRAC_DECL_2(_b); _FP_FRAC_DECL_2(_c); \
185 \
186 doit(_FP_FRAC_WORD_4(_z,1), _FP_FRAC_WORD_4(_z,0), X##_f0, Y##_f0); \
187 doit(_b_f1, _b_f0, X##_f0, Y##_f1); \
188 doit(_c_f1, _c_f0, X##_f1, Y##_f0); \
189 doit(_FP_FRAC_WORD_4(_z,3), _FP_FRAC_WORD_4(_z,2), X##_f1, Y##_f1); \
190 \
191 __FP_FRAC_ADD_4(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
192 _FP_FRAC_WORD_4(_z,1),_FP_FRAC_WORD_4(_z,0), \
193 0, _b_f1, _b_f0, 0, \
194 _FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
195 _FP_FRAC_WORD_4(_z,1),_FP_FRAC_WORD_4(_z,0)); \
196 __FP_FRAC_ADD_4(_FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
197 _FP_FRAC_WORD_4(_z,1),_FP_FRAC_WORD_4(_z,0), \
198 0, _c_f1, _c_f0, 0, \
199 _FP_FRAC_WORD_4(_z,3),_FP_FRAC_WORD_4(_z,2), \
200 _FP_FRAC_WORD_4(_z,1),_FP_FRAC_WORD_4(_z,0)); \
201 \
202 /* Normalize since we know where the msb of the multiplicands \
203 were (bit B), we know that the msb of the of the product is \
204 at either 2B or 2B-1. */ \
205 _FP_FRAC_SRS_4(_z, _FP_WFRACBITS_##fs-1, 2*_FP_WFRACBITS_##fs); \
206 R##_f0 = _FP_FRAC_WORD_4(_z,0); \
207 R##_f1 = _FP_FRAC_WORD_4(_z,1); \
208 } while (0)
209
210/* This next macro appears to be totally broken. Fortunately nowhere
211 * seems to use it :-> The problem is that we define _z[4] but
212 * then use it in _FP_FRAC_SRS_4, which will attempt to access
213 * _z_f[n] which will cause an error. The fix probably involves
214 * declaring it with _FP_FRAC_DECL_4, see previous macro. -- PMM 02/1998
215 */
216#define _FP_MUL_MEAT_2_gmp(fs, R, X, Y) \
217 do { \
218 _FP_W_TYPE _x[2], _y[2], _z[4]; \
219 _x[0] = X##_f0; _x[1] = X##_f1; \
220 _y[0] = Y##_f0; _y[1] = Y##_f1; \
221 \
222 mpn_mul_n(_z, _x, _y, 2); \
223 \
224 /* Normalize since we know where the msb of the multiplicands \
225 were (bit B), we know that the msb of the of the product is \
226 at either 2B or 2B-1. */ \
227 _FP_FRAC_SRS_4(_z, _FP_WFRACBITS##_fs-1, 2*_FP_WFRACBITS_##fs); \
228 R##_f0 = _z[0]; \
229 R##_f1 = _z[1]; \
230 } while (0)
231
232
233/*
234 * Division algorithms:
235 * This seems to be giving me difficulties -- PMM
236 * Look, NetBSD seems to be able to comment algorithms. Can't you?
237 * I've thrown printks at the problem.
238 * This now appears to work, but I still don't really know why.
239 * Also, I don't think the result is properly normalised...
240 */
241
242#define _FP_DIV_MEAT_2_udiv_64(fs, R, X, Y) \
243 do { \
244 extern void _fp_udivmodti4(_FP_W_TYPE q[2], _FP_W_TYPE r[2], \
245 _FP_W_TYPE n1, _FP_W_TYPE n0, \
246 _FP_W_TYPE d1, _FP_W_TYPE d0); \
247 _FP_W_TYPE _n_f3, _n_f2, _n_f1, _n_f0, _r_f1, _r_f0; \
248 _FP_W_TYPE _q_f1, _q_f0, _m_f1, _m_f0; \
249 _FP_W_TYPE _rmem[2], _qmem[2]; \
250 /* I think this check is to ensure that the result is normalised. \
251 * Assuming X,Y normalised (ie in [1.0,2.0)) X/Y will be in \
252 * [0.5,2.0). Furthermore, it will be less than 1.0 iff X < Y. \
253 * In this case we tweak things. (this is based on comments in \
254 * the NetBSD FPU emulation code. ) \
255 * We know X,Y are normalised because we ensure this as part of \
256 * the unpacking process. -- PMM \
257 */ \
258 if (_FP_FRAC_GT_2(X, Y)) \
259 { \
260/* R##_e++; */ \
261 _n_f3 = X##_f1 >> 1; \
262 _n_f2 = X##_f1 << (_FP_W_TYPE_SIZE - 1) | X##_f0 >> 1; \
263 _n_f1 = X##_f0 << (_FP_W_TYPE_SIZE - 1); \
264 _n_f0 = 0; \
265 } \
266 else \
267 { \
268 R##_e--; \
269 _n_f3 = X##_f1; \
270 _n_f2 = X##_f0; \
271 _n_f1 = _n_f0 = 0; \
272 } \
273 \
274 /* Normalize, i.e. make the most significant bit of the \
275 denominator set. CHANGED: - 1 to nothing -- PMM */ \
276 _FP_FRAC_SLL_2(Y, _FP_WFRACXBITS_##fs /* -1 */); \
277 \
278 /* Do the 256/128 bit division given the 128-bit _fp_udivmodtf4 \
279 primitive snagged from libgcc2.c. */ \
280 \
281 _fp_udivmodti4(_qmem, _rmem, _n_f3, _n_f2, 0, Y##_f1); \
282 _q_f1 = _qmem[0]; \
283 umul_ppmm(_m_f1, _m_f0, _q_f1, Y##_f0); \
284 _r_f1 = _rmem[0]; \
285 _r_f0 = _n_f1; \
286 if (_FP_FRAC_GT_2(_m, _r)) \
287 { \
288 _q_f1--; \
289 _FP_FRAC_ADD_2(_r, _r, Y); \
290 if (_FP_FRAC_GE_2(_r, Y) && _FP_FRAC_GT_2(_m, _r)) \
291 { \
292 _q_f1--; \
293 _FP_FRAC_ADD_2(_r, _r, Y); \
294 } \
295 } \
296 _FP_FRAC_SUB_2(_r, _r, _m); \
297 \
298 _fp_udivmodti4(_qmem, _rmem, _r_f1, _r_f0, 0, Y##_f1); \
299 _q_f0 = _qmem[0]; \
300 umul_ppmm(_m_f1, _m_f0, _q_f0, Y##_f0); \
301 _r_f1 = _rmem[0]; \
302 _r_f0 = _n_f0; \
303 if (_FP_FRAC_GT_2(_m, _r)) \
304 { \
305 _q_f0--; \
306 _FP_FRAC_ADD_2(_r, _r, Y); \
307 if (_FP_FRAC_GE_2(_r, Y) && _FP_FRAC_GT_2(_m, _r)) \
308 { \
309 _q_f0--; \
310 _FP_FRAC_ADD_2(_r, _r, Y); \
311 } \
312 } \
313 _FP_FRAC_SUB_2(_r, _r, _m); \
314 \
315 R##_f1 = _q_f1; \
316 R##_f0 = _q_f0 | ((_r_f1 | _r_f0) != 0); \
317 /* adjust so answer is normalized again. I'm not sure what the \
318 * final sz param should be. In practice it's never used since \
319 * N is 1 which is always going to be < _FP_W_TYPE_SIZE... \
320 */ \
321 /* _FP_FRAC_SRS_2(R,1,_FP_WFRACBITS_##fs); */ \
322 } while (0)
323
324
325#define _FP_DIV_MEAT_2_gmp(fs, R, X, Y) \
326 do { \
327 _FP_W_TYPE _x[4], _y[2], _z[4]; \
328 _y[0] = Y##_f0; _y[1] = Y##_f1; \
329 _x[0] = _x[3] = 0; \
330 if (_FP_FRAC_GT_2(X, Y)) \
331 { \
332 R##_e++; \
333 _x[1] = (X##_f0 << (_FP_WFRACBITS-1 - _FP_W_TYPE_SIZE) | \
334 X##_f1 >> (_FP_W_TYPE_SIZE - \
335 (_FP_WFRACBITS-1 - _FP_W_TYPE_SIZE))); \
336 _x[2] = X##_f1 << (_FP_WFRACBITS-1 - _FP_W_TYPE_SIZE); \
337 } \
338 else \
339 { \
340 _x[1] = (X##_f0 << (_FP_WFRACBITS - _FP_W_TYPE_SIZE) | \
341 X##_f1 >> (_FP_W_TYPE_SIZE - \
342 (_FP_WFRACBITS - _FP_W_TYPE_SIZE))); \
343 _x[2] = X##_f1 << (_FP_WFRACBITS - _FP_W_TYPE_SIZE); \
344 } \
345 \
346 (void) mpn_divrem (_z, 0, _x, 4, _y, 2); \
347 R##_f1 = _z[1]; \
348 R##_f0 = _z[0] | ((_x[0] | _x[1]) != 0); \
349 } while (0)
350
351
352/*
353 * Square root algorithms:
354 * We have just one right now, maybe Newton approximation
355 * should be added for those machines where division is fast.
356 */
357
358#define _FP_SQRT_MEAT_2(R, S, T, X, q) \
359 do { \
360 while (q) \
361 { \
362 T##_f1 = S##_f1 + q; \
363 if (T##_f1 <= X##_f1) \
364 { \
365 S##_f1 = T##_f1 + q; \
366 X##_f1 -= T##_f1; \
367 R##_f1 += q; \
368 } \
369 _FP_FRAC_SLL_2(X, 1); \
370 q >>= 1; \
371 } \
372 q = (_FP_W_TYPE)1 << (_FP_W_TYPE_SIZE - 1); \
373 while (q) \
374 { \
375 T##_f0 = S##_f0 + q; \
376 T##_f1 = S##_f1; \
377 if (T##_f1 < X##_f1 || \
378 (T##_f1 == X##_f1 && T##_f0 < X##_f0)) \
379 { \
380 S##_f0 = T##_f0 + q; \
381 if (((_FP_WS_TYPE)T##_f0) < 0 && \
382 ((_FP_WS_TYPE)S##_f0) >= 0) \
383 S##_f1++; \
384 _FP_FRAC_SUB_2(X, X, T); \
385 R##_f0 += q; \
386 } \
387 _FP_FRAC_SLL_2(X, 1); \
388 q >>= 1; \
389 } \
390 } while (0)
391
392
393/*
394 * Assembly/disassembly for converting to/from integral types.
395 * No shifting or overflow handled here.
