1 | /* Copyright 2020 Google LLC. All Rights Reserved. |
2 | |
3 | Licensed under the Apache License, Version 2.0 (the "License"); |
4 | you may not use this file except in compliance with the License. |
5 | You may obtain a copy of the License at |
6 | |
7 | http://www.apache.org/licenses/LICENSE-2.0 |
8 | |
9 | Unless required by applicable law or agreed to in writing, software |
10 | distributed under the License is distributed on an "AS IS" BASIS, |
11 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
12 | See the License for the specific language governing permissions and |
13 | limitations under the License. |
14 | ==============================================================================*/ |
15 | |
16 | #include "ruy/apply_multiplier.h" |
17 | |
18 | #include <cmath> |
19 | #include <cstdint> |
20 | #include <cstdlib> |
21 | #include <limits> |
22 | |
23 | namespace ruy { |
24 | namespace detail { |
25 | |
26 | // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! |
27 | // Warning: this code is not meant to be bit-exact-normative. |
28 | // Please refer to the class comment of ruy::MulParams, in mul_params.h. |
29 | // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! |
30 | // Simplified multiplier application function |
31 | // |
32 | // Double rounding and symmetric rounding are removed compared to reference. |
33 | // Double rounding seems unnecessary and can complicate implementations. |
34 | // Symmetric rounding also adds implementation complexity. |
35 | // |
36 | // Composed of a single rounding shift right and can lead to more HW |
37 | // friendly implementations. |
38 | // |
39 | // On NEON this can be translated to a SQDMULH + rounding shift right sequence. |
40 | // The use of SQDMULH rather than SQRDMULH gives a result that is |
41 | // equivalent to a single rounded shift since the truncating shift of SQDMULH |
42 | // can be combined with the rounding right shift via the formula (for k>=1): |
43 | // ((x>>31)+(1<<(k-1)))>>k = (x + (1<<(30+k))>>(31+k) |
44 | // |
45 | // Preconditions: |
46 | // - quantized_multiplier >= 0 |
47 | // - shift is -31 to +7 (negative for right shift) |
48 | std::int32_t MultiplyByQuantizedMultiplier(std::int32_t x, |
49 | std::int32_t quantized_multiplier, |
50 | int shift) { |
51 | RUY_CHECK_GE(shift, -31); |
52 | |
53 | int total_shift = 31 - shift; |
54 | |
55 | std::int64_t x_64(x); |
56 | std::int64_t quantized_multiplier_64(quantized_multiplier); |
57 | std::int64_t round = (int64_t)1 << (total_shift - 1); |
58 | int64_t result = x_64 * quantized_multiplier_64 + round; |
59 | result = result >> total_shift; |
60 | |
61 | RUY_DCHECK_GE(result, std::numeric_limits<std::int32_t>::lowest()); |
62 | RUY_DCHECK_LE(result, std::numeric_limits<std::int32_t>::max()); |
63 | |
64 | return static_cast<std::int32_t>(result); |
65 | } |
66 | |
67 | } // namespace detail |
68 | |
69 | } // namespace ruy |
70 | |