1 /*
2 * Copyright (c) 2020 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #ifndef ARM_COMPUTE_BFLOAT16_H
25 #define ARM_COMPUTE_BFLOAT16_H
26
27 #include <cstdint>
28
29 namespace arm_compute
30 {
31 namespace
32 {
33 /** Convert float to bfloat16
34 *
35 * @param[in] v Floating-point value to convert to bfloat
36 *
37 * @return Converted value
38 */
float_to_bf16(const float v)39 inline uint16_t float_to_bf16(const float v)
40 {
41 const uint32_t *fromptr = reinterpret_cast<const uint32_t *>(&v);
42 #if defined(__ARM_FEATURE_BF16_VECTOR_ARITHMETIC) || defined(ARM_COMPUTE_FORCE_BF16)
43 uint16_t res;
44
45 __asm __volatile(
46 "ldr s0, [%[fromptr]]\n"
47 ".inst 0x1e634000\n" // BFCVT h0, s0
48 "str h0, [%[toptr]]\n"
49 :
50 : [fromptr] "r"(fromptr), [toptr] "r"(&res)
51 : "v0", "memory");
52 #else /* defined(__ARM_FEATURE_BF16_VECTOR_ARITHMETIC) || defined(ARM_COMPUTE_FORCE_BF16) */
53 uint16_t res = (*fromptr >> 16);
54 const uint16_t error = (*fromptr & 0x0000ffff);
55 uint16_t bf_l = res & 0x0001;
56 if((error > 0x8000) || ((error == 0x8000) && (bf_l != 0)))
57 {
58 res += 1;
59 }
60 #endif /* defined(__ARM_FEATURE_BF16_VECTOR_ARITHMETIC) || defined(ARM_COMPUTE_FORCE_BF16) */
61 return res;
62 }
63
64 /** Convert bfloat16 to float
65 *
66 * @param[in] v Bfloat16 value to convert to float
67 *
68 * @return Converted value
69 */
bf16_to_float(const uint16_t & v)70 inline float bf16_to_float(const uint16_t &v)
71 {
72 const uint32_t lv = (v << 16);
73 const float *fp = reinterpret_cast<const float *>(&lv);
74
75 return *fp;
76 }
77 }
78
79 /** Brain floating point representation class */
80 class bfloat16 final
81 {
82 public:
83 /** Default Constructor */
bfloat16()84 bfloat16()
85 : value(0)
86 {
87 }
88 /** Constructor
89 *
90 * @param[in] v Floating-point value
91 */
bfloat16(float v)92 explicit bfloat16(float v)
93 : value(float_to_bf16(v))
94 {
95 }
96 /** Assignment operator
97 *
98 * @param[in] v Floating point value to assign
99 *
100 * @return The updated object
101 */
102 bfloat16 &operator=(float v)
103 {
104 value = float_to_bf16(v);
105 return *this;
106 }
107 /** Floating point conversion operator
108 *
109 * @return Floating point representation of the value
110 */
111 operator float() const
112 {
113 return bf16_to_float(value);
114 }
115 /** Lowest representative value
116 *
117 * @return Returns the lowest finite value representable by bfloat16
118 */
lowest()119 static bfloat16 lowest()
120 {
121 bfloat16 val;
122 val.value = 0xFF7F;
123 return val;
124 }
125 /** Largest representative value
126 *
127 * @return Returns the largest finite value representable by bfloat16
128 */
max()129 static bfloat16 max()
130 {
131 bfloat16 val;
132 val.value = 0x7F7F;
133 return val;
134 }
135
136 private:
137 uint16_t value;
138 };
139 } // namespace arm_compute
140 #endif /* ARM_COMPUTE_BFLOAT16_H */