396 */
397
398#define _FP_FRAC_ASSEMBLE_2(r, X, rsize) \
399 do { \
400 if (rsize <= _FP_W_TYPE_SIZE) \
401 r = X##_f0; \
402 else \
403 { \
404 r = X##_f1; \
405 r <<= _FP_W_TYPE_SIZE; \
406 r += X##_f0; \
407 } \
408 } while (0)
409
410#define _FP_FRAC_DISASSEMBLE_2(X, r, rsize) \
411 do { \
412 X##_f0 = r; \
413 X##_f1 = (rsize <= _FP_W_TYPE_SIZE ? 0 : r >> _FP_W_TYPE_SIZE); \
414 } while (0)
415
416/*
417 * Convert FP values between word sizes
418 */
419
420#define _FP_FRAC_CONV_1_2(dfs, sfs, D, S) \
421 do { \
422 _FP_FRAC_SRS_2(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs), \
423 _FP_WFRACBITS_##sfs); \
424 D##_f = S##_f0; \
425 } while (0)
426
427#define _FP_FRAC_CONV_2_1(dfs, sfs, D, S) \
428 do { \
429 D##_f0 = S##_f; \
430 D##_f1 = 0; \
431 _FP_FRAC_SLL_2(D, (_FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs)); \
432 } while (0)
433
diff --git a/arch/powerpc/math-emu/op-4.h b/arch/powerpc/math-emu/op-4.h
new file mode 100644
index 000000000000..fcdd6d064c54
--- /dev/null
+++ b/arch/powerpc/math-emu/op-4.h
@@ -0,0 +1,297 @@
1/*
2 * Basic four-word fraction declaration and manipulation.
3 *
4 * When adding quadword support for 32 bit machines, we need
5 * to be a little careful as double multiply uses some of these
6 * macros: (in op-2.h)
7 * _FP_MUL_MEAT_2_wide() uses _FP_FRAC_DECL_4, _FP_FRAC_WORD_4,
8 * _FP_FRAC_ADD_4, _FP_FRAC_SRS_4
9 * _FP_MUL_MEAT_2_gmp() uses _FP_FRAC_SRS_4 (and should use
10 * _FP_FRAC_DECL_4: it appears to be broken and is not used
11 * anywhere anyway. )
12 *
13 * I've now fixed all the macros that were here from the sparc64 code.
14 * [*none* of the shift macros were correct!] -- PMM 02/1998
15 *
16 * The only quadword stuff that remains to be coded is:
17 * 1) the conversion to/from ints, which requires
18 * that we check (in op-common.h) that the following do the right thing
19 * for quadwords: _FP_TO_INT(Q,4,r,X,rsz,rsg), _FP_FROM_INT(Q,4,X,r,rs,rt)
20 * 2) multiply, divide and sqrt, which require:
21 * _FP_MUL_MEAT_4_*(R,X,Y), _FP_DIV_MEAT_4_*(R,X,Y), _FP_SQRT_MEAT_4(R,S,T,X,q),
22 * This also needs _FP_MUL_MEAT_Q and _FP_DIV_MEAT_Q to be defined to
23 * some suitable _FP_MUL_MEAT_4_* macros in sfp-machine.h.
24 * [we're free to choose whatever FP_MUL_MEAT_4_* macros we need for
25 * these; they are used nowhere else. ]
26 */
27
28#define _FP_FRAC_DECL_4(X) _FP_W_TYPE X##_f[4]
29#define _FP_FRAC_COPY_4(D,S) \
30 (D##_f[0] = S##_f[0], D##_f[1] = S##_f[1], \
31 D##_f[2] = S##_f[2], D##_f[3] = S##_f[3])
32/* The _FP_FRAC_SET_n(X,I) macro is intended for use with another
33 * macro such as _FP_ZEROFRAC_n which returns n comma separated values.
34 * The result is that we get an expansion of __FP_FRAC_SET_n(X,I0,I1,I2,I3)
35 * which just assigns the In values to the array X##_f[].
36 * This is why the number of parameters doesn't appear to match
37 * at first glance... -- PMM
38 */
39#define _FP_FRAC_SET_4(X,I) __FP_FRAC_SET_4(X, I)
40#define _FP_FRAC_HIGH_4(X) (X##_f[3])
41#define _FP_FRAC_LOW_4(X) (X##_f[0])
42#define _FP_FRAC_WORD_4(X,w) (X##_f[w])
43
44#define _FP_FRAC_SLL_4(X,N) \
45 do { \
46 _FP_I_TYPE _up, _down, _skip, _i; \
47 _skip = (N) / _FP_W_TYPE_SIZE; \
48 _up = (N) % _FP_W_TYPE_SIZE; \
49 _down = _FP_W_TYPE_SIZE - _up; \
50 for (_i = 3; _i > _skip; --_i) \
51 X##_f[_i] = X##_f[_i-_skip] << _up | X##_f[_i-_skip-1] >> _down; \
52/* bugfixed: was X##_f[_i] <<= _up; -- PMM 02/1998 */ \
53 X##_f[_i] = X##_f[0] << _up; \
54 for (--_i; _i >= 0; --_i) \
55 X##_f[_i] = 0; \
56 } while (0)
57
58/* This one was broken too */
59#define _FP_FRAC_SRL_4(X,N) \
60 do { \
61 _FP_I_TYPE _up, _down, _skip, _i; \
62 _skip = (N) / _FP_W_TYPE_SIZE; \
63 _down = (N) % _FP_W_TYPE_SIZE; \
64 _up = _FP_W_TYPE_SIZE - _down; \
65 for (_i = 0; _i < 3-_skip; ++_i) \
66 X##_f[_i] = X##_f[_i+_skip] >> _down | X##_f[_i+_skip+1] << _up; \
67 X##_f[_i] = X##_f[3] >> _down; \
68 for (++_i; _i < 4; ++_i) \
69 X##_f[_i] = 0; \
70 } while (0)
71
72
73/* Right shift with sticky-lsb.
74 * What this actually means is that we do a standard right-shift,
75 * but that if any of the bits that fall off the right hand side
76 * were one then we always set the LSbit.
77 */
78#define _FP_FRAC_SRS_4(X,N,size) \
79 do { \
80 _FP_I_TYPE _up, _down, _skip, _i; \
81 _FP_W_TYPE _s; \
82 _skip = (N) / _FP_W_TYPE_SIZE; \
83 _down = (N) % _FP_W_TYPE_SIZE; \
84 _up = _FP_W_TYPE_SIZE - _down; \
85 for (_s = _i = 0; _i < _skip; ++_i) \
86 _s |= X##_f[_i]; \
87 _s |= X##_f[_i] << _up; \
88/* s is now != 0 if we want to set the LSbit */ \
89 for (_i = 0; _i < 3-_skip; ++_i) \
90 X##_f[_i] = X##_f[_i+_skip] >> _down | X##_f[_i+_skip+1] << _up; \
91 X##_f[_i] = X##_f[3] >> _down; \
92 for (++_i; _i < 4; ++_i) \
93 X##_f[_i] = 0; \
94 /* don't fix the LSB until the very end when we're sure f[0] is stable */ \
95 X##_f[0] |= (_s != 0); \
96 } while (0)
97
98#define _FP_FRAC_ADD_4(R,X,Y) \
99 __FP_FRAC_ADD_4(R##_f[3], R##_f[2], R##_f[1], R##_f[0], \
100 X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
101 Y##_f[3], Y##_f[2], Y##_f[1], Y##_f[0])
102
103#define _FP_FRAC_SUB_4(R,X,Y) \
104 __FP_FRAC_SUB_4(R##_f[3], R##_f[2], R##_f[1], R##_f[0], \
105 X##_f[3], X##_f[2], X##_f[1], X##_f[0], \
106 Y##_f[3], Y##_f[2], Y##_f[1], Y##_f[0])
107
108#define _FP_FRAC_ADDI_4(X,I) \
109 __FP_FRAC_ADDI_4(X##_f[3], X##_f[2], X##_f[1], X##_f[0], I)
110
111#define _FP_ZEROFRAC_4 0,0,0,0
112#define _FP_MINFRAC_4 0,0,0,1
113
114#define _FP_FRAC_ZEROP_4(X) ((X##_f[0] | X##_f[1] | X##_f[2] | X##_f[3]) == 0)
115#define _FP_FRAC_NEGP_4(X) ((_FP_WS_TYPE)X##_f[3] < 0)
116#define _FP_FRAC_OVERP_4(fs,X) (X##_f[0] & _FP_OVERFLOW_##fs)
117
118#define _FP_FRAC_EQ_4(X,Y) \
119 (X##_f[0] == Y##_f[0] && X##_f[1] == Y##_f[1] \
120 && X##_f[2] == Y##_f[2] && X##_f[3] == Y##_f[3])
121
122#define _FP_FRAC_GT_4(X,Y) \
123 (X##_f[3] > Y##_f[3] || \
124 (X##_f[3] == Y##_f[3] && (X##_f[2] > Y##_f[2] || \
125 (X##_f[2] == Y##_f[2] && (X##_f[1] > Y##_f[1] || \
126 (X##_f[1] == Y##_f[1] && X##_f[0] > Y##_f[0]) \
127 )) \
128 )) \
129 )
130
131#define _FP_FRAC_GE_4(X,Y) \
132 (X##_f[3] > Y##_f[3] || \
133 (X##_f[3] == Y##_f[3] && (X##_f[2] > Y##_f[2] || \
134 (X##_f[2] == Y##_f[2] && (X##_f[1] > Y##_f[1] || \
135 (X##_f[1] == Y##_f[1] && X##_f[0] >= Y##_f[0]) \
136 )) \
137 )) \
138 )
139
140
141#define _FP_FRAC_CLZ_4(R,X) \
142 do { \
143 if (X##_f[3]) \
144 { \
145 __FP_CLZ(R,X##_f[3]); \
146 } \
147 else if (X##_f[2]) \
148 { \
149 __FP_CLZ(R,X##_f[2]); \
150 R += _FP_W_TYPE_SIZE; \
151 } \
152 else if (X##_f[1]) \
153 { \
154 __FP_CLZ(R,X##_f[2]); \
155 R += _FP_W_TYPE_SIZE*2; \
156 } \
157 else \
158 { \
159 __FP_CLZ(R,X##_f[0]); \
160 R += _FP_W_TYPE_SIZE*3; \
161 } \
162 } while(0)
163
164
165#define _FP_UNPACK_RAW_4(fs, X, val) \
166 do { \
167 union _FP_UNION_##fs _flo; _flo.flt = (val); \
168 X##_f[0] = _flo.bits.frac0; \
169 X##_f[1] = _flo.bits.frac1; \
170 X##_f[2] = _flo.bits.frac2; \
171 X##_f[3] = _flo.bits.frac3; \
172 X##_e = _flo.bits.exp; \
173 X##_s = _flo.bits.sign; \
174 } while (0)
175
176#define _FP_PACK_RAW_4(fs, val, X) \
177 do { \
178 union _FP_UNION_##fs _flo; \
179 _flo.bits.frac0 = X##_f[0]; \
180 _flo.bits.frac1 = X##_f[1]; \
181 _flo.bits.frac2 = X##_f[2]; \
182 _flo.bits.frac3 = X##_f[3]; \
183 _flo.bits.exp = X##_e; \
184 _flo.bits.sign = X##_s; \
185 (val) = _flo.flt; \
186 } while (0)
187
188
189/*
190 * Internals
191 */
192
193#define __FP_FRAC_SET_4(X,I3,I2,I1,I0) \
194 (X##_f[3] = I3, X##_f[2] = I2, X##_f[1] = I1, X##_f[0] = I0)
195
196#ifndef __FP_FRAC_ADD_4
197#define __FP_FRAC_ADD_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \
198 (r0 = x0 + y0, \
199 r1 = x1 + y1 + (r0 < x0), \
200 r2 = x2 + y2 + (r1 < x1), \
201 r3 = x3 + y3 + (r2 < x2))
202#endif
203
204#ifndef __FP_FRAC_SUB_4
205#define __FP_FRAC_SUB_4(r3,r2,r1,r0,x3,x2,x1,x0,y3,y2,y1,y0) \
206 (r0 = x0 - y0, \
207 r1 = x1 - y1 - (r0 > x0), \
208 r2 = x2 - y2 - (r1 > x1), \
209 r3 = x3 - y3 - (r2 > x2))
210#endif
211
212#ifndef __FP_FRAC_ADDI_4
213/* I always wanted to be a lisp programmer :-> */
214#define __FP_FRAC_ADDI_4(x3,x2,x1,x0,i) \
215 (x3 += ((x2 += ((x1 += ((x0 += i) < x0)) < x1) < x2)))
216#endif
217
218/* Convert FP values between word sizes. This appears to be more
219 * complicated than I'd have expected it to be, so these might be
220 * wrong... These macros are in any case somewhat bogus because they
221 * use information about what various FRAC_n variables look like
222 * internally [eg, that 2 word vars are X_f0 and x_f1]. But so do
223 * the ones in op-2.h and op-1.h.
