simd 0.0.1 → 0.1.0
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_floatarray.c +172 -29
- data/ext/simd/simd_floatarray.h +14 -14
- data/ext/simd/simd_types.h +20 -0
- metadata +3 -2
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA1:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 8f0bff704f3580fee55205c32ada6a1844dde4ae
|
4
|
+
data.tar.gz: 9e4f9d097fa20a349b77279535d39ea695a881a4
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: ccaa110b5ff7708c5e14dd5355f4361544832b412c2d150e9849f8106c55e1183252cbe540163941b65968037369b8618b5e66c9267e75807553aa2e7e8ccda5
|
7
|
+
data.tar.gz: e107fe3e242f6baa6d93ab6edf2886f58499d70fd690f6f1c72d7bff165932ce953026d480b0ac3a41c5e35d1ec4dc8b96c19d38d19857f0fdc8c7e759c3fa44
|
data/ext/simd/simd_floatarray.c
CHANGED
@@ -2,87 +2,123 @@
|
|
2
2
|
|
3
3
|
VALUE SIMD_FloatArray = Qnil;
|
4
4
|
|
5
|
+
/* Internal: Create the SIMD::FloatArray class. */
|
5
6
|
void Init_SIMD_FloatArray(VALUE parent)
|
6
7
|
{
|
7
8
|
SIMD_FloatArray = rb_define_class_under(parent, "FloatArray", rb_cObject);
|
8
9
|
rb_define_alloc_func(SIMD_FloatArray, allocate);
|
9
10
|
rb_define_method(SIMD_FloatArray, "initialize", method_initialize, 1);
|
10
11
|
rb_define_method(SIMD_FloatArray, "*", method_multiply, 1);
|
12
|
+
rb_define_method(SIMD_FloatArray, "/", method_divide, 1);
|
13
|
+
rb_define_method(SIMD_FloatArray, "+", method_add, 1);
|
14
|
+
rb_define_method(SIMD_FloatArray, "-", method_subtract, 1);
|
11
15
|
rb_define_method(SIMD_FloatArray, "length", method_length, 0);
|
12
16
|
rb_define_method(SIMD_FloatArray, "to_a", method_to_a, 0);
|
13
17
|
}
|
14
18
|
|
19
|
+
/* Internal: Allocate memory for the vector container. */
|
15
20
|
static VALUE allocate(VALUE klass)
|
16
21
|
{
|
17
|
-
|
18
|
-
vector
|
22
|
+
d2v_container *vector = malloc(sizeof(d2v_container));
|
23
|
+
if(vector == NULL)
|
24
|
+
{
|
25
|
+
rb_raise(rb_eNoMemError, "Unable to allocate memory");
|
26
|
+
}
|
19
27
|
|
28
|
+
vector->data = NULL; /* Avoid potentially freeing unitialized memory. */
|
20
29
|
return(Data_Wrap_Struct(klass, NULL, deallocate, vector));
|
21
30
|
}
|
22
31
|
|
23
|
-
|
32
|
+
/* Internal: Free memory from the vector container and the data array. */
|
33
|
+
static void deallocate(d2v_container *vector)
|
24
34
|
{
|
25
|
-
|
35
|
+
if(vector)
|
36
|
+
{
|
37
|
+
if(vector->data)
|
38
|
+
{
|
39
|
+
free(vector->data);
|
40
|
+
}
|
41
|
+
free(vector);
|
42
|
+
}
|
26
43
|
}
|
27
44
|
|
45
|
+
/* Public: Initialize the FloatArray object given a Ruby Array of values
|
46
|
+
* which can be cast to a double. */
|
28
47
|
static VALUE method_initialize(VALUE self, VALUE rb_array)
|
29
48
|
{
|
30
|
-
|
31
|
-
long n,m,i;
|
49
|
+
d2v_container *vector;
|
50
|
+
unsigned long n,m,i;
|
32
51
|
|
33
52
|
Check_Type(rb_array, T_ARRAY);
|
53
|
+
Data_Get_Struct(self, d2v_container, vector);
|
34
54
|
|
35
|
-
Data_Get_Struct(self, double_vector_wrapper, vector);
|
36
55
|
vector->len = n = RARRAY_LEN(rb_array);
|
37
|
-
|
38
|
-
vector->
|
56
|
+
|
57
|
+
