simd 0.3.0 → 0.4.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/ext/simd/simd.c +1 -0
- data/ext/simd/simd.h +1 -0
- data/ext/simd/simd_intarray.c +42 -0
- data/ext/simd/simd_intarray.h +6 -0
- data/ext/simd/simd_longarray.c +220 -0
- data/ext/simd/simd_longarray.h +22 -0
- data/ext/simd/simd_types.h +23 -13
- metadata +3 -1
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA1:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: c15d248c872e4369c45e4151faf81e9761ea1587
|
4
|
+
data.tar.gz: cdfc5170454be15be1bc75ccc5eba08aa96c884d
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 38df7b30a113b325bba7be834fcbe24c44738706495579af5a9e4b508e4db4375e2f36b16a97612eab4dd8da379c450d062c6e0bceb6e8ddf308a952ecf6419c
|
7
|
+
data.tar.gz: 342680d7f4f6690c8b3338de32ef192aaadb2952461e62e86a848650414b09d793d71e70c5b38909e0c882f062e346995f86f40d3dbf1de692cbe3eb0cfb0c73
|
data/ext/simd/simd.c
CHANGED
data/ext/simd/simd.h
CHANGED
data/ext/simd/simd_intarray.c
CHANGED
@@ -12,6 +12,9 @@ void Init_SIMD_IntArray(VALUE parent)
|
|
12
12
|
rb_define_method(SIMD_IntArray, "/", method_divide, 1);
|
13
13
|
rb_define_method(SIMD_IntArray, "+", method_add, 1);
|
14
14
|
rb_define_method(SIMD_IntArray, "-", method_subtract, 1);
|
15
|
+
rb_define_method(SIMD_IntArray, "&", method_and, 1);
|
16
|
+
rb_define_method(SIMD_IntArray, "|", method_or, 1);
|
17
|
+
rb_define_method(SIMD_IntArray, "^", method_xor, 1);
|
15
18
|
rb_define_method(SIMD_IntArray, "length", method_length, 0);
|
16
19
|
rb_define_method(SIMD_IntArray, "to_a", method_to_a, 0);
|
17
20
|
}
|
@@ -73,6 +76,27 @@ static VALUE method_add(VALUE self, VALUE obj)
|
|
73
76
|
return(internal_apply_operation(self, obj, func_add));
|
74
77
|
}
|
75
78
|
|
79
|
+
/* Public: and values contained in the data array with those contained in
|
80
|
+
* another FloatArray object, returning a new FloatArray. */
|
81
|
+
static VALUE method_and(VALUE self, VALUE obj)
|
82
|
+
{
|
83
|
+
return(internal_apply_operation(self, obj, func_and));
|
84
|
+
}
|
85
|
+
|
86
|
+
/* Public: or values contained in the data array with those contained in
|
87
|
+
* another FloatArray object, returning a new FloatArray. */
|
88
|
+
static VALUE method_or(VALUE self, VALUE obj)
|
89
|
+
{
|
90
|
+
return(internal_apply_operation(self, obj, func_or));
|
91
|
+
}
|
92
|
+
|
93
|
+
/* Public: xor values contained in the data array with those contained in
|
94
|
+
* another FloatArray object, returning a new FloatArray. */
|
95
|
+
static VALUE method_xor(VALUE self, VALUE obj)
|
96
|
+
{
|
97
|
+
return(internal_apply_operation(self, obj, func_xor));
|
98
|
+
}
|
99
|
+
|
76
100
|
/* Public: Subtract values contained in another FloatArray object from those
|
77
101
|
* contained in the current data array object, returning a new FloatArray. */
|
78
102
|
static VALUE method_subtract(VALUE self, VALUE obj)
|
@@ -179,3 +203,21 @@ static void func_subtract(void *v1, void *v2, void *r)
|
|
179
203
|
{
|
180
204
|
*(i4v *)r = *(i4v *)v1 - *(i4v *)v2;
|
181
205
|
}
|
206
|
+
|
207
|
+
/* Function: Perform a binary AND on two vectors. */
|
208
|
+
static void func_and(void *v1, void *v2, void *r)