224 */
225#define _FP_FRAC_CONV_1_4(dfs, sfs, D, S) \
226 do { \
227 _FP_FRAC_SRS_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs), \
228 _FP_WFRACBITS_##sfs); \
229 D##_f = S##_f[0]; \
230 } while (0)
231
232#define _FP_FRAC_CONV_2_4(dfs, sfs, D, S) \
233 do { \
234 _FP_FRAC_SRS_4(S, (_FP_WFRACBITS_##sfs - _FP_WFRACBITS_##dfs), \
235 _FP_WFRACBITS_##sfs); \
236 D##_f0 = S##_f[0]; \
237 D##_f1 = S##_f[1]; \
238 } while (0)
239
240/* Assembly/disassembly for converting to/from integral types.
241 * No shifting or overflow handled here.
242 */
243/* Put the FP value X into r, which is an integer of size rsize. */
244#define _FP_FRAC_ASSEMBLE_4(r, X, rsize) \
245 do { \
246 if (rsize <= _FP_W_TYPE_SIZE) \
247 r = X##_f[0]; \
248 else if (rsize <= 2*_FP_W_TYPE_SIZE) \
249 { \
250 r = X##_f[1]; \
251 r <<= _FP_W_TYPE_SIZE; \
252 r += X##_f[0]; \
253 } \
254 else \
255 { \
256 /* I'm feeling lazy so we deal with int == 3words (implausible)*/ \
257 /* and int == 4words as a single case. */ \
258 r = X##_f[3]; \
259 r <<= _FP_W_TYPE_SIZE; \
260 r += X##_f[2]; \
261 r <<= _FP_W_TYPE_SIZE; \
262 r += X##_f[1]; \
263 r <<= _FP_W_TYPE_SIZE; \
264 r += X##_f[0]; \
265 } \
266 } while (0)
267
268/* "No disassemble Number Five!" */
269/* move an integer of size rsize into X's fractional part. We rely on
270 * the _f[] array consisting of words of size _FP_W_TYPE_SIZE to avoid
271 * having to mask the values we store into it.
272 */
273#define _FP_FRAC_DISASSEMBLE_4(X, r, rsize) \
274 do { \
275 X##_f[0] = r; \
276 X##_f[1] = (rsize <= _FP_W_TYPE_SIZE ? 0 : r >> _FP_W_TYPE_SIZE); \
277 X##_f[2] = (rsize <= 2*_FP_W_TYPE_SIZE ? 0 : r >> 2*_FP_W_TYPE_SIZE); \
278 X##_f[3] = (rsize <= 3*_FP_W_TYPE_SIZE ? 0 : r >> 3*_FP_W_TYPE_SIZE); \
279 } while (0)
280
281#define _FP_FRAC_CONV_4_1(dfs, sfs, D, S) \
282 do { \
283 D##_f[0] = S##_f; \
284 D##_f[1] = D##_f[2] = D##_f[3] = 0; \
285 _FP_FRAC_SLL_4(D, (_FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs)); \
286 } while (0)
287
288#define _FP_FRAC_CONV_4_2(dfs, sfs, D, S) \
289 do { \
290 D##_f[0] = S##_f0; \
291 D##_f[1] = S##_f1; \
292 D##_f[2] = D##_f[3] = 0; \
293 _FP_FRAC_SLL_4(D, (_FP_WFRACBITS_##dfs - _FP_WFRACBITS_##sfs)); \
294 } while (0)
295
296/* FIXME! This has to be written */
297#define _FP_SQRT_MEAT_4(R, S, T, X, q)
diff --git a/arch/powerpc/math-emu/op-common.h b/arch/powerpc/math-emu/op-common.h
new file mode 100644
index 000000000000..afb82b6498ce
--- /dev/null
+++ b/arch/powerpc/math-emu/op-common.h
@@ -0,0 +1,688 @@
1#define _FP_DECL(wc, X) \
2 _FP_I_TYPE X##_c, X##_s, X##_e; \
3 _FP_FRAC_DECL_##wc(X)
4
5/*
6 * Finish truely unpacking a native fp value by classifying the kind
7 * of fp value and normalizing both the exponent and the fraction.
8 */
9
10#define _FP_UNPACK_CANONICAL(fs, wc, X) \
11do { \
12 switch (X##_e) \
13 { \
14 default: \
15 _FP_FRAC_HIGH_##wc(X) |= _FP_IMPLBIT_##fs; \
16 _FP_FRAC_SLL_##wc(X, _FP_WORKBITS); \
17 X##_e -= _FP_EXPBIAS_##fs; \
18 X##_c = FP_CLS_NORMAL; \
19 break; \
20 \
21 case 0: \
22 if (_FP_FRAC_ZEROP_##wc(X)) \
23 X##_c = FP_CLS_ZERO; \
24 else \
25 { \
26 /* a denormalized number */ \
27 _FP_I_TYPE _shift; \
28 _FP_FRAC_CLZ_##wc(_shift, X); \
29 _shift -= _FP_FRACXBITS_##fs; \
30 _FP_FRAC_SLL_##wc(X, (_shift+_FP_WORKBITS)); \
31 X##_e -= _FP_EXPBIAS_##fs - 1 + _shift; \
32 X##_c = FP_CLS_NORMAL; \
33 } \
34 break; \
35 \
36 case _FP_EXPMAX_##fs: \
37 if (_FP_FRAC_ZEROP_##wc(X)) \
38 X##_c = FP_CLS_INF; \
39 else \
40 /* we don't differentiate between signaling and quiet nans */ \
41 X##_c = FP_CLS_NAN; \
42 break; \
43 } \
44} while (0)
45
46
47/*
48 * Before packing the bits back into the native fp result, take care
49 * of such mundane things as rounding and overflow. Also, for some
50 * kinds of fp values, the original parts may not have been fully
51 * extracted -- but that is ok, we can regenerate them now.
52 */
53
54#define _FP_PACK_CANONICAL(fs, wc, X) \
55({int __ret = 0; \
56 switch (X##_c) \
57 { \
58 case FP_CLS_NORMAL: \
59 X##_e += _FP_EXPBIAS_##fs; \
60 if (X##_e > 0) \
61 { \
62 __ret |= _FP_ROUND(wc, X); \
63 if (_FP_FRAC_OVERP_##wc(fs, X)) \
64 { \
65 _FP_FRAC_SRL_##wc(X, (_FP_WORKBITS+1)); \
66 X##_e++; \
67 } \
68 else \
69 _FP_FRAC_SRL_##wc(X, _FP_WORKBITS); \
70 if (X##_e >= _FP_EXPMAX_##fs) \
71 { \
72 /* overflow to infinity */ \
73 X##_e = _FP_EXPMAX_##fs; \
74 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
75 __ret |= EFLAG_OVERFLOW; \
76 } \
77 } \
78 else \
79 { \
80 /* we've got a denormalized number */ \
81 X##_e = -X##_e + 1; \
82 if (X##_e <= _FP_WFRACBITS_##fs) \
83 { \
84 _FP_FRAC_SRS_##wc(X, X##_e, _FP_WFRACBITS_##fs); \
85 _FP_FRAC_SLL_##wc(X, 1); \
86 if (_FP_FRAC_OVERP_##wc(fs, X)) \
87 { \
88 X##_e = 1; \
89 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
90 } \
91 else \
92 { \
93 X##_e = 0; \
94 _FP_FRAC_SRL_##wc(X, _FP_WORKBITS+1); \
95 __ret |= EFLAG_UNDERFLOW; \
96 } \
97 } \
98 else \
99 { \
100 /* underflow to zero */ \
101 X##_e = 0; \
102 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
103 __ret |= EFLAG_UNDERFLOW; \
104 } \
105 } \
106 break; \
107 \
108 case FP_CLS_ZERO: \
109 X##_e = 0; \
110 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
111 break; \
112 \
113 case FP_CLS_INF: \
114 X##_e = _FP_EXPMAX_##fs; \
115 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
116 break; \
117 \
118 case FP_CLS_NAN: \
119 X##_e = _FP_EXPMAX_##fs; \
120 if (!_FP_KEEPNANFRACP) \
121 { \
122 _FP_FRAC_SET_##wc(X, _FP_NANFRAC_##fs); \
123 X##_s = 0; \
124 } \
125 else \
126 _FP_FRAC_HIGH_##wc(X) |= _FP_QNANBIT_##fs; \
127 break; \
128 } \
129 __ret; \
130})
131
132
133/*
134 * Main addition routine. The input values should be cooked.
