simd 0.2.0 → 0.3.0
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_floatarray.c +16 -11
- data/ext/simd/simd_floatarray.h +4 -13
- data/ext/simd/simd_intarray.c +181 -0
- data/ext/simd/simd_intarray.h +16 -0
- data/ext/simd/simd_smallfloatarray.c +20 -12
- data/ext/simd/simd_smallfloatarray.h +5 -14
- data/ext/simd/simd_types.h +13 -4
- 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: e2917ec7ede39c14aaac0a85f4e0c6cd6cf204cd
|
4
|
+
data.tar.gz: 8bf4b61b6337ba5e5d833797dcffec0e2de32480
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: e7ffa5a5ca85d1affc0326871517ba1dfde05d46544c0c8d803edf307d10ffe87dfdd8f78bbfd74b672f488d266037718dfd0c7d59b96967cb69ab3d8943b64e
|
7
|
+
data.tar.gz: 69a5db3a4baf9fa3b68ff13a7d418337b20a237b150253d2cfd09421b8f63f2a11a50f51cd86d5412ab5a71bc46841143132cbe6af74f31d618408b9b1040d03
|
data/ext/simd/simd.c
CHANGED
data/ext/simd/simd.h
CHANGED
data/ext/simd/simd_floatarray.c
CHANGED
@@ -129,45 +129,50 @@ static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func)
|
|
129
129
|
case 0: /* Same size arrays */
|
130
130
|
for(i = 0; i < size; i++)
|
131
131
|
{
|
132
|
-
|
132
|
+
func(&d1[i].v, &d2[i].v, &r[i].v);
|
133
133
|
}
|
134
134
|
break;
|
135
135
|
case 1: /* Operand is exactly 2 long (size of 1 sse register) */
|
136
136
|
for(i = 0; i < size; i++)
|
137
137
|
{
|
138
|
-
|
138
|
+
func(&d1[i].v, &d2[0].v, &r[i].v);
|
139
139
|
}
|
140
140
|
break;
|
141
141
|
default: /* Self is a multiple of operand's length long */
|
142
142
|
for(i = 0; i < size; i++)
|
143
143
|
{
|
144
|
-
|
144
|
+
func(&d1[i].v, &d2[i % v2->len].v, &r[i].v);
|
145
145
|
}
|
146
146
|
}
|
147
147
|
|
148
|
+
if(rv->len != rv->len + (rv->len % 2))
|
149
|
+
{
|
150
|
+
r[size].f[1] = 1;
|
151
|
+
}
|
152
|
+
|
148
153
|
return(result_obj);
|
149
154
|
}
|
150
155
|
|
151
156
|
/* Function: Multiply two vectors. */
|
152
|
-
static
|
157
|
+
static void func_multiply(void *v1, void *v2, void *r)
|
153
158
|
{
|
154
|
-
|
159
|
+
*(d2v *)r = *(d2v *)v1 * *(d2v *)v2;
|
155
160
|
}
|
156
161
|
|
157
162
|
/* Function: Divide two vectors. */
|
158
|
-
static
|
163
|
+
static void func_divide(void *v1, void *v2, void *r)
|
159
164
|
{
|
160
|
-
|
165
|
+
*(d2v *)r = *(d2v *)v1 / *(d2v *)v2;
|
161
166
|
}
|
162
167
|
|
163
168
|
/* Function: Add two vectors. */
|
164
|
-
static
|
169
|
+
static void func_add(void *v1, void *v2, void *r)
|
165
170
|
{
|
166
|
-
|
171
|
+
*(d2v *)r = *(d2v *)v1 + *(d2v *)v2;
|
167
172
|
}
|
168
173
|
|
169
174
|
/* Function: Subtract two vectors. */
|
170
|
-
static
|
175
|
+
static void func_subtract(void *v1, void *v2, void *r)
|
171
176
|
{
|
172
|
-
|
177
|
+
*(d2v *)r = *(d2v *)v1 - *(d2v *)v2;
|
173
178
|
}
|
data/ext/simd/simd_floatarray.h
CHANGED
@@ -1,11 +1,6 @@
|
|
1
1
|
#include "ruby.h"
|
2
2
|
#include "simd_common.h"
|
3
3
|
|
4
|
-
/*
|
5
|
-
static VALUE allocate(VALUE klass);
|
6
|
-
static void deallocate(d2v_container *floatarray);
|
7
|
-
*/
|
8
|
-
|
9
4
|
static VALUE method_initialize(VALUE self, VALUE rb_array);
|
10
5
|
static VALUE method_multiply(VALUE self, VALUE obj);
|
11
6
|
static VALUE method_divide(VALUE self, VALUE obj);
|
