Commit 20495218834824723ef81306c6e1fd27fc3ae560
1 parent
158142c2
use the generic soft float code
git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@1333 c046a42c-6fe2-441c-8c8c-71466251a162
Showing
13 changed files
with
84 additions
and
1352 deletions
Too many changes to show.
To preserve performance only 13 of 15 files are displayed.
target-arm/nwfpe/ARM-gcc.h deleted
100644 → 0
1 | -/* | |
2 | -------------------------------------------------------------------------------- | |
3 | -The macro `BITS64' can be defined to indicate that 64-bit integer types are | |
4 | -supported by the compiler. | |
5 | -------------------------------------------------------------------------------- | |
6 | -*/ | |
7 | -#define BITS64 | |
8 | - | |
9 | -/* | |
10 | -------------------------------------------------------------------------------- | |
11 | -Each of the following `typedef's defines the most convenient type that holds | |
12 | -integers of at least as many bits as specified. For example, `uint8' should | |
13 | -be the most convenient type that can hold unsigned integers of as many as | |
14 | -8 bits. The `flag' type must be able to hold either a 0 or 1. For most | |
15 | -implementations of C, `flag', `uint8', and `int8' should all be `typedef'ed | |
16 | -to the same as `int'. | |
17 | -------------------------------------------------------------------------------- | |
18 | -*/ | |
19 | -typedef char flag; | |
20 | -typedef unsigned char uint8; | |
21 | -typedef signed char int8; | |
22 | -typedef int uint16; | |
23 | -typedef int int16; | |
24 | -typedef unsigned int uint32; | |
25 | -typedef signed int int32; | |
26 | -#ifdef BITS64 | |
27 | -typedef unsigned long long int bits64; | |
28 | -typedef signed long long int sbits64; | |
29 | -#endif | |
30 | - | |
31 | -/* | |
32 | -------------------------------------------------------------------------------- | |
33 | -Each of the following `typedef's defines a type that holds integers | |
34 | -of _exactly_ the number of bits specified. For instance, for most | |
35 | -implementation of C, `bits16' and `sbits16' should be `typedef'ed to | |
36 | -`unsigned short int' and `signed short int' (or `short int'), respectively. | |
37 | -------------------------------------------------------------------------------- | |
38 | -*/ | |
39 | -typedef unsigned char bits8; | |
40 | -typedef signed char sbits8; | |
41 | -typedef unsigned short int bits16; | |
42 | -typedef signed short int sbits16; | |
43 | -typedef unsigned int bits32; | |
44 | -typedef signed int sbits32; | |
45 | -#ifdef BITS64 | |
46 | -typedef unsigned long long int uint64; | |
47 | -typedef signed long long int int64; | |
48 | -#endif | |
49 | - | |
50 | -#ifdef BITS64 | |
51 | -/* | |
52 | -------------------------------------------------------------------------------- | |
53 | -The `LIT64' macro takes as its argument a textual integer literal and if | |
54 | -necessary ``marks'' the literal as having a 64-bit integer type. For | |
55 | -example, the Gnu C Compiler (`gcc') requires that 64-bit literals be | |
56 | -appended with the letters `LL' standing for `long long', which is `gcc's | |
57 | -name for the 64-bit integer type. Some compilers may allow `LIT64' to be | |
58 | -defined as the identity macro: `#define LIT64( a ) a'. | |
59 | -------------------------------------------------------------------------------- | |
60 | -*/ | |
61 | -#define LIT64( a ) a##LL | |
62 | -#endif | |
63 | - | |
64 | -/* | |
65 | -------------------------------------------------------------------------------- | |
66 | -The macro `INLINE' can be used before functions that should be inlined. If | |
67 | -a compiler does not support explicit inlining, this macro should be defined | |
68 | -to be `static'. | |
69 | -------------------------------------------------------------------------------- | |
70 | -*/ | |
71 | -#define INLINE extern __inline__ | |
72 | - | |
73 | - | |
74 | -/* For use as a GCC soft-float library we need some special function names. */ | |
75 | - | |
76 | -#ifdef __LIBFLOAT__ | |
77 | - | |
78 | -/* Some 32-bit ops can be mapped straight across by just changing the name. */ | |
79 | -#define float32_add __addsf3 | |
80 | -#define float32_sub __subsf3 | |
81 | -#define float32_mul __mulsf3 | |
82 | -#define float32_div __divsf3 | |
83 | -#define int32_to_float32 __floatsisf | |
84 | -#define float32_to_int32_round_to_zero __fixsfsi | |
85 | -#define float32_to_uint32_round_to_zero __fixunssfsi | |
86 | - | |
87 | -/* These ones go through the glue code. To avoid namespace pollution | |
88 | - we rename the internal functions too. */ | |
89 | -#define float32_eq ___float32_eq | |
90 | -#define float32_le ___float32_le | |
91 | -#define float32_lt ___float32_lt | |
92 | - | |
93 | -/* All the 64-bit ops have to go through the glue, so we pull the same | |
94 | - trick. */ | |
95 | -#define float64_add ___float64_add | |
96 | -#define float64_sub ___float64_sub | |
97 | -#define float64_mul ___float64_mul | |
98 | -#define float64_div ___float64_div | |
99 | -#define int32_to_float64 ___int32_to_float64 | |
100 | -#define float64_to_int32_round_to_zero ___float64_to_int32_round_to_zero | |
101 | -#define float64_to_uint32_round_to_zero ___float64_to_uint32_round_to_zero | |
102 | -#define float64_to_float32 ___float64_to_float32 | |
103 | -#define float32_to_float64 ___float32_to_float64 | |
104 | -#define float64_eq ___float64_eq | |
105 | -#define float64_le ___float64_le | |
106 | -#define float64_lt ___float64_lt | |
107 | - | |
108 | -#if 0 | |
109 | -#define float64_add __adddf3 | |
110 | -#define float64_sub __subdf3 | |
111 | -#define float64_mul __muldf3 | |
112 | -#define float64_div __divdf3 | |
113 | -#define int32_to_float64 __floatsidf | |
114 | -#define float64_to_int32_round_to_zero __fixdfsi | |
115 | -#define float64_to_uint32_round_to_zero __fixunsdfsi | |
116 | -#define float64_to_float32 __truncdfsf2 | |
117 | -#define float32_to_float64 __extendsfdf2 | |
118 | -#endif | |
119 | - | |
120 | -#endif |
target-arm/nwfpe/double_cpdo.c
... | ... | @@ -53,7 +53,7 @@ unsigned int DoubleCPDO(const unsigned int opcode) |
53 | 53 | switch (fpa11->fType[Fm]) |
54 | 54 | { |
55 | 55 | case typeSingle: |
56 | - rFm = float32_to_float64(fpa11->fpreg[Fm].fSingle); | |
56 | + rFm = float32_to_float64(fpa11->fpreg[Fm].fSingle, &fpa11->fp_status); | |
57 | 57 | break; |
58 | 58 | |
59 | 59 | case typeDouble: |
... | ... | @@ -79,7 +79,7 @@ unsigned int DoubleCPDO(const unsigned int opcode) |
79 | 79 | switch (fpa11->fType[Fn]) |
80 | 80 | { |
81 | 81 | case typeSingle: |
82 | - rFn = float32_to_float64(fpa11->fpreg[Fn].fSingle); | |
82 | + rFn = float32_to_float64(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status); | |
83 | 83 | break; |
84 | 84 | |
85 | 85 | case typeDouble: |
... | ... | @@ -96,30 +96,30 @@ unsigned int DoubleCPDO(const unsigned int opcode) |
96 | 96 | { |
97 | 97 | /* dyadic opcodes */ |
98 | 98 | case ADF_CODE: |
99 | - fpa11->fpreg[Fd].fDouble = float64_add(rFn,rFm); | |
99 | + fpa11->fpreg[Fd].fDouble = float64_add(rFn,rFm, &fpa11->fp_status); | |
100 | 100 | break; |
101 | 101 | |
102 | 102 | case MUF_CODE: |
103 | 103 | case FML_CODE: |
104 | - fpa11->fpreg[Fd].fDouble = float64_mul(rFn,rFm); | |
104 | + fpa11->fpreg[Fd].fDouble = float64_mul(rFn,rFm, &fpa11->fp_status); | |
105 | 105 | break; |
106 | 106 | |
107 | 107 | case SUF_CODE: |
108 | - fpa11->fpreg[Fd].fDouble = float64_sub(rFn,rFm); | |
108 | + fpa11->fpreg[Fd].fDouble = float64_sub(rFn,rFm, &fpa11->fp_status); | |
109 | 109 | break; |
110 | 110 | |
111 | 111 | case RSF_CODE: |
112 | - fpa11->fpreg[Fd].fDouble = float64_sub(rFm,rFn); | |
112 | + fpa11->fpreg[Fd].fDouble = float64_sub(rFm,rFn, &fpa11->fp_status); | |
113 | 113 | break; |
114 | 114 | |
115 | 115 | case DVF_CODE: |
116 | 116 | case FDV_CODE: |
117 | - fpa11->fpreg[Fd].fDouble = float64_div(rFn,rFm); | |
117 | + fpa11->fpreg[Fd].fDouble = float64_div(rFn,rFm, &fpa11->fp_status); | |
118 | 118 | break; |
119 | 119 | |
120 | 120 | case RDF_CODE: |
121 | 121 | case FRD_CODE: |
122 | - fpa11->fpreg[Fd].fDouble = float64_div(rFm,rFn); | |
122 | + fpa11->fpreg[Fd].fDouble = float64_div(rFm,rFn, &fpa11->fp_status); | |
123 | 123 | break; |
124 | 124 | |
125 | 125 | #if 0 |
... | ... | @@ -133,7 +133,7 @@ unsigned int DoubleCPDO(const unsigned int opcode) |
133 | 133 | #endif |
134 | 134 | |
135 | 135 | case RMF_CODE: |
136 | - fpa11->fpreg[Fd].fDouble = float64_rem(rFn,rFm); | |
136 | + fpa11->fpreg[Fd].fDouble = float64_rem(rFn,rFm, &fpa11->fp_status); | |
137 | 137 | break; |
138 | 138 | |
139 | 139 | #if 0 |
... | ... | @@ -173,11 +173,11 @@ unsigned int DoubleCPDO(const unsigned int opcode) |
173 | 173 | |
174 | 174 | case RND_CODE: |
175 | 175 | case URD_CODE: |
176 | - fpa11->fpreg[Fd].fDouble = float64_round_to_int(rFm); | |
176 | + fpa11->fpreg[Fd].fDouble = float64_round_to_int(rFm, &fpa11->fp_status); | |
177 | 177 | break; |
178 | 178 | |
179 | 179 | case SQT_CODE: |
180 | - fpa11->fpreg[Fd].fDouble = float64_sqrt(rFm); | |
180 | + fpa11->fpreg[Fd].fDouble = float64_sqrt(rFm, &fpa11->fp_status); | |
181 | 181 | break; |
182 | 182 | |
183 | 183 | #if 0 | ... | ... |
target-arm/nwfpe/extended_cpdo.c
... | ... | @@ -53,11 +53,11 @@ unsigned int ExtendedCPDO(const unsigned int opcode) |
53 | 53 | switch (fpa11->fType[Fm]) |
54 | 54 | { |
55 | 55 | case typeSingle: |
56 | - rFm = float32_to_floatx80(fpa11->fpreg[Fm].fSingle); | |
56 | + rFm = float32_to_floatx80(fpa11->fpreg[Fm].fSingle, &fpa11->fp_status); | |
57 | 57 | break; |
58 | 58 | |
59 | 59 | case typeDouble: |