135 */
136
137#define _FP_ADD(fs, wc, R, X, Y) \
138do { \
139 switch (_FP_CLS_COMBINE(X##_c, Y##_c)) \
140 { \
141 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NORMAL): \
142 { \
143 /* shift the smaller number so that its exponent matches the larger */ \
144 _FP_I_TYPE diff = X##_e - Y##_e; \
145 \
146 if (diff < 0) \
147 { \
148 diff = -diff; \
149 if (diff <= _FP_WFRACBITS_##fs) \
150 _FP_FRAC_SRS_##wc(X, diff, _FP_WFRACBITS_##fs); \
151 else if (!_FP_FRAC_ZEROP_##wc(X)) \
152 _FP_FRAC_SET_##wc(X, _FP_MINFRAC_##wc); \
153 else \
154 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
155 R##_e = Y##_e; \
156 } \
157 else \
158 { \
159 if (diff > 0) \
160 { \
161 if (diff <= _FP_WFRACBITS_##fs) \
162 _FP_FRAC_SRS_##wc(Y, diff, _FP_WFRACBITS_##fs); \
163 else if (!_FP_FRAC_ZEROP_##wc(Y)) \
164 _FP_FRAC_SET_##wc(Y, _FP_MINFRAC_##wc); \
165 else \
166 _FP_FRAC_SET_##wc(Y, _FP_ZEROFRAC_##wc); \
167 } \
168 R##_e = X##_e; \
169 } \
170 \
171 R##_c = FP_CLS_NORMAL; \
172 \
173 if (X##_s == Y##_s) \
174 { \
175 R##_s = X##_s; \
176 _FP_FRAC_ADD_##wc(R, X, Y); \
177 if (_FP_FRAC_OVERP_##wc(fs, R)) \
178 { \
179 _FP_FRAC_SRS_##wc(R, 1, _FP_WFRACBITS_##fs); \
180 R##_e++; \
181 } \
182 } \
183 else \
184 { \
185 R##_s = X##_s; \
186 _FP_FRAC_SUB_##wc(R, X, Y); \
187 if (_FP_FRAC_ZEROP_##wc(R)) \
188 { \
189 /* return an exact zero */ \
190 if (FP_ROUNDMODE == FP_RND_MINF) \
191 R##_s |= Y##_s; \
192 else \
193 R##_s &= Y##_s; \
194 R##_c = FP_CLS_ZERO; \
195 } \
196 else \
197 { \
198 if (_FP_FRAC_NEGP_##wc(R)) \
199 { \
200 _FP_FRAC_SUB_##wc(R, Y, X); \
201 R##_s = Y##_s; \
202 } \
203 \
204 /* renormalize after subtraction */ \
205 _FP_FRAC_CLZ_##wc(diff, R); \
206 diff -= _FP_WFRACXBITS_##fs; \
207 if (diff) \
208 { \
209 R##_e -= diff; \
210 _FP_FRAC_SLL_##wc(R, diff); \
211 } \
212 } \
213 } \
214 break; \
215 } \
216 \
217 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NAN): \
218 _FP_CHOOSENAN(fs, wc, R, X, Y); \
219 break; \
220 \
221 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_ZERO): \
222 R##_e = X##_e; \
223 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NORMAL): \
224 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_INF): \
225 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_ZERO): \
226 _FP_FRAC_COPY_##wc(R, X); \
227 R##_s = X##_s; \
228 R##_c = X##_c; \
229 break; \
230 \
231 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NORMAL): \
232 R##_e = Y##_e; \
233 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NAN): \
234 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NAN): \
235 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NAN): \
236 _FP_FRAC_COPY_##wc(R, Y); \
237 R##_s = Y##_s; \
238 R##_c = Y##_c; \
239 break; \
240 \
241 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_INF): \
242 if (X##_s != Y##_s) \
243 { \
244 /* +INF + -INF => NAN */ \
245 _FP_FRAC_SET_##wc(R, _FP_NANFRAC_##fs); \
246 R##_s = X##_s ^ Y##_s; \
247 R##_c = FP_CLS_NAN; \
248 break; \
249 } \
250 /* FALLTHRU */ \
251 \
252 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NORMAL): \
253 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_ZERO): \
254 R##_s = X##_s; \
255 R##_c = FP_CLS_INF; \
256 break; \
257 \
258 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_INF): \
259 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_INF): \
260 R##_s = Y##_s; \
261 R##_c = FP_CLS_INF; \
262 break; \
263 \
264 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_ZERO): \
265 /* make sure the sign is correct */ \
266 if (FP_ROUNDMODE == FP_RND_MINF) \
267 R##_s = X##_s | Y##_s; \
268 else \
269 R##_s = X##_s & Y##_s; \
270 R##_c = FP_CLS_ZERO; \
271 break; \
272 \
273 default: \
274 abort(); \
275 } \
276} while (0)
277
278
279/*
280 * Main negation routine. FIXME -- when we care about setting exception
281 * bits reliably, this will not do. We should examine all of the fp classes.
282 */
283
284#define _FP_NEG(fs, wc, R, X) \
285 do { \
286 _FP_FRAC_COPY_##wc(R, X); \
287 R##_c = X##_c; \
288 R##_e = X##_e; \
289 R##_s = 1 ^ X##_s; \
290 } while (0)
291
292
293/*
294 * Main multiplication routine. The input values should be cooked.
295 */
296
297#define _FP_MUL(fs, wc, R, X, Y) \
298do { \
299 R##_s = X##_s ^ Y##_s; \
300 switch (_FP_CLS_COMBINE(X##_c, Y##_c)) \
301 { \
302 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NORMAL): \
303 R##_c = FP_CLS_NORMAL; \
304 R##_e = X##_e + Y##_e + 1; \
305 \
306 _FP_MUL_MEAT_##fs(R,X,Y); \
307 \
308 if (_FP_FRAC_OVERP_##wc(fs, R)) \
309 _FP_FRAC_SRS_##wc(R, 1, _FP_WFRACBITS_##fs); \
310 else \
311 R##_e--; \
312 break; \
313 \
314 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NAN): \
315 _FP_CHOOSENAN(fs, wc, R, X, Y); \
316 break; \
317 \
318 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NORMAL): \
319 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_INF): \
320 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_ZERO): \
321 R##_s = X##_s; \
322 \
323 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_INF): \
324 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NORMAL): \
325 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NORMAL): \
326 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_ZERO): \
327 _FP_FRAC_COPY_##wc(R, X); \
328 R##_c = X##_c; \
329 break; \
330 \
331 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NAN): \
332 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NAN): \
333 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NAN): \
334 R##_s = Y##_s; \
335 \
336 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_INF): \
337 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_ZERO): \
338 _FP_FRAC_COPY_##wc(R, Y); \
339 R##_c = Y##_c; \
340 break; \
341 \
342 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_ZERO): \
343 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_INF): \
344 R##_c = FP_CLS_NAN; \
345 _FP_FRAC_SET_##wc(R, _FP_NANFRAC_##fs); \
346 break; \
347 \
348 default: \
349 abort(); \
350 } \
351} while (0)
352
353
354/*
355 * Main division routine. The input values should be cooked.
356 */
357
358#define _FP_DIV(fs, wc, R, X, Y) \
359do { \
360 R##_s = X##_s ^ Y##_s; \
361 switch (_FP_CLS_COMBINE(X##_c, Y##_c)) \
362 { \
363 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NORMAL): \
364 R##_c = FP_CLS_NORMAL; \
365 R##_e = X##_e - Y##_e; \
366 \
367 _FP_DIV_MEAT_##fs(R,X,Y); \
368 break; \
369 \
370 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NAN): \
371 _FP_CHOOSENAN(fs, wc, R, X, Y); \
372 break; \
373 \
374 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_NORMAL): \
375 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_INF): \
376 case _FP_CLS_COMBINE(FP_CLS_NAN,FP_CLS_ZERO): \
377 R##_s = X##_s; \
378 _FP_FRAC_COPY_##wc(R, X); \
379 R##_c = X##_c; \
380 break; \
381 \
382 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_NAN): \
383 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NAN): \
384 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NAN): \
385 R##_s = Y##_s; \
386 _FP_FRAC_COPY_##wc(R, Y); \
387 R##_c = Y##_c; \
388 break; \
389 \
390 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_INF): \
391 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_INF): \
392 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_NORMAL): \
393 R##_c = FP_CLS_ZERO; \
394 break; \
395 \
396 case _FP_CLS_COMBINE(FP_CLS_NORMAL,FP_CLS_ZERO): \
397 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_ZERO): \
398 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_NORMAL): \
399 R##_c = FP_CLS_INF; \
400 break; \
401 \
402 case _FP_CLS_COMBINE(FP_CLS_INF,FP_CLS_INF): \
403 case _FP_CLS_COMBINE(FP_CLS_ZERO,FP_CLS_ZERO): \
404 R##_c = FP_CLS_NAN; \
405 _FP_FRAC_SET_##wc(R, _FP_NANFRAC_##fs); \
406 break; \
407 \
408 default: \
409 abort(); \
410 } \
411} while (0)
412
413
414/*
415 * Main differential comparison routine. The inputs should be raw not
416 * cooked. The return is -1,0,1 for normal values, 2 otherwise.
417 */
418
419#define _FP_CMP(fs, wc, ret, X, Y, un) \
420 do { \
421 /* NANs are unordered */ \
422 if ((X##_e == _FP_EXPMAX_##fs && !_FP_FRAC_ZEROP_##wc(X)) \
423 || (Y##_e == _FP_EXPMAX_##fs && !_FP_FRAC_ZEROP_##wc(Y))) \
424 { \
425 ret = un; \
426 } \
427 else \
428 { \
429 int __x_zero = (!X##_e && _FP_FRAC_ZEROP_##wc(X)) ? 1 : 0; \
430 int __y_zero = (!Y##_e && _FP_FRAC_ZEROP_##wc(Y)) ? 1 : 0; \
431 \
432 if (__x_zero && __y_zero) \
433 ret = 0; \
434 else if (__x_zero) \
435 ret = Y##_s ? 1 : -1; \
436 else if (__y_zero) \
437 ret = X##_s ? -1 : 1; \
438 else if (X##_s != Y##_s) \
439 ret = X##_s ? -1 : 1; \
440 else if (X##_e > Y##_e) \
441 ret = X##_s ? -1 : 1; \
442 else if (X##_e < Y##_e) \
443 ret = X##_s ? 1 : -1; \
444 else if (_FP_FRAC_GT_##wc(X, Y)) \
445 ret = X##_s ? -1 : 1; \
446 else if (_FP_FRAC_GT_##wc(Y, X)) \
447 ret = X##_s ? 1 : -1; \
448 else \
449 ret = 0; \
450 } \
451 } while (0)
452
453
454/* Simplification for strict equality. */
455
456#define _FP_CMP_EQ(fs, wc, ret, X, Y) \
457 do { \
458 /* NANs are unordered */ \
459 if ((X##_e == _FP_EXPMAX_##fs && !_FP_FRAC_ZEROP_##wc(X)) \
460 || (Y##_e == _FP_EXPMAX_##fs && !_FP_FRAC_ZEROP_##wc(Y))) \
461 { \
462 ret = 1; \
463 } \
464 else \
465 { \
466 ret = !(X##_e == Y##_e \
467 && _FP_FRAC_EQ_##wc(X, Y) \
468 && (X##_s == Y##_s || !X##_e && _FP_FRAC_ZEROP_##wc(X))); \
469 } \
470 } while (0)
471
472/*
473 * Main square root routine. The input value should be cooked.