|
209
|
+
{
|
210
|
+
*(i4v *)r = *(i4v *)v1 & *(i4v *)v2;
|
211
|
+
}
|
212
|
+
|
213
|
+
/* Function: Perform a binary OR on two vectors. */
|
214
|
+
static void func_or(void *v1, void *v2, void *r)
|
215
|
+
{
|
216
|
+
*(i4v *)r = *(i4v *)v1 | *(i4v *)v2;
|
217
|
+
}
|
218
|
+
|
219
|
+
/* Function: Perform a binary XOR on two vectors. */
|
220
|
+
static void func_xor(void *v1, void *v2, void *r)
|
221
|
+
{
|
222
|
+
*(i4v *)r = *(i4v *)v1 ^ *(i4v *)v2;
|
223
|
+
}
|
data/ext/simd/simd_intarray.h
CHANGED
@@ -5,6 +5,9 @@ static VALUE method_initialize(VALUE self, VALUE rb_array);
|
|
5
5
|
static VALUE method_multiply(VALUE self, VALUE obj);
|
6
6
|
static VALUE method_divide(VALUE self, VALUE obj);
|
7
7
|
static VALUE method_add(VALUE self, VALUE obj);
|
8
|
+
static VALUE method_and(VALUE self, VALUE obj);
|
9
|
+
static VALUE method_or(VALUE self, VALUE obj);
|
10
|
+
static VALUE method_xor(VALUE self, VALUE obj);
|
8
11
|
static VALUE method_subtract(VALUE self, VALUE obj);
|
9
12
|
static VALUE method_to_a(VALUE self);
|
10
13
|
|
@@ -14,3 +17,6 @@ static void func_multiply(void *v1, void *v2, void *r);
|
|
14
17
|
static void func_divide(void *v1, void *v2, void *r);
|
15
18
|
static void func_add(void *v1, void *v2, void *r);
|
16
19
|
static void func_subtract(void *v1, void *v2, void *r);
|
20
|
+
static void func_and(void *v1, void *v2, void *r);
|
21
|
+
static void func_or(void *v1, void *v2, void *r);
|
22
|
+
static void func_xor(void *v1, void *v2, void *r);
|
@@ -0,0 +1,220 @@
|
|
1
|
+
#include "simd_longarray.h"
|
2
|
+
|
3
|
+
VALUE SIMD_LongArray = Qnil;
|
4
|
+
|
5
|
+
/* Internal: Create the SIMD::FloatArray class. */
|
6
|
+
void Init_SIMD_LongArray(VALUE parent)
|
7
|
+
{
|
8
|
+
SIMD_LongArray = rb_define_class_under(parent, "LongArray", rb_cObject);
|
9
|
+
rb_define_alloc_func(SIMD_LongArray, allocate);
|
10
|
+
rb_define_method(SIMD_LongArray, "initialize", method_initialize, 1);
|
11
|
+
rb_define_method(SIMD_LongArray, "*", method_multiply, 1);
|
12
|
+
rb_define_method(SIMD_LongArray, "/", method_divide, 1);
|
13
|
+
rb_define_method(SIMD_LongArray, "+", method_add, 1);
|
14
|
+
rb_define_method(SIMD_LongArray, "-", method_subtract, 1);
|
15
|
+
rb_define_method(SIMD_LongArray, "&", method_and, 1);
|
16
|
+
rb_define_method(SIMD_LongArray, "|", method_or, 1);
|
17
|
+
rb_define_method(SIMD_LongArray, "^", method_xor, 1);
|
18
|
+
rb_define_method(SIMD_LongArray, "length", method_length, 0);
|
19
|
+
rb_define_method(SIMD_LongArray, "to_a", method_to_a, 0);
|
20
|
+
}
|
21
|
+
|
22
|
+
/* Public: Initialize the FloatArray object given a Ruby Array of values
|
23
|
+
* which can be cast to a double. */
|
24
|
+
static VALUE method_initialize(VALUE self, VALUE rb_array)
|
25
|
+
{
|
26
|
+
vector_t *vector;
|
27
|
+
l2v_t *data;
|
28
|
+
unsigned long n,m,i;
|
29
|
+
|
30
|
+
Check_Type(rb_array, T_ARRAY);
|
31
|
+
Data_Get_Struct(self, vector_t, vector);
|
32
|
+
|
33
|
+
vector->len = n = RARRAY_LEN(rb_array);
|
34
|
+
|
35
|
+
if(vector->len < 2)
|
36
|
+
{
|
37
|
+
rb_raise(rb_eArgError, "Vectors must be at least 2 long");
|
38
|
+
}
|
39
|
+
|
