diff options
Diffstat (limited to 'include/EASTL/internal/fill_help.h')
| -rw-r--r-- | include/EASTL/internal/fill_help.h | 484 |
1 files changed, 484 insertions, 0 deletions
diff --git a/include/EASTL/internal/fill_help.h b/include/EASTL/internal/fill_help.h new file mode 100644 index 0000000..235a24e --- /dev/null +++ b/include/EASTL/internal/fill_help.h @@ -0,0 +1,484 @@ +///////////////////////////////////////////////////////////////////////////// +// Copyright (c) Electronic Arts Inc. All rights reserved. +///////////////////////////////////////////////////////////////////////////// + + +#ifndef EASTL_INTERNAL_FILL_HELP_H +#define EASTL_INTERNAL_FILL_HELP_H + + +#if defined(EA_PRAGMA_ONCE_SUPPORTED) + #pragma once +#endif + +#include <EASTL/internal/config.h> + +#if defined(EA_COMPILER_MICROSOFT) && (defined(EA_PROCESSOR_X86) || defined(EA_PROCESSOR_X86_64)) +#include <intrin.h> +#endif + +namespace eastl +{ + // fill + // + // We implement some fill helper functions in order to allow us to optimize it + // where possible. + // + template <bool bIsScalar> + struct fill_imp + { + template <typename ForwardIterator, typename T> + static void do_fill(ForwardIterator first, ForwardIterator last, const T& value) + { + // The C++ standard doesn't specify whether we need to create a temporary + // or not, but all std STL implementations are written like what we have here. + for(; first != last; ++first) + *first = value; + } + }; + + template <> + struct fill_imp<true> + { + template <typename ForwardIterator, typename T> + static void do_fill(ForwardIterator first, ForwardIterator last, const T& value) + { + typedef typename eastl::iterator_traits<ForwardIterator>::value_type value_type; + // We create a temp and fill from that because value might alias to the + // destination range and so the compiler would be forced into generating + // less efficient code. + for(const T temp = value; first != last; ++first) + { + EA_UNUSED(temp); + *first = static_cast<value_type>(temp); + } + } + }; + + /// fill + /// + /// fill is like memset in that it assigns a single value repeatedly to a + /// destination range. It allows for any type of iterator (not just an array) + /// and the source value can be any type, not just a byte. + /// Note that the source value (which is a reference) can come from within + /// the destination range. + /// + /// Effects: Assigns value through all the iterators in the range [first, last). + /// + /// Complexity: Exactly 'last - first' assignments. + /// + /// Note: The C++ standard doesn't specify anything about the value parameter + /// coming from within the first-last range. All std STL implementations act + /// as if the standard specifies that value must not come from within this range. + /// + template <typename ForwardIterator, typename T> + inline void fill(ForwardIterator first, ForwardIterator last, const T& value) + { + eastl::fill_imp< is_scalar<T>::value >::do_fill(first, last, value); + + // Possibly better implementation, as it will deal with small PODs as well as scalars: + // bEasyCopy is true if the type has a trivial constructor (e.g. is a POD) and if + // it is small. Thus any built-in type or any small user-defined struct will qualify. + //const bool bEasyCopy = eastl::type_and<eastl::has_trivial_constructor<T>::value, + // eastl::integral_constant<bool, (sizeof(T) <= 16)>::value; + //eastl::fill_imp<bEasyCopy>::do_fill(first, last, value); + + } + + #if(defined(EA_COMPILER_GNUC) || defined(EA_COMPILER_CLANG)) && (defined(EA_PROCESSOR_X86) || defined(EA_PROCESSOR_X86_64)) + #if defined(EA_PROCESSOR_X86_64) + template <typename Value> + inline void fill(uint64_t* first, uint64_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + uint64_t value = (uint64_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosq\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + + + template <typename Value> + inline void fill(int64_t* first, int64_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + int64_t value = (int64_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosq\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + #endif + + template <typename Value> + inline void fill(uint32_t* first, uint32_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + uint32_t value = (uint32_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosl\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + + + template <typename Value> + inline void fill(int32_t* first, int32_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + int32_t value = (int32_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosl\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + + + template <typename Value> + inline void fill(uint16_t* first, uint16_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + uint16_t value = (uint16_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosw\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + + + template <typename Value> + inline void fill(int16_t* first, int16_t* last, Value c) + { + uintptr_t count = (uintptr_t)(last - first); + int16_t value = (int16_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosw\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + } + + #elif defined(EA_COMPILER_MICROSOFT) && (defined(EA_PROCESSOR_X86) || defined(EA_PROCESSOR_X86_64)) + #if defined(EA_PROCESSOR_X86_64) + template <typename Value> + inline void fill(uint64_t* first, uint64_t* last, Value c) + { + __stosq(first, (uint64_t)c, (size_t)(last - first)); + } + + template <typename Value> + inline void fill(int64_t* first, int64_t* last, Value c) + { + __stosq((uint64_t*)first, (uint64_t)c, (size_t)(last - first)); + } + #endif + + template <typename Value> + inline void fill(uint32_t* first, uint32_t* last, Value c) + { + __stosd((unsigned long*)first, (unsigned long)c, (size_t)(last - first)); + } + + template <typename Value> + inline void fill(int32_t* first, int32_t* last, Value c) + { + __stosd((unsigned long*)first, (unsigned long)c, (size_t)(last - first)); + } + + template <typename Value> + inline void fill(uint16_t* first, uint16_t* last, Value c) + { + __stosw(first, (uint16_t)c, (size_t)(last - first)); + } + + template <typename Value> + inline void fill(int16_t* first, int16_t* last, Value c) + { + __stosw((uint16_t*)first, (uint16_t)c, (size_t)(last - first)); + } + #endif + + + inline void fill(char* first, char* last, const char& c) // It's debateable whether we should use 'char& c' or 'char c' here. + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + inline void fill(char* first, char* last, const int c) // This is used for cases like 'fill(first, last, 0)'. + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + inline void fill(unsigned char* first, unsigned char* last, const unsigned char& c) + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + inline void fill(unsigned char* first, unsigned char* last, const int c) + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + inline void fill(signed char* first, signed char* last, const signed char& c) + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + inline void fill(signed char* first, signed char* last, const int c) + { + memset(first, (unsigned char)c, (size_t)(last - first)); + } + + #if defined(_MSC_VER) || defined(__BORLANDC__) || defined(__ICL) // ICL = Intel compiler + inline void fill(bool* first, bool* last, const bool& b) + { + memset(first, (char)b, (size_t)(last - first)); + } + #endif + + + + + // fill_n + // + // We implement some fill helper functions in order to allow us to optimize it + // where possible. + // + template <bool bIsScalar> + struct fill_n_imp + { + template <typename OutputIterator, typename Size, typename T> + static OutputIterator do_fill(OutputIterator first, Size n, const T& value) + { + for(; n-- > 0; ++first) + *first = value; + return first; + } + }; + + template <> + struct fill_n_imp<true> + { + template <typename OutputIterator, typename Size, typename T> + static OutputIterator do_fill(OutputIterator first, Size n, const T& value) + { + typedef typename eastl::iterator_traits<OutputIterator>::value_type value_type; + + // We create a temp and fill from that because value might alias to + // the destination range and so the compiler would be forced into + // generating less efficient code. + for(const T temp = value; n-- > 0; ++first) + *first = static_cast<value_type>(temp); + return first; + } + }; + + /// fill_n + /// + /// The fill_n function is very much like memset in that a copies a source value + /// n times into a destination range. The source value may come from within + /// the destination range. + /// + /// Effects: Assigns value