vstore.cl
9.24 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
#include <clc/clc.h>
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
#define VSTORE_VECTORIZE(PRIM_TYPE, ADDR_SPACE) \
typedef PRIM_TYPE##2 less_aligned_##ADDR_SPACE##PRIM_TYPE##2 __attribute__ ((aligned (sizeof(PRIM_TYPE))));\
_CLC_OVERLOAD _CLC_DEF void vstore2(PRIM_TYPE##2 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
*((ADDR_SPACE less_aligned_##ADDR_SPACE##PRIM_TYPE##2*) (&mem[2*offset])) = vec; \
} \
\
_CLC_OVERLOAD _CLC_DEF void vstore3(PRIM_TYPE##3 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
*((ADDR_SPACE less_aligned_##ADDR_SPACE##PRIM_TYPE##2*) (&mem[3*offset])) = (PRIM_TYPE##2)(vec.s0, vec.s1); \
mem[3 * offset + 2] = vec.s2;\
} \
\
typedef PRIM_TYPE##4 less_aligned_##ADDR_SPACE##PRIM_TYPE##4 __attribute__ ((aligned (sizeof(PRIM_TYPE))));\
_CLC_OVERLOAD _CLC_DEF void vstore4(PRIM_TYPE##4 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
*((ADDR_SPACE less_aligned_##ADDR_SPACE##PRIM_TYPE##4*) (&mem[4*offset])) = vec; \
} \
\
typedef PRIM_TYPE##8 less_aligned_##ADDR_SPACE##PRIM_TYPE##8 __attribute__ ((aligned (sizeof(PRIM_TYPE))));\
_CLC_OVERLOAD _CLC_DEF void vstore8(PRIM_TYPE##8 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
*((ADDR_SPACE less_aligned_##ADDR_SPACE##PRIM_TYPE##8*) (&mem[8*offset])) = vec; \
} \
\
typedef PRIM_TYPE##16 less_aligned_##ADDR_SPACE##PRIM_TYPE##16 __attribute__ ((aligned (sizeof(PRIM_TYPE))));\
_CLC_OVERLOAD _CLC_DEF void vstore16(PRIM_TYPE##16 vec, size_t offset, ADDR_SPACE PRIM_TYPE *mem) { \
*((ADDR_SPACE less_aligned_##ADDR_SPACE##PRIM_TYPE##16*) (&mem[16*offset])) = vec; \
} \
#define VSTORE_ADDR_SPACES(__CLC_SCALAR___CLC_GENTYPE) \
VSTORE_VECTORIZE(__CLC_SCALAR___CLC_GENTYPE, __private) \
VSTORE_VECTORIZE(__CLC_SCALAR___CLC_GENTYPE, __local) \
VSTORE_VECTORIZE(__CLC_SCALAR___CLC_GENTYPE, __global) \
VSTORE_ADDR_SPACES(char)
VSTORE_ADDR_SPACES(uchar)
VSTORE_ADDR_SPACES(short)
VSTORE_ADDR_SPACES(ushort)
VSTORE_ADDR_SPACES(int)
VSTORE_ADDR_SPACES(uint)
VSTORE_ADDR_SPACES(long)
VSTORE_ADDR_SPACES(ulong)
VSTORE_ADDR_SPACES(float)
#ifdef cl_khr_fp64
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
VSTORE_ADDR_SPACES(double)
#endif
#ifdef cl_khr_fp16
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
VSTORE_ADDR_SPACES(half)
#endif
/* vstore_half are legal even without cl_khr_fp16 */
#if __clang_major__ < 6
#define DECLARE_HELPER(STYPE, AS, builtin) void __clc_vstore_half_##STYPE##_helper##AS(STYPE, AS half *);
#else
#define DECLARE_HELPER(STYPE, AS, __builtin) \
_CLC_DEF void __clc_vstore_half_##STYPE##_helper##AS(STYPE s, AS half *d) \
{ \
__builtin(s, d); \
}
#endif
DECLARE_HELPER(float, __private, __builtin_store_halff);
DECLARE_HELPER(float, __global, __builtin_store_halff);
DECLARE_HELPER(float, __local, __builtin_store_halff);
#ifdef cl_khr_fp64
DECLARE_HELPER(double, __private, __builtin_store_half);
DECLARE_HELPER(double, __global, __builtin_store_half);
DECLARE_HELPER(double, __local, __builtin_store_half);
#endif