40
|
+
vector->data = internal_allocate_vector_array(vector->len, sizeof(l2v_t));
|
41
|
+
|
42
|
+
data = (l2v_t *)vector->data;
|
43
|
+
for(i = 0; i < vector->len; i++)
|
44
|
+
{
|
45
|
+
data[i/2].f[i%2] = NUM2LONG(rb_ary_entry(rb_array, i));
|
46
|
+
}
|
47
|
+
|
48
|
+
/* If the array is an odd number of elements, set the final element to 1 */
|
49
|
+
m = n + (n % 2);
|
50
|
+
for(i = n % 2; i > 0; i--)
|
51
|
+
{
|
52
|
+
data[m/2].f[i] = 1;
|
53
|
+
}
|
54
|
+
|
55
|
+
return(self);
|
56
|
+
}
|
57
|
+
|
58
|
+
/* Public: Multiply values contained in the data array with those contained in
|
59
|
+
* another FloatArray object, returning a new FloatArray. */
|
60
|
+
static VALUE method_multiply(VALUE self, VALUE obj)
|
61
|
+
{
|
62
|
+
return(internal_apply_operation(self, obj, func_multiply));
|
63
|
+
}
|
64
|
+
|
65
|
+
/* Public: Divide values contained in the data array by those contained in
|
66
|
+
* another FloatArray object, returning a new FloatArray. */
|
67
|
+
static VALUE method_divide(VALUE self, VALUE obj)
|
68
|
+
{
|
69
|
+
return(internal_apply_operation(self, obj, func_divide));
|
70
|
+
}
|
71
|
+
|
72
|
+
/* Public: add values contained in the data array with those contained in
|
73
|
+
* another FloatArray object, returning a new FloatArray. */
|
74
|
+
static VALUE method_add(VALUE self, VALUE obj)
|
75
|
+
{
|
76
|
+
return(internal_apply_operation(self, obj, func_add));
|
77
|
+
}
|
78
|
+
|
79
|
+
/* Public: and values contained in the data array with those contained in
|
80
|
+
* another FloatArray object, returning a new FloatArray. */
|
81
|
+
static VALUE method_and(VALUE self, VALUE obj)
|
82
|
+
{
|
83
|
+
return(internal_apply_operation(self, obj, func_and));
|
84
|
+
}
|
85
|
+
|
86
|
+
/* Public: or values contained in the data array with those contained in
|
87
|
+
* another FloatArray object, returning a new FloatArray. */
|
88
|
+
static VALUE method_or(VALUE self, VALUE obj)
|
89
|
+
{
|
90
|
+
return(internal_apply_operation(self, obj, func_or));
|
91
|
+
}
|
92
|
+
|
93
|
+
/* Public: xor values contained in the data array with those contained in
|
94
|
+
* another FloatArray object, returning a new FloatArray. */
|
95
|
+
static VALUE method_xor(VALUE self, VALUE obj)
|
96
|
+
{
|
97
|
+
return(internal_apply_operation(self, obj, func_xor));
|
98
|
+
}
|
99
|
+
|
100
|
+
/* Public: Subtract values contained in another FloatArray object from those
|
101
|
+
* contained in the current data array object, returning a new FloatArray. */
|
102
|
+
static VALUE method_subtract(VALUE self, VALUE obj)
|
103
|
+
{
|
104
|
+
return(internal_apply_operation(self, obj, func_subtract));
|
105
|
+
}
|
106
|
+
|
107
|
+
/* Public: Return a Ruby Array containing the doubles within the data array. */
|
108
|
+
static VALUE method_to_a(VALUE self)
|
109
|
+
{
|
110
|
+
unsigned long i;
|
111
|
+
vector_t *vector;
|
112
|
+
l2v_t *data;
|
113
|
+
VALUE rb_array = rb_ary_new();
|
114
|
+
|
115
|
+
Data_Get_Struct(self, vector_t, vector);
|
116
|
+
data = (l2v_t *)vector->data;
|
117
|
+
for(i = 0; i < vector->len; i++)
|
118
|
+
{
|
119
|
+