60 | - rFm = float64_to_floatx80(fpa11->fpreg[Fm].fDouble); | |
60 | + rFm = float64_to_floatx80(fpa11->fpreg[Fm].fDouble, &fpa11->fp_status); | |
61 | 61 | break; |
62 | 62 | |
63 | 63 | case typeExtended: |
... | ... | @@ -74,11 +74,11 @@ unsigned int ExtendedCPDO(const unsigned int opcode) |
74 | 74 | switch (fpa11->fType[Fn]) |
75 | 75 | { |
76 | 76 | case typeSingle: |
77 | - rFn = float32_to_floatx80(fpa11->fpreg[Fn].fSingle); | |
77 | + rFn = float32_to_floatx80(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status); | |
78 | 78 | break; |
79 | 79 | |
80 | 80 | case typeDouble: |
81 | - rFn = float64_to_floatx80(fpa11->fpreg[Fn].fDouble); | |
81 | + rFn = float64_to_floatx80(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status); | |
82 | 82 | break; |
83 | 83 | |
84 | 84 | case typeExtended: |
... | ... | @@ -94,30 +94,30 @@ unsigned int ExtendedCPDO(const unsigned int opcode) |
94 | 94 | { |
95 | 95 | /* dyadic opcodes */ |
96 | 96 | case ADF_CODE: |
97 | - fpa11->fpreg[Fd].fExtended = floatx80_add(rFn,rFm); | |
97 | + fpa11->fpreg[Fd].fExtended = floatx80_add(rFn,rFm, &fpa11->fp_status); | |
98 | 98 | break; |
99 | 99 | |
100 | 100 | case MUF_CODE: |
101 | 101 | case FML_CODE: |
102 | - fpa11->fpreg[Fd].fExtended = floatx80_mul(rFn,rFm); | |
102 | + fpa11->fpreg[Fd].fExtended = floatx80_mul(rFn,rFm, &fpa11->fp_status); | |
103 | 103 | break; |
104 | 104 | |
105 | 105 | case SUF_CODE: |
106 | - fpa11->fpreg[Fd].fExtended = floatx80_sub(rFn,rFm); | |
106 | + fpa11->fpreg[Fd].fExtended = floatx80_sub(rFn,rFm, &fpa11->fp_status); | |
107 | 107 | break; |
108 | 108 | |
109 | 109 | case RSF_CODE: |
110 | - fpa11->fpreg[Fd].fExtended = floatx80_sub(rFm,rFn); | |
110 | + fpa11->fpreg[Fd].fExtended = floatx80_sub(rFm,rFn, &fpa11->fp_status); | |
111 | 111 | break; |
112 | 112 | |
113 | 113 | case DVF_CODE: |
114 | 114 | case FDV_CODE: |
115 | - fpa11->fpreg[Fd].fExtended = floatx80_div(rFn,rFm); | |
115 | + fpa11->fpreg[Fd].fExtended = floatx80_div(rFn,rFm, &fpa11->fp_status); | |
116 | 116 | break; |
117 | 117 | |
118 | 118 | case RDF_CODE: |
119 | 119 | case FRD_CODE: |
120 | - fpa11->fpreg[Fd].fExtended = floatx80_div(rFm,rFn); | |
120 | + fpa11->fpreg[Fd].fExtended = floatx80_div(rFm,rFn, &fpa11->fp_status); | |
121 | 121 | break; |
122 | 122 | |
123 | 123 | #if 0 |
... | ... | @@ -131,7 +131,7 @@ unsigned int ExtendedCPDO(const unsigned int opcode) |
131 | 131 | #endif |
132 | 132 | |
133 | 133 | case RMF_CODE: |
134 | - fpa11->fpreg[Fd].fExtended = floatx80_rem(rFn,rFm); | |
134 | + fpa11->fpreg[Fd].fExtended = floatx80_rem(rFn,rFm, &fpa11->fp_status); | |
135 | 135 | break; |
136 | 136 | |
137 | 137 | #if 0 |
... | ... | @@ -157,11 +157,11 @@ unsigned int ExtendedCPDO(const unsigned int opcode) |
157 | 157 | |
158 | 158 | case RND_CODE: |
159 | 159 | case URD_CODE: |
160 | - fpa11->fpreg[Fd].fExtended = floatx80_round_to_int(rFm); | |
160 | + fpa11->fpreg[Fd].fExtended = floatx80_round_to_int(rFm, &fpa11->fp_status); | |
161 | 161 | break; |
162 | 162 | |
163 | 163 | case SQT_CODE: |
164 | - fpa11->fpreg[Fd].fExtended = floatx80_sqrt(rFm); | |
164 | + fpa11->fpreg[Fd].fExtended = floatx80_sqrt(rFm, &fpa11->fp_status); | |
165 | 165 | break; |
166 | 166 | |
167 | 167 | #if 0 | ... | ... |
target-arm/nwfpe/fpa11.c
... | ... | @@ -61,74 +61,79 @@ void resetFPA11(void) |
61 | 61 | |
62 | 62 | void SetRoundingMode(const unsigned int opcode) |
63 | 63 | { |
64 | -#if MAINTAIN_FPCR | |
64 | + int rounding_mode; | |
65 | 65 | FPA11 *fpa11 = GET_FPA11(); |
66 | + | |
67 | +#if MAINTAIN_FPCR | |
66 | 68 | fpa11->fpcr &= ~MASK_ROUNDING_MODE; |
67 | 69 | #endif |
68 | 70 | switch (opcode & MASK_ROUNDING_MODE) |
69 | 71 | { |
70 | 72 | default: |
71 | 73 | case ROUND_TO_NEAREST: |
72 | - float_rounding_mode = float_round_nearest_even; | |
74 | + rounding_mode = float_round_nearest_even; | |
73 | 75 | #if MAINTAIN_FPCR |
74 | 76 | fpa11->fpcr |= ROUND_TO_NEAREST; |
75 | 77 | #endif |
76 | 78 | break; |
77 | 79 | |
78 | 80 | case ROUND_TO_PLUS_INFINITY: |
79 | - float_rounding_mode = float_round_up; | |
81 | + rounding_mode = float_round_up; | |
80 | 82 | #if MAINTAIN_FPCR |
81 | 83 | fpa11->fpcr |= ROUND_TO_PLUS_INFINITY; |
82 | 84 | #endif |
83 | 85 | break; |
84 | 86 | |
85 | 87 | case ROUND_TO_MINUS_INFINITY: |
86 | - float_rounding_mode = float_round_down; | |
88 | + rounding_mode = float_round_down; | |
87 | 89 | #if MAINTAIN_FPCR |
88 | 90 | fpa11->fpcr |= ROUND_TO_MINUS_INFINITY; |
89 | 91 | #endif |
90 | 92 | break; |
91 | 93 | |
92 | 94 | case ROUND_TO_ZERO: |
93 | - float_rounding_mode = float_round_to_zero; | |
95 | + rounding_mode = float_round_to_zero; | |
94 | 96 | #if MAINTAIN_FPCR |
95 | 97 | fpa11->fpcr |= ROUND_TO_ZERO; |
96 | 98 | #endif |
97 | 99 | break; |
98 | 100 | } |
101 | + set_float_rounding_mode(rounding_mode, &fpa11->fp_status); | |
99 | 102 | } |
100 | 103 | |
101 | 104 | void SetRoundingPrecision(const unsigned int opcode) |
102 | 105 | { |
103 | -#if MAINTAIN_FPCR | |
106 | + int rounding_precision; | |
104 | 107 | FPA11 *fpa11 = GET_FPA11(); |
108 | +#if MAINTAIN_FPCR | |
105 | 109 | fpa11->fpcr &= ~MASK_ROUNDING_PRECISION; |
106 | 110 | #endif |
107 | 111 | switch (opcode & MASK_ROUNDING_PRECISION) |
108 | 112 | { |
109 | 113 | case ROUND_SINGLE: |
110 | - floatx80_rounding_precision = 32; | |
114 | + rounding_precision = 32; | |
111 | 115 | #if MAINTAIN_FPCR |
112 | 116 | fpa11->fpcr |= ROUND_SINGLE; |
113 | 117 | #endif |
114 | 118 | break; |
115 | 119 | |
116 | 120 | case ROUND_DOUBLE: |
117 | - floatx80_rounding_precision = 64; | |
121 | + rounding_precision = 64; | |
118 | 122 | #if MAINTAIN_FPCR |
119 | 123 | fpa11->fpcr |= ROUND_DOUBLE; |
120 | 124 | #endif |
121 | 125 | break; |
122 | 126 | |
123 | 127 | case ROUND_EXTENDED: |
124 | - floatx80_rounding_precision = 80; | |
128 | + rounding_precision = 80; | |
125 | 129 | #if MAINTAIN_FPCR |
126 | 130 | fpa11->fpcr |= ROUND_EXTENDED; |
127 | 131 | #endif |
128 | 132 | break; |
129 | 133 | |
130 | - default: floatx80_rounding_precision = 80; | |
134 | + default: rounding_precision = 80; | |
131 | 135 | } |
136 | + set_floatx80_rounding_precision(rounding_precision, &fpa11->fp_status); | |
132 | 137 | } |
133 | 138 | |
134 | 139 | /* Emulate the instruction in the opcode. */ | ... | ... |
target-arm/nwfpe/fpa11.h
target-arm/nwfpe/fpa11_cpdo.c
... | ... | @@ -80,10 +80,10 @@ unsigned int EmulateCPDO(const unsigned int opcode) |
80 | 80 | { |
81 | 81 | if (typeDouble == nType) |
82 | 82 | fpa11->fpreg[Fd].fSingle = |
83 | - float64_to_float32(fpa11->fpreg[Fd].fDouble); | |
83 | + float64_to_float32(fpa11->fpreg[Fd].fDouble, &fpa11->fp_status); | |
84 | 84 | else |
85 | 85 | fpa11->fpreg[Fd].fSingle = |
86 | - floatx80_to_float32(fpa11->fpreg[Fd].fExtended); | |
86 | + floatx80_to_float32(fpa11->fpreg[Fd].fExtended, &fpa11->fp_status); | |
87 | 87 | } |
88 | 88 | break; |
89 | 89 | |
... | ... | @@ -91,10 +91,10 @@ unsigned int EmulateCPDO(const unsigned int opcode) |
91 | 91 | { |
92 | 92 | if (typeSingle == nType) |
93 | 93 | fpa11->fpreg[Fd].fDouble = |
94 | - float32_to_float64(fpa11->fpreg[Fd].fSingle); | |
94 | + float32_to_float64(fpa11->fpreg[Fd].fSingle, &fpa11->fp_status); | |
95 | 95 | else |
96 | 96 | fpa11->fpreg[Fd].fDouble = |
97 | - floatx80_to_float64(fpa11->fpreg[Fd].fExtended); | |
97 | + floatx80_to_float64(fpa11->fpreg[Fd].fExtended, &fpa11->fp_status); | |
98 | 98 | } |
99 | 99 | break; |
100 | 100 | |
... | ... | @@ -102,10 +102,10 @@ unsigned int EmulateCPDO(const unsigned int opcode) |
102 | 102 | { |
103 | 103 | if (typeSingle == nType) |
104 | 104 | fpa11->fpreg[Fd].fExtended = |
105 | - float32_to_floatx80(fpa11->fpreg[Fd].fSingle); | |
105 | + float32_to_floatx80(fpa11->fpreg[Fd].fSingle, &fpa11->fp_status); | |
106 | 106 | else |
107 | 107 | fpa11->fpreg[Fd].fExtended = |
108 | - float64_to_floatx80(fpa11->fpreg[Fd].fDouble); | |
108 | + float64_to_floatx80(fpa11->fpreg[Fd].fDouble, &fpa11->fp_status); | |
109 | 109 | } |
110 | 110 | break; |
111 | 111 | } | ... | ... |
target-arm/nwfpe/fpa11_cpdt.c
... | ... | @@ -106,11 +106,11 @@ void storeSingle(const unsigned int Fn,unsigned int *pMem) |
106 | 106 | switch (fpa11->fType[Fn]) |
107 | 107 | { |
108 | 108 | case typeDouble: |
109 | - val = float64_to_float32(fpa11->fpreg[Fn].fDouble); | |
109 | + val = float64_to_float32(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status); | |
110 | 110 | break; |
111 | 111 | |
112 | 112 | case typeExtended: |
113 | - val = floatx80_to_float32(fpa11->fpreg[Fn].fExtended); | |
113 | + val = floatx80_to_float32(fpa11->fpreg[Fn].fExtended, &fpa11->fp_status); | |
114 | 114 | break; |
115 | 115 | |
116 | 116 | default: val = fpa11->fpreg[Fn].fSingle; |
... | ... | @@ -129,11 +129,11 @@ void storeDouble(const unsigned int Fn,unsigned int *pMem) |
129 | 129 | switch (fpa11->fType[Fn]) |
130 | 130 | { |
131 | 131 | case typeSingle: |
132 | - val = float32_to_float64(fpa11->fpreg[Fn].fSingle); | |
132 | + val = float32_to_float64(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status); | |
133 | 133 | break; |
134 | 134 | |
135 | 135 | case typeExtended: |
136 | - val = floatx80_to_float64(fpa11->fpreg[Fn].fExtended); | |
136 | + val = floatx80_to_float64(fpa11->fpreg[Fn].fExtended, &fpa11->fp_status); | |
137 | 137 | break; |
138 | 138 | |
139 | 139 | default: val = fpa11->fpreg[Fn].fDouble; |
... | ... | @@ -157,11 +157,11 @@ void storeExtended(const unsigned int Fn,unsigned int *pMem) |
157 | 157 | switch (fpa11->fType[Fn]) |
158 | 158 | { |
159 | 159 | case typeSingle: |
160 | - val = float32_to_floatx80(fpa11->fpreg[Fn].fSingle); | |
