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
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
|
/*========================== begin_copyright_notice ============================
Copyright (C) 2021 Intel Corporation
SPDX-License-Identifier: MIT
============================= end_copyright_notice ===========================*/
//
// This file defines SPIRV vstoren, vstore_half, vstore_half_r, vstore_halfn,
// vstore_halfn_r, vstorea_halfn and vstorea_halfn_r built-ins.
//
//*****************************************************************************/
// helper functions
//*****************************************************************************/
static OVERLOADABLE half __intel_spirv_float2half(float f, RoundingMode_t r)
{
switch (r)
{
case rte:
return SPIRV_BUILTIN(FConvert, _RTE_f16_f32, _Rhalf_rte)(f);
case rtz:
return SPIRV_BUILTIN(FConvert, _RTZ_f16_f32, _Rhalf_rtz)(f);
case rtp:
return SPIRV_BUILTIN(FConvert, _RTP_f16_f32, _Rhalf_rtp)(f);
case rtn:
return SPIRV_BUILTIN(FConvert, _RTN_f16_f32, _Rhalf_rtn)(f);
}
}
GENERATE_VECTOR_FUNCTIONS_2ARGS_VS_NO_MANG(__intel_spirv_float2half, half, float, RoundingMode_t)
#if defined(cl_khr_fp64)
static OVERLOADABLE half __intel_spirv_double2half(double d, RoundingMode_t r)
{
switch (r)
{
case rte:
return SPIRV_BUILTIN(FConvert, _RTE_f16_f64, _Rhalf_rte)(d);
case rtz:
return SPIRV_BUILTIN(FConvert, _RTZ_f16_f64, _Rhalf_rtz)(d);
case rtp:
return SPIRV_BUILTIN(FConvert, _RTP_f16_f64, _Rhalf_rtp)(d);
case rtn:
return SPIRV_BUILTIN(FConvert, _RTN_f16_f64, _Rhalf_rtn)(d);
}
}
GENERATE_VECTOR_FUNCTIONS_2ARGS_VS_NO_MANG(__intel_spirv_double2half, half, double, RoundingMode_t)
#endif //defined(cl_khr_fp64)
//*****************************************************************************/
// vstoren
// "Writes n components from the data vector value to the address computed as (p + (offset * n)),
// where n is equal to the component count of the vector data."
//*****************************************************************************/
#define VSTOREN_DEF(addressSpace, scalarType, numElements, offsetType, mangle) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(vstore, numElements##_##mangle, n)( \
scalarType##numElements data, offsetType offset, addressSpace scalarType *p) { \
addressSpace scalarType *pOffset = p + offset * numElements; \
scalarType##numElements ret = data; \
__builtin_IB_memcpy_private_to_##addressSpace((addressSpace uchar *)pOffset, (private uchar *)&ret, \
sizeof(scalarType) * numElements, sizeof(scalarType)); \
}
#define VSTOREN_AS(addressSpace, scalarType, typemang, mang) \
VSTOREN_DEF(addressSpace, scalarType, 2, long, v2##typemang##_i64_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 2, int, v2##typemang##_i32_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 3, long, v3##typemang##_i64_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 3, int, v3##typemang##_i32_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 4, long, v4##typemang##_i64_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 4, int, v4##typemang##_i32_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 8, long, v8##typemang##_i64_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 8, int, v8##typemang##_i32_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 16, long, v16##typemang##_i64_##mang) \
VSTOREN_DEF(addressSpace, scalarType, 16, int, v16##typemang##_i32_##mang)
#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
#define VSTOREN_TYPE(TYPE, TYPEMANG) \
VSTOREN_AS(global, TYPE, TYPEMANG, p1##TYPEMANG) \
VSTOREN_AS(local, TYPE, TYPEMANG, p3##TYPEMANG) \
VSTOREN_AS(private, TYPE, TYPEMANG, p0##TYPEMANG) \
VSTOREN_AS(generic, TYPE, TYPEMANG, p4##TYPEMANG)
#else
#define VSTOREN_TYPE(TYPE, TYPEMANG) \
VSTOREN_AS(global, TYPE, TYPEMANG, p1##TYPEMANG) \
VSTOREN_AS(local, TYPE, TYPEMANG, p3##TYPEMANG) \
VSTOREN_AS(private, TYPE, TYPEMANG, p0##TYPEMANG)
#endif // __OPENCL_C_VERSION__ >= CL_VERSION_2_0
VSTOREN_TYPE(char, i8)
VSTOREN_TYPE(short, i16)
VSTOREN_TYPE(int, i32)
VSTOREN_TYPE(long, i64)
VSTOREN_TYPE(half, f16)
VSTOREN_TYPE(float, f32)
#if defined(cl_khr_fp64)
VSTOREN_TYPE(double, f64)
#endif
//*****************************************************************************/
// vstore_half
// "Converts the data float or double value to a half value using the default rounding mode
// and writes the half value to the address computed as (p + offset)."