474 */
475
476#define _FP_SQRT(fs, wc, R, X) \
477do { \
478 _FP_FRAC_DECL_##wc(T); _FP_FRAC_DECL_##wc(S); \
479 _FP_W_TYPE q; \
480 switch (X##_c) \
481 { \
482 case FP_CLS_NAN: \
483 R##_s = 0; \
484 R##_c = FP_CLS_NAN; \
485 _FP_FRAC_SET_##wc(X, _FP_ZEROFRAC_##wc); \
486 break; \
487 case FP_CLS_INF: \
488 if (X##_s) \
489 { \
490 R##_s = 0; \
491 R##_c = FP_CLS_NAN; /* sNAN */ \
492 } \
493 else \
494 { \
495 R##_s = 0; \
496 R##_c = FP_CLS_INF; /* sqrt(+inf) = +inf */ \
497 } \
498 break; \
499 case FP_CLS_ZERO: \
500 R##_s = X##_s; \
501 R##_c = FP_CLS_ZERO; /* sqrt(+-0) = +-0 */ \
502 break; \
503 case FP_CLS_NORMAL: \
504 R##_s = 0; \
505 if (X##_s) \
506 { \
507 R##_c = FP_CLS_NAN; /* sNAN */ \
508 break; \
509 } \
510 R##_c = FP_CLS_NORMAL; \
511 if (X##_e & 1) \
512 _FP_FRAC_SLL_##wc(X, 1); \
513 R##_e = X##_e >> 1; \
514 _FP_FRAC_SET_##wc(S, _FP_ZEROFRAC_##wc); \
515 _FP_FRAC_SET_##wc(R, _FP_ZEROFRAC_##wc); \
516 q = _FP_OVERFLOW_##fs; \
517 _FP_FRAC_SLL_##wc(X, 1); \
518 _FP_SQRT_MEAT_##wc(R, S, T, X, q); \
519 _FP_FRAC_SRL_##wc(R, 1); \
520 } \
521 } while (0)
522
523/*
524 * Convert from FP to integer
525 */
526
527/* "When a NaN, infinity, large positive argument >= 2147483648.0, or
528 * large negative argument <= -2147483649.0 is converted to an integer,
529 * the invalid_current bit...should be set and fp_exception_IEEE_754 should
530 * be raised. If the floating point invalid trap is disabled, no trap occurs
531 * and a numerical result is generated: if the sign bit of the operand
532 * is 0, the result is 2147483647; if the sign bit of the operand is 1,
533 * the result is -2147483648."
534 * Similarly for conversion to extended ints, except that the boundaries
535 * are >= 2^63, <= -(2^63 + 1), and the results are 2^63 + 1 for s=0 and
536 * -2^63 for s=1.
537 * -- SPARC Architecture Manual V9, Appendix B, which specifies how
538 * SPARCs resolve implementation dependencies in the IEEE-754 spec.
539 * I don't believe that the code below follows this. I'm not even sure
540 * it's right!
541 * It doesn't cope with needing to convert to an n bit integer when there
542 * is no n bit integer type. Fortunately gcc provides long long so this
543 * isn't a problem for sparc32.
544 * I have, however, fixed its NaN handling to conform as above.
545 * -- PMM 02/1998
546 * NB: rsigned is not 'is r declared signed?' but 'should the value stored
547 * in r be signed or unsigned?'. r is always(?) declared unsigned.
548 * Comments below are mine, BTW -- PMM
549 */
550#define _FP_TO_INT(fs, wc, r, X, rsize, rsigned) \
551 do { \
552 switch (X##_c) \
553 { \
554 case FP_CLS_NORMAL: \
555 if (X##_e < 0) \
556 { \
557 /* case FP_CLS_NAN: see above! */ \
558 case FP_CLS_ZERO: \
559 r = 0; \
560 } \
561 else if (X##_e >= rsize - (rsigned != 0)) \
562 { /* overflow */ \
563 case FP_CLS_NAN: \
564 case FP_CLS_INF: \
565 if (rsigned) \
566 { \
567 r = 1; \
568 r <<= rsize - 1; \
569 r -= 1 - X##_s; \
570 } \
571 else \
572 { \
573 r = 0; \
574 if (!X##_s) \
575 r = ~r; \
576 } \
577 } \
578 else \
579 { \
580 if (_FP_W_TYPE_SIZE*wc < rsize) \
581 { \
582 _FP_FRAC_ASSEMBLE_##wc(r, X, rsize); \
583 r <<= X##_e - _FP_WFRACBITS_##fs; \
584 } \
585 else \
586 { \
587 if (X##_e >= _FP_WFRACBITS_##fs) \
588 _FP_FRAC_SLL_##wc(X, (X##_e - _FP_WFRACBITS_##fs + 1));\
589 else \
590 _FP_FRAC_SRL_##wc(X, (_FP_WFRACBITS_##fs - X##_e - 1));\
591 _FP_FRAC_ASSEMBLE_##wc(r, X, rsize); \
592 } \
593 if (rsigned && X##_s) \
594 r = -r; \
595 } \
596 break; \
597 } \
598 } while (0)
599
600#define _FP_FROM_INT(fs, wc, X, r, rsize, rtype) \
601 do { \
602 if (r) \
603 { \
604 X##_c = FP_CLS_NORMAL; \
605 \
606 if ((X##_s = (r < 0))) \
607 r = -r; \
608 /* Note that `r' is now considered unsigned, so we don't have \
609 to worry about the single signed overflow case. */ \
610 \
611 if (rsize <= _FP_W_TYPE_SIZE) \
612 __FP_CLZ(X##_e, r); \
613 else \
614 __FP_CLZ_2(X##_e, (_FP_W_TYPE)(r >> _FP_W_TYPE_SIZE), \
615 (_FP_W_TYPE)r); \
616 if (rsize < _FP_W_TYPE_SIZE) \
617 X##_e -= (_FP_W_TYPE_SIZE - rsize); \
618 X##_e = rsize - X##_e - 1; \
619 \
620 if (_FP_FRACBITS_##fs < rsize && _FP_WFRACBITS_##fs < X##_e) \
621 __FP_FRAC_SRS_1(r, (X##_e - _FP_WFRACBITS_##fs), rsize); \
622 r &= ~((_FP_W_TYPE)1 << X##_e); \
623 _FP_FRAC_DISASSEMBLE_##wc(X, ((unsigned rtype)r), rsize); \
624 _FP_FRAC_SLL_##wc(X, (_FP_WFRACBITS_##fs - X##_e - 1)); \
625 } \
626 else \
627 { \
628 X##_c = FP_CLS_ZERO, X##_s = 0; \
629 } \
630 } while (0)
631
632
633#define FP_CONV(dfs,sfs,dwc,swc,D,S) \
634 do { \
635 _FP_FRAC_CONV_##dwc##_##swc(dfs, sfs, D, S); \
636 D##_e = S##_e; \
637 D##_c = S##_c; \
638 D##_s = S##_s; \
639 } while (0)
640
641/*
642 * Helper primitives.
643 */
644
645/* Count leading zeros in a word. */
646
647#ifndef __FP_CLZ
648#if _FP_W_TYPE_SIZE < 64
649/* this is just to shut the compiler up about shifts > word length -- PMM 02/1998 */
650#define __FP_CLZ(r, x) \
651 do { \
652 _FP_W_TYPE _t = (x); \
653 r = _FP_W_TYPE_SIZE - 1; \
654 if (_t > 0xffff) r -= 16; \
655 if (_t > 0xffff) _t >>= 16; \
656 if (_t > 0xff) r -= 8; \
657 if (_t > 0xff) _t >>= 8; \
658 if (_t & 0xf0) r -= 4; \
659 if (_t & 0xf0) _t >>= 4; \
660 if (_t & 0xc) r -= 2; \
661 if (_t & 0xc) _t >>= 2; \
662 if (_t & 0x2) r -= 1; \
663 } while (0)
664#else /* not _FP_W_TYPE_SIZE < 64 */
665#define __FP_CLZ(r, x) \
666 do { \
667 _FP_W_TYPE _t = (x); \
668 r = _FP_W_TYPE_SIZE - 1; \
669 if (_t > 0xffffffff) r -= 32; \
670 if (_t > 0xffffffff) _t >>= 32; \
671 if (_t > 0xffff) r -= 16; \
672 if (_t > 0xffff) _t >>= 16; \
673 if (_t > 0xff) r -= 8; \
674 if (_t > 0xff) _t >>= 8; \
675 if (_t & 0xf0) r -= 4; \
676 if (_t & 0xf0) _t >>= 4; \
677 if (_t & 0xc) r -= 2; \
678 if (_t & 0xc) _t >>= 2; \
679 if (_t & 0x2) r -= 1; \
680 } while (0)
681#endif /* not _FP_W_TYPE_SIZE < 64 */
682#endif /* ndef __FP_CLZ */
683
684#define _FP_DIV_HELP_imm(q, r, n, d) \
685 do { \
686 q = n / d, r = n % d; \
687 } while (0)
688
diff --git a/arch/powerpc/math-emu/sfp-machine.h b/arch/powerpc/math-emu/sfp-machine.h
new file mode 100644
index 000000000000..4b17d83cfcdd
--- /dev/null
+++ b/arch/powerpc/math-emu/sfp-machine.h
@@ -0,0 +1,377 @@
1/* Machine-dependent software floating-point definitions. PPC version.
2 Copyright (C) 1997 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Library General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Library General Public License for more details.
14
15 You should have received a copy of the GNU Library General Public
16 License along with the GNU C Library; see the file COPYING.LIB. If
17 not, write to the Free Software Foundation, Inc.,
18 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19
20 Actually, this is a PPC (32bit) version, written based on the
21 i386, sparc, and sparc64 versions, by me,
22 Peter Maydell (pmaydell@chiark.greenend.org.uk).
23 Comments are by and large also mine, although they may be inaccurate.
24
25 In picking out asm fragments I've gone with the lowest common
26 denominator, which also happens to be the hardware I have :->
27 That is, a SPARC without hardware multiply and divide.
28 */
29
30/* basic word size definitions */
31#define _FP_W_TYPE_SIZE 32
32#define _FP_W_TYPE unsigned long
33#define _FP_WS_TYPE signed long
34#define _FP_I_TYPE long
35
36#define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2))
37#define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1))
38#define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2))
39
40/* You can optionally code some things like addition in asm. For
41 * example, i386 defines __FP_FRAC_ADD_2 as asm. If you don't
42 * then you get a fragment of C code [if you change an #ifdef 0
43 * in op-2.h] or a call to add_ssaaaa (see below).
44 * Good places to look for asm fragments to use are gcc and glibc.
45 * gcc's longlong.h is useful.
46 */
47
48/* We need to know how to multiply and divide. If the host word size
49 * is >= 2*fracbits you can use FP_MUL_MEAT_n_imm(t,R,X,Y) which
50 * codes the multiply with whatever gcc does to 'a * b'.
51 * _FP_MUL_MEAT_n_wide(t,R,X,Y,f) is used when you have an asm
52 * function that can multiply two 1W values and get a 2W result.
53 * Otherwise you're stuck with _FP_MUL_MEAT_n_hard(t,R,X,Y) which
54 * does bitshifting to avoid overflow.
55 * For division there is FP_DIV_MEAT_n_imm(t,R,X,Y,f) for word size
56 * >= 2*fracbits, where f is either _FP_DIV_HELP_imm or
57 * _FP_DIV_HELP_ldiv (see op-1.h).
58 * _FP_DIV_MEAT_udiv() is if you have asm to do 2W/1W => (1W, 1W).
59 * [GCC and glibc have longlong.h which has the asm macro udiv_qrnnd
60 * to do this.]
61 * In general, 'n' is the number of words required to hold the type,
62 * and 't' is either S, D or Q for single/double/quad.