160 | + val = float32_to_floatx80(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status); | |
161 | 161 | break; |
162 | 162 | |
163 | 163 | case typeDouble: |
164 | - val = float64_to_floatx80(fpa11->fpreg[Fn].fDouble); | |
164 | + val = float64_to_floatx80(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status); | |
165 | 165 | break; |
166 | 166 | |
167 | 167 | default: val = fpa11->fpreg[Fn].fExtended; | ... | ... |
target-arm/nwfpe/fpa11_cprt.c
... | ... | @@ -21,7 +21,6 @@ |
21 | 21 | */ |
22 | 22 | |
23 | 23 | #include "fpa11.h" |
24 | -#include "milieu.h" | |
25 | 24 | #include "softfloat.h" |
26 | 25 | #include "fpopcode.h" |
27 | 26 | #include "fpa11.inl" |
... | ... | @@ -89,7 +88,7 @@ unsigned int PerformFLT(const unsigned int opcode) |
89 | 88 | { |
90 | 89 | fpa11->fType[getFn(opcode)] = typeSingle; |
91 | 90 | fpa11->fpreg[getFn(opcode)].fSingle = |
92 | - int32_to_float32(readRegister(getRd(opcode))); | |
91 | + int32_to_float32(readRegister(getRd(opcode)), &fpa11->fp_status); | |
93 | 92 | } |
94 | 93 | break; |
95 | 94 | |
... | ... | @@ -97,7 +96,7 @@ unsigned int PerformFLT(const unsigned int opcode) |
97 | 96 | { |
98 | 97 | fpa11->fType[getFn(opcode)] = typeDouble; |
99 | 98 | fpa11->fpreg[getFn(opcode)].fDouble = |
100 | - int32_to_float64(readRegister(getRd(opcode))); | |
99 | + int32_to_float64(readRegister(getRd(opcode)), &fpa11->fp_status); | |
101 | 100 | } |
102 | 101 | break; |
103 | 102 | |
... | ... | @@ -105,7 +104,7 @@ unsigned int PerformFLT(const unsigned int opcode) |
105 | 104 | { |
106 | 105 | fpa11->fType[getFn(opcode)] = typeExtended; |
107 | 106 | fpa11->fpreg[getFn(opcode)].fExtended = |
108 | - int32_to_floatx80(readRegister(getRd(opcode))); | |
107 | + int32_to_floatx80(readRegister(getRd(opcode)), &fpa11->fp_status); | |
109 | 108 | } |
110 | 109 | break; |
111 | 110 | |
... | ... | @@ -128,7 +127,7 @@ unsigned int PerformFIX(const unsigned int opcode) |
128 | 127 | case typeSingle: |
129 | 128 | { |
130 | 129 | writeRegister(getRd(opcode), |
131 | - float32_to_int32(fpa11->fpreg[Fn].fSingle)); | |
130 | + float32_to_int32(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status)); | |
132 | 131 | } |
133 | 132 | break; |
134 | 133 | |
... | ... | @@ -136,14 +135,14 @@ unsigned int PerformFIX(const unsigned int opcode) |
136 | 135 | { |
137 | 136 | //printf("F%d is 0x%llx\n",Fn,fpa11->fpreg[Fn].fDouble); |
138 | 137 | writeRegister(getRd(opcode), |
139 | - float64_to_int32(fpa11->fpreg[Fn].fDouble)); | |
138 | + float64_to_int32(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status)); | |
140 | 139 | } |
141 | 140 | break; |
142 | 141 | |
143 | 142 | case typeExtended: |
144 | 143 | { |
145 | 144 | writeRegister(getRd(opcode), |
146 | - floatx80_to_int32(fpa11->fpreg[Fn].fExtended)); | |
145 | + floatx80_to_int32(fpa11->fpreg[Fn].fExtended, &fpa11->fp_status)); | |
147 | 146 | } |
148 | 147 | break; |
149 | 148 | |
... | ... | @@ -157,22 +156,23 @@ unsigned int PerformFIX(const unsigned int opcode) |
157 | 156 | static unsigned int __inline__ |
158 | 157 | PerformComparisonOperation(floatx80 Fn, floatx80 Fm) |
159 | 158 | { |
159 | + FPA11 *fpa11 = GET_FPA11(); | |
160 | 160 | unsigned int flags = 0; |
161 | 161 | |
162 | 162 | /* test for less than condition */ |
163 | - if (floatx80_lt(Fn,Fm)) | |
163 | + if (floatx80_lt(Fn,Fm, &fpa11->fp_status)) | |
164 | 164 | { |
165 | 165 | flags |= CC_NEGATIVE; |
166 | 166 | } |
167 | 167 | |
168 | 168 | /* test for equal condition */ |
169 | - if (floatx80_eq(Fn,Fm)) | |
169 | + if (floatx80_eq(Fn,Fm, &fpa11->fp_status)) | |
170 | 170 | { |
171 | 171 | flags |= CC_ZERO; |
172 | 172 | } |
173 | 173 | |
174 | 174 | /* test for greater than or equal condition */ |
175 | - if (floatx80_lt(Fm,Fn)) | |
175 | + if (floatx80_lt(Fm,Fn, &fpa11->fp_status)) | |
176 | 176 | { |
177 | 177 | flags |= CC_CARRY; |
178 | 178 | } |
... | ... | @@ -208,14 +208,14 @@ static unsigned int PerformComparison(const unsigned int opcode) |
208 | 208 | //printk("single.\n"); |
209 | 209 | if (float32_is_nan(fpa11->fpreg[Fn].fSingle)) |
210 | 210 | goto unordered; |
211 | - rFn = float32_to_floatx80(fpa11->fpreg[Fn].fSingle); | |
211 | + rFn = float32_to_floatx80(fpa11->fpreg[Fn].fSingle, &fpa11->fp_status); | |
212 | 212 | break; |
213 | 213 | |
214 | 214 | case typeDouble: |
215 | 215 | //printk("double.\n"); |
216 | 216 | if (float64_is_nan(fpa11->fpreg[Fn].fDouble)) |
217 | 217 | goto unordered; |
218 | - rFn = float64_to_floatx80(fpa11->fpreg[Fn].fDouble); | |
218 | + rFn = float64_to_floatx80(fpa11->fpreg[Fn].fDouble, &fpa11->fp_status); | |
219 | 219 | break; |
220 | 220 | |
221 | 221 | case typeExtended: |
... | ... | @@ -244,14 +244,14 @@ static unsigned int PerformComparison(const unsigned int opcode) |
244 | 244 | //printk("single.\n"); |
245 | 245 | if (float32_is_nan(fpa11->fpreg[Fm].fSingle)) |
246 | 246 | goto unordered; |
247 | - rFm = float32_to_floatx80(fpa11->fpreg[Fm].fSingle); | |
247 | + rFm = float32_to_floatx80(fpa11->fpreg[Fm].fSingle, &fpa11->fp_status); | |
248 | 248 | break; |
249 | 249 | |
250 | 250 | case typeDouble: |
251 | 251 | //printk("double.\n"); |
252 | 252 | if (float64_is_nan(fpa11->fpreg[Fm].fDouble)) |
253 | 253 | goto unordered; |
254 | - rFm = float64_to_floatx80(fpa11->fpreg[Fm].fDouble); | |
254 | + rFm = float64_to_floatx80(fpa11->fpreg[Fm].fDouble, &fpa11->fp_status); | |
255 | 255 | break; |
256 | 256 | |
257 | 257 | case typeExtended: |
... | ... | @@ -283,7 +283,7 @@ static unsigned int PerformComparison(const unsigned int opcode) |
283 | 283 | |
284 | 284 | if (BIT_AC & readFPSR()) flags |= CC_CARRY; |
285 | 285 | |
286 | - if (e_flag) float_raise(float_flag_invalid); | |
286 | + if (e_flag) float_raise(float_flag_invalid, &fpa11->fp_status); | |
287 | 287 | |
288 | 288 | writeConditionCodes(flags); |
289 | 289 | return 1; | ... | ... |
target-arm/nwfpe/fpopcode.c
... | ... | @@ -27,14 +27,14 @@ |
27 | 27 | //#include "fpmodule.inl" |
28 | 28 | |
29 | 29 | const floatx80 floatx80Constant[] = { |
30 | - { 0x0000, 0x0000000000000000ULL}, /* extended 0.0 */ | |
31 | - { 0x3fff, 0x8000000000000000ULL}, /* extended 1.0 */ | |
32 | - { 0x4000, 0x8000000000000000ULL}, /* extended 2.0 */ | |
33 | - { 0x4000, 0xc000000000000000ULL}, /* extended 3.0 */ | |
34 | - { 0x4001, 0x8000000000000000ULL}, /* extended 4.0 */ | |
35 | - { 0x4001, 0xa000000000000000ULL}, /* extended 5.0 */ | |
36 | - { 0x3ffe, 0x8000000000000000ULL}, /* extended 0.5 */ | |
37 | - { 0x4002, 0xa000000000000000ULL} /* extended 10.0 */ | |
30 | + { 0x0000000000000000ULL, 0x0000}, /* extended 0.0 */ | |
31 | + { 0x8000000000000000ULL, 0x3fff}, /* extended 1.0 */ | |
32 | + { 0x8000000000000000ULL, 0x4000}, /* extended 2.0 */ | |
33 | + { 0xc000000000000000ULL, 0x4000}, /* extended 3.0 */ | |
34 | + { 0x8000000000000000ULL, 0x4001}, /* extended 4.0 */ | |
35 | + { 0xa000000000000000ULL, 0x4001}, /* extended 5.0 */ | |
36 | + { 0x8000000000000000ULL, 0x3ffe}, /* extended 0.5 */ | |
37 | + { 0xa000000000000000ULL, 0x4002} /* extended 10.0 */ | |
38 | 38 | }; |
39 | 39 | |
40 | 40 | const float64 float64Constant[] = { | ... | ... |
target-arm/nwfpe/milieu.h deleted
100644 → 0
1 | - | |
2 | -/* | |
3 | -=============================================================================== | |
4 | - | |
5 | -This C header file is part of the SoftFloat IEC/IEEE Floating-point | |
6 | -Arithmetic Package, Release 2. | |
7 | - | |
8 | -Written by John R. Hauser. This work was made possible in part by the | |
9 | -International Computer Science Institute, located at Suite 600, 1947 Center | |
10 | -Street, Berkeley, California 94704. Funding was partially provided by the | |
11 | -National Science Foundation under grant MIP-9311980. The original version | |
12 | -of this code was written as part of a project to build a fixed-point vector | |
13 | -processor in collaboration with the University of California at Berkeley, | |
14 | -overseen by Profs. Nelson Morgan and John Wawrzynek. More information | |
15 | -is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ | |
16 | -arithmetic/softfloat.html'. | |
17 | - | |
18 | -THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort | |
19 | -has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT | |
20 | -TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO | |
21 | -PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY | |
22 | -AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. | |
23 | - | |
24 | -Derivative works are acceptable, even for commercial purposes, so long as | |
25 | -(1) they include prominent notice that the work is derivative, and (2) they | |
26 | -include prominent notice akin to these three paragraphs for those parts of | |
27 | -this code that are retained. | |
28 | - | |
29 | -=============================================================================== | |
30 | -*/ | |
31 | - | |
32 | -/* | |
33 | -------------------------------------------------------------------------------- | |
34 | -Include common integer types and flags. | |
35 | -------------------------------------------------------------------------------- | |
36 | -*/ | |
37 | -#include "ARM-gcc.h" | |
38 | - | |
39 | -/* | |
40 | -------------------------------------------------------------------------------- | |
41 | -Symbolic Boolean literals. | |
42 | -------------------------------------------------------------------------------- | |
43 | -*/ | |
44 | -enum { | |
45 | - FALSE = 0, | |
46 | - TRUE = 1 | |
47 | -}; | |
48 | - |
target-arm/nwfpe/single_cpdo.c
... | ... | @@ -76,30 +76,30 @@ unsigned int SingleCPDO(const unsigned int opcode) |
76 | 76 | { |
77 | 77 | /* dyadic opcodes */ |