// vstore_half_r
// "Converts the data float or double value to a half value using the specified rounding mode
// mode and writes the half value to the address computed as (p + offset)."
//*****************************************************************************/
#define VSTORE_HALF_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(vstore_half, _##MANGSRC##_##MANGSIZE##_p##ASNUM##f16, )( \
srcType data, SIZETYPE offset, addressSpace half * p) { \
addressSpace half *pHalf = (addressSpace half *)(p + offset); \
*pHalf = __intel_spirv_##srcType##2half(data, rte); \
}
#define VSTORE_HALF_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN(vstore_half_r, _##MANGSRC##_##MANGSIZE##_p##ASNUM##f16_i32, )( \
srcType data, SIZETYPE offset, addressSpace half * p, RoundingMode_t r) { \
addressSpace half *pHalf = (addressSpace half *)(p + offset); \
*pHalf = __intel_spirv_##srcType##2half(data, r); \
}
#define VSTORE_HALF_BOTH(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType) \
VSTORE_HALF_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType) \
VSTORE_HALF_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType)
#if defined(cl_khr_fp64)
#define VSTORE_HALF_ALL_TYPES(addressSpace, ASNUM) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i64, long, f32, float) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i64, long, f64, double) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i32, int, f32, float) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i32, int, f64, double)
#else
#define VSTORE_HALF_ALL_TYPES(addressSpace, ASNUM) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i64, long, f32, float) \
VSTORE_HALF_BOTH(addressSpace, ASNUM, i32, int, f32, float)
#endif
VSTORE_HALF_ALL_TYPES(private, 0)
VSTORE_HALF_ALL_TYPES(global, 1)
VSTORE_HALF_ALL_TYPES(local, 3)
#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
VSTORE_HALF_ALL_TYPES(generic, 4)
#endif // __OPENCL_C_VERSION__ >= CL_VERSION_2_0
//*****************************************************************************/
// vstore_halfn
// "Converts the data vector of float or vector of double values to a vector of half values
// using the default rounding mode and writes the half values to memory."
// vstore_halfn_r
// "Converts the data vector of float or vector of double values to a vector of half values
// using the specified rounding mode mode and writes the half values to memory."