#define VEC_STORE1(STYPE, AS, val, ROUNDF) __clc_vstore_half_##STYPE##_helper##AS (ROUNDF(val), &mem[offset++]);
#define VEC_STORE2(STYPE, AS, val, ROUNDF) \
VEC_STORE1(STYPE, AS, val.lo, ROUNDF) \
VEC_STORE1(STYPE, AS, val.hi, ROUNDF)
#define VEC_STORE3(STYPE, AS, val, ROUNDF) \
VEC_STORE1(STYPE, AS, val.s0, ROUNDF) \
VEC_STORE1(STYPE, AS, val.s1, ROUNDF) \
VEC_STORE1(STYPE, AS, val.s2, ROUNDF)
#define VEC_STORE4(STYPE, AS, val, ROUNDF) \
VEC_STORE2(STYPE, AS, val.lo, ROUNDF) \
VEC_STORE2(STYPE, AS, val.hi, ROUNDF)
#define VEC_STORE8(STYPE, AS, val, ROUNDF) \
VEC_STORE4(STYPE, AS, val.lo, ROUNDF) \
VEC_STORE4(STYPE, AS, val.hi, ROUNDF)
#define VEC_STORE16(STYPE, AS, val, ROUNDF) \
VEC_STORE8(STYPE, AS, val.lo, ROUNDF) \
VEC_STORE8(STYPE, AS, val.hi, ROUNDF)
#define __FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS, ROUNDF) \
_CLC_OVERLOAD _CLC_DEF void vstore_half##SUFFIX(TYPE vec, size_t offset, AS half *mem) { \
offset *= VEC_SIZE; \
VEC_STORE##VEC_SIZE(STYPE, AS, vec, ROUNDF) \
} \
_CLC_OVERLOAD _CLC_DEF void vstorea_half##SUFFIX(TYPE vec, size_t offset, AS half *mem) { \
offset *= OFFSET; \
VEC_STORE##VEC_SIZE(STYPE, AS, vec, ROUNDF) \
}
_CLC_DEF _CLC_OVERLOAD float __clc_noop(float x)
{
return x;
}
_CLC_DEF _CLC_OVERLOAD float __clc_rtz(float x)
{
/* Remove lower 13 bits to make sure the number is rounded down */
int mask = 0xffffe000;
const int exp = (as_uint(x) >> 23 & 0xff) - 127;
/* Denormals cannot be flushed, and they use different bit for rounding */
if (exp < -14)
mask <<= min(-(exp + 14), 10);
/* RTZ does not produce Inf for large numbers */
if (fabs(x) > 65504.0f && !isinf(x))
return copysign(65504.0f, x);
/* Handle nan corner case */
if (isnan(x))
return x;
return as_float(as_uint(x) & mask);
}
_CLC_DEF _CLC_OVERLOAD float __clc_rti(float x)
{
const float inf = copysign(INFINITY, x);
/* Set lower 13 bits */
int mask = (1 << 13) - 1;
const int exp = (as_uint(x) >> 23 & 0xff) - 127;
/* Denormals cannot be flushed, and they use different bit for rounding */
if (exp < -14)
mask = (1 << (13 + min(-(exp + 14), 10))) - 1;
/* Handle nan corner case */
if (isnan(x))
return x;
const float next = nextafter(as_float(as_uint(x) | mask), inf);
return ((as_uint(x) & mask) == 0) ? x : next;
}
_CLC_DEF _CLC_OVERLOAD float __clc_rtn(float x)
{
return ((as_uint(x) & 0x80000000) == 0) ? __clc_rtz(x) : __clc_rti(x);
}
_CLC_DEF _CLC_OVERLOAD float __clc_rtp(float x)
{
return ((as_uint(x) & 0x80000000) == 0) ? __clc_rti(x) : __clc_rtz(x);
}
_CLC_DEF _CLC_OVERLOAD float __clc_rte(float x)
{
/* Mantisa + implicit bit */
const uint mantissa = (as_uint(x) & 0x7fffff) | (1u << 23);
const int exp = (as_uint(x) >> 23 & 0xff) - 127;
int shift = 13;
if (exp < -14) {
/* The default assumes lower 13 bits are rounded,
* but it might be more for denormals.
* Shifting beyond last == 0b, and qr == 00b is not necessary */
shift += min(-(exp + 14), 15);
}
int mask = (1 << shift) - 1;
const uint grs = mantissa & mask;
const uint last = mantissa & (1 << shift);
/* IEEE round up rule is: grs > 101b or grs == 100b and last == 1.