@@ -13,13 +8,9 @@ static VALUE method_add(VALUE self, VALUE obj);
|
|
13
8
|
static VALUE method_subtract(VALUE self, VALUE obj);
|
14
9
|
static VALUE method_to_a(VALUE self);
|
15
10
|
|
16
|
-
/*
|
17
|
-
static d2v_t *internal_allocate_vector_array(unsigned long size);
|
18
|
-
static int internal_align_vectors(unsigned long v1, unsigned long v2);
|
19
|
-
*/
|
20
11
|
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func);
|
21
12
|
|
22
|
-
static
|
23
|
-
static
|
24
|
-
static
|
25
|
-
static
|
13
|
+
static void func_multiply(void *v1, void *v2, void *r);
|
14
|
+
static void func_divide(void *v1, void *v2, void *r);
|
15
|
+
static void func_add(void *v1, void *v2, void *r);
|
16
|
+
static void func_subtract(void *v1, void *v2, void *r);
|
@@ -0,0 +1,181 @@
|
|
1
|
+
#include "simd_intarray.h"
|
2
|
+
|
3
|
+
VALUE SIMD_IntArray = Qnil;
|
4
|
+
|
5
|
+
/* Internal: Create the SIMD::FloatArray class. */
|
6
|
+
void Init_SIMD_IntArray(VALUE parent)
|
7
|
+
{
|
8
|
+
SIMD_IntArray = rb_define_class_under(parent, "IntArray", rb_cObject);
|
9
|
+
rb_define_alloc_func(SIMD_IntArray, allocate);
|
10
|
+
rb_define_method(SIMD_IntArray, "initialize", method_initialize, 1);
|
11
|
+
rb_define_method(SIMD_IntArray, "*", method_multiply, 1);
|
12
|
+
rb_define_method(SIMD_IntArray, "/", method_divide, 1);
|
13
|
+
rb_define_method(SIMD_IntArray, "+", method_add, 1);
|
14
|
+
rb_define_method(SIMD_IntArray, "-", method_subtract, 1);
|
15
|
+
rb_define_method(SIMD_IntArray, "length", method_length, 0);
|
16
|
+
rb_define_method(SIMD_IntArray, "to_a", method_to_a, 0);
|
17
|
+
}
|
18
|
+
|
19
|
+
/* Public: Initialize the FloatArray object given a Ruby Array of values
|
20
|
+
* which can be cast to a double. */
|
21
|
+
static VALUE method_initialize(VALUE self, VALUE rb_array)
|
22
|
+
{
|
23
|
+
vector_t *vector;
|
24
|
+
i4v_t *data;
|
25
|
+
unsigned long n,m,i;
|
26
|
+
|
27
|
+
Check_Type(rb_array, T_ARRAY);
|
28
|
+
Data_Get_Struct(self, vector_t, vector);
|
29
|
+
|
30
|
+
vector->len = n = RARRAY_LEN(rb_array);
|
31
|
+
|
32
|
+
if(vector->len < 4)
|
33
|
+
{
|
34
|
+
rb_raise(rb_eArgError, "Vectors must be at least 4 long");
|
35
|
+
}
|
36
|
+
|
37
|
+
vector->data = internal_allocate_vector_array(vector->len, sizeof(i4v_t));
|
38
|
+
|
39
|
+
data = (i4v_t *)vector->data;
|
40
|
+
for(i = 0; i < vector->len; i++)
|
41
|
+
{
|
42
|
+
data[i/4].f[i%4] = NUM2INT(rb_ary_entry(rb_array, i));
|
43
|
+
}
|
44
|
+
|
45
|
+
/* If the array is an odd number of elements, set the final element to 1 */
|
46
|
+
m = n + (n % 4);
|
47
|
+
for(i = n % 4; i > 0; i--)
|
48
|
+
{
|
49
|
+
data[m/4].f[i] = 1.0;
|
50
|
+
}
|
51
|
+
|
52
|
+
return(self);
|
53
|
+
}
|
54
|
+
|
55
|
+
/* Public: Multiply values contained in the data array with those contained in
|
56
|
+
* another FloatArray object, returning a new FloatArray. */
|
57
|
+
static VALUE method_multiply(VALUE self, VALUE obj)
|
58
|
+
{
|
59
|
+
return(internal_apply_operation(self, obj, func_multiply));
|
60
|
+
}
|
61
|
+
|
62
|
+
/* Public: Divide values contained in the data array by those contained in
|
63
|
+
* another FloatArray object, returning a new FloatArray. */