if(vector->len < 2)
|
58
|
+
{
|
59
|
+
rb_raise(rb_eArgError, "Vectors must be at least 2 long");
|
60
|
+
}
|
61
|
+
|
62
|
+
vector->data = internal_allocate_vector_array(vector->len);
|
39
63
|
|
40
64
|
for(i = 0; i < n; i++)
|
41
65
|
vector->data[i/2].f[i%2] = NUM2DBL(rb_ary_entry(rb_array, i));
|
42
66
|
|
67
|
+
/* If the array is an odd number of elements, set the final element to 1 */
|
68
|
+
m = n + (n % 2);
|
43
69
|
if(n < m)
|
70
|
+
{
|
44
71
|
vector->data[m/2].f[1] = 1.0;
|
72
|
+
}
|
45
73
|
|
46
74
|
return(self);
|
47
75
|
}
|
48
76
|
|
77
|
+
/* Public: Multiply values contained in the data array with those contained in
|
78
|
+
* another FloatArray object, returning a new FloatArray. */
|
49
79
|
static VALUE method_multiply(VALUE self, VALUE obj)
|
50
80
|
{
|
51
|
-
|
52
|
-
|
53
|
-
|
54
|
-
Data_Get_Struct(self, double_vector_wrapper, vector);
|
55
|
-
Data_Get_Struct(obj, double_vector_wrapper, vector2);
|
56
|
-
|
57
|
-
if(vector->len != vector2->len && vector2->len != 2)
|
58
|
-
rb_raise(rb_eArgError, "Vectors must be the same size, or 2.");
|
81
|
+
return(internal_apply_operation(self, obj, func_multiply));
|
82
|
+
}
|
59
83
|
|
60
|
-
|
61
|
-
|
62
|
-
|
63
|
-
|
84
|
+
/* Public: Divide values contained in the data array by those contained in
|
85
|
+
* another FloatArray object, returning a new FloatArray. */
|
86
|
+
static VALUE method_divide(VALUE self, VALUE obj)
|
87
|
+
{
|
88
|
+
return(internal_apply_operation(self, obj, func_divide));
|
89
|
+
}
|
64
90
|
|
65
|
-
|
66
|
-
|
91
|
+
/* Public: add values contained in the data array with those contained in
|
92
|
+
* another FloatArray object, returning a new FloatArray. */
|
93
|
+
static VALUE method_add(VALUE self, VALUE obj)
|
94
|
+
{
|
95
|
+
return(internal_apply_operation(self, obj, func_add));
|
96
|
+
}
|
67
97
|
|
68
|
-
|
98
|
+
/* Public: Subtract values contained in another FloatArray object from those
|
99
|
+
* contained in the current data array object, returning a new FloatArray. */
|
100
|
+
static VALUE method_subtract(VALUE self, VALUE obj)
|
101
|
+
{
|
102
|
+
return(internal_apply_operation(self, obj, func_subtract));
|
69
103
|
}
|
70
104
|
|
105
|
+
/* Public: Return the number of elements in the Array. */
|
71
106
|
static VALUE method_length(VALUE self)
|
72
107
|
{
|
73
|
-
|
74
|
-
Data_Get_Struct(self,
|
108
|
+
d2v_container *vector;
|
109
|
+
Data_Get_Struct(self, d2v_container, vector);
|
75
110
|
|
76
111
|
return(INT2NUM(vector->len));
|
77
112
|
}
|
78
113
|
|
114
|
+
/* Public: Return a Ruby Array containing the doubles within the data array. */
|
79
115
|
static VALUE method_to_a(VALUE self)
|
80
116
|
{
|
81
|
-
long i;
|
82
|
-
|
117
|
+
unsigned long i;
|
118
|
+
d2v_container *vector;
|
83
119
|
VALUE rb_array = rb_ary_new();
|
84
120
|
|
85
|
-
Data_Get_Struct(self,
|
121
|
+
Data_Get_Struct(self, d2v_container, vector);
|
86
122
|
for(i = 0; i < vector->len; i++)
|
87
123
|
{
|
88
124
|
rb_ary_store(rb_array, i, DBL2NUM(vector->data[i/2].f[i%2]));
|
