1 /****************************************************************************
2 * Copyright (C) 2014-2017 Intel Corporation. All Rights Reserved.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 ****************************************************************************/
28 #include "core/knobs.h"
30 #if (defined(FORCE_WINDOWS) || defined(_WIN32)) && !defined(FORCE_LINUX)
32 #define SWR_API __cdecl
33 #define SWR_VISIBLE __declspec(dllexport)
45 #if defined(MemoryFence)
46 // Windows.h defines MemoryFence as _mm_mfence, but this conflicts with llvm::sys::MemoryFence
50 #define OSALIGN(RWORD, WIDTH) __declspec(align(WIDTH)) RWORD
53 // We compile Debug builds with inline function expansion enabled. This allows
54 // functions compiled with __forceinline to be inlined even in Debug builds.
55 // The inline_depth(0) pragma below will disable inline function expansion for
56 // normal INLINE / inline functions, but not for __forceinline functions.
57 // Our SIMD function wrappers (see simdlib.hpp) use __forceinline even in
60 #pragma inline_depth(0)
62 #define INLINE __forceinline
64 #define DEBUGBREAK __debugbreak()
66 #define PRAGMA_WARNING_PUSH_DISABLE(...) \
67 __pragma(warning(push));\
68 __pragma(warning(disable:__VA_ARGS__));
70 #define PRAGMA_WARNING_POP() __pragma(warning(pop))
72 static inline void *AlignedMalloc(size_t _Size
, size_t _Alignment
)
74 return _aligned_malloc(_Size
, _Alignment
);
77 static inline void AlignedFree(void* p
)
79 return _aligned_free(p
);
83 #define BitScanReverseSizeT BitScanReverse64
84 #define BitScanForwardSizeT BitScanForward64
85 #define _mm_popcount_sizeT _mm_popcnt_u64
87 #define BitScanReverseSizeT BitScanReverse
88 #define BitScanForwardSizeT BitScanForward
89 #define _mm_popcount_sizeT _mm_popcnt_u32
92 #elif defined(__APPLE__) || defined(FORCE_LINUX) || defined(__linux__) || defined(__gnu_linux__)
95 #define SWR_VISIBLE __attribute__((visibility("default")))
99 #include <x86intrin.h>
101 #include <sys/types.h>
103 #include <sys/stat.h>
108 typedef void* LPVOID
;
110 typedef unsigned int UINT
;
111 typedef void* HANDLE
;
113 typedef unsigned int DWORD
;
121 #define MAX_PATH PATH_MAX
123 #define OSALIGN(RWORD, WIDTH) RWORD __attribute__((aligned(WIDTH)))
125 #define INLINE __inline
127 #define DEBUGBREAK asm ("int $3")
129 #if !defined(__CYGWIN__)
138 #if defined(__GNUC__) && !defined(__INTEL_COMPILER)
139 #define __declspec(x) __declspec_##x
140 #define __declspec_align(y) __attribute__((aligned(y)))
141 #define __declspec_deprecated __attribute__((deprecated))
142 #define __declspec_dllexport
143 #define __declspec_dllimport
144 #define __declspec_noinline __attribute__((__noinline__))
145 #define __declspec_nothrow __attribute__((nothrow))
146 #define __declspec_novtable
147 #define __declspec_thread __thread
149 #define __declspec(X)
154 #define GCC_VERSION (__GNUC__ * 10000 \
155 + __GNUC_MINOR__ * 100 \
156 + __GNUC_PATCHLEVEL__)
158 #if !defined(__clang__) && (__GNUC__) && (GCC_VERSION < 40500)
163 asm volatile("rdtsc" : "=a"(low
), "=d"(high
));
164 return (low
| ((uint64_t)high
<< 32));
168 #if !defined( __clang__) && !defined(__INTEL_COMPILER)
169 // Intrinsic not defined in gcc
171 void _mm256_storeu2_m128i(__m128i
*hi
, __m128i
