src/gallium/drivers/swr/rasterizer/common/os.h

   1 /****************************************************************************
   2  * Copyright (C) 2014-2017 Intel Corporation.   All Rights Reserved.
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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
  21  * IN THE SOFTWARE.
  22  ****************************************************************************/
  23
  24 #ifndef __SWR_OS_H__
  25 #define __SWR_OS_H__
  26
  27 #include <cstddef>
  28 #include "core/knobs.h"
  29
  30 #if (defined(FORCE_WINDOWS) || defined(_WIN32)) && !defined(FORCE_LINUX)
  31
  32 #define SWR_API __cdecl
  33 #define SWR_VISIBLE __declspec(dllexport)
  34
  35 #ifndef NOMINMAX
  36 #undef UNICODE
  37 #define NOMINMAX
  38 #include <windows.h>
  39 #undef NOMINMAX
  40 #define UNICODE
  41 #else
  42 #include <windows.h>
  43 #endif
  44 #include <intrin.h>
  45 #include <cstdint>
  46
  47 #if defined(MemoryFence)
  48 // Windows.h defines MemoryFence as _mm_mfence, but this conflicts with llvm::sys::MemoryFence
  49 #undef MemoryFence
  50 #endif
  51
  52 #define OSALIGN(RWORD, WIDTH) __declspec(align(WIDTH)) RWORD
  53
  54 #if defined(_DEBUG)
  55 // We compile Debug builds with inline function expansion enabled.  This allows
  56 // functions compiled with __forceinline to be inlined even in Debug builds.
  57 // The inline_depth(0) pragma below will disable inline function expansion for
  58 // normal INLINE / inline functions, but not for __forceinline functions.
  59 // Our SIMD function wrappers (see simdlib.hpp) use __forceinline even in
  60 // Debug builds.
  61 #define INLINE inline
  62 #pragma inline_depth(0)
  63 #else
  64 // Use of __forceinline increases compile time dramatically in release builds
  65 // and provides almost 0 measurable benefit.  Disable until we have a compelling
  66 // use-case
  67 // #define INLINE __forceinline
  68 #define INLINE inline
  69 #endif
  70 #ifndef FORCEINLINE
  71 #define FORCEINLINE __forceinline
  72 #endif
  73
  74 #define DEBUGBREAK __debugbreak()
  75
  76 #define PRAGMA_WARNING_PUSH_DISABLE(...) \
  77     __pragma(warning(push));             \
  78     __pragma(warning(disable : __VA_ARGS__));
  79
  80 #define PRAGMA_WARNING_POP() __pragma(warning(pop))
  81
  82 static inline void* AlignedMalloc(size_t _Size, size_t _Alignment)
  83 {
  84     return _aligned_malloc(_Size, _Alignment);
  85 }
  86
  87 static inline void AlignedFree(void* p)
  88 {
  89     return _aligned_free(p);
  90 }
  91
  92 #if defined(_WIN64)
  93 #define BitScanReverseSizeT BitScanReverse64
  94 #define BitScanForwardSizeT BitScanForward64
  95 #define _mm_popcount_sizeT _mm_popcnt_u64
  96 #else
  97 #define BitScanReverseSizeT BitScanReverse
  98 #define BitScanForwardSizeT BitScanForward
  99 #define _mm_popcount_sizeT _mm_popcnt_u32
 100 #endif
 101
 102 #elif defined(__APPLE__) || defined(FORCE_LINUX) || defined(__linux__) || defined(__gnu_linux__)
 103
 104 #define SWR_API
 105 #define SWR_VISIBLE __attribute__((visibility("default")))
 106
 107 #include <stdlib.h>
 108 #include <string.h>
 109 #include <x86intrin.h>
 110 #include <stdint.h>
 111 #include <sys/types.h>
 112 #include <unistd.h>
 113 #include <sys/stat.h>
 114 #include <stdio.h>
 115 #include <limits.h>
 116
 117 typedef void         VOID;
 118 typedef void*        LPVOID;
 119 typedef int          INT;
 120 typedef unsigned int UINT;
 121 typedef void*        HANDLE;
 122 typedef int          LONG;
 123 typedef unsigned int DWORD;
 124
 125 #undef FALSE
 126 #define FALSE 0
 127
 128 #undef TRUE
 129 #define TRUE 1
 130
 131 #define MAX_PATH PATH_MAX
 132
 133 #define OSALIGN(RWORD, WIDTH) RWORD __attribute__((aligned(WIDTH)))
 134 #ifndef INLINE
 135 #define INLINE __inline