@@ -90,3 +126,110 @@ static VALUE method_to_a(VALUE self)
|
|
90
126
|
|
91
127
|
return(rb_array);
|
92
128
|
}
|
129
|
+
|
130
|
+
/* Internal: Given another FloatArray object, perform an action specified via a
|
131
|
+
* function pointer against both. */
|
132
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func)
|
133
|
+
{
|
134
|
+
unsigned long size, i;
|
135
|
+
int align;
|
136
|
+
d2v_container *v1, *v2, *r;
|
137
|
+
VALUE result_obj = allocate(SIMD_FloatArray);
|
138
|
+
|
139
|
+
Data_Get_Struct(self, d2v_container, v1);
|
140
|
+
Data_Get_Struct(obj, d2v_container, v2);
|
141
|
+
Data_Get_Struct(result_obj, d2v_container, r);
|
142
|
+
|
143
|
+
align = internal_align_vectors(v1->len, v2->len);
|
144
|
+
|
145
|
+
/* Ensure that size will be the result of ceil(len / 2.0) */
|
146
|
+
size = (v1->len + 1) / 2;
|
147
|
+
r->data = internal_allocate_vector_array(v1->len);
|
148
|
+
r->len = v1->len;
|
149
|
+
|
150
|
+
switch(align)
|
151
|
+
{
|
152
|
+
case 0: /* Same size arrays */
|
153
|
+
for(i = 0; i < size; i++)
|
154
|
+
{
|
155
|
+
r->data[i].v = func(v1->data[i].v, v2->data[i].v);
|
156
|
+
}
|
157
|
+
break;
|
158
|
+
case 1: /* Operand is exactly 2 long (size of 1 sse register) */
|
159
|
+
for(i = 0; i < size; i++)
|
160
|
+
{
|
161
|
+
r->data[i].v = func(v1->data[i].v, v2->data[0].v);
|
162
|
+
}
|
163
|
+
break;
|
164
|
+
default: /* Self is a multiple of operand's length long */
|
165
|
+
for(i = 0; i < size; i++)
|
166
|
+
{
|
167
|
+
r->data[i].v = func(v1->data[i].v, v2->data[i % v2->len].v);
|
168
|
+
}
|
169
|
+
}
|
170
|
+
|
171
|
+
return(result_obj);
|
172
|
+
}
|
173
|
+
|
174
|
+
/* Internal: Allocate memory for the data array. */
|
175
|
+
static d2v_t *internal_allocate_vector_array(unsigned long size)
|
176
|
+
{
|
177
|
+
d2v_t *vector = malloc(((size + (size % 2)) / 2 + 1) * sizeof(d2v_t));
|
178
|
+
if(vector == NULL)
|
179
|
+
{
|
180
|
+
rb_raise(rb_eNoMemError, "Unable to allocate memory");
|
181
|
+
}
|
182
|
+
|
183
|
+
return(vector);
|
184
|
+
}
|
185
|
+
|
186
|
+
/* Internal: Determine if two arrays can be acted upon, by being of equal
|
187
|
+
* lengths or with the operand's length being a multiple of the data array's. */
|
188
|
+
static int internal_align_vectors(unsigned long v1, unsigned long v2)
|
189
|
+
{
|
190
|
+
if((v1 % 2) != (v2 % 2))
|
191
|
+
{
|
192
|
+
rb_raise(rb_eArgError, "Both Vectors must be of even or odd length.");
|
193
|
+
}
|
194
|
+
|
195
|
+
if(v1 == v2)
|
196
|
+
{
|
197
|
+
return(0);
|
198
|
+
}
|
199
|
+
if(v2 == 2)
|
200
|
+
{
|
201
|
+
return(1);
|
202
|
+
}
|
203
|
+
if(v1 % v2 == 0)
|
204
|
+
{
|
205
|
+
return(2);
|
206
|
+
}
|
207
|
+
|
208
|
+
rb_raise(rb_eArgError, "Vector length must be evenly divisible by operand.");
|
209
|
+
/* Never reached */
|
210
|
+
return(-1);
|
211
|
+
}
|
212
|
+
|
213
|
+
/* Function: Multiply two vectors. */
|
214
|
+
static d2v func_multiply(d2v v1, d2v v2)
|
215
|
+
{
|
216
|
+
return(v1 * v2);
|
217
|
+
}
|
218
|
+
|
219
|
+
/* Function: Divide two vectors. */
|
220
|
+