|
64
|
+
static VALUE method_divide(VALUE self, VALUE obj)
|
65
|
+
{
|
66
|
+
return(internal_apply_operation(self, obj, func_divide));
|
67
|
+
}
|
68
|
+
|
69
|
+
/* Public: add values contained in the data array with those contained in
|
70
|
+
* another FloatArray object, returning a new FloatArray. */
|
71
|
+
static VALUE method_add(VALUE self, VALUE obj)
|
72
|
+
{
|
73
|
+
return(internal_apply_operation(self, obj, func_add));
|
74
|
+
}
|
75
|
+
|
76
|
+
/* Public: Subtract values contained in another FloatArray object from those
|
77
|
+
* contained in the current data array object, returning a new FloatArray. */
|
78
|
+
static VALUE method_subtract(VALUE self, VALUE obj)
|
79
|
+
{
|
80
|
+
return(internal_apply_operation(self, obj, func_subtract));
|
81
|
+
}
|
82
|
+
|
83
|
+
/* Public: Return a Ruby Array containing the doubles within the data array. */
|
84
|
+
static VALUE method_to_a(VALUE self)
|
85
|
+
{
|
86
|
+
unsigned long i;
|
87
|
+
vector_t *vector;
|
88
|
+
i4v_t *data;
|
89
|
+
VALUE rb_array = rb_ary_new();
|
90
|
+
|
91
|
+
Data_Get_Struct(self, vector_t, vector);
|
92
|
+
data = (i4v_t *)vector->data;
|
93
|
+
for(i = 0; i < vector->len; i++)
|
94
|
+
{
|
95
|
+
rb_ary_store(rb_array, i, INT2NUM(data[i/4].f[i%4]));
|
96
|
+
}
|
97
|
+
|
98
|
+
return(rb_array);
|
99
|
+
}
|
100
|
+
|
101
|
+
/* Internal: Given another FloatArray object, perform an action specified via a
|
102
|
+
* function pointer against both. */
|
103
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func)
|
104
|
+
{
|
105
|
+
unsigned long size, i;
|
106
|
+
int align;
|
107
|
+
vector_t *v1, *v2, *rv;
|
108
|
+
i4v_t *d1, *d2, *r;
|
109
|
+
VALUE result_obj = allocate(SIMD_IntArray);
|
110
|
+
|
111
|
+
Data_Get_Struct(self, vector_t, v1);
|
112
|
+
Data_Get_Struct(obj, vector_t, v2);
|
113
|
+
Data_Get_Struct(result_obj, vector_t, rv);
|
114
|
+
rv->data = internal_allocate_vector_array(v1->len, sizeof(i4v_t));
|
115
|
+
|
116
|
+
align = internal_align_vectors(v1->len, v2->len, 4);
|
117
|
+
|
118
|
+
/* Ensure that size will be the result of ceil(len / 4.0) */
|
119
|
+
size = (v1->len + 3) / 4;
|
120
|
+
|
121
|
+
d1 = (i4v_t *)v1->data;
|
122
|
+
d2 = (i4v_t *)v2->data;
|
123
|
+
r = (i4v_t *)rv->data;
|
124
|
+
|
125
|
+
rv->len = v1->len;
|
126
|
+
|
127
|
+
switch(align)
|
128
|
+
{
|
129
|
+
case 0: /* Same size arrays */
|
130
|
+
for(i = 0; i < size; i++)
|
131
|
+
{
|
132
|
+
func(&d1[i].v, &d2[i].v, &r[i].v);
|
133
|
+
}
|
134
|
+
break;
|
135
|
+
case 1: /* Operand is exactly 4 long (size of 1 sse register) */
|
136
|
+
for(i = 0; i < size; i++)
|
137
|
+
{
|
138
|
+
func(&d1[i].v, &d2[0].v, &r[i].v);
|
139
|
+
}
|
140
|
+
break;
|
141
|
+
default: /* Self is a multiple of operand's length long */
|
142
|
+
for(i = 0; i < size; i++)
|
143
|
+
{
|
144
|
+
func(&d1[i].v, &d2[i % v2->len].v, &r[i].v);
|
145
|
+
}
|
146
|
+
}
|
147
|
+
|
148
|
+
if(rv->len != rv->len + (rv->len % 4))
|
149
|
+
{
|
150
|
+
for(i = 3; i > rv->len + (rv->len % 4); i--)
|
151
|
+
{
|
152
|
+
r[size].f[i] = 1;
|
153
|
+
}
|
154
|
+
}
|
155
|
+
|
156
|
+
return(result_obj);
|
157
|
+
}
|
158
|
+
|
159