78 | 78 | case ADF_CODE: |
79 | - fpa11->fpreg[Fd].fSingle = float32_add(rFn,rFm); | |
79 | + fpa11->fpreg[Fd].fSingle = float32_add(rFn,rFm, &fpa11->fp_status); | |
80 | 80 | break; |
81 | 81 | |
82 | 82 | case MUF_CODE: |
83 | 83 | case FML_CODE: |
84 | - fpa11->fpreg[Fd].fSingle = float32_mul(rFn,rFm); | |
84 | + fpa11->fpreg[Fd].fSingle = float32_mul(rFn,rFm, &fpa11->fp_status); | |
85 | 85 | break; |
86 | 86 | |
87 | 87 | case SUF_CODE: |
88 | - fpa11->fpreg[Fd].fSingle = float32_sub(rFn,rFm); | |
88 | + fpa11->fpreg[Fd].fSingle = float32_sub(rFn,rFm, &fpa11->fp_status); | |
89 | 89 | break; |
90 | 90 | |
91 | 91 | case RSF_CODE: |
92 | - fpa11->fpreg[Fd].fSingle = float32_sub(rFm,rFn); | |
92 | + fpa11->fpreg[Fd].fSingle = float32_sub(rFm,rFn, &fpa11->fp_status); | |
93 | 93 | break; |
94 | 94 | |
95 | 95 | case DVF_CODE: |
96 | 96 | case FDV_CODE: |
97 | - fpa11->fpreg[Fd].fSingle = float32_div(rFn,rFm); | |
97 | + fpa11->fpreg[Fd].fSingle = float32_div(rFn,rFm, &fpa11->fp_status); | |
98 | 98 | break; |
99 | 99 | |
100 | 100 | case RDF_CODE: |
101 | 101 | case FRD_CODE: |
102 | - fpa11->fpreg[Fd].fSingle = float32_div(rFm,rFn); | |
102 | + fpa11->fpreg[Fd].fSingle = float32_div(rFm,rFn, &fpa11->fp_status); | |
103 | 103 | break; |
104 | 104 | |
105 | 105 | #if 0 |
... | ... | @@ -113,7 +113,7 @@ unsigned int SingleCPDO(const unsigned int opcode) |
113 | 113 | #endif |
114 | 114 | |
115 | 115 | case RMF_CODE: |
116 | - fpa11->fpreg[Fd].fSingle = float32_rem(rFn,rFm); | |
116 | + fpa11->fpreg[Fd].fSingle = float32_rem(rFn,rFm, &fpa11->fp_status); | |
117 | 117 | break; |
118 | 118 | |
119 | 119 | #if 0 |
... | ... | @@ -139,11 +139,11 @@ unsigned int SingleCPDO(const unsigned int opcode) |
139 | 139 | |
140 | 140 | case RND_CODE: |
141 | 141 | case URD_CODE: |
142 | - fpa11->fpreg[Fd].fSingle = float32_round_to_int(rFm); | |
142 | + fpa11->fpreg[Fd].fSingle = float32_round_to_int(rFm, &fpa11->fp_status); | |
143 | 143 | break; |
144 | 144 | |
145 | 145 | case SQT_CODE: |
146 | - fpa11->fpreg[Fd].fSingle = float32_sqrt(rFm); | |
146 | + fpa11->fpreg[Fd].fSingle = float32_sqrt(rFm, &fpa11->fp_status); | |
147 | 147 | break; |
148 | 148 | |
149 | 149 | #if 0 | ... | ... |
target-arm/nwfpe/softfloat-macros deleted
100644 → 0
1 | - | |
2 | -/* | |
3 | -=============================================================================== | |
4 | - | |
5 | -This C source fragment is part of the SoftFloat IEC/IEEE Floating-point | |
6 | -Arithmetic Package, Release 2. | |
7 | - | |
8 | -Written by John R. Hauser. This work was made possible in part by the | |
9 | -International Computer Science Institute, located at Suite 600, 1947 Center | |
10 | -Street, Berkeley, California 94704. Funding was partially provided by the | |
11 | -National Science Foundation under grant MIP-9311980. The original version | |
12 | -of this code was written as part of a project to build a fixed-point vector | |
13 | -processor in collaboration with the University of California at Berkeley, | |
14 | -overseen by Profs. Nelson Morgan and John Wawrzynek. More information | |
15 | -is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ | |
16 | -arithmetic/softfloat.html'. | |
17 | - | |
18 | -THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort | |
19 | -has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT | |
20 | -TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO | |
21 | -PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY | |
22 | -AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. | |
23 | - | |
24 | -Derivative works are acceptable, even for commercial purposes, so long as | |
25 | -(1) they include prominent notice that the work is derivative, and (2) they | |
26 | -include prominent notice akin to these three paragraphs for those parts of | |
27 | -this code that are retained. | |
28 | - | |
29 | -=============================================================================== | |
30 | -*/ | |
31 | - | |
32 | -/* | |
33 | -------------------------------------------------------------------------------- | |
34 | -Shifts `a' right by the number of bits given in `count'. If any nonzero | |
35 | -bits are shifted off, they are ``jammed'' into the least significant bit of | |
36 | -the result by setting the least significant bit to 1. The value of `count' | |
37 | -can be arbitrarily large; in particular, if `count' is greater than 32, the | |
38 | -result will be either 0 or 1, depending on whether `a' is zero or nonzero. | |
39 | -The result is stored in the location pointed to by `zPtr'. | |
40 | -------------------------------------------------------------------------------- | |
41 | -*/ | |
42 | -INLINE void shift32RightJamming( bits32 a, int16 count, bits32 *zPtr ) | |
43 | -{ | |
44 | - bits32 z; | |
45 | - if ( count == 0 ) { | |
46 | - z = a; | |
47 | - } | |
48 | - else if ( count < 32 ) { | |
49 | - z = ( a>>count ) | ( ( a<<( ( - count ) & 31 ) ) != 0 ); | |
50 | - } | |
51 | - else { | |
52 | - z = ( a != 0 ); | |
53 | - } | |
54 | - *zPtr = z; | |
55 | -} | |
56 | - | |
57 | -/* | |
58 | -------------------------------------------------------------------------------- | |
59 | -Shifts `a' right by the number of bits given in `count'. If any nonzero | |
60 | -bits are shifted off, they are ``jammed'' into the least significant bit of | |
61 | -the result by setting the least significant bit to 1. The value of `count' | |
62 | -can be arbitrarily large; in particular, if `count' is greater than 64, the | |
63 | -result will be either 0 or 1, depending on whether `a' is zero or nonzero. | |
64 | -The result is stored in the location pointed to by `zPtr'. | |
65 | -------------------------------------------------------------------------------- | |
66 | -*/ | |
67 | -INLINE void shift64RightJamming( bits64 a, int16 count, bits64 *zPtr ) | |
68 | -{ | |
69 | - bits64 z; | |
70 | - | |
71 | -// __asm__("@shift64RightJamming -- start"); | |
72 | - if ( count == 0 ) { | |
73 | - z = a; | |
74 | - } | |
75 | - else if ( count < 64 ) { | |
76 | - z = ( a>>count ) | ( ( a<<( ( - count ) & 63 ) ) != 0 ); | |
77 | - } | |
78 | - else { | |
79 | - z = ( a != 0 ); | |
80 | - } | |
81 | -// __asm__("@shift64RightJamming -- end"); | |
82 | - *zPtr = z; | |
83 | -} | |
84 | - | |
85 | -/* | |
86 | -------------------------------------------------------------------------------- | |
87 | -Shifts the 128-bit value formed by concatenating `a0' and `a1' right by 64 | |
88 | -_plus_ the number of bits given in `count'. The shifted result is at most | |
89 | -64 nonzero bits; this is stored at the location pointed to by `z0Ptr'. The | |
90 | -bits shifted off form a second 64-bit result as follows: The _last_ bit | |
91 | -shifted off is the most-significant bit of the extra result, and the other | |
92 | -63 bits of the extra result are all zero if and only if _all_but_the_last_ | |
93 | -bits shifted off were all zero. This extra result is stored in the location | |
94 | -pointed to by `z1Ptr'. The value of `count' can be arbitrarily large. | |
95 | - (This routine makes more sense if `a0' and `a1' are considered to form a | |
96 | -fixed-point value with binary point between `a0' and `a1'. This fixed-point | |
97 | -value is shifted right by the number of bits given in `count', and the | |
98 | -integer part of the result is returned at the location pointed to by | |
99 | -`z0Ptr'. The fractional part of the result may be slightly corrupted as | |
100 | -described above, and is returned at the location pointed to by `z1Ptr'.) | |
101 | -------------------------------------------------------------------------------- | |
102 | -*/ | |
103 | -INLINE void | |
104 | - shift64ExtraRightJamming( | |
105 | - bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) | |
106 | -{ | |
107 | - bits64 z0, z1; | |
108 | - int8 negCount = ( - count ) & 63; | |
109 | - | |
110 | - if ( count == 0 ) { | |
111 | - z1 = a1; | |
112 | - z0 = a0; | |
113 | - } | |
114 | - else if ( count < 64 ) { | |
115 | - z1 = ( a0<<negCount ) | ( a1 != 0 ); | |
116 | - z0 = a0>>count; | |
117 | - } | |
118 | - else { | |
119 | - if ( count == 64 ) { | |
120 | - z1 = a0 | ( a1 != 0 ); | |
121 | - } | |
122 | - else { | |
123 | - z1 = ( ( a0 | a1 ) != 0 ); | |
124 | - } | |
125 | - z0 = 0; | |
126 | - } | |
127 | - *z1Ptr = z1; | |
128 | - *z0Ptr = z0; | |
129 | - | |
130 | -} | |
131 | - | |
132 | -/* | |
133 | -------------------------------------------------------------------------------- | |
134 | -Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the | |
135 | -number of bits given in `count'. Any bits shifted off are lost. The value | |
136 | -of `count' can be arbitrarily large; in particular, if `count' is greater | |
137 | -than 128, the result will be 0. The result is broken into two 64-bit pieces | |
138 | -which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. | |
139 | -------------------------------------------------------------------------------- | |
140 | -*/ | |
141 | -INLINE void | |
142 | - shift128Right( | |
143 | - bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) | |
144 | -{ | |
145 | - bits64 z0, z1; | |
146 | - int8 negCount = ( - count ) & 63; | |
147 | - | |
148 | - if ( count == 0 ) { | |
149 | - z1 = a1; | |
150 | - z0 = a0; | |
151 | - } | |
152 | - else if ( count < 64 ) { | |
153 | - z1 = ( a0<<negCount ) | ( a1>>count ); | |
154 | - z0 = a0>>count; | |
155 | - } | |
156 | - else { | |
157 | - z1 = ( count < 64 ) ? ( a0>>( count & 63 ) ) : 0; | |
158 | - z0 = 0; | |
159 | - } | |
160 | - *z1Ptr = z1; | |
161 | - *z0Ptr = z0; | |
162 | - | |
163 | -} | |
164 | - | |
165 | -/* | |