rb_ary_store(rb_array, i, LONG2NUM(data[i/2].f[i%2]));
|
120
|
+
}
|
121
|
+
|
122
|
+
return(rb_array);
|
123
|
+
}
|
124
|
+
|
125
|
+
/* Internal: Given another FloatArray object, perform an action specified via a
|
126
|
+
* function pointer against both. */
|
127
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func)
|
128
|
+
{
|
129
|
+
unsigned long size, i;
|
130
|
+
int align;
|
131
|
+
vector_t *v1, *v2, *rv;
|
132
|
+
l2v_t *d1, *d2, *r;
|
133
|
+
VALUE result_obj = allocate(SIMD_LongArray);
|
134
|
+
|
135
|
+
Data_Get_Struct(self, vector_t, v1);
|
136
|
+
Data_Get_Struct(obj, vector_t, v2);
|
137
|
+
Data_Get_Struct(result_obj, vector_t, rv);
|
138
|
+
rv->data = internal_allocate_vector_array(v1->len, sizeof(l2v_t));
|
139
|
+
|
140
|
+
align = internal_align_vectors(v1->len, v2->len, 2);
|
141
|
+
|
142
|
+
/* Ensure that size will be the result of ceil(len / 4.0) */
|
143
|
+
size = (v1->len + 1) / 2;
|
144
|
+
|
145
|
+
d1 = (l2v_t *)v1->data;
|
146
|
+
d2 = (l2v_t *)v2->data;
|
147
|
+
r = (l2v_t *)rv->data;
|
148
|
+
|
149
|
+
rv->len = v1->len;
|
150
|
+
|
151
|
+
switch(align)
|
152
|
+
{
|
153
|
+
case 0: /* Same size arrays */
|
154
|
+
for(i = 0; i < size; i++)
|
155
|
+
{
|
156
|
+
func(&d1[i].v, &d2[i].v, &r[i].v);
|
157
|
+
}
|
158
|
+
break;
|
159
|
+
case 1: /* Operand is exactly 4 long (size of 1 sse register) */
|
160
|
+
for(i = 0; i < size; i++)
|
161
|
+
{
|
162
|
+
func(&d1[i].v, &d2[0].v, &r[i].v);
|
163
|
+
}
|
164
|
+
break;
|
165
|
+
default: /* Self is a multiple of operand's length long */
|
166
|
+
for(i = 0; i < size; i++)
|
167
|
+
{
|
168
|
+
func(&d1[i].v, &d2[i % v2->len].v, &r[i].v);
|
169
|
+
}
|
170
|
+
}
|
171
|
+
|
172
|
+
if(rv->len != rv->len + (rv->len % 2))
|
173
|
+
{
|
174
|
+
r[size].f[1] = 1;
|
175
|
+
}
|
176
|
+
|
177
|
+
return(result_obj);
|
178
|
+
}
|
179
|
+
|
180
|
+
/* Function: Multiply two vectors. */
|
181
|
+
static void func_multiply(void *v1, void *v2, void *r)
|
182
|
+
{
|
183
|
+
*(l2v *)r = *(l2v *)v1 * *(l2v *)v2;
|
184
|
+
}
|
185
|
+
|
186
|
+
/* Function: Divide two vectors. */
|
187
|
+
static void func_divide(void *v1, void *v2, void *r)
|
188
|
+
{
|
189
|
+
*(l2v *)r = *(l2v *)v1 / *(l2v *)v2;
|
190
|
+
}
|
191
|
+
|
192
|
+
/* Function: Add two vectors. */
|
193
|
+
static void func_add(void *v1, void *v2, void *r)
|
194
|
+
{
|
195
|
+
*(l2v *)r = *(l2v *)v1 + *(l2v *)v2;
|
196
|
+
}
|
197
|
+
|
198
|
+
/* Function: Subtract two vectors. */
|
199
|
+
static void func_subtract(void *v1, void *v2, void *r)
|
200
|
+
{
|
201
|
+
*(l2v *)r = *(l2v *)v1 - *(l2v *)v2;
|
202
|
+
}
|
203
|
+
|
204
|
+
/* Function: Perform a binary AND on two vectors. */
|
205
|
+
static void func_and(void *v1, void *v2, void *r)
|
206
|
+
{
|
207
|
+
*(l2v *)r = *(l2v *)v1 & *(l2v *)v2;
|
208
|
+
}
|
209
|
+
|
210
|
+
/* Function: Perform a binary OR on two vectors. */
|
211
|
+
static void func_or(void *v1, void *v2, void *r)
|
212
|
+
{
|
213
|
+
*(l2v *)r = *(l2v *)v1 | *(l2v *)v2;
|
214
|
+
}
|
215
|
+
|
216
|
+
/* Function: Perform a binary XOR on two vectors. */
|
217
|
+
static void func_xor(void *v1, void *v2, void *r)
|
218
|
+
{
|
219
|
+
