third_party/highway/hwy/examples/skeleton.cc

   1 // Copyright 2020 Google LLC
   2 // SPDX-License-Identifier: Apache-2.0
   3 //
   4 // Licensed under the Apache License, Version 2.0 (the "License");
   5 // you may not use this file except in compliance with the License.
   6 // You may obtain a copy of the License at
   7 //
   8 //      http://www.apache.org/licenses/LICENSE-2.0
   9 //
  10 // Unless required by applicable law or agreed to in writing, software
  11 // distributed under the License is distributed on an "AS IS" BASIS,
  12 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 // See the License for the specific language governing permissions and
  14 // limitations under the License.
  15
  16 #include "hwy/examples/skeleton.h"
  17
  18 #include <stdio.h>
  19
  20 // >>>> for dynamic dispatch only, skip if you want static dispatch
  21
  22 // First undef to prevent error when re-included.
  23 #undef HWY_TARGET_INCLUDE
  24 // For dynamic dispatch, specify the name of the current file (unfortunately
  25 // __FILE__ is not reliable) so that foreach_target.h can re-include it.
  26 #define HWY_TARGET_INCLUDE "hwy/examples/skeleton.cc"
  27 // Generates code for each enabled target by re-including this source file.
  28 #include "hwy/foreach_target.h"  // IWYU pragma: keep
  29
  30 // <<<< end of dynamic dispatch
  31
  32 // Must come after foreach_target.h to avoid redefinition errors.
  33 #include "hwy/highway.h"
  34
  35 // Optional, can instead add HWY_ATTR to all functions.
  36 HWY_BEFORE_NAMESPACE();
  37
  38 namespace skeleton {
  39 // This namespace name is unique per target, which allows code for multiple
  40 // targets to co-exist in the same translation unit. Required when using dynamic
  41 // dispatch, otherwise optional.
  42 namespace HWY_NAMESPACE {
  43
  44 // Highway ops reside here; ADL does not find templates nor builtins.
  45 namespace hn = hwy::HWY_NAMESPACE;
  46
  47 // Computes log2 by converting to a vector of floats. Compiled once per target.
  48 template <class DF>
  49 HWY_ATTR_NO_MSAN void OneFloorLog2(const DF df,
  50                                    const uint8_t* HWY_RESTRICT values,
  51                                    uint8_t* HWY_RESTRICT log2) {
  52   // Type tags for converting to other element types (Rebind = same count).
  53   const hn::RebindToSigned<DF> d32;
  54   const hn::Rebind<uint8_t, DF> d8;
  55   using VI32 = hn::Vec<decltype(d32)>;
  56
  57   const VI32 vi32 = hn::PromoteTo(d32, hn::Load(d8, values));
  58   const VI32 bits = hn::BitCast(d32, hn::ConvertTo(df, vi32));
  59   const VI32 exponent = hn::Sub(hn::ShiftRight<23>(bits), hn::Set(d32, 127));
  60   hn::Store(hn::DemoteTo(d8, exponent), d8, log2);
  61 }
  62
  63 void CodepathDemo() {
  64   // Highway defaults to portability, but per-target codepaths may be selected
  65   // via #if HWY_TARGET == HWY_SSE4 or by testing capability macros:
  66 #if HWY_HAVE_INTEGER64
  67   const char* gather = "Has int64";
  68 #else
  69   const char* gather = "No int64";
  70 #endif
  71   printf("Target %s: %s\n", hwy::TargetName(HWY_TARGET), gather);
  72 }
  73
  74 void FloorLog2(const uint8_t* HWY_RESTRICT values, size_t count,
  75                uint8_t* HWY_RESTRICT log2) {
  76   CodepathDemo();
  77
  78   const hn::ScalableTag<float> df;
  79   const size_t N = hn::Lanes(df);
  80   size_t i = 0;
  81   for (; i + N <= count; i += N) {
  82     OneFloorLog2(df, values + i, log2 + i);
  83   }
  84   for (; i < count; ++i) {
  85     hn::CappedTag<float, 1> d1;
  86     OneFloorLog2(d1, values + i, log2 + i);
  87   }
  88 }
  89
  90 // NOLINTNEXTLINE(google-readability-namespace-comments)
  91 }  // namespace HWY_NAMESPACE
  92 }  // namespace skeleton
  93 HWY_AFTER_NAMESPACE();
  94
  95 // The table of pointers to the various implementations in HWY_NAMESPACE must
  96 // be compiled only once (foreach_target #includes this file multiple times).
  97 // HWY_ONCE is true for only one of these 'compilation passes'.
  98 #if HWY_ONCE
  99
 100 namespace skeleton {
 101
 102 // This macro declares a static array used for dynamic dispatch; it resides in
 103 // the same outer namespace that contains FloorLog2.
 104 HWY_EXPORT(FloorLog2);
 105
 106 // This function is optional and only needed in the case of exposing it in the
 107 // header file. Otherwise using HWY_DYNAMIC_DISPATCH(FloorLog2) in this module
 108 // is equivalent to inlining this function.
 109 HWY_DLLEXPORT void CallFloorLog2(const uint8_t* HWY_RESTRICT in,
 110                                  const size_t count,
 111                                  uint8_t* HWY_RESTRICT out) {
 112   // This must reside outside of HWY_NAMESPACE because it references (calls the
 113   // appropriate one from) the per-target implementations there.
 114   // For static dispatch, use HWY_STATIC_DISPATCH.
 115   return HWY_DYNAMIC_DISPATCH(FloorLog2)(in, count, out);
 116 }
 117
 118 // Optional: anything to compile only once, e.g. non-SIMD implementations of
 119 // public functions provided by this module, can go inside #if HWY_ONCE.
 120
 121 }  // namespace skeleton
 122 #endif  // HWY_ONCE