166 | -------------------------------------------------------------------------------- | |
167 | -Shifts the 128-bit value formed by concatenating `a0' and `a1' right by the | |
168 | -number of bits given in `count'. If any nonzero bits are shifted off, they | |
169 | -are ``jammed'' into the least significant bit of the result by setting the | |
170 | -least significant bit to 1. The value of `count' can be arbitrarily large; | |
171 | -in particular, if `count' is greater than 128, the result will be either 0 | |
172 | -or 1, depending on whether the concatenation of `a0' and `a1' is zero or | |
173 | -nonzero. The result is broken into two 64-bit pieces which are stored at | |
174 | -the locations pointed to by `z0Ptr' and `z1Ptr'. | |
175 | -------------------------------------------------------------------------------- | |
176 | -*/ | |
177 | -INLINE void | |
178 | - shift128RightJamming( | |
179 | - bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) | |
180 | -{ | |
181 | - bits64 z0, z1; | |
182 | - int8 negCount = ( - count ) & 63; | |
183 | - | |
184 | - if ( count == 0 ) { | |
185 | - z1 = a1; | |
186 | - z0 = a0; | |
187 | - } | |
188 | - else if ( count < 64 ) { | |
189 | - z1 = ( a0<<negCount ) | ( a1>>count ) | ( ( a1<<negCount ) != 0 ); | |
190 | - z0 = a0>>count; | |
191 | - } | |
192 | - else { | |
193 | - if ( count == 64 ) { | |
194 | - z1 = a0 | ( a1 != 0 ); | |
195 | - } | |
196 | - else if ( count < 128 ) { | |
197 | - z1 = ( a0>>( count & 63 ) ) | ( ( ( a0<<negCount ) | a1 ) != 0 ); | |
198 | - } | |
199 | - else { | |
200 | - z1 = ( ( a0 | a1 ) != 0 ); | |
201 | - } | |
202 | - z0 = 0; | |
203 | - } | |
204 | - *z1Ptr = z1; | |
205 | - *z0Ptr = z0; | |
206 | - | |
207 | -} | |
208 | - | |
209 | -/* | |
210 | -------------------------------------------------------------------------------- | |
211 | -Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' right | |
212 | -by 64 _plus_ the number of bits given in `count'. The shifted result is | |
213 | -at most 128 nonzero bits; these are broken into two 64-bit pieces which are | |
214 | -stored at the locations pointed to by `z0Ptr' and `z1Ptr'. The bits shifted | |
215 | -off form a third 64-bit result as follows: The _last_ bit shifted off is | |
216 | -the most-significant bit of the extra result, and the other 63 bits of the | |
217 | -extra result are all zero if and only if _all_but_the_last_ bits shifted off | |
218 | -were all zero. This extra result is stored in the location pointed to by | |
219 | -`z2Ptr'. The value of `count' can be arbitrarily large. | |
220 | - (This routine makes more sense if `a0', `a1', and `a2' are considered | |
221 | -to form a fixed-point value with binary point between `a1' and `a2'. This | |
222 | -fixed-point value is shifted right by the number of bits given in `count', | |
223 | -and the integer part of the result is returned at the locations pointed to | |
224 | -by `z0Ptr' and `z1Ptr'. The fractional part of the result may be slightly | |
225 | -corrupted as described above, and is returned at the location pointed to by | |
226 | -`z2Ptr'.) | |
227 | -------------------------------------------------------------------------------- | |
228 | -*/ | |
229 | -INLINE void | |
230 | - shift128ExtraRightJamming( | |
231 | - bits64 a0, | |
232 | - bits64 a1, | |
233 | - bits64 a2, | |
234 | - int16 count, | |
235 | - bits64 *z0Ptr, | |
236 | - bits64 *z1Ptr, | |
237 | - bits64 *z2Ptr | |
238 | - ) | |
239 | -{ | |
240 | - bits64 z0, z1, z2; | |
241 | - int8 negCount = ( - count ) & 63; | |
242 | - | |
243 | - if ( count == 0 ) { | |
244 | - z2 = a2; | |
245 | - z1 = a1; | |
246 | - z0 = a0; | |
247 | - } | |
248 | - else { | |
249 | - if ( count < 64 ) { | |
250 | - z2 = a1<<negCount; | |
251 | - z1 = ( a0<<negCount ) | ( a1>>count ); | |
252 | - z0 = a0>>count; | |
253 | - } | |
254 | - else { | |
255 | - if ( count == 64 ) { | |
256 | - z2 = a1; | |
257 | - z1 = a0; | |
258 | - } | |
259 | - else { | |
260 | - a2 |= a1; | |
261 | - if ( count < 128 ) { | |
262 | - z2 = a0<<negCount; | |
263 | - z1 = a0>>( count & 63 ); | |
264 | - } | |
265 | - else { | |
266 | - z2 = ( count == 128 ) ? a0 : ( a0 != 0 ); | |
267 | - z1 = 0; | |
268 | - } | |
269 | - } | |
270 | - z0 = 0; | |
271 | - } | |
272 | - z2 |= ( a2 != 0 ); | |
273 | - } | |
274 | - *z2Ptr = z2; | |
275 | - *z1Ptr = z1; | |
276 | - *z0Ptr = z0; | |
277 | - | |
278 | -} | |
279 | - | |
280 | -/* | |
281 | -------------------------------------------------------------------------------- | |
282 | -Shifts the 128-bit value formed by concatenating `a0' and `a1' left by the | |
283 | -number of bits given in `count'. Any bits shifted off are lost. The value | |
284 | -of `count' must be less than 64. The result is broken into two 64-bit | |
285 | -pieces which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. | |
286 | -------------------------------------------------------------------------------- | |
287 | -*/ | |
288 | -INLINE void | |
289 | - shortShift128Left( | |
290 | - bits64 a0, bits64 a1, int16 count, bits64 *z0Ptr, bits64 *z1Ptr ) | |
291 | -{ | |
292 | - | |
293 | - *z1Ptr = a1<<count; | |
294 | - *z0Ptr = | |
295 | - ( count == 0 ) ? a0 : ( a0<<count ) | ( a1>>( ( - count ) & 63 ) ); | |
296 | - | |
297 | -} | |
298 | - | |
299 | -/* | |
300 | -------------------------------------------------------------------------------- | |
301 | -Shifts the 192-bit value formed by concatenating `a0', `a1', and `a2' left | |
302 | -by the number of bits given in `count'. Any bits shifted off are lost. | |
303 | -The value of `count' must be less than 64. The result is broken into three | |
304 | -64-bit pieces which are stored at the locations pointed to by `z0Ptr', | |
305 | -`z1Ptr', and `z2Ptr'. | |
306 | -------------------------------------------------------------------------------- | |
307 | -*/ | |
308 | -INLINE void | |
309 | - shortShift192Left( | |
310 | - bits64 a0, | |
311 | - bits64 a1, | |
312 | - bits64 a2, | |
313 | - int16 count, | |
314 | - bits64 *z0Ptr, | |
315 | - bits64 *z1Ptr, | |
316 | - bits64 *z2Ptr | |
317 | - ) | |
318 | -{ | |
319 | - bits64 z0, z1, z2; | |
320 | - int8 negCount; | |
321 | - | |
322 | - z2 = a2<<count; | |
323 | - z1 = a1<<count; | |
324 | - z0 = a0<<count; | |
325 | - if ( 0 < count ) { | |
326 | - negCount = ( ( - count ) & 63 ); | |
327 | - z1 |= a2>>negCount; | |
328 | - z0 |= a1>>negCount; | |
329 | - } | |
330 | - *z2Ptr = z2; | |
331 | - *z1Ptr = z1; | |
332 | - *z0Ptr = z0; | |
333 | - | |
334 | -} | |
335 | - | |
336 | -/* | |
337 | -------------------------------------------------------------------------------- | |
338 | -Adds the 128-bit value formed by concatenating `a0' and `a1' to the 128-bit | |
339 | -value formed by concatenating `b0' and `b1'. Addition is modulo 2^128, so | |
340 | -any carry out is lost. The result is broken into two 64-bit pieces which | |
341 | -are stored at the locations pointed to by `z0Ptr' and `z1Ptr'. | |
342 | -------------------------------------------------------------------------------- | |
343 | -*/ | |
344 | -INLINE void | |
345 | - add128( | |
346 | - bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr ) | |
347 | -{ | |
348 | - bits64 z1; | |
349 | - | |
350 | - z1 = a1 + b1; | |
351 | - *z1Ptr = z1; | |
352 | - *z0Ptr = a0 + b0 + ( z1 < a1 ); | |
353 | - | |
354 | -} | |
355 | - | |
356 | -/* | |
357 | -------------------------------------------------------------------------------- | |
358 | -Adds the 192-bit value formed by concatenating `a0', `a1', and `a2' to the | |
359 | -192-bit value formed by concatenating `b0', `b1', and `b2'. Addition is | |
360 | -modulo 2^192, so any carry out is lost. The result is broken into three | |
361 | -64-bit pieces which are stored at the locations pointed to by `z0Ptr', | |
362 | -`z1Ptr', and `z2Ptr'. | |
363 | -------------------------------------------------------------------------------- | |
364 | -*/ | |
365 | -INLINE void | |
366 | - add192( | |
367 | - bits64 a0, | |
368 | - bits64 a1, | |
369 | - bits64 a2, | |
370 | - bits64 b0, | |
371 | - bits64 b1, | |
372 | - bits64 b2, | |
373 | - bits64 *z0Ptr, | |
374 | - bits64 *z1Ptr, | |
375 | - bits64 *z2Ptr | |
376 | - ) | |
377 | -{ | |
378 | - bits64 z0, z1, z2; | |
379 | - int8 carry0, carry1; | |
380 | - | |
381 | - z2 = a2 + b2; | |
382 | - carry1 = ( z2 < a2 ); | |
383 | - z1 = a1 + b1; | |
384 | - carry0 = ( z1 < a1 ); | |
385 | - z0 = a0 + b0; | |
386 | - z1 += carry1; | |
387 | - z0 += ( z1 < carry1 ); | |
388 | - z0 += carry0; | |
389 | - *z2Ptr = z2; | |
390 | - *z1Ptr = z1; | |
391 | - *z0Ptr = z0; | |
392 | - | |
393 | -} | |
394 | - | |
395 | -/* | |
396 | -------------------------------------------------------------------------------- | |
397 | -Subtracts the 128-bit value formed by concatenating `b0' and `b1' from the | |
398 | -128-bit value formed by concatenating `a0' and `a1'. Subtraction is modulo | |
399 | -2^128, so any borrow out (carry out) is lost. The result is broken into two | |
400 | -64-bit pieces which are stored at the locations pointed to by `z0Ptr' and | |
401 | -`z1Ptr'. | |
402 | -------------------------------------------------------------------------------- | |
403 | -*/ | |
404 | -INLINE void | |
405 | - sub128( | |
406 | - bits64 a0, bits64 a1, bits64 b0, bits64 b1, bits64 *z0Ptr, bits64 *z1Ptr ) | |
407 | -{ | |
408 | - | |
409 | - *z1Ptr = a1 - b1; | |
410 | - *z0Ptr = a0 - b0 - ( a1 < b1 ); | |
411 | - | |
412 | -} | |
413 | - | |
414 | -/* | |
415 | -------------------------------------------------------------------------------- | |