*(l2v *)r = *(l2v *)v1 ^ *(l2v *)v2;
|
220
|
+
}
|
@@ -0,0 +1,22 @@
|
|
1
|
+
#include "ruby.h"
|
2
|
+
#include "simd_common.h"
|
3
|
+
|
4
|
+
static VALUE method_initialize(VALUE self, VALUE rb_array);
|
5
|
+
static VALUE method_multiply(VALUE self, VALUE obj);
|
6
|
+
static VALUE method_divide(VALUE self, VALUE obj);
|
7
|
+
static VALUE method_add(VALUE self, VALUE obj);
|
8
|
+
static VALUE method_and(VALUE self, VALUE obj);
|
9
|
+
static VALUE method_or(VALUE self, VALUE obj);
|
10
|
+
static VALUE method_xor(VALUE self, VALUE obj);
|
11
|
+
static VALUE method_subtract(VALUE self, VALUE obj);
|
12
|
+
static VALUE method_to_a(VALUE self);
|
13
|
+
|
14
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func);
|
15
|
+
|
16
|
+
static void func_multiply(void *v1, void *v2, void *r);
|
17
|
+
static void func_divide(void *v1, void *v2, void *r);
|
18
|
+
static void func_add(void *v1, void *v2, void *r);
|
19
|
+
static void func_subtract(void *v1, void *v2, void *r);
|
20
|
+
static void func_and(void *v1, void *v2, void *r);
|
21
|
+
static void func_or(void *v1, void *v2, void *r);
|
22
|
+
static void func_xor(void *v1, void *v2, void *r);
|
data/ext/simd/simd_types.h
CHANGED
@@ -1,18 +1,5 @@
|
|
1
1
|
#pragma once
|
2
2
|
|
3
|
-
/*
|
4
|
-
* Types for FloatArray
|
5
|
-
*
|
6
|
-
* Since ruby internally uses doubles for the Float type, SIMD::FloatArray will
|
7
|
-
* use packed double operations by default.
|
8
|
-
*/
|
9
|
-
typedef double __attribute__ ((vector_size (16))) d2v;
|
10
|
-
typedef union d2v_t
|
11
|
-
{
|
12
|
-
d2v v;
|
13
|
-
double f[2];
|
14
|
-
} d2v_t;
|
15
|
-
|
16
3
|
/*
|
17
4
|
* Types for SmallFloatArray
|
18
5
|
*
|
@@ -27,6 +14,19 @@ typedef union f4v_t
|
|
27
14
|
float f[4];
|
28
15
|
} f4v_t;
|
29
16
|
|
17
|
+
/*
|
18
|
+
* Types for FloatArray
|
19
|
+
*
|
20
|
+
* Since ruby internally uses doubles for the Float type, SIMD::FloatArray will
|
21
|
+
* use packed double operations by default.
|
22
|
+
*/
|
23
|
+
typedef double __attribute__ ((vector_size (16))) d2v;
|
24
|
+
typedef union d2v_t
|
25
|
+
{
|
26
|
+
d2v v;
|
27
|
+
double f[2];
|
28
|
+
} d2v_t;
|
29
|
+
|
30
30
|
/*
|
31
31
|
* Types for IntArray
|
32
32
|
*/
|
@@ -37,6 +37,16 @@ typedef union i4v_t
|
|
37
37
|
int f[4];
|
38
38
|
} i4v_t;
|
39
39
|
|
40
|
+
/*
|
41
|
+
* Types for LongArray
|
42
|
+
*/
|
43
|
+
typedef long int __attribute__ ((vector_size (16))) l2v;
|
44
|
+
typedef union l2v_t
|
45
|
+
{
|
46
|
+
l2v v;
|
47
|
+
long int f[2];
|
48
|
+
} l2v_t;
|
49
|
+
|
40
50
|
typedef struct vector_t
|
41
51
|
{
|
42
52
|
void *data;
|
metadata
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: simd
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.
|
4
|
+
version: 0.4.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Tina Wuest
|
@@ -40,6 +40,8 @@ files:
|
|
40
40
|
- ext/simd/simd_floatarray.h
|
41
41
|
- ext/simd/simd_intarray.c
|
42
42
|
- ext/simd/simd_intarray.h
|
43
|
+
- ext/simd/simd_longarray.c
|
44
|
+
- ext/simd/simd_longarray.h
|
43
45
|
- ext/simd/simd_smallfloatarray.c
|
44
46
|
- ext/simd/simd_smallfloatarray.h
|
45
47
|
- ext/simd/simd_types.h
|