//*****************************************************************************/
#define VSTORE_HALFN_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, numElements) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN( \
vstore_half, numElements##_v##numElements##MANGSRC##_##MANGSIZE##_p##ASNUM##f16, \
n)(srcType##numElements data, SIZETYPE offset, addressSpace half * p) { \
SPIRV_OCL_BUILTIN(vstore, numElements##_v##numElements##f16##_##MANGSIZE##_p##ASNUM##f16, n) \
(__intel_spirv_##srcType##2half(data, rte), offset, p); \
}
#define VSTORE_HALFN_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, numElements) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN( \
vstore_half, numElements##_r_v##numElements##MANGSRC##_##MANGSIZE##_p##ASNUM##f16_i32, \
n_r)(srcType##numElements data, SIZETYPE offset, addressSpace half * p, RoundingMode_t r) { \
SPIRV_OCL_BUILTIN(vstore, numElements##_v##numElements##f16##_##MANGSIZE##_p##ASNUM##f16, n) \
(__intel_spirv_##srcType##2half(data, r), offset, p); \
}
#define VSTORE_HALFN_BOTH(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, numElements) \
VSTORE_HALFN_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, numElements) \
VSTORE_HALFN_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, numElements)
#if defined(cl_khr_fp64)
#define VSTORE_HALFN_ALL_TYPES(addressSpace, ASNUM) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 16) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 16) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 16) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 16)
#else
#define VSTORE_HALFN_ALL_TYPES(addressSpace, ASNUM) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 16) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 2) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 3) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 8) \
VSTORE_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 16)
#endif //defined(cl_khr_fp64)
VSTORE_HALFN_ALL_TYPES(private, 0)
VSTORE_HALFN_ALL_TYPES(global, 1)
VSTORE_HALFN_ALL_TYPES(local, 3)
#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
VSTORE_HALFN_ALL_TYPES(generic, 4)
#endif // __OPENCL_C_VERSION__ >= CL_VERSION_2_0
//*****************************************************************************/
// vstorea_halfn
// "Converts the data vector of float or vector of double values to a vector of half values using the default rounding mode,
// and then writes the converted vector of half values to aligned memory."
// vstorea_halfn_r
// "Converts the data vector of float or vector of double values to a vector of half values using the specified rounding mode mode,
// and then write the converted vector of half values to aligned memory."
//*****************************************************************************/
#define VSTOREA_HALFN_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, step, numElements) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN( \
vstorea_half, numElements##_v##numElements##MANGSRC##_##MANGSIZE##_p##ASNUM##f16, \
n)(srcType##numElements data, SIZETYPE offset, addressSpace half * p) { \
addressSpace half##numElements *pHalf = (addressSpace half##numElements *)(p + offset * step); \
*pHalf = __intel_spirv_##srcType##2half(data, rte); \
}
#define VSTOREA_HALFN_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, step, numElements) \
INLINE void SPIRV_OVERLOADABLE SPIRV_OCL_BUILTIN( \
vstorea_half, numElements##_r_v##numElements##MANGSRC##_##MANGSIZE##_p##ASNUM##f16_i32, \
n_r)(srcType##numElements data, SIZETYPE offset, addressSpace half * p, RoundingMode_t r) { \
addressSpace half##numElements *pHalf = (addressSpace half##numElements *)(p + offset * step); \
*pHalf = __intel_spirv_##srcType##2half(data, r); \
}
#define VSTOREA_HALFN_BOTH(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, step, numElements) \
VSTOREA_HALFN_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, step, numElements) \
VSTOREA_HALFN_R_DEF(addressSpace, ASNUM, MANGSIZE, SIZETYPE, MANGSRC, srcType, step, numElements)
#if defined(cl_khr_fp64)
#define VSTOREA_HALFN_ALL_TYPES(addressSpace, ASNUM) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 16, 16) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f64, double, 16, 16) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 16, 16) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f64, double, 16, 16)
#else
#define VSTOREA_HALFN_ALL_TYPES(addressSpace, ASNUM) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i64, long, f32, float, 16, 16) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 2, 2) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4, 3) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 4, 4) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 8, 8) \
VSTOREA_HALFN_BOTH(addressSpace, ASNUM, i32, int, f32, float, 16, 16)
#endif //defined(cl_khr_fp64)
VSTOREA_HALFN_ALL_TYPES(private, 0)
VSTOREA_HALFN_ALL_TYPES(global, 1)
VSTOREA_HALFN_ALL_TYPES(local, 3)
#if (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
VSTOREA_HALFN_ALL_TYPES(generic, 4)
#endif // __OPENCL_C_VERSION__ >= CL_VERSION_2_0
|
