| 1 | /*  Copyright (C) 2013  Povilas Kanapickas <povilas@radix.lt> | 
|---|
| 2 |  | 
|---|
| 3 | Distributed under the Boost Software License, Version 1.0. | 
|---|
| 4 | (See accompanying file LICENSE_1_0.txt or copy at | 
|---|
| 5 | http://www.boost.org/LICENSE_1_0.txt) | 
|---|
| 6 | */ | 
|---|
| 7 |  | 
|---|
| 8 | #ifndef LIBSIMDPP_DISPATCHER_H | 
|---|
| 9 | #define LIBSIMDPP_DISPATCHER_H | 
|---|
| 10 |  | 
|---|
| 11 | #ifndef LIBSIMDPP_SIMD_H | 
|---|
| 12 | #error "This file must be included through simd.h" | 
|---|
| 13 | #endif | 
|---|
| 14 |  | 
|---|
| 15 | #include <algorithm> | 
|---|
| 16 | #include <cstdlib> | 
|---|
| 17 | #include <simdpp/dispatch/arch.h> | 
|---|
| 18 |  | 
|---|
| 19 | namespace simdpp { | 
|---|
| 20 |  | 
|---|
| 21 | /** @def SIMDPP_USER_ARCH_INFO | 
|---|
| 22 | The user must define this macro if he wants to use the dispatcher | 
|---|
| 23 | infrastructure. The macro must evaluate to a constant expression that could | 
|---|
| 24 | implicitly initialize an object of type @c std::function<Arch()>. | 
|---|
| 25 |  | 
|---|
| 26 | The macro is used by @c SIMDPP_MAKE_DISPATCHER_* to specify function which | 
|---|
| 27 | is invoked to determine which version of the dispatched function to | 
|---|
| 28 | dispatch. The macro is just a more convenient method to pass a parameter | 
|---|
| 29 | which has high chance to be the same in the majority of use cases. The user | 
|---|
| 30 | may redefine the macro and use different definitions for each site of | 
|---|
| 31 | @c SIMDPP_MAKE_DISPATCHER_* expansion. | 
|---|
| 32 |  | 
|---|
| 33 | The function identified by the @c SIMDPP_USER_ARCH_INFO is called at the | 
|---|
| 34 | first time the specific dispatcher is invoked. The user must ensure that | 
|---|
| 35 | proper synchronization is used if the dispatcher is called concurrently. | 
|---|
| 36 |  | 
|---|
| 37 | The user must ensure that the returned information is sensible: e.g. SSE2 | 
|---|
| 38 | must be supported if SSE3 support is indicated. | 
|---|
| 39 |  | 
|---|
| 40 | The @c simdpp/dispatch/get_arch_*.h files provide several ready | 
|---|
| 41 | implementations of CPU features detection. | 
|---|
| 42 | */ | 
|---|
| 43 |  | 
|---|
| 44 | namespace detail { | 
|---|
| 45 |  | 
|---|
| 46 | using VoidFunPtr = void (*)(); | 
|---|
| 47 |  | 
|---|
| 48 | struct FnVersion { | 
|---|
| 49 | /*  Identifies the instruction support that is needed for this version to | 
|---|
| 50 | run. | 
|---|
| 51 | */ | 
|---|
| 52 | Arch needed_arch; | 
|---|
| 53 |  | 
|---|
| 54 | /*  The type-erased function pointer. Only non-member functions are | 
|---|
| 55 | supported. | 
|---|
| 56 |  | 
|---|
| 57 | We type-erase this pointer to reduce the code bloat by allowing @c | 
|---|
| 58 | select_version to be shared across all dispatchers. | 
|---|
| 59 |  | 
|---|
| 60 | Note, that we can't reinterpret_cast to void*, as that is undefined | 
|---|
| 61 | behavior in C++. reinterpret_cast to another function pointer | 
|---|
| 62 | type is allowed fortunately. | 
|---|
| 63 | */ | 
|---|
| 64 | VoidFunPtr fun_ptr; | 
|---|
| 65 |  | 
|---|
| 66 | /*  Optional string identifier identifying the architecture. */ | 
|---|
| 67 | const char* arch_name; | 
|---|
| 68 | }; | 
|---|
| 69 |  | 
|---|
| 70 | inline FnVersion select_version_any(FnVersion* versions, unsigned size, | 
|---|
| 71 | Arch arch) | 
|---|
| 72 | { | 
|---|
| 73 | // No need to try to be very efficient here. | 
|---|
| 74 | std::sort(versions, versions + size, | 
|---|
| 75 | [](const FnVersion& lhs, const FnVersion& rhs) { | 
|---|
| 76 | return lhs.needed_arch > rhs.needed_arch; | 
|---|
| 77 | }); | 
|---|
| 78 |  | 
|---|
| 79 | unsigned i; | 
|---|
| 80 | for (i = 0; i < size; ++i) { | 
|---|
| 81 | if (versions[i].fun_ptr == nullptr) | 
|---|
| 82 | continue; | 
|---|
| 83 | if (test_arch_subset(arch, versions[i].needed_arch)) { | 
|---|
| 84 | break; | 
|---|
| 85 | } | 
|---|
| 86 | } | 
|---|
| 87 | if (i == size) { | 
|---|
| 88 | // The user didn't provide the NONE_NULL version and no SIMD | 
|---|
| 89 | // architecture is supported. We can't do anything except to abort | 
|---|
| 90 | std::abort(); | 
|---|
| 91 | } | 
|---|
| 92 | return versions[i]; | 
|---|
| 93 | } | 
|---|
| 94 |  | 
|---|
| 95 | } // namespace detail | 
|---|
| 96 | } // namespace simdpp | 
|---|
| 97 |  | 
|---|
| 98 | #endif | 
|---|
| 99 |  | 
|---|