416 | -Subtracts the 192-bit value formed by concatenating `b0', `b1', and `b2' | |
417 | -from the 192-bit value formed by concatenating `a0', `a1', and `a2'. | |
418 | -Subtraction is modulo 2^192, so any borrow out (carry out) is lost. The | |
419 | -result is broken into three 64-bit pieces which are stored at the locations | |
420 | -pointed to by `z0Ptr', `z1Ptr', and `z2Ptr'. | |
421 | -------------------------------------------------------------------------------- | |
422 | -*/ | |
423 | -INLINE void | |
424 | - sub192( | |
425 | - bits64 a0, | |
426 | - bits64 a1, | |
427 | - bits64 a2, | |
428 | - bits64 b0, | |
429 | - bits64 b1, | |
430 | - bits64 b2, | |
431 | - bits64 *z0Ptr, | |
432 | - bits64 *z1Ptr, | |
433 | - bits64 *z2Ptr | |
434 | - ) | |
435 | -{ | |
436 | - bits64 z0, z1, z2; | |
437 | - int8 borrow0, borrow1; | |
438 | - | |
439 | - z2 = a2 - b2; | |
440 | - borrow1 = ( a2 < b2 ); | |
441 | - z1 = a1 - b1; | |
442 | - borrow0 = ( a1 < b1 ); | |
443 | - z0 = a0 - b0; | |
444 | - z0 -= ( z1 < borrow1 ); | |
445 | - z1 -= borrow1; | |
446 | - z0 -= borrow0; | |
447 | - *z2Ptr = z2; | |
448 | - *z1Ptr = z1; | |
449 | - *z0Ptr = z0; | |
450 | - | |
451 | -} | |
452 | - | |
453 | -/* | |
454 | -------------------------------------------------------------------------------- | |
455 | -Multiplies `a' by `b' to obtain a 128-bit product. The product is broken | |
456 | -into two 64-bit pieces which are stored at the locations pointed to by | |
457 | -`z0Ptr' and `z1Ptr'. | |
458 | -------------------------------------------------------------------------------- | |
459 | -*/ | |
460 | -INLINE void mul64To128( bits64 a, bits64 b, bits64 *z0Ptr, bits64 *z1Ptr ) | |
461 | -{ | |
462 | - bits32 aHigh, aLow, bHigh, bLow; | |
463 | - bits64 z0, zMiddleA, zMiddleB, z1; | |
464 | - | |
465 | - aLow = a; | |
466 | - aHigh = a>>32; | |
467 | - bLow = b; | |
468 | - bHigh = b>>32; | |
469 | - z1 = ( (bits64) aLow ) * bLow; | |
470 | - zMiddleA = ( (bits64) aLow ) * bHigh; | |
471 | - zMiddleB = ( (bits64) aHigh ) * bLow; | |
472 | - z0 = ( (bits64) aHigh ) * bHigh; | |
473 | - zMiddleA += zMiddleB; | |
474 | - z0 += ( ( (bits64) ( zMiddleA < zMiddleB ) )<<32 ) + ( zMiddleA>>32 ); | |
475 | - zMiddleA <<= 32; | |
476 | - z1 += zMiddleA; | |
477 | - z0 += ( z1 < zMiddleA ); | |
478 | - *z1Ptr = z1; | |
479 | - *z0Ptr = z0; | |
480 | - | |
481 | -} | |
482 | - | |
483 | -/* | |
484 | -------------------------------------------------------------------------------- | |
485 | -Multiplies the 128-bit value formed by concatenating `a0' and `a1' by `b' to | |
486 | -obtain a 192-bit product. The product is broken into three 64-bit pieces | |
487 | -which are stored at the locations pointed to by `z0Ptr', `z1Ptr', and | |
488 | -`z2Ptr'. | |
489 | -------------------------------------------------------------------------------- | |
490 | -*/ | |
491 | -INLINE void | |
492 | - mul128By64To192( | |
493 | - bits64 a0, | |
494 | - bits64 a1, | |
495 | - bits64 b, | |
496 | - bits64 *z0Ptr, | |
497 | - bits64 *z1Ptr, | |
498 | - bits64 *z2Ptr | |
499 | - ) | |
500 | -{ | |
501 | - bits64 z0, z1, z2, more1; | |
502 | - | |
503 | - mul64To128( a1, b, &z1, &z2 ); | |
504 | - mul64To128( a0, b, &z0, &more1 ); | |
505 | - add128( z0, more1, 0, z1, &z0, &z1 ); | |
506 | - *z2Ptr = z2; | |
507 | - *z1Ptr = z1; | |
508 | - *z0Ptr = z0; | |
509 | - | |
510 | -} | |
511 | - | |
512 | -/* | |
513 | -------------------------------------------------------------------------------- | |
514 | -Multiplies the 128-bit value formed by concatenating `a0' and `a1' to the | |
515 | -128-bit value formed by concatenating `b0' and `b1' to obtain a 256-bit | |
516 | -product. The product is broken into four 64-bit pieces which are stored at | |
517 | -the locations pointed to by `z0Ptr', `z1Ptr', `z2Ptr', and `z3Ptr'. | |
518 | -------------------------------------------------------------------------------- | |
519 | -*/ | |
520 | -INLINE void | |
521 | - mul128To256( | |
522 | - bits64 a0, | |
523 | - bits64 a1, | |
524 | - bits64 b0, | |
525 | - bits64 b1, | |
526 | - bits64 *z0Ptr, | |
527 | - bits64 *z1Ptr, | |
528 | - bits64 *z2Ptr, | |
529 | - bits64 *z3Ptr | |
530 | - ) | |
531 | -{ | |
532 | - bits64 z0, z1, z2, z3; | |
533 | - bits64 more1, more2; | |
534 | - | |
535 | - mul64To128( a1, b1, &z2, &z3 ); | |
536 | - mul64To128( a1, b0, &z1, &more2 ); | |
537 | - add128( z1, more2, 0, z2, &z1, &z2 ); | |
538 | - mul64To128( a0, b0, &z0, &more1 ); | |
539 | - add128( z0, more1, 0, z1, &z0, &z1 ); | |
540 | - mul64To128( a0, b1, &more1, &more2 ); | |
541 | - add128( more1, more2, 0, z2, &more1, &z2 ); | |
542 | - add128( z0, z1, 0, more1, &z0, &z1 ); | |
543 | - *z3Ptr = z3; | |
544 | - *z2Ptr = z2; | |
545 | - *z1Ptr = z1; | |
546 | - *z0Ptr = z0; | |
547 | - | |
548 | -} | |
549 | - | |
550 | -/* | |
551 | -------------------------------------------------------------------------------- | |
552 | -Returns an approximation to the 64-bit integer quotient obtained by dividing | |
553 | -`b' into the 128-bit value formed by concatenating `a0' and `a1'. The | |
554 | -divisor `b' must be at least 2^63. If q is the exact quotient truncated | |
555 | -toward zero, the approximation returned lies between q and q + 2 inclusive. | |
556 | -If the exact quotient q is larger than 64 bits, the maximum positive 64-bit | |
557 | -unsigned integer is returned. | |
558 | -------------------------------------------------------------------------------- | |
559 | -*/ | |
560 | -static bits64 estimateDiv128To64( bits64 a0, bits64 a1, bits64 b ) | |
561 | -{ | |
562 | - bits64 b0, b1; | |
563 | - bits64 rem0, rem1, term0, term1; | |
564 | - bits64 z; | |
565 | - if ( b <= a0 ) return LIT64( 0xFFFFFFFFFFFFFFFF ); | |
566 | - b0 = b>>32; | |
567 | - z = ( b0<<32 <= a0 ) ? LIT64( 0xFFFFFFFF00000000 ) : ( a0 / b0 )<<32; | |
568 | - mul64To128( b, z, &term0, &term1 ); | |
569 | - sub128( a0, a1, term0, term1, &rem0, &rem1 ); | |
570 | - while ( ( (sbits64) rem0 ) < 0 ) { | |
571 | - z -= LIT64( 0x100000000 ); | |
572 | - b1 = b<<32; | |
573 | - add128( rem0, rem1, b0, b1, &rem0, &rem1 ); | |
574 | - } | |
575 | - rem0 = ( rem0<<32 ) | ( rem1>>32 ); | |
576 | - z |= ( b0<<32 <= rem0 ) ? 0xFFFFFFFF : rem0 / b0; | |
577 | - return z; | |
578 | - | |
579 | -} | |
580 | - | |
581 | -/* | |
582 | -------------------------------------------------------------------------------- | |
583 | -Returns an approximation to the square root of the 32-bit significand given | |
584 | -by `a'. Considered as an integer, `a' must be at least 2^31. If bit 0 of | |
585 | -`aExp' (the least significant bit) is 1, the integer returned approximates | |
586 | -2^31*sqrt(`a'/2^31), where `a' is considered an integer. If bit 0 of `aExp' | |
587 | -is 0, the integer returned approximates 2^31*sqrt(`a'/2^30). In either | |
588 | -case, the approximation returned lies strictly within +/-2 of the exact | |
589 | -value. | |
590 | -------------------------------------------------------------------------------- | |
591 | -*/ | |
592 | -static bits32 estimateSqrt32( int16 aExp, bits32 a ) | |
593 | -{ | |
594 | - static const bits16 sqrtOddAdjustments[] = { | |
595 | - 0x0004, 0x0022, 0x005D, 0x00B1, 0x011D, 0x019F, 0x0236, 0x02E0, | |
596 | - 0x039C, 0x0468, 0x0545, 0x0631, 0x072B, 0x0832, 0x0946, 0x0A67 | |
597 | - }; | |
598 | - static const bits16 sqrtEvenAdjustments[] = { | |
599 | - 0x0A2D, 0x08AF, 0x075A, 0x0629, 0x051A, 0x0429, 0x0356, 0x029E, | |
600 | - 0x0200, 0x0179, 0x0109, 0x00AF, 0x0068, 0x0034, 0x0012, 0x0002 | |
601 | - }; | |
602 | - int8 index; | |
603 | - bits32 z; | |
604 | - | |
605 | - index = ( a>>27 ) & 15; | |
606 | - if ( aExp & 1 ) { | |
607 | - z = 0x4000 + ( a>>17 ) - sqrtOddAdjustments[ index ]; | |
608 | - z = ( ( a / z )<<14 ) + ( z<<15 ); | |
609 | - a >>= 1; | |
610 | - } | |
611 | - else { | |
612 | - z = 0x8000 + ( a>>17 ) - sqrtEvenAdjustments[ index ]; | |
613 | - z = a / z + z; | |
614 | - z = ( 0x20000 <= z ) ? 0xFFFF8000 : ( z<<15 ); | |
615 | - if ( z <= a ) return (bits32) ( ( (sbits32) a )>>1 ); | |
616 | - } | |
617 | - return ( (bits32) ( ( ( (bits64) a )<<31 ) / z ) ) + ( z>>1 ); | |
618 | - | |
619 | -} | |
620 | - | |
621 | -/* | |
622 | -------------------------------------------------------------------------------- | |
623 | -Returns the number of leading 0 bits before the most-significant 1 bit | |
624 | -of `a'. If `a' is zero, 32 is returned. | |
625 | -------------------------------------------------------------------------------- | |
626 | -*/ | |
627 | -static int8 countLeadingZeros32( bits32 a ) | |
628 | -{ | |
629 | - static const int8 countLeadingZerosHigh[] = { | |
630 | - 8, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, | |
631 | - 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, | |
632 | - 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
633 | - 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, | |
634 | - 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
635 | - 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
636 | - 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
637 | - 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, | |
638 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
639 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
640 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
641 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
642 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
643 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
644 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, | |
645 | - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 | |