63 * -- PMM
64 */
65/* Example: SPARC64:
66 * #define _FP_MUL_MEAT_S(R,X,Y) _FP_MUL_MEAT_1_imm(S,R,X,Y)
67 * #define _FP_MUL_MEAT_D(R,X,Y) _FP_MUL_MEAT_1_wide(D,R,X,Y,umul_ppmm)
68 * #define _FP_MUL_MEAT_Q(R,X,Y) _FP_MUL_MEAT_2_wide(Q,R,X,Y,umul_ppmm)
69 *
70 * #define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_imm(S,R,X,Y,_FP_DIV_HELP_imm)
71 * #define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_1_udiv(D,R,X,Y)
72 * #define _FP_DIV_MEAT_Q(R,X,Y) _FP_DIV_MEAT_2_udiv_64(Q,R,X,Y)
73 *
74 * Example: i386:
75 * #define _FP_MUL_MEAT_S(R,X,Y) _FP_MUL_MEAT_1_wide(S,R,X,Y,_i386_mul_32_64)
76 * #define _FP_MUL_MEAT_D(R,X,Y) _FP_MUL_MEAT_2_wide(D,R,X,Y,_i386_mul_32_64)
77 *
78 * #define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_udiv(S,R,X,Y,_i386_div_64_32)
79 * #define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_2_udiv_64(D,R,X,Y)
80 */
81
82#define _FP_MUL_MEAT_S(R,X,Y) _FP_MUL_MEAT_1_wide(S,R,X,Y,umul_ppmm)
83#define _FP_MUL_MEAT_D(R,X,Y) _FP_MUL_MEAT_2_wide(D,R,X,Y,umul_ppmm)
84
85#define _FP_DIV_MEAT_S(R,X,Y) _FP_DIV_MEAT_1_udiv(S,R,X,Y)
86#define _FP_DIV_MEAT_D(R,X,Y) _FP_DIV_MEAT_2_udiv_64(D,R,X,Y)
87
88/* These macros define what NaN looks like. They're supposed to expand to
89 * a comma-separated set of 32bit unsigned ints that encode NaN.
90 */
91#define _FP_NANFRAC_S _FP_QNANBIT_S
92#define _FP_NANFRAC_D _FP_QNANBIT_D, 0
93#define _FP_NANFRAC_Q _FP_QNANBIT_Q, 0, 0, 0
94
95#define _FP_KEEPNANFRACP 1
96
97/* This macro appears to be called when both X and Y are NaNs, and
98 * has to choose one and copy it to R. i386 goes for the larger of the
99 * two, sparc64 just picks Y. I don't understand this at all so I'll
100 * go with sparc64 because it's shorter :-> -- PMM
101 */
102#define _FP_CHOOSENAN(fs, wc, R, X, Y) \
103 do { \
104 R##_s = Y##_s; \
105 _FP_FRAC_COPY_##wc(R,Y); \
106 R##_c = FP_CLS_NAN; \
107 } while (0)
108
109
110extern void fp_unpack_d(long *, unsigned long *, unsigned long *,
111 long *, long *, void *);
112extern int fp_pack_d(void *, long, unsigned long, unsigned long, long, long);
113extern int fp_pack_ds(void *, long, unsigned long, unsigned long, long, long);
114
115#define __FP_UNPACK_RAW_1(fs, X, val) \
116 do { \
117 union _FP_UNION_##fs *_flo = \
118 (union _FP_UNION_##fs *)val; \
119 \
120 X##_f = _flo->bits.frac; \
121 X##_e = _flo->bits.exp; \
122 X##_s = _flo->bits.sign; \
123 } while (0)
124
125#define __FP_UNPACK_RAW_2(fs, X, val) \
126 do { \
127 union _FP_UNION_##fs *_flo = \
128 (union _FP_UNION_##fs *)val; \
129 \
130 X##_f0 = _flo->bits.frac0; \
131 X##_f1 = _flo->bits.frac1; \
132 X##_e = _flo->bits.exp; \
133 X##_s = _flo->bits.sign; \
134 } while (0)
135
136#define __FP_UNPACK_S(X,val) \
137 do { \
138 __FP_UNPACK_RAW_1(S,X,val); \
139 _FP_UNPACK_CANONICAL(S,1,X); \
140 } while (0)
141
142#define __FP_UNPACK_D(X,val) \
143 fp_unpack_d(&X##_s, &X##_f1, &X##_f0, &X##_e, &X##_c, val)
144
145#define __FP_PACK_RAW_1(fs, val, X) \
146 do { \
147 union _FP_UNION_##fs *_flo = \
148 (union _FP_UNION_##fs *)val; \
149 \
150 _flo->bits.frac = X##_f; \
151 _flo->bits.exp = X##_e; \
152 _flo->bits.sign = X##_s; \
153 } while (0)
154
155#define __FP_PACK_RAW_2(fs, val, X) \
156 do { \
157 union _FP_UNION_##fs *_flo = \
158 (union _FP_UNION_##fs *)val; \
159 \
160 _flo->bits.frac0 = X##_f0; \
161 _flo->bits.frac1 = X##_f1; \
162 _flo->bits.exp = X##_e; \
163 _flo->bits.sign = X##_s; \
164 } while (0)
165
166#include <linux/kernel.h>
167#include <linux/sched.h>
168
169#define __FPU_FPSCR (current->thread.fpscr.val)
170
171/* We only actually write to the destination register
172 * if exceptions signalled (if any) will not trap.
173 */
174#define __FPU_ENABLED_EXC \
175({ \
176 (__FPU_FPSCR >> 3) & 0x1f; \
177})
178
179#define __FPU_TRAP_P(bits) \
180 ((__FPU_ENABLED_EXC & (bits)) != 0)
181
182#define __FP_PACK_S(val,X) \
183({ int __exc = _FP_PACK_CANONICAL(S,1,X); \
184 if(!__exc || !__FPU_TRAP_P(__exc)) \
185 __FP_PACK_RAW_1(S,val,X); \
186 __exc; \
187})
188
189#define __FP_PACK_D(val,X) \
190 fp_pack_d(val, X##_s, X##_f1, X##_f0, X##_e, X##_c)
191
192#define __FP_PACK_DS(val,X) \
193 fp_pack_ds(val, X##_s, X##_f1, X##_f0, X##_e, X##_c)
194
195/* Obtain the current rounding mode. */
196#define FP_ROUNDMODE \
197({ \
198 __FPU_FPSCR & 0x3; \
199})
200
201/* the asm fragments go here: all these are taken from glibc-2.0.5's
202 * stdlib/longlong.h
203 */
204
205#include <linux/types.h>
206#include <asm/byteorder.h>
207
208/* add_ssaaaa is used in op-2.h and should be equivalent to
209 * #define add_ssaaaa(sh,sl,ah,al,bh,bl) (sh = ah+bh+ (( sl = al+bl) < al))
210 * add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1,
211 * high_addend_2, low_addend_2) adds two UWtype integers, composed by
212 * HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2
213 * respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow
214 * (i.e. carry out) is not stored anywhere, and is lost.
215 */
216#define add_ssaaaa(sh, sl, ah, al, bh, bl) \
217 do { \
218 if (__builtin_constant_p (bh) && (bh) == 0) \
219 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{aze|addze} %0,%2" \
220 : "=r" ((USItype)(sh)), \
221 "=&r" ((USItype)(sl)) \
222 : "%r" ((USItype)(ah)), \
223 "%r" ((USItype)(al)), \
224 "rI" ((USItype)(bl))); \
225 else if (__builtin_constant_p (bh) && (bh) ==~(USItype) 0) \
226 __asm__ ("{a%I4|add%I4c} %1,%3,%4\n\t{ame|addme} %0,%2" \
227 : "=r" ((USItype)(sh)), \
228 "=&r" ((USItype)(sl)) \
229 : "%r" ((USItype)(ah)), \
230 "%r" ((USItype)(al)), \
231 "rI" ((USItype)(bl))); \
232 else \
233 __asm__ ("{a%I5|add%I5c} %1,%4,%5\n\t{ae|adde} %0,%2,%3" \
234 : "=r" ((USItype)(sh)), \
235 "=&r" ((USItype)(sl)) \
236 : "%r" ((USItype)(ah)), \
237 "r" ((USItype)(bh)), \
238 "%r" ((USItype)(al)), \
239 "rI" ((USItype)(bl))); \
240 } while (0)
241
242/* sub_ddmmss is used in op-2.h and udivmodti4.c and should be equivalent to
243 * #define sub_ddmmss(sh, sl, ah, al, bh, bl) (sh = ah-bh - ((sl = al-bl) > al))
244 * sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend,
245 * high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers,
246 * composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and
247 * LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE
248 * and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere,
249 * and is lost.
250 */
251#define sub_ddmmss(sh, sl, ah, al, bh, bl) \
252 do { \
253 if (__builtin_constant_p (ah) && (ah) == 0) \
254 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfze|subfze} %0,%2" \
255 : "=r" ((USItype)(sh)), \
256 "=&r" ((USItype)(sl)) \
257 : "r" ((USItype)(bh)), \
258 "rI" ((USItype)(al)), \
259 "r" ((USItype)(bl))); \
260 else if (__builtin_constant_p (ah) && (ah) ==~(USItype) 0) \
261 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{sfme|subfme} %0,%2" \
262 : "=r" ((USItype)(sh)), \
263 "=&r" ((USItype)(sl)) \
264 : "r" ((USItype)(bh)), \
265 "rI" ((USItype)(al)), \
266 "r" ((USItype)(bl))); \
267 else if (__builtin_constant_p (bh) && (bh) == 0) \
268 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{ame|addme} %0,%2" \
269 : "=r" ((USItype)(sh)), \
270 "=&r" ((USItype)(sl)) \
271 : "r" ((USItype)(ah)), \
272 "rI" ((USItype)(al)), \
273 "r" ((USItype)(bl))); \
274 else if (__builtin_constant_p (bh) && (bh) ==~(USItype) 0) \
275 __asm__ ("{sf%I3|subf%I3c} %1,%4,%3\n\t{aze|addze} %0,%2" \
276 : "=r" ((USItype)(sh)), \
277 "=&r" ((USItype)(sl)) \
278 : "r" ((USItype)(ah)), \
279 "rI" ((USItype)(al)), \
280 "r" ((USItype)(bl))); \
281 else \
282 __asm__ ("{sf%I4|subf%I4c} %1,%5,%4\n\t{sfe|subfe} %0,%3,%2" \
283 : "=r" ((USItype)(sh)), \
284 "=&r" ((USItype)(sl)) \
285 : "r" ((USItype)(ah)), \
286 "r" ((USItype)(bh)), \
287 "rI" ((USItype)(al)), \
288 "r" ((USItype)(bl))); \
289 } while (0)
290
291/* asm fragments for mul and div */
292
293/* umul_ppmm(high_prod, low_prod, multipler, multiplicand) multiplies two
294 * UWtype integers MULTIPLER and MULTIPLICAND, and generates a two UWtype
295 * word product in HIGH_PROD and LOW_PROD.