static d2v func_divide(d2v v1, d2v v2)
|
221
|
+
{
|
222
|
+
return(v1 / v2);
|
223
|
+
}
|
224
|
+
|
225
|
+
/* Function: Add two vectors. */
|
226
|
+
static d2v func_add(d2v v1, d2v v2)
|
227
|
+
{
|
228
|
+
return(v1 + v2);
|
229
|
+
}
|
230
|
+
|
231
|
+
/* Function: Subtract two vectors. */
|
232
|
+
static d2v func_subtract(d2v v1, d2v v2)
|
233
|
+
{
|
234
|
+
return(v1 - v2);
|
235
|
+
}
|
data/ext/simd/simd_floatarray.h
CHANGED
@@ -1,22 +1,22 @@
|
|
1
1
|
#include "ruby.h"
|
2
|
-
|
3
|
-
typedef double __attribute__ ((vector_size (16))) v2d;
|
4
|
-
typedef union d2v_t
|
5
|
-
{
|
6
|
-
v2d v;
|
7
|
-
double f[2];
|
8
|
-
} d2v_t;
|
9
|
-
|
10
|
-
typedef struct double_vector_wrapper
|
11
|
-
{
|
12
|
-
d2v_t *data;
|
13
|
-
long len;
|
14
|
-
} double_vector_wrapper;
|
2
|
+
#include "simd_types.h"
|
15
3
|
|
16
4
|
static VALUE allocate(VALUE klass);
|
17
|
-
static void deallocate(
|
5
|
+
static void deallocate(d2v_container *floatarray);
|
18
6
|
|
19
7
|
static VALUE method_initialize(VALUE self, VALUE rb_array);
|
20
8
|
static VALUE method_multiply(VALUE self, VALUE obj);
|
9
|
+
static VALUE method_divide(VALUE self, VALUE obj);
|
10
|
+
static VALUE method_add(VALUE self, VALUE obj);
|
11
|
+
static VALUE method_subtract(VALUE self, VALUE obj);
|
21
12
|
static VALUE method_length(VALUE self);
|
22
13
|
static VALUE method_to_a(VALUE self);
|
14
|
+
|
15
|
+
static d2v_t *internal_allocate_vector_array(unsigned long size);
|
16
|
+
static int internal_align_vectors(unsigned long v1, unsigned long v2);
|
17
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func);
|
18
|
+
|
19
|
+
static d2v func_multiply(d2v v1, d2v v2);
|
20
|
+
static d2v func_divide(d2v v1, d2v v2);
|
21
|
+
static d2v func_add(d2v v1, d2v v2);
|
22
|
+
static d2v func_subtract(d2v v1, d2v v2);
|
@@ -0,0 +1,20 @@
|
|
1
|
+
/*
|
2
|
+
* Types for FloatArray
|
3
|
+
*
|
4
|
+
* Since ruby internally uses doubles for the Float type, SIMD::FloatArray will
|
5
|
+
* use packed double operations by default.
|
6
|
+
*/
|
7
|
+
typedef double __attribute__ ((vector_size (16))) d2v;
|
8
|
+
typedef union d2v_t
|
9
|
+
{
|
10
|
+
d2v v;
|
11
|
+
double f[2];
|
12
|
+
} d2v_t;
|
13
|
+
|
14
|
+
typedef struct d2v_container
|
15
|
+
{
|
16
|
+
d2v_t *data;
|
17
|
+
unsigned long len;
|
18
|
+
} d2v_container;
|
19
|
+
|
20
|
+
typedef d2v (*b_operation)(d2v v1, d2v v2);
|
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: simd
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.0
|
4
|
+
version: 0.1.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Tina Wuest
|
8
8
|
autorequire:
|
9
9
|
bindir: bin
|
10
10
|
cert_chain: []
|
11
|
-
date: 2014-12-
|
11
|
+
date: 2014-12-11 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: rake-compiler
|
@@ -36,6 +36,7 @@ files:
|
|
36
36
|
- ext/simd/simd.h
|
37
37
|
- ext/simd/simd_floatarray.c
|
38
38
|
- ext/simd/simd_floatarray.h
|
39
|
+
- ext/simd/simd_types.h
|
39
40
|
homepage: https://gitlab.com/wuest/simd-ruby
|
40
41
|
licenses: []
|
41
42
|
metadata: {}
|