646 | - }; | |
647 | - int8 shiftCount; | |
648 | - | |
649 | - shiftCount = 0; | |
650 | - if ( a < 0x10000 ) { | |
651 | - shiftCount += 16; | |
652 | - a <<= 16; | |
653 | - } | |
654 | - if ( a < 0x1000000 ) { | |
655 | - shiftCount += 8; | |
656 | - a <<= 8; | |
657 | - } | |
658 | - shiftCount += countLeadingZerosHigh[ a>>24 ]; | |
659 | - return shiftCount; | |
660 | - | |
661 | -} | |
662 | - | |
663 | -/* | |
664 | -------------------------------------------------------------------------------- | |
665 | -Returns the number of leading 0 bits before the most-significant 1 bit | |
666 | -of `a'. If `a' is zero, 64 is returned. | |
667 | -------------------------------------------------------------------------------- | |
668 | -*/ | |
669 | -static int8 countLeadingZeros64( bits64 a ) | |
670 | -{ | |
671 | - int8 shiftCount; | |
672 | - | |
673 | - shiftCount = 0; | |
674 | - if ( a < ( (bits64) 1 )<<32 ) { | |
675 | - shiftCount += 32; | |
676 | - } | |
677 | - else { | |
678 | - a >>= 32; | |
679 | - } | |
680 | - shiftCount += countLeadingZeros32( a ); | |
681 | - return shiftCount; | |
682 | - | |
683 | -} | |
684 | - | |
685 | -/* | |
686 | -------------------------------------------------------------------------------- | |
687 | -Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' | |
688 | -is equal to the 128-bit value formed by concatenating `b0' and `b1'. | |
689 | -Otherwise, returns 0. | |
690 | -------------------------------------------------------------------------------- | |
691 | -*/ | |
692 | -INLINE flag eq128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) | |
693 | -{ | |
694 | - | |
695 | - return ( a0 == b0 ) && ( a1 == b1 ); | |
696 | - | |
697 | -} | |
698 | - | |
699 | -/* | |
700 | -------------------------------------------------------------------------------- | |
701 | -Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less | |
702 | -than or equal to the 128-bit value formed by concatenating `b0' and `b1'. | |
703 | -Otherwise, returns 0. | |
704 | -------------------------------------------------------------------------------- | |
705 | -*/ | |
706 | -INLINE flag le128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) | |
707 | -{ | |
708 | - | |
709 | - return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 <= b1 ) ); | |
710 | - | |
711 | -} | |
712 | - | |
713 | -/* | |
714 | -------------------------------------------------------------------------------- | |
715 | -Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is less | |
716 | -than the 128-bit value formed by concatenating `b0' and `b1'. Otherwise, | |
717 | -returns 0. | |
718 | -------------------------------------------------------------------------------- | |
719 | -*/ | |
720 | -INLINE flag lt128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) | |
721 | -{ | |
722 | - | |
723 | - return ( a0 < b0 ) || ( ( a0 == b0 ) && ( a1 < b1 ) ); | |
724 | - | |
725 | -} | |
726 | - | |
727 | -/* | |
728 | -------------------------------------------------------------------------------- | |
729 | -Returns 1 if the 128-bit value formed by concatenating `a0' and `a1' is | |
730 | -not equal to the 128-bit value formed by concatenating `b0' and `b1'. | |
731 | -Otherwise, returns 0. | |
732 | -------------------------------------------------------------------------------- | |
733 | -*/ | |
734 | -INLINE flag ne128( bits64 a0, bits64 a1, bits64 b0, bits64 b1 ) | |
735 | -{ | |
736 | - | |
737 | - return ( a0 != b0 ) || ( a1 != b1 ); | |
738 | - | |
739 | -} | |
740 | - |
target-arm/nwfpe/softfloat-specialize deleted
100644 → 0
1 | - | |
2 | -/* | |
3 | -=============================================================================== | |
4 | - | |
5 | -This C source fragment is part of the SoftFloat IEC/IEEE Floating-point | |
6 | -Arithmetic Package, Release 2. | |
7 | - | |
8 | -Written by John R. Hauser. This work was made possible in part by the | |
9 | -International Computer Science Institute, located at Suite 600, 1947 Center | |
10 | -Street, Berkeley, California 94704. Funding was partially provided by the | |
11 | -National Science Foundation under grant MIP-9311980. The original version | |
12 | -of this code was written as part of a project to build a fixed-point vector | |
13 | -processor in collaboration with the University of California at Berkeley, | |
14 | -overseen by Profs. Nelson Morgan and John Wawrzynek. More information | |
15 | -is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ | |
16 | -arithmetic/softfloat.html'. | |
17 | - | |
18 | -THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort | |
19 | -has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT | |
20 | -TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO | |
21 | -PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY | |
22 | -AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. | |
23 | - | |
24 | -Derivative works are acceptable, even for commercial purposes, so long as | |
25 | -(1) they include prominent notice that the work is derivative, and (2) they | |
26 | -include prominent notice akin to these three paragraphs for those parts of | |
27 | -this code that are retained. | |
28 | - | |
29 | -=============================================================================== | |
30 | -*/ | |
31 | - | |
32 | -/* | |
33 | -------------------------------------------------------------------------------- | |
34 | -Underflow tininess-detection mode, statically initialized to default value. | |
35 | -(The declaration in `softfloat.h' must match the `int8' type here.) | |
36 | -------------------------------------------------------------------------------- | |
37 | -*/ | |
38 | -int8 float_detect_tininess = float_tininess_after_rounding; | |
39 | - | |
40 | -/* | |
41 | -------------------------------------------------------------------------------- | |
42 | -Raises the exceptions specified by `flags'. Floating-point traps can be | |
43 | -defined here if desired. It is currently not possible for such a trap to | |
44 | -substitute a result value. If traps are not implemented, this routine | |
45 | -should be simply `float_exception_flags |= flags;'. | |
46 | - | |
47 | -ScottB: November 4, 1998 | |
48 | -Moved this function out of softfloat-specialize into fpmodule.c. | |
49 | -This effectively isolates all the changes required for integrating with the | |
50 | -Linux kernel into fpmodule.c. Porting to NetBSD should only require modifying | |
51 | -fpmodule.c to integrate with the NetBSD kernel (I hope!). | |
52 | -------------------------------------------------------------------------------- | |
53 | -*/ | |
54 | -void float_raise( int8 flags ) | |
55 | -{ | |
56 | - float_exception_flags |= flags; | |
57 | -} | |
58 | - | |
59 | -/* | |
60 | -------------------------------------------------------------------------------- | |
61 | -Internal canonical NaN format. | |
62 | -------------------------------------------------------------------------------- | |
63 | -*/ | |
64 | -typedef struct { | |
65 | - flag sign; | |
66 | - bits64 high, low; | |
67 | -} commonNaNT; | |
68 | - | |
69 | -/* | |
70 | -------------------------------------------------------------------------------- | |
71 | -The pattern for a default generated single-precision NaN. | |
72 | -------------------------------------------------------------------------------- | |
73 | -*/ | |
74 | -#define float32_default_nan 0xFFFFFFFF | |
75 | - | |
76 | -/* | |
77 | -------------------------------------------------------------------------------- | |
78 | -Returns 1 if the single-precision floating-point value `a' is a NaN; | |
79 | -otherwise returns 0. | |
80 | -------------------------------------------------------------------------------- | |
81 | -*/ | |
82 | -flag float32_is_nan( float32 a ) | |
83 | -{ | |
84 | - | |
85 | - return ( 0xFF000000 < (bits32) ( a<<1 ) ); | |
86 | - | |
87 | -} | |
88 | - | |
89 | -/* | |
90 | -------------------------------------------------------------------------------- | |
91 | -Returns 1 if the single-precision floating-point value `a' is a signaling | |
92 | -NaN; otherwise returns 0. | |
93 | -------------------------------------------------------------------------------- | |
94 | -*/ | |
95 | -flag float32_is_signaling_nan( float32 a ) | |
96 | -{ | |
97 | - | |
98 | - return ( ( ( a>>22 ) & 0x1FF ) == 0x1FE ) && ( a & 0x003FFFFF ); | |
99 | - | |
100 | -} | |
101 | - | |
102 | -/* | |
103 | -------------------------------------------------------------------------------- | |
104 | -Returns the result of converting the single-precision floating-point NaN | |
105 | -`a' to the canonical NaN format. If `a' is a signaling NaN, the invalid | |
106 | -exception is raised. | |
107 | -------------------------------------------------------------------------------- | |
108 | -*/ | |
109 | -static commonNaNT float32ToCommonNaN( float32 a ) | |
110 | -{ | |
111 | - commonNaNT z; | |
112 | - | |
113 | - if ( float32_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); | |
114 | - z.sign = a>>31; | |
115 | - z.low = 0; | |
116 | - z.high = ( (bits64) a )<<41; | |
117 | - return z; | |
118 | - | |
119 | -} | |
120 | - | |
121 | -/* | |
122 | -------------------------------------------------------------------------------- | |
123 | -Returns the result of converting the canonical NaN `a' to the single- | |
124 | -precision floating-point format. | |
125 | -------------------------------------------------------------------------------- | |
126 | -*/ | |
127 | -static float32 commonNaNToFloat32( commonNaNT a ) | |
128 | -{ | |
129 | - | |
130 | - return ( ( (bits32) a.sign )<<31 ) | 0x7FC00000 | ( a.high>>41 ); | |