296 */
297#define umul_ppmm(ph, pl, m0, m1) \
298 do { \
299 USItype __m0 = (m0), __m1 = (m1); \
300 __asm__ ("mulhwu %0,%1,%2" \
301 : "=r" ((USItype)(ph)) \
302 : "%r" (__m0), \
303 "r" (__m1)); \
304 (pl) = __m0 * __m1; \
305 } while (0)
306
307/* udiv_qrnnd(quotient, remainder, high_numerator, low_numerator,
308 * denominator) divides a UDWtype, composed by the UWtype integers
309 * HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient
310 * in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less
311 * than DENOMINATOR for correct operation. If, in addition, the most
312 * significant bit of DENOMINATOR must be 1, then the pre-processor symbol
313 * UDIV_NEEDS_NORMALIZATION is defined to 1.
314 */
315#define udiv_qrnnd(q, r, n1, n0, d) \
316 do { \
317 UWtype __d1, __d0, __q1, __q0, __r1, __r0, __m; \
318 __d1 = __ll_highpart (d); \
319 __d0 = __ll_lowpart (d); \
320 \
321 __r1 = (n1) % __d1; \
322 __q1 = (n1) / __d1; \
323 __m = (UWtype) __q1 * __d0; \
324 __r1 = __r1 * __ll_B | __ll_highpart (n0); \
325 if (__r1 < __m) \
326 { \
327 __q1--, __r1 += (d); \
328 if (__r1 >= (d)) /* we didn't get carry when adding to __r1 */ \
329 if (__r1 < __m) \
330 __q1--, __r1 += (d); \
331 } \
332 __r1 -= __m; \
333 \
334 __r0 = __r1 % __d1; \
335 __q0 = __r1 / __d1; \
336 __m = (UWtype) __q0 * __d0; \
337 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \
338 if (__r0 < __m) \
339 { \
340 __q0--, __r0 += (d); \
341 if (__r0 >= (d)) \
342 if (__r0 < __m) \
343 __q0--, __r0 += (d); \
344 } \
345 __r0 -= __m; \
346 \
347 (q) = (UWtype) __q1 * __ll_B | __q0; \
348 (r) = __r0; \
349 } while (0)
350
351#define UDIV_NEEDS_NORMALIZATION 1
352
353#define abort() \
354 return 0
355
356#ifdef __BIG_ENDIAN
357#define __BYTE_ORDER __BIG_ENDIAN
358#else
359#define __BYTE_ORDER __LITTLE_ENDIAN
360#endif
361
362/* Exception flags. */
363#define EFLAG_INVALID (1 << (31 - 2))
364#define EFLAG_OVERFLOW (1 << (31 - 3))
365#define EFLAG_UNDERFLOW (1 << (31 - 4))
366#define EFLAG_DIVZERO (1 << (31 - 5))
367#define EFLAG_INEXACT (1 << (31 - 6))
368
369#define EFLAG_VXSNAN (1 << (31 - 7))
370#define EFLAG_VXISI (1 << (31 - 8))
371#define EFLAG_VXIDI (1 << (31 - 9))
372#define EFLAG_VXZDZ (1 << (31 - 10))
373#define EFLAG_VXIMZ (1 << (31 - 11))
374#define EFLAG_VXVC (1 << (31 - 12))
375#define EFLAG_VXSOFT (1 << (31 - 21))
376#define EFLAG_VXSQRT (1 << (31 - 22))
377#define EFLAG_VXCVI (1 << (31 - 23))
diff --git a/arch/powerpc/math-emu/single.h b/arch/powerpc/math-emu/single.h
new file mode 100644
index 000000000000..f19d99451815
--- /dev/null
+++ b/arch/powerpc/math-emu/single.h
@@ -0,0 +1,66 @@
1/*
2 * Definitions for IEEE Single Precision
3 */
4
5#if _FP_W_TYPE_SIZE < 32
6#error "Here's a nickel kid. Go buy yourself a real computer."
7#endif
8
9#define _FP_FRACBITS_S 24
10#define _FP_FRACXBITS_S (_FP_W_TYPE_SIZE - _FP_FRACBITS_S)
11#define _FP_WFRACBITS_S (_FP_WORKBITS + _FP_FRACBITS_S)
12#define _FP_WFRACXBITS_S (_FP_W_TYPE_SIZE - _FP_WFRACBITS_S)
13#define _FP_EXPBITS_S 8
14#define _FP_EXPBIAS_S 127
15#define _FP_EXPMAX_S 255
16#define _FP_QNANBIT_S ((_FP_W_TYPE)1 << (_FP_FRACBITS_S-2))
17#define _FP_IMPLBIT_S ((_FP_W_TYPE)1 << (_FP_FRACBITS_S-1))
18#define _FP_OVERFLOW_S ((_FP_W_TYPE)1 << (_FP_WFRACBITS_S))
19
20/* The implementation of _FP_MUL_MEAT_S and _FP_DIV_MEAT_S should be
21 chosen by the target machine. */
22
23union _FP_UNION_S
24{
25 float flt;
26 struct {
27#if __BYTE_ORDER == __BIG_ENDIAN
28 unsigned sign : 1;
29 unsigned exp : _FP_EXPBITS_S;
30 unsigned frac : _FP_FRACBITS_S - (_FP_IMPLBIT_S != 0);
31#else
32 unsigned frac : _FP_FRACBITS_S - (_FP_IMPLBIT_S != 0);
33 unsigned exp : _FP_EXPBITS_S;
34 unsigned sign : 1;
35#endif
36 } bits __attribute__((packed));
37};
38
39#define FP_DECL_S(X) _FP_DECL(1,X)
40#define FP_UNPACK_RAW_S(X,val) _FP_UNPACK_RAW_1(S,X,val)
41#define FP_PACK_RAW_S(val,X) _FP_PACK_RAW_1(S,val,X)
42
43#define FP_UNPACK_S(X,val) \
44 do { \
45 _FP_UNPACK_RAW_1(S,X,val); \
46 _FP_UNPACK_CANONICAL(S,1,X); \
47 } while (0)
48
49#define FP_PACK_S(val,X) \
50 do { \
51 _FP_PACK_CANONICAL(S,1,X); \
52 _FP_PACK_RAW_1(S,val,X); \
53 } while (0)
54
55#define FP_NEG_S(R,X) _FP_NEG(S,1,R,X)
56#define FP_ADD_S(R,X,Y) _FP_ADD(S,1,R,X,Y)
57#define FP_SUB_S(R,X,Y) _FP_SUB(S,1,R,X,Y)
58#define FP_MUL_S(R,X,Y) _FP_MUL(S,1,R,X,Y)
59#define FP_DIV_S(R,X,Y) _FP_DIV(S,1,R,X,Y)
60#define FP_SQRT_S(R,X) _FP_SQRT(S,1,R,X)
61
62#define FP_CMP_S(r,X,Y,un) _FP_CMP(S,1,r,X,Y,un)
63#define FP_CMP_EQ_S(r,X,Y) _FP_CMP_EQ(S,1,r,X,Y)
64
65#define FP_TO_INT_S(r,X,rsz,rsg) _FP_TO_INT(S,1,r,X,rsz,rsg)
66#define FP_FROM_INT_S(X,r,rs,rt) _FP_FROM_INT(S,1,X,r,rs,rt)
diff --git a/arch/powerpc/math-emu/soft-fp.h b/arch/powerpc/math-emu/soft-fp.h
new file mode 100644
index 000000000000..cca39598f873
--- /dev/null
+++ b/arch/powerpc/math-emu/soft-fp.h
@@ -0,0 +1,104 @@
1#ifndef SOFT_FP_H
2#define SOFT_FP_H
3
4#include "sfp-machine.h"
5
6#define _FP_WORKBITS 3
7#define _FP_WORK_LSB ((_FP_W_TYPE)1 << 3)
8#define _FP_WORK_ROUND ((_FP_W_TYPE)1 << 2)
9#define _FP_WORK_GUARD ((_FP_W_TYPE)1 << 1)
10#define _FP_WORK_STICKY ((_FP_W_TYPE)1 << 0)
11
12#ifndef FP_RND_NEAREST
13# define FP_RND_NEAREST 0
14# define FP_RND_ZERO 1
15# define FP_RND_PINF 2
16# define FP_RND_MINF 3
17#ifndef FP_ROUNDMODE
18# define FP_ROUNDMODE FP_RND_NEAREST
19#endif
20#endif
21
22#define _FP_ROUND_NEAREST(wc, X) \
23({ int __ret = 0; \
24 int __frac = _FP_FRAC_LOW_##wc(X) & 15; \
25 if (__frac & 7) { \
26 __ret = EFLAG_INEXACT; \
27 if ((__frac & 7) != _FP_WORK_ROUND) \
28 _FP_FRAC_ADDI_##wc(X, _FP_WORK_ROUND); \
29 else if (__frac & _FP_WORK_LSB) \
30 _FP_FRAC_ADDI_##wc(X, _FP_WORK_ROUND); \
31 } \
32 __ret; \
33})
34
35#define _FP_ROUND_ZERO(wc, X) \
36({ int __ret = 0; \
37 if (_FP_FRAC_LOW_##wc(X) & 7) \
38 __ret = EFLAG_INEXACT; \
39 __ret; \
40})
41
42#define _FP_ROUND_PINF(wc, X) \
43({ int __ret = EFLAG_INEXACT; \
44 if (!X##_s && (_FP_FRAC_LOW_##wc(X) & 7)) \
45 _FP_FRAC_ADDI_##wc(X, _FP_WORK_LSB); \
46 else __ret = 0; \
47 __ret; \
48})
49
50#define _FP_ROUND_MINF(wc, X) \
51({ int __ret = EFLAG_INEXACT; \
52 if (X##_s && (_FP_FRAC_LOW_##wc(X) & 7)) \
53 _FP_FRAC_ADDI_##wc(X, _FP_WORK_LSB); \
54 else __ret = 0; \
55 __ret; \
56})
57
58#define _FP_ROUND(wc, X) \
59({ int __ret = 0; \
60 switch (FP_ROUNDMODE) \
61 { \
62 case FP_RND_NEAREST: \
63 __ret |= _FP_ROUND_NEAREST(wc,X); \
64 break; \
65 case FP_RND_ZERO: \
66 __ret |= _FP_ROUND_ZERO(wc,X); \
67 break; \
68 case FP_RND_PINF: \
69 __ret |= _FP_ROUND_PINF(wc,X); \
70 break; \
71 case FP_RND_MINF: \
72 __ret |= _FP_ROUND_MINF(wc,X); \
73 break; \
74 }; \
75 __ret; \
76})
77
78#define FP_CLS_NORMAL 0
79#define FP_CLS_ZERO 1
80#define FP_CLS_INF 2
81#define FP_CLS_NAN 3
82
83#define _FP_CLS_COMBINE(x,y) (((x) << 2) | (y))
84
85#include "op-1.h"
86#include "op-2.h"
87#include "op-4.h"
88#include "op-common.h"
89
90/* Sigh. Silly things longlong.h needs. */
91#define UWtype _FP_W_TYPE
92#define W_TYPE_SIZE _FP_W_TYPE_SIZE
93
94typedef int SItype __attribute__((mode(SI)));
95typedef int DItype __attribute__((mode(DI)));
96typedef unsigned int USItype __attribute__((mode(SI)));
