Merge 94c7e3a into 691f020

randomgen/src/aesctr/aesctr.c

@@ -13,7 +13,7 @@ int aes_capable(void)
     return RANDOMGEN_USE_AESNI;
 }
 
-#if (defined(HAVE_SSE2) && HAVE_SSE2)
+#if defined(HAVE_IMMINTRIN) && !defined(RANDOMGEN_FORCE_SOFTAES)
 #define AES_ROUND(rcon, index)                                                 \
     do                                                                         \
     {                                                                          \
@@ -38,7 +38,7 @@ void aesctr_seed_r(aesctr_state_t *state, uint64_t *seed)
     };*/
     if (RANDOMGEN_USE_AESNI)
     {
-#if (defined(HAVE_SSE2) && HAVE_SSE2)
+#if defined(HAVE_IMMINTRIN) && !defined(RANDOMGEN_FORCE_SOFTAES)
         __m128i k = _mm_set_epi64x(seed[1], seed[0]);
         state->seed[0].m128 = k;
         // D. Lemire manually unrolled following loop since

randomgen/src/aesctr/aesctr.h

@@ -10,59 +10,18 @@
 // #ifdef __AES__
 // contributed by Samuel Neves
 
-#if defined(RANDOMGEN_FORCE_SOFTAES)
-#define HAVE_SSE2 0
-#endif
-
-#undef HAVE_IMMINTRIN
-#if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) ||             \
-    defined(_M_IX86)
-#if defined(_MSC_VER) && defined(_WIN32)
-#if _MSC_VER >= 1900
-#include <immintrin.h>
-#define HAVE_IMMINTRIN 1
-#endif
-#else
-#include <immintrin.h>
-#define HAVE_IMMINTRIN 1
-#endif
-#endif
-
-#ifdef _WIN32
-#define ALIGN_WINDOWS __declspec(align(16))
-#define ALIGN_GCC_CLANG
-#if _MSC_VER == 1500
-#include "../common/inttypes.h"
-#define INLINE __forceinline
-#else
-#include <inttypes.h>
-#define INLINE __inline __forceinline
-#endif
-#else
-#define ALIGN_WINDOWS
-#define ALIGN_GCC_CLANG __attribute__((aligned(16)))
-#include <inttypes.h>
-#define INLINE inline
-#endif
+#include "../common/randomgen_config.h"
+#include "../common/randomgen_immintrin.h"
 
 #include "softaes.h"
-#include <stddef.h>
-#include <stdio.h>
-#include <string.h>
 
 #define AESCTR_UNROLL 4
 #define AESCTR_ROUNDS 10
 
-#ifdef _WIN32
-#define UNLIKELY(x) ((x))
-#else
-#define UNLIKELY(x) (__builtin_expect((x), 0))
-#endif
-
 extern int RANDOMGEN_USE_AESNI;
 
 union AES128_T {
-#if (defined(HAVE_SSE2) && HAVE_SSE2)
+#if defined(HAVE_IMMINTRIN) && !defined(RANDOMGEN_FORCE_SOFTAES)
   __m128i m128;
 #endif
   uint64_t u64[2];
@@ -87,7 +46,7 @@ static INLINE uint64_t aesctr_r(aesctr_state_t *state) {
   uint64_t output;
   if (UNLIKELY(state->offset >= 16 * AESCTR_UNROLL)) {
     if (RANDOMGEN_USE_AESNI) {
-#if (defined(HAVE_SSE2) && HAVE_SSE2)
+#if defined(HAVE_IMMINTRIN) && !defined(RANDOMGEN_FORCE_SOFTAES)
       __m128i work[AESCTR_UNROLL];
       for (int i = 0; i < AESCTR_UNROLL; ++i) {
         work[i] = _mm_xor_si128(state->ctr[i].m128, state->seed[0].m128);

randomgen/src/chacha/chacha.h

@@ -21,31 +21,16 @@
 #define _RANDOMDGEN__CHACHA_H_
 
 
-#ifdef _WIN32
-#if _MSC_VER == 1500
-#include "../common/inttypes.h"
-#define INLINE __forceinline
-#else
-#include <inttypes.h>
-#define INLINE __inline __forceinline
-#endif
-#else
-#include <inttypes.h>
-#define INLINE inline
-#endif
+#include "../common/randomgen_config.h"
 