131 | - | |
132 | -} | |
133 | - | |
134 | -/* | |
135 | -------------------------------------------------------------------------------- | |
136 | -Takes two single-precision floating-point values `a' and `b', one of which | |
137 | -is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a | |
138 | -signaling NaN, the invalid exception is raised. | |
139 | -------------------------------------------------------------------------------- | |
140 | -*/ | |
141 | -static float32 propagateFloat32NaN( float32 a, float32 b ) | |
142 | -{ | |
143 | - flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; | |
144 | - | |
145 | - aIsNaN = float32_is_nan( a ); | |
146 | - aIsSignalingNaN = float32_is_signaling_nan( a ); | |
147 | - bIsNaN = float32_is_nan( b ); | |
148 | - bIsSignalingNaN = float32_is_signaling_nan( b ); | |
149 | - a |= 0x00400000; | |
150 | - b |= 0x00400000; | |
151 | - if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); | |
152 | - if ( aIsNaN ) { | |
153 | - return ( aIsSignalingNaN & bIsNaN ) ? b : a; | |
154 | - } | |
155 | - else { | |
156 | - return b; | |
157 | - } | |
158 | - | |
159 | -} | |
160 | - | |
161 | -/* | |
162 | -------------------------------------------------------------------------------- | |
163 | -The pattern for a default generated double-precision NaN. | |
164 | -------------------------------------------------------------------------------- | |
165 | -*/ | |
166 | -#define float64_default_nan LIT64( 0xFFFFFFFFFFFFFFFF ) | |
167 | - | |
168 | -/* | |
169 | -------------------------------------------------------------------------------- | |
170 | -Returns 1 if the double-precision floating-point value `a' is a NaN; | |
171 | -otherwise returns 0. | |
172 | -------------------------------------------------------------------------------- | |
173 | -*/ | |
174 | -flag float64_is_nan( float64 a ) | |
175 | -{ | |
176 | - | |
177 | - return ( LIT64( 0xFFE0000000000000 ) < (bits64) ( a<<1 ) ); | |
178 | - | |
179 | -} | |
180 | - | |
181 | -/* | |
182 | -------------------------------------------------------------------------------- | |
183 | -Returns 1 if the double-precision floating-point value `a' is a signaling | |
184 | -NaN; otherwise returns 0. | |
185 | -------------------------------------------------------------------------------- | |
186 | -*/ | |
187 | -flag float64_is_signaling_nan( float64 a ) | |
188 | -{ | |
189 | - | |
190 | - return | |
191 | - ( ( ( a>>51 ) & 0xFFF ) == 0xFFE ) | |
192 | - && ( a & LIT64( 0x0007FFFFFFFFFFFF ) ); | |
193 | - | |
194 | -} | |
195 | - | |
196 | -/* | |
197 | -------------------------------------------------------------------------------- | |
198 | -Returns the result of converting the double-precision floating-point NaN | |
199 | -`a' to the canonical NaN format. If `a' is a signaling NaN, the invalid | |
200 | -exception is raised. | |
201 | -------------------------------------------------------------------------------- | |
202 | -*/ | |
203 | -static commonNaNT float64ToCommonNaN( float64 a ) | |
204 | -{ | |
205 | - commonNaNT z; | |
206 | - | |
207 | - if ( float64_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); | |
208 | - z.sign = a>>63; | |
209 | - z.low = 0; | |
210 | - z.high = a<<12; | |
211 | - return z; | |
212 | - | |
213 | -} | |
214 | - | |
215 | -/* | |
216 | -------------------------------------------------------------------------------- | |
217 | -Returns the result of converting the canonical NaN `a' to the double- | |
218 | -precision floating-point format. | |
219 | -------------------------------------------------------------------------------- | |
220 | -*/ | |
221 | -static float64 commonNaNToFloat64( commonNaNT a ) | |
222 | -{ | |
223 | - | |
224 | - return | |
225 | - ( ( (bits64) a.sign )<<63 ) | |
226 | - | LIT64( 0x7FF8000000000000 ) | |
227 | - | ( a.high>>12 ); | |
228 | - | |
229 | -} | |
230 | - | |
231 | -/* | |
232 | -------------------------------------------------------------------------------- | |
233 | -Takes two double-precision floating-point values `a' and `b', one of which | |
234 | -is a NaN, and returns the appropriate NaN result. If either `a' or `b' is a | |
235 | -signaling NaN, the invalid exception is raised. | |
236 | -------------------------------------------------------------------------------- | |
237 | -*/ | |
238 | -static float64 propagateFloat64NaN( float64 a, float64 b ) | |
239 | -{ | |
240 | - flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; | |
241 | - | |
242 | - aIsNaN = float64_is_nan( a ); | |
243 | - aIsSignalingNaN = float64_is_signaling_nan( a ); | |
244 | - bIsNaN = float64_is_nan( b ); | |
245 | - bIsSignalingNaN = float64_is_signaling_nan( b ); | |
246 | - a |= LIT64( 0x0008000000000000 ); | |
247 | - b |= LIT64( 0x0008000000000000 ); | |
248 | - if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); | |
249 | - if ( aIsNaN ) { | |
250 | - return ( aIsSignalingNaN & bIsNaN ) ? b : a; | |
251 | - } | |
252 | - else { | |
253 | - return b; | |
254 | - } | |
255 | - | |
256 | -} | |
257 | - | |
258 | -#ifdef FLOATX80 | |
259 | - | |
260 | -/* | |
261 | -------------------------------------------------------------------------------- | |
262 | -The pattern for a default generated extended double-precision NaN. The | |
263 | -`high' and `low' values hold the most- and least-significant bits, | |
264 | -respectively. | |
265 | -------------------------------------------------------------------------------- | |
266 | -*/ | |
267 | -#define floatx80_default_nan_high 0xFFFF | |
268 | -#define floatx80_default_nan_low LIT64( 0xFFFFFFFFFFFFFFFF ) | |
269 | - | |
270 | -/* | |
271 | -------------------------------------------------------------------------------- | |
272 | -Returns 1 if the extended double-precision floating-point value `a' is a | |
273 | -NaN; otherwise returns 0. | |
274 | -------------------------------------------------------------------------------- | |
275 | -*/ | |
276 | -flag floatx80_is_nan( floatx80 a ) | |
277 | -{ | |
278 | - | |
279 | - return ( ( a.high & 0x7FFF ) == 0x7FFF ) && (bits64) ( a.low<<1 ); | |
280 | - | |
281 | -} | |
282 | - | |
283 | -/* | |
284 | -------------------------------------------------------------------------------- | |
285 | -Returns 1 if the extended double-precision floating-point value `a' is a | |
286 | -signaling NaN; otherwise returns 0. | |
287 | -------------------------------------------------------------------------------- | |
288 | -*/ | |
289 | -flag floatx80_is_signaling_nan( floatx80 a ) | |
290 | -{ | |
291 | - //register int lr; | |
292 | - bits64 aLow; | |
293 | - | |
294 | - //__asm__("mov %0, lr" : : "g" (lr)); | |
295 | - //fp_printk("floatx80_is_signalling_nan() called from 0x%08x\n",lr); | |
296 | - aLow = a.low & ~ LIT64( 0x4000000000000000 ); | |
297 | - return | |
298 | - ( ( a.high & 0x7FFF ) == 0x7FFF ) | |
299 | - && (bits64) ( aLow<<1 ) | |
300 | - && ( a.low == aLow ); | |
301 | - | |
302 | -} | |
303 | - | |
304 | -/* | |
305 | -------------------------------------------------------------------------------- | |
306 | -Returns the result of converting the extended double-precision floating- | |
307 | -point NaN `a' to the canonical NaN format. If `a' is a signaling NaN, the | |
308 | -invalid exception is raised. | |
309 | -------------------------------------------------------------------------------- | |
310 | -*/ | |
311 | -static commonNaNT floatx80ToCommonNaN( floatx80 a ) | |
312 | -{ | |
313 | - commonNaNT z; | |
314 | - | |
315 | - if ( floatx80_is_signaling_nan( a ) ) float_raise( float_flag_invalid ); | |
316 | - z.sign = a.high>>15; | |
317 | - z.low = 0; | |
318 | - z.high = a.low<<1; | |
319 | - return z; | |
320 | - | |
321 | -} | |
322 | - | |
323 | -/* | |
324 | -------------------------------------------------------------------------------- | |
325 | -Returns the result of converting the canonical NaN `a' to the extended | |
326 | -double-precision floating-point format. | |
327 | -------------------------------------------------------------------------------- | |
328 | -*/ | |
329 | -static floatx80 commonNaNToFloatx80( commonNaNT a ) | |
330 | -{ | |
331 | - floatx80 z; | |
332 | - | |
333 | - z.low = LIT64( 0xC000000000000000 ) | ( a.high>>1 ); | |
334 | - z.high = ( ( (bits16) a.sign )<<15 ) | 0x7FFF; | |
335 | - return z; | |
336 | - | |
337 | -} | |
338 | - | |
339 | -/* | |
340 | -------------------------------------------------------------------------------- | |
341 | -Takes two extended double-precision floating-point values `a' and `b', one | |
342 | -of which is a NaN, and returns the appropriate NaN result. If either `a' or | |
343 | -`b' is a signaling NaN, the invalid exception is raised. | |
344 | -------------------------------------------------------------------------------- | |
345 | -*/ | |
346 | -static floatx80 propagateFloatx80NaN( floatx80 a, floatx80 b ) | |
347 | -{ | |
348 | - flag aIsNaN, aIsSignalingNaN, bIsNaN, bIsSignalingNaN; | |
349 | - | |
350 | - aIsNaN = floatx80_is_nan( a ); | |
351 | - aIsSignalingNaN = floatx80_is_signaling_nan( a ); | |
352 | - bIsNaN = floatx80_is_nan( b ); | |
353 | - bIsSignalingNaN = floatx80_is_signaling_nan( b ); | |
354 | - a.low |= LIT64( 0xC000000000000000 ); | |
355 | - b.low |= LIT64( 0xC000000000000000 ); | |
356 | - if ( aIsSignalingNaN | bIsSignalingNaN ) float_raise( float_flag_invalid ); | |
357 | - if ( aIsNaN ) { | |
358 | - return ( aIsSignalingNaN & bIsNaN ) ? b : a; | |
359 | - } | |
360 | - else { | |
361 | - return b; | |
362 | - } | |
363 | - | |
364 | -} | |
365 | - | |
366 | -#endif |