|
+
/* Function: Multiply two vectors. */
|
160
|
+
static void func_multiply(void *v1, void *v2, void *r)
|
161
|
+
{
|
162
|
+
*(i4v *)r = *(i4v *)v1 * *(i4v *)v2;
|
163
|
+
}
|
164
|
+
|
165
|
+
/* Function: Divide two vectors. */
|
166
|
+
static void func_divide(void *v1, void *v2, void *r)
|
167
|
+
{
|
168
|
+
*(i4v *)r = *(i4v *)v1 / *(i4v *)v2;
|
169
|
+
}
|
170
|
+
|
171
|
+
/* Function: Add two vectors. */
|
172
|
+
static void func_add(void *v1, void *v2, void *r)
|
173
|
+
{
|
174
|
+
*(i4v *)r = *(i4v *)v1 + *(i4v *)v2;
|
175
|
+
}
|
176
|
+
|
177
|
+
/* Function: Subtract two vectors. */
|
178
|
+
static void func_subtract(void *v1, void *v2, void *r)
|
179
|
+
{
|
180
|
+
*(i4v *)r = *(i4v *)v1 - *(i4v *)v2;
|
181
|
+
}
|
@@ -0,0 +1,16 @@
|
|
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_subtract(VALUE self, VALUE obj);
|
9
|
+
static VALUE method_to_a(VALUE self);
|
10
|
+
|
11
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func);
|
12
|
+
|
13
|
+
static void func_multiply(void *v1, void *v2, void *r);
|
14
|
+
static void func_divide(void *v1, void *v2, void *r);
|
15
|
+
static void func_add(void *v1, void *v2, void *r);
|
16
|
+
static void func_subtract(void *v1, void *v2, void *r);
|
@@ -100,7 +100,7 @@ static VALUE method_to_a(VALUE self)
|
|
100
100
|
|
101
101
|
/* Internal: Given another FloatArray object, perform an action specified via a
|
102
102
|
* function pointer against both. */
|
103
|
-
static VALUE internal_apply_operation(VALUE self, VALUE obj,
|
103
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func)
|
104
104
|
{
|
105
105
|
unsigned long size, i;
|
106
106
|
int align;
|
@@ -129,45 +129,53 @@ static VALUE internal_apply_operation(VALUE self, VALUE obj, bf_operation func)
|
|
129
129
|
case 0: /* Same size arrays */
|
130
130
|
for(i = 0; i < size; i++)
|
131
131
|
{
|
132
|
-
|
132
|
+
func(&d1[i].v, &d2[i].v, &r[i].v);
|
133
133
|
}
|
134
134
|
break;
|
135
135
|
case 1: /* Operand is exactly 4 long (size of 1 sse register) */
|
136
136
|
for(i = 0; i < size; i++)
|
137
137
|
{
|
138
|
-
|
138
|
+
func(&d1[i].v, &d2[0].v, &r[i].v);
|
139
139
|
}
|
140
140
|
break;
|
141
141
|
default: /* Self is a multiple of operand's length long */
|
142
142
|
for(i = 0; i < size; i++)
|
143
143
|
{
|
144
|
-
|
144
|
+
func(&d1[i].v, &d2[i % v2->len].v, &r[i].v);
|
145
145
|
}
|
146
146
|
}
|
147
147
|
|
148
|
+
if(rv->len != rv->len + (rv->len % 4))
|
149
|
+
{
|
150
|
+
for(i = 3; i > rv->len + (rv->len % 4); i--)
|
151
|
+
{
|
152
|
+
r[size].f[i] = 1;
|
153
|
+
}
|
154
|
+
}
|
155
|
+
|
148
156
|
return(result_obj);
|
149
157
|
}
|
150
158
|
|
151
159
|
/* Function: Multiply two vectors. */
|
152
|
-
static
|
160
|
+
static void func_multiply(void *v1, void *v2, void *r)
|
153
161
|
{
|
154
|
-
|
162
|
+
*(f4v *)r = *(f4v *)v1 * *(f4v *)v2;
|
155
163
|
}
|
156
164
|
|
157
165
|
/* Function: Divide two vectors. */
|
158
|
-
static
|
166
|
+
static void func_divide(void *v1, void *v2, void *r)
|
159
167
|
{
|
160
|
-
|
168
|
+
*(f4v *)r = *(f4v *)v1 / *(f4v *)v2;
|
161
169
|
}
|
162
170
|
|
163
171
|
/* Function: Add two vectors. */
|
164
|
-
static
|
172
|
+