 136 #endif
 137 #ifndef FORCEINLINE
 138 #define FORCEINLINE INLINE
 139 #endif
 140 #define DEBUGBREAK asm("int $3")
 141
 142 #if !defined(__CYGWIN__)
 143
 144 #ifndef __cdecl
 145 #define __cdecl
 146 #endif
 147 #ifndef __stdcall
 148 #define __stdcall
 149 #endif
 150
 151 #if defined(__GNUC__) && !defined(__INTEL_COMPILER)
 152 #define __declspec(x) __declspec_##x
 153 #define __declspec_align(y) __attribute__((aligned(y)))
 154 #define __declspec_deprecated __attribute__((deprecated))
 155 #define __declspec_dllexport
 156 #define __declspec_dllimport
 157 #define __declspec_noinline __attribute__((__noinline__))
 158 #define __declspec_nothrow __attribute__((nothrow))
 159 #define __declspec_novtable
 160 #define __declspec_thread __thread
 161 #else
 162 #define __declspec(X)
 163 #endif
 164
 165 #endif
 166
 167 #define GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)
 168
 169 #if !defined(__clang__) && (__GNUC__) && (GCC_VERSION < 40500)
 170 inline uint64_t      __rdtsc()
 171 {
 172     long low, high;
 173     asm volatile("rdtsc" : "=a"(low), "=d"(high));
 174     return (low | ((uint64_t)high << 32));
 175 }
 176 #endif
 177
 178 #if !defined(__clang__) && !defined(__INTEL_COMPILER)
 179 // Intrinsic not defined in gcc < 10
 180 #if (__GNUC__) && (GCC_VERSION < 100000)
 181 static INLINE void _mm256_storeu2_m128i(__m128i* hi, __m128i* lo, __m256i a)
 182 {
 183     _mm_storeu_si128((__m128i*)lo, _mm256_castsi256_si128(a));
 184     _mm_storeu_si128((__m128i*)hi, _mm256_extractf128_si256(a, 0x1));
 185 }
 186 #endif
 187
 188 // gcc prior to 4.9 doesn't have _mm*_undefined_*
 189 #if (__GNUC__) && (GCC_VERSION < 40900)
 190 #define _mm_undefined_si128 _mm_setzero_si128
 191 #define _mm256_undefined_ps _mm256_setzero_ps
 192 #endif
 193 #endif
 194
 195 inline unsigned char _BitScanForward(unsigned long* Index, unsigned long Mask)
 196 {
 197     *Index = __builtin_ctz(Mask);
 198     return (Mask != 0);
 199 }
 200
 201 inline unsigned char _BitScanForward(unsigned int* Index, unsigned int Mask)
 202 {
 203     *Index = __builtin_ctz(Mask);
 204     return (Mask != 0);
 205 }
 206
 207 inline unsigned char _BitScanReverse(unsigned long* Index, unsigned long Mask)
 208 {
 209     *Index = 63 - __builtin_clz(Mask);
 210     return (Mask != 0);
 211 }
 212
 213 inline unsigned char _BitScanReverse(unsigned int* Index, unsigned int Mask)
 214 {
 215     *Index = 31 - __builtin_clz(Mask);
 216     return (Mask != 0);
 217 }
 218
 219 #define _BitScanForward64 _BitScanForward
 220 #define _BitScanReverse64 _BitScanReverse
 221
 222 inline void* AlignedMalloc(size_t size, size_t alignment)
 223 {
 224     void* ret;
 225     if (posix_memalign(&ret, alignment, size))
 226     {
 227         return NULL;
 228     }
 229     return ret;
 230 }
 231
 232 static inline void AlignedFree(void* p)
 233 {
 234     free(p);
 235 }
 236
 237 #define _countof(a) (sizeof(a) / sizeof(*(a)))
 238
 239 #define sprintf_s sprintf
 240 #define strcpy_s(dst, size, src) strncpy(dst, src, size)
 241 #define GetCurrentProcessId getpid
 242
 243 #define InterlockedCompareExchange(Dest, Exchange, Comparand) \
 244     __sync_val_compare_and_swap(Dest, Comparand, Exchange)
 245 #define InterlockedExchangeAdd(Addend, Value) __sync_fetch_and_add(Addend, Value)
 246 #define InterlockedDecrement(Append) __sync_sub_and_fetch(Append, 1)
 247 #define InterlockedDecrement64(Append) __sync_sub_and_fetch(Append, 1)
 248 #define InterlockedIncrement(Append) __sync_add_and_fetch(Append, 1)
 249 #define InterlockedAdd(Addend, Value) __sync_add_and_fetch(Addend, Value)
 250 #define InterlockedAdd64(Addend, Value) __sync_add_and_fetch(Addend, Value)
 251 #define _ReadWriteBarrier() asm volatile("" ::: "memory")
 252
 253 #define PRAGMA_WARNING_PUSH_DISABLE(...)