97typedef unsigned int UDItype __attribute__((mode(DI)));
98#if _FP_W_TYPE_SIZE == 32
99typedef unsigned int UHWtype __attribute__((mode(HI)));
100#elif _FP_W_TYPE_SIZE == 64
101typedef USItype UHWtype;
102#endif
103
104#endif
diff --git a/arch/powerpc/math-emu/stfd.c b/arch/powerpc/math-emu/stfd.c
new file mode 100644
index 000000000000..3f8c2558a9e8
--- /dev/null
+++ b/arch/powerpc/math-emu/stfd.c
@@ -0,0 +1,20 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6stfd(void *frS, void *ea)
7{
8#if 0
9#ifdef DEBUG
10 printk("%s: S %p, ea %p: ", __FUNCTION__, frS, ea);
11 dump_double(frS);
12 printk("\n");
13#endif
14#endif
15
16 if (copy_to_user(ea, frS, sizeof(double)))
17 return -EFAULT;
18
19 return 0;
20}
diff --git a/arch/powerpc/math-emu/stfiwx.c b/arch/powerpc/math-emu/stfiwx.c
new file mode 100644
index 000000000000..95caaeec6a08
--- /dev/null
+++ b/arch/powerpc/math-emu/stfiwx.c
@@ -0,0 +1,16 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5int
6stfiwx(u32 *frS, void *ea)
7{
8#ifdef DEBUG
9 printk("%s: %p %p\n", __FUNCTION__, frS, ea);
10#endif
11
12 if (copy_to_user(ea, &frS[1], sizeof(frS[1])))
13 return -EFAULT;
14
15 return 0;
16}
diff --git a/arch/powerpc/math-emu/stfs.c b/arch/powerpc/math-emu/stfs.c
new file mode 100644
index 000000000000..e87ca23c6dc3
--- /dev/null
+++ b/arch/powerpc/math-emu/stfs.c
@@ -0,0 +1,41 @@
1#include <linux/types.h>
2#include <linux/errno.h>
3#include <asm/uaccess.h>
4
5#include "soft-fp.h"
6#include "double.h"
7#include "single.h"
8
9int
10stfs(void *frS, void *ea)
11{
12 FP_DECL_D(A);
13 FP_DECL_S(R);
14 float f;
15 int err;
16
17#ifdef DEBUG
18 printk("%s: S %p, ea %p\n", __FUNCTION__, frS, ea);
19#endif
20
21 __FP_UNPACK_D(A, frS);
22
23#ifdef DEBUG
24 printk("A: %ld %lu %lu %ld (%ld)\n", A_s, A_f1, A_f0, A_e, A_c);
25#endif
26
27 FP_CONV(S, D, 1, 2, R, A);
28
29#ifdef DEBUG
30 printk("R: %ld %lu %ld (%ld)\n", R_s, R_f, R_e, R_c);
31#endif
32
33 err = _FP_PACK_CANONICAL(S, 1, R);
34 if (!err || !__FPU_TRAP_P(err)) {
35 __FP_PACK_RAW_1(S, &f, R);
36 if (copy_to_user(ea, &f, sizeof(float)))
37 return -EFAULT;
38 }
39
40 return err;
41}
diff --git a/arch/powerpc/math-emu/types.c b/arch/powerpc/math-emu/types.c
new file mode 100644
index 000000000000..e1ed15d829db
--- /dev/null
+++ b/arch/powerpc/math-emu/types.c
@@ -0,0 +1,51 @@
1#include "soft-fp.h"
2#include "double.h"
3#include "single.h"
4
5void
6fp_unpack_d(long *_s, unsigned long *_f1, unsigned long *_f0,
7 long *_e, long *_c, void *val)
8{
9 FP_DECL_D(X);
10
11 __FP_UNPACK_RAW_2(D, X, val);
12
13 _FP_UNPACK_CANONICAL(D, 2, X);
14
15 *_s = X_s;
16 *_f1 = X_f1;
17 *_f0 = X_f0;
18 *_e = X_e;
19 *_c = X_c;
20}
21
22int
23fp_pack_d(void *val, long X_s, unsigned long X_f1,
24 unsigned long X_f0, long X_e, long X_c)
25{
26 int exc;
27
28 exc = _FP_PACK_CANONICAL(D, 2, X);
29 if (!exc || !__FPU_TRAP_P(exc))
30 __FP_PACK_RAW_2(D, val, X);
31 return exc;
32}
33
34int
35fp_pack_ds(void *val, long X_s, unsigned long X_f1,
36 unsigned long X_f0, long X_e, long X_c)
37{
38 FP_DECL_S(__X);
39 int exc;
40
41 FP_CONV(S, D, 1, 2, __X, X);
42 exc = _FP_PACK_CANONICAL(S, 1, __X);
43 if (!exc || !__FPU_TRAP_P(exc)) {
44 _FP_UNPACK_CANONICAL(S, 1, __X);
45 FP_CONV(D, S, 2, 1, X, __X);
46 exc |= _FP_PACK_CANONICAL(D, 2, X);
47 if (!exc || !__FPU_TRAP_P(exc))
48 __FP_PACK_RAW_2(D, val, X);
49 }
50 return exc;
51}
diff --git a/arch/powerpc/math-emu/udivmodti4.c b/arch/powerpc/math-emu/udivmodti4.c
new file mode 100644
index 000000000000..7e112dc1e2f2
--- /dev/null
+++ b/arch/powerpc/math-emu/udivmodti4.c
@@ -0,0 +1,191 @@
1/* This has so very few changes over libgcc2's __udivmoddi4 it isn't funny. */
2
3#include "soft-fp.h"
4
5#undef count_leading_zeros
6#define count_leading_zeros __FP_CLZ
7
8void
9_fp_udivmodti4(_FP_W_TYPE q[2], _FP_W_TYPE r[2],
10 _FP_W_TYPE n1, _FP_W_TYPE n0,
11 _FP_W_TYPE d1, _FP_W_TYPE d0)
12{
13 _FP_W_TYPE q0, q1, r0, r1;
14 _FP_I_TYPE b, bm;
15
16 if (d1 == 0)
17 {
18#if !UDIV_NEEDS_NORMALIZATION
19 if (d0 > n1)
20 {
21 /* 0q = nn / 0D */
22
23 udiv_qrnnd (q0, n0, n1, n0, d0);
24 q1 = 0;
25
26 /* Remainder in n0. */
27 }
28 else
29 {
30 /* qq = NN / 0d */
31
32 if (d0 == 0)
33 d0 = 1 / d0; /* Divide intentionally by zero. */
34
35 udiv_qrnnd (q1, n1, 0, n1, d0);
36 udiv_qrnnd (q0, n0, n1, n0, d0);
37
38 /* Remainder in n0. */
39 }
40
41 r0 = n0;
42 r1 = 0;
43
44#else /* UDIV_NEEDS_NORMALIZATION */
45
46 if (d0 > n1)
47 {
48 /* 0q = nn / 0D */
49
50 count_leading_zeros (bm, d0);
51
52 if (bm != 0)
53 {
54 /* Normalize, i.e. make the most significant bit of the
55 denominator set. */
56
57 d0 = d0 << bm;
58 n1 = (n1 << bm) | (n0 >> (_FP_W_TYPE_SIZE - bm));
59 n0 = n0 << bm;
60 }
61
62 udiv_qrnnd (q0, n0, n1, n0, d0);
63 q1 = 0;
64
65 /* Remainder in n0 >> bm. */
66 }
67 else
68 {
69 /* qq = NN / 0d */
70
71 if (d0 == 0)
72 d0 = 1 / d0; /* Divide intentionally by zero. */
73
74 count_leading_zeros (bm, d0);
75
76 if (bm == 0)
77 {
78 /* From (n1 >= d0) /\ (the most significant bit of d0 is set),
79 conclude (the most significant bit of n1 is set) /\ (the
80 leading quotient digit q1 = 1).
81
82 This special case is necessary, not an optimization.
83 (Shifts counts of SI_TYPE_SIZE are undefined.) */
84
85 n1 -= d0;
86 q1 = 1;
87 }
88 else
89 {
90 _FP_W_TYPE n2;
91
92 /* Normalize. */
93
94 b = _FP_W_TYPE_SIZE - bm;
95
96 d0 = d0 << bm;
97 n2 = n1 >> b;
98 n1 = (n1 << bm) | (n0 >> b);
99 n0 = n0 << bm;
100
101 udiv_qrnnd (q1, n1, n2, n1, d0);
102 }
103
104 /* n1 != d0... */
105
106 udiv_qrnnd (q0, n0, n1, n0, d0);
107
108 /* Remainder in n0 >> bm. */
109 }
110
111 r0 = n0 >> bm;
112 r1 = 0;
113#endif /* UDIV_NEEDS_NORMALIZATION */
114 }
115 else
116 {
117 if (d1 > n1)
118 {
119 /* 00 = nn / DD */
120
121 q0 = 0;
122 q1 = 0;
123
124 /* Remainder in n1n0. */
125 r0 = n0;
126 r1 = n1;
127 }
128 else
129 {
130 /* 0q = NN / dd */
131
132 count_leading_zeros (bm, d1);
133 if (bm == 0)
134 {
135 /* From (n1 >= d1) /\ (the most significant bit of d1 is set),
136 conclude (the most significant bit of n1 is set) /\ (the
137 quotient digit q0 = 0 or 1).
138
139 This special case is necessary, not an optimization. */
140
141 /* The condition on the next line takes advantage of that
142 n1 >= d1 (true due to program flow). */
143 if (n1 > d1 || n0 >= d0)
144 {
145 q0 = 1;
146 sub_ddmmss (n1, n0, n1, n0, d1, d0);
147 }
148 else
149 q0 = 0;
150
151 q1 = 0;
152
153 r0 = n0;
154 r1 = n1;
155 }
156 else
157 {
158 _FP_W_TYPE m1, m0, n2;
159
160 /* Normalize. */
161
162 b = _FP_W_TYPE_SIZE - bm;
163
164 d1 = (d1 << bm) | (d0 >> b);
165 d0 = d0 << bm;
166 n2 = n1 >> b;
167 n1 = (n1 << bm) | (n0 >> b);
168 n0 = n0 << bm;
169
170 udiv_qrnnd (q0, n1, n2, n1, d1);
171 umul_ppmm (m1, m0, q0, d0);
172
173 if (m1 > n1 || (m1 == n1 && m0 > n0))
174 {
175 q0--;
176 sub_ddmmss (m1, m0, m1, m0, d1, d0);
177 }
178
179 q1 = 0;
180
181 /* Remainder in (n1n0 - m1m0) >> bm. */
182 sub_ddmmss (n1, n0, n1, n0, m1, m0);
183 r0 = (n1 << b) | (n0 >> bm);
184 r1 = n1 >> bm;
185 }
186 }
187 }
188
189 q[0] = q0; q[1] = q1;
190 r[0] = r0, r[1] = r1;
191}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-14 23:25:22 -0500