static void func_add(void *v1, void *v2, void *r)
|
165
173
|
{
|
166
|
-
|
174
|
+
*(f4v *)r = *(f4v *)v1 + *(f4v *)v2;
|
167
175
|
}
|
168
176
|
|
169
177
|
/* Function: Subtract two vectors. */
|
170
|
-
static
|
178
|
+
static void func_subtract(void *v1, void *v2, void *r)
|
171
179
|
{
|
172
|
-
|
180
|
+
*(f4v *)r = *(f4v *)v1 - *(f4v *)v2;
|
173
181
|
}
|
@@ -1,11 +1,6 @@
|
|
1
1
|
#include "ruby.h"
|
2
2
|
#include "simd_common.h"
|
3
3
|
|
4
|
-
/*
|
5
|
-
static VALUE allocate(VALUE klass);
|
6
|
-
static void deallocate(vector_t *vector);
|
7
|
-
*/
|
8
|
-
|
9
4
|
static VALUE method_initialize(VALUE self, VALUE rb_array);
|
10
5
|
static VALUE method_multiply(VALUE self, VALUE obj);
|
11
6
|
static VALUE method_divide(VALUE self, VALUE obj);
|
@@ -13,13 +8,9 @@ static VALUE method_add(VALUE self, VALUE obj);
|
|
13
8
|
static VALUE method_subtract(VALUE self, VALUE obj);
|
14
9
|
static VALUE method_to_a(VALUE self);
|
15
10
|
|
16
|
-
|
17
|
-
static f4v_t *internal_allocate_vector_array(unsigned long size);
|
18
|
-
static int internal_align_vectors(unsigned long v1, unsigned long v2);
|
19
|
-
*/
|
20
|
-
static VALUE internal_apply_operation(VALUE self, VALUE obj, bf_operation func);
|
11
|
+
static VALUE internal_apply_operation(VALUE self, VALUE obj, b_operation func);
|
21
12
|
|
22
|
-
static
|
23
|
-
static
|
24
|
-
static
|
25
|
-
static
|
13
|
+
static void func_multiply(void *v1, void *v2, void *r);
|
14
|
+
static void func_divide(void *v1, void *v2, void *r);
|
15
|
+
static void func_add(void *v1, void *v2, void *r);
|
16
|
+
static void func_subtract(void *v1, void *v2, void *r);
|
data/ext/simd/simd_types.h
CHANGED
@@ -1,6 +1,6 @@
|
|
1
1
|
#pragma once
|
2
2
|
|
3
|
-
/*
|
3
|
+
/*
|
4
4
|
* Types for FloatArray
|
5
5
|
*
|
6
6
|
* Since ruby internally uses doubles for the Float type, SIMD::FloatArray will
|
@@ -13,7 +13,7 @@ typedef union d2v_t
|
|
13
13
|
double f[2];
|
14
14
|
} d2v_t;
|
15
15
|
|
16
|
-
/*
|
16
|
+
/*
|
17
17
|
* Types for SmallFloatArray
|
18
18
|
*
|
19
19
|
* Since ruby internally uses doubles for the Float type, SIMD::SmallFloatArray
|
@@ -27,11 +27,20 @@ typedef union f4v_t
|
|
27
27
|
float f[4];
|
28
28
|
} f4v_t;
|
29
29
|
|
30
|
+
/*
|
31
|
+
* Types for IntArray
|
32
|
+
*/
|
33
|
+
typedef int __attribute__ ((vector_size (16))) i4v;
|
34
|
+
typedef union i4v_t
|
35
|
+
{
|
36
|
+
i4v v;
|
37
|
+
int f[4];
|
38
|
+
} i4v_t;
|
39
|
+
|
30
40
|
typedef struct vector_t
|
31
41
|
{
|
32
42
|
void *data;
|
33
43
|
unsigned long len;
|
34
44
|
} vector_t;
|
35
45
|
|
36
|
-
typedef
|
37
|
-
typedef f4v (*bf_operation)(f4v v1, f4v v2);
|
46
|
+
typedef void (b_operation)(void *v1, void *v2, void *r);
|
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: 0.3.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Tina Wuest
|
@@ -38,6 +38,8 @@ files:
|
|
38
38
|
- ext/simd/simd_common.h
|
39
39
|
- ext/simd/simd_floatarray.c
|
40
40
|
- ext/simd/simd_floatarray.h
|
41
|
+
- ext/simd/simd_intarray.c
|
42
|
+
- ext/simd/simd_intarray.h
|
41
43
|
- ext/simd/simd_smallfloatarray.c
|
42
44
|
- ext/simd/simd_smallfloatarray.h
|
43
45
|
- ext/simd/simd_types.h
|