through all the iterators in the range [first, first + n). + /// + /// Complexity: Exactly n assignments. + /// + template <typename OutputIterator, typename Size, typename T> + OutputIterator fill_n(OutputIterator first, Size n, const T& value) + { + return eastl::fill_n_imp<is_scalar<T>::value>::do_fill(first, n, value); + } + + template <typename Size> + inline char* fill_n(char* first, Size n, const char& c) + { + return (char*)memset(first, (char)c, (size_t)n) + n; + } + + template <typename Size> + inline unsigned char* fill_n(unsigned char* first, Size n, const unsigned char& c) + { + return (unsigned char*)memset(first, (unsigned char)c, (size_t)n) + n; + } + + template <typename Size> + inline signed char* fill_n(signed char* first, Size n, const signed char& c) + { + return (signed char*)memset(first, (signed char)c, n) + (size_t)n; + } + + #if defined(_MSC_VER) || defined(__BORLANDC__) || defined(__ICL) // ICL = Intel compiler + template <typename Size> + inline bool* fill_n(bool* first, Size n, const bool& b) + { + return (bool*)memset(first, (char)b, n) + (size_t)n; + } + #endif + + #if(defined(EA_COMPILER_GNUC) || defined(EA_COMPILER_CLANG)) && (defined(EA_PROCESSOR_X86) || defined(EA_PROCESSOR_X86_64)) + #if defined(EA_PROCESSOR_X86_64) + template <typename Size, typename Value> + inline uint64_t* fill_n(uint64_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + uint64_t value = (uint64_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosq\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + + + template <typename Size, typename Value> + inline int64_t* fill_n(int64_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + int64_t value = (int64_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosq\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + #endif + + template <typename Size, typename Value> + inline uint32_t* fill_n(uint32_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + uint32_t value = (uint32_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosl\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + + + template <typename Size, typename Value> + inline int32_t* fill_n(int32_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + int32_t value = (int32_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosl\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + + + template <typename Size, typename Value> + inline uint16_t* fill_n(uint16_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + uint16_t value = (uint16_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosw\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + + + template <typename Size, typename Value> + inline int16_t* fill_n(int16_t* first, Size n, Value c) + { + uintptr_t count = (uintptr_t)(n); + int16_t value = (int16_t)(c); + + __asm__ __volatile__ ("cld\n\t" + "rep stosw\n\t" + : "+c" (count), "+D" (first), "=m" (first) + : "a" (value) + : "cc" ); + return first; // first is updated by the code above. + } + + #elif defined(EA_COMPILER_MICROSOFT) && (defined(EA_PROCESSOR_X86) || defined(EA_PROCESSOR_X86_64)) + #if defined(EA_PROCESSOR_X86_64) + template <typename Size, typename Value> + inline uint64_t* fill_n(uint64_t* first, Size n, Value c) + { + __stosq(first, (uint64_t)c, (size_t)n); + return first + n; + } + + template <typename Size, typename Value> + inline int64_t* fill_n(int64_t* first, Size n, Value c) + { + __stosq((uint64_t*)first, (uint64_t)c, (size_t)n); + return first + n; + } + #endif + + template <typename Size, typename Value> + inline uint32_t* fill_n(uint32_t* first, Size n, Value c) + { + __stosd((unsigned long*)first, (unsigned long)c, (size_t)n); + return first + n; + } + + template <typename Size, typename Value> + inline int32_t* fill_n(int32_t* first, Size n, Value c) + { + __stosd((unsigned long*)first, (unsigned long)c, (size_t)n); + return first + n; + } + + template <typename Size, typename Value> + inline uint16_t* fill_n(uint16_t* first, Size n, Value c) + { + __stosw(first, (uint16_t)c, (size_t)n); + return first + n; + } + + template <typename Size, typename Value> + inline int16_t* fill_n(int16_t* first, Size n, Value c) + { + __stosw((uint16_t*)first, (uint16_t)c, (size_t)n); + return first + n; + } + #endif + +} // namespace eastl + +#endif // Header include guard + + + + + + + + + + + + + + + |