 254 #define PRAGMA_WARNING_POP()
 255
 256 #define ZeroMemory(dst, size) memset(dst, 0, size)
 257 #else
 258
 259 #error Unsupported OS/system.
 260
 261 #endif
 262
 263 #define THREAD thread_local
 264
 265 // Universal types
 266 typedef uint8_t  KILOBYTE[1024];
 267 typedef KILOBYTE MEGABYTE[1024];
 268 typedef MEGABYTE GIGABYTE[1024];
 269
 270 #define OSALIGNLINE(RWORD) OSALIGN(RWORD, 64)
 271 #define OSALIGNSIMD(RWORD) OSALIGN(RWORD, KNOB_SIMD_BYTES)
 272 #define OSALIGNSIMD16(RWORD) OSALIGN(RWORD, KNOB_SIMD16_BYTES)
 273
 274 #include "common/swr_assert.h"
 275
 276 #ifdef __GNUC__
 277 #define ATTR_UNUSED __attribute__((unused))
 278 #else
 279 #define ATTR_UNUSED
 280 #endif
 281
 282 #define SWR_FUNC(_retType, _funcName, /* args */...)        \
 283     typedef _retType(SWR_API* PFN##_funcName)(__VA_ARGS__); \
 284     _retType SWR_API _funcName(__VA_ARGS__);
 285
 286 // Defined in os.cpp
 287 void SWR_API SetCurrentThreadName(const char* pThreadName);
 288 void SWR_API CreateDirectoryPath(const std::string& path);
 289
 290 /// Execute Command (block until finished)
 291 /// @returns process exit value
 292 int SWR_API
 293     ExecCmd(const std::string& cmd,                ///< (In) Command line string
 294             const char*  pOptEnvStrings = nullptr, ///< (Optional In) Environment block for new process
 295             std::string* pOptStdOut     = nullptr,   ///< (Optional Out) Standard Output text
 296             std::string* pOptStdErr     = nullptr,   ///< (Optional Out) Standard Error text
 297             const std::string* pOptStdIn = nullptr); ///< (Optional In) Standard Input text
 298
 299
 300 /// Helper for setting up FP state
 301 /// @returns old csr state
 302 static INLINE uint32_t SetOptimalVectorCSR()
 303 {
 304     uint32_t oldCSR = _mm_getcsr();
 305
 306     uint32_t newCSR = (oldCSR & ~(_MM_ROUND_MASK | _MM_DENORMALS_ZERO_MASK | _MM_FLUSH_ZERO_MASK));
 307     newCSR |= (_MM_ROUND_NEAREST | _MM_FLUSH_ZERO_ON | _MM_DENORMALS_ZERO_ON);
 308     _mm_setcsr(newCSR);
 309
 310     return oldCSR;
 311 }
 312
 313 /// Set Vector CSR state.
 314 /// @param csrState - should be value returned from SetOptimalVectorCSR()
 315 static INLINE void RestoreVectorCSR(uint32_t csrState)
 316 {
 317     _mm_setcsr(csrState);
 318 }
 319
 320 #endif //__SWR_OS_H__