*lo
, __m256i a
)
173 _mm_storeu_si128((__m128i
*)lo
, _mm256_castsi256_si128(a
));
174 _mm_storeu_si128((__m128i
*)hi
, _mm256_extractf128_si256(a
, 0x1));
177 // gcc prior to 4.9 doesn't have _mm*_undefined_*
178 #if (__GNUC__) && (GCC_VERSION < 409000)
179 #define _mm_undefined_si128 _mm_setzero_si128
180 #define _mm256_undefined_ps _mm256_setzero_ps
185 unsigned char _BitScanForward(unsigned long *Index
, unsigned long Mask
)
187 *Index
= __builtin_ctz(Mask
);
192 unsigned char _BitScanForward(unsigned int *Index
, unsigned int Mask
)
194 *Index
= __builtin_ctz(Mask
);
199 unsigned char _BitScanReverse(unsigned long *Index
, unsigned long Mask
)
201 *Index
= __builtin_clz(Mask
);
206 unsigned char _BitScanReverse(unsigned int *Index
, unsigned int Mask
)
208 *Index
= __builtin_clz(Mask
);
212 #define _BitScanForward64 _BitScanForward
213 #define _BitScanReverse64 _BitScanReverse
216 void *AlignedMalloc(size_t size
, size_t alignment
)
219 if (posix_memalign(&ret
, alignment
, size
))
227 void AlignedFree(void* p
)
232 #define _countof(a) (sizeof(a)/sizeof(*(a)))
234 #define sprintf_s sprintf
235 #define strcpy_s(dst,size,src) strncpy(dst,src,size)
236 #define GetCurrentProcessId getpid
238 #define InterlockedCompareExchange(Dest, Exchange, Comparand) __sync_val_compare_and_swap(Dest, Comparand, Exchange)
239 #define InterlockedExchangeAdd(Addend, Value) __sync_fetch_and_add(Addend, Value)
240 #define InterlockedDecrement(Append) __sync_sub_and_fetch(Append, 1)
241 #define InterlockedDecrement64(Append) __sync_sub_and_fetch(Append, 1)
242 #define InterlockedIncrement(Append) __sync_add_and_fetch(Append, 1)
243 #define InterlockedAdd(Addend, Value) __sync_add_and_fetch(Addend, Value)
244 #define InterlockedAdd64(Addend, Value) __sync_add_and_fetch(Addend, Value)
245 #define _ReadWriteBarrier() asm volatile("" ::: "memory")
247 #define PRAGMA_WARNING_PUSH_DISABLE(...)
248 #define PRAGMA_WARNING_POP()
250 #define ZeroMemory(dst, size) memset(dst, 0, size)
253 #error Unsupported OS/system.
257 #define THREAD thread_local
260 typedef uint8_t KILOBYTE
[1024];
261 typedef KILOBYTE MEGABYTE
[1024];
262 typedef MEGABYTE GIGABYTE
[1024];
264 #define OSALIGNLINE(RWORD) OSALIGN(RWORD, 64)
265 #define OSALIGNSIMD(RWORD) OSALIGN(RWORD, KNOB_SIMD_BYTES)
266 #if ENABLE_AVX512_SIMD16
267 #define OSALIGNSIMD16(RWORD) OSALIGN(RWORD, KNOB_SIMD16_BYTES)
270 #include "common/swr_assert.h"
273 #define ATTR_UNUSED __attribute__((unused))
278 #define SWR_FUNC(_retType, _funcName, /* args */...) \
279 typedef _retType (SWR_API * PFN##_funcName)(__VA_ARGS__); \
280 _retType SWR_API _funcName(__VA_ARGS__);
283 void SWR_API
SetCurrentThreadName(const char* pThreadName
);
284 void SWR_API
CreateDirectoryPath(const std::string
& path
);
286 /// Execute Command (block until finished)
287 /// @returns process exit value
289 const std::string
& cmd
, ///< (In) Command line string
290 const char* pOptEnvStrings
= nullptr, ///< (Optional In) Environment block for new process
291 std::string
* pOptStdOut
= nullptr, ///< (Optional Out) Standard Output text
292 std::string
* pOptStdErr
= nullptr, ///< (Optional Out) Standard Error text
293 const std::string
* pOptStdIn
= nullptr); ///< (Optional In) Standard Input text