* exp > 15 should round to inf. */
bool roundup = (grs > (1 << (shift - 1))) ||
(grs == (1 << (shift - 1)) && last != 0) || (exp > 15);
return roundup ? __clc_rti(x) : __clc_rtz(x);
}
#ifdef cl_khr_fp64
_CLC_DEF _CLC_OVERLOAD double __clc_noop(double x)
{
return x;
}
_CLC_DEF _CLC_OVERLOAD double __clc_rtz(double x)
{
/* Remove lower 42 bits to make sure the number is rounded down */
ulong mask = 0xfffffc0000000000UL;
const int exp = (as_ulong(x) >> 52 & 0x7ff) - 1023;
/* Denormals cannot be flushed, and they use different bit for rounding */
if (exp < -14)
mask <<= min(-(exp + 14), 10);
/* RTZ does not produce Inf for large numbers */
if (fabs(x) > 65504.0 && !isinf(x))
return copysign(65504.0, x);
/* Handle nan corner case */
if (isnan(x))
return x;
return as_double(as_ulong(x) & mask);
}
_CLC_DEF _CLC_OVERLOAD double __clc_rti(double x)
{
const double inf = copysign((double)INFINITY, x);
/* Set lower 42 bits */
long mask = (1UL << 42UL) - 1UL;
const int exp = (as_ulong(x) >> 52 & 0x7ff) - 1023;
/* Denormals cannot be flushed, and they use different bit for rounding */
if (exp < -14)
mask = (1UL << (42UL + min(-(exp + 14), 10))) - 1;
/* Handle nan corner case */
if (isnan(x))
return x;
const double next = nextafter(as_double(as_ulong(x) | mask), inf);
return ((as_ulong(x) & mask) == 0) ? x : next;
}
_CLC_DEF _CLC_OVERLOAD double __clc_rtn(double x)
{
return ((as_ulong(x) & 0x8000000000000000UL) == 0) ? __clc_rtz(x) : __clc_rti(x);
}
_CLC_DEF _CLC_OVERLOAD double __clc_rtp(double x)
{
return ((as_ulong(x) & 0x8000000000000000UL) == 0) ? __clc_rti(x) : __clc_rtz(x);
}
_CLC_DEF _CLC_OVERLOAD double __clc_rte(double x)
{
/* Mantisa + implicit bit */
const ulong mantissa = (as_ulong(x) & 0xfffffffffffff) | (1UL << 52);
const int exp = (as_ulong(x) >> 52 & 0x7ff) - 1023;
int shift = 42;
if (exp < -14) {
/* The default assumes lower 13 bits are rounded,
* but it might be more for denormals.
* Shifting beyond last == 0b, and qr == 00b is not necessary */
shift += min(-(exp + 14), 15);
}
ulong mask = (1UL << shift) - 1UL;
const ulong grs = mantissa & mask;
const ulong last = mantissa & (1UL << shift);
/* IEEE round up rule is: grs > 101b or grs == 100b and last == 1.
* exp > 15 should round to inf. */
bool roundup = (grs > (1UL << (shift - 1UL))) ||
(grs == (1UL << (shift - 1UL)) && last != 0) || (exp > 15);
return roundup ? __clc_rti(x) : __clc_rtz(x);
}
#endif
#define __XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) \
__FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_noop) \
__FUNC(SUFFIX ## _rtz, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtz) \
__FUNC(SUFFIX ## _rtn, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtn) \
__FUNC(SUFFIX ## _rtp, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rtp) \
__FUNC(SUFFIX ## _rte, VEC_SIZE, OFFSET, TYPE, STYPE, AS, __clc_rte)
#define FUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS) \
__XFUNC(SUFFIX, VEC_SIZE, OFFSET, TYPE, STYPE, AS)
#define __CLC_BODY "vstore_half.inc"
#include <clc/math/gentype.inc>
#undef __CLC_BODY
#undef FUNC
#undef __XFUNC
#undef __FUNC
#undef VEC_LOAD16
#undef VEC_LOAD8
#undef VEC_LOAD4
#undef VEC_LOAD3
#undef VEC_LOAD2
#undef VEC_LOAD1
#undef DECLARE_HELPER
#undef VSTORE_ADDR_SPACES
#undef VSTORE_VECTORIZE