-#ifdef _MSC_VER
+#define USE_128BIT_COUNTER 1
+
+#if defined(_WIN32) && defined(_MSC_VER)
 #define M128I_CAST
-#define ALIGN_WINDOWS __declspec(align(16))
-#define ALIGN_GCC_CLANG
 #else
-#define ALIGN_WINDOWS
-#define ALIGN_GCC_CLANG __attribute__((aligned(16)))
 #define M128I_CAST (__m128i)
 #endif
 
-#define USE_128BIT_COUNTER 1
-
 typedef double * aligned_double_ptr ;
 
 ALIGN_WINDOWS struct CHACHA_STATE_T {

randomgen/src/common/cpu_features.c

@@ -15,7 +15,7 @@ void feature_flags(int flags[])
 #elif defined(_MSC_VER) && defined(_WIN32)
     int cpu_info[4];
     int num_ids, ecx = 0;
-    __cpuid(&cpu_info, 0);
+    __cpuid(cpu_info, 0);
     num_ids = (int)cpu_info[0];
     if (num_ids >= 1)
     {

randomgen/src/common/cpu_features.h

@@ -1,11 +1,7 @@
 #ifndef _RANDOMGEN_CPU_FEATURES_H_
 #define _RANDOMGEN_CPU_FEATURES_H_
 
-#if defined(_WIN32) && defined(_MSC_VER) && _MSC_VER == 1500
-#include "../common/inttypes.h"
-#else
-#include <inttypes.h>
-#endif
+#include "randomgen_config.h"
 
 #undef HAVE_CPUID
 #if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) ||             \

randomgen/src/common/randomgen_config.h

@@ -0,0 +1,83 @@
+#ifndef _RANDOMDGEN__CONFIG_H_
+#define _RANDOMDGEN__CONFIG_H_
+
+#include <math.h>
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+
+#if defined(_WIN32) && defined(_MSC_VER)
+
+/* windows msvc */
+
+#ifndef MSVCFORCEINLINE
+#ifndef _DEBUG
+#define MSVCFORCEINLINE __forceinline
+#else
+#define MSVCFORCEINLINE
+#endif
+#endif
+
+#ifndef inline
+#if _MSC_VER < 1600
+/* msvs 2008 and earlier */
+#define inline _inline MSVCFORCEINLINE
+#else
+/* msvs 2010 and later */
+#define inline __inline MSVCFORCEINLINE
+#endif
+#define INLINE inline
+#else
+#define INLINE __inline MSVCFORCEINLINE
+#endif
+
+#if _MSC_VER < 1600
+
+/* msvs 2008 and earlier */
+#include "inttypes.h"
+#include "stdbool.h"
+#include "stdint.h"
+
+#elif _MSC_VER < 1800
+
+/* msvs 2010-2012 */
+#include "inttypes.h"
+#include "stdbool.h"
+#include <stdint.h>
+
+#else
+
+/* msvs 2013 and later */
+#include <inttypes.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+#endif
+
+#define ALIGN_WINDOWS __declspec(align(16))
+#define ALIGN_GCC_CLANG
+
+#else
+
+#include <inttypes.h>
+#include <stdbool.h>
+#include <stdint.h>
+
+#define INLINE inline
+
+#define ALIGN_WINDOWS
+#define ALIGN_GCC_CLANG __attribute__((aligned(16)))
+
+#endif
+
+#if defined(__MINGW32__)
+#include <x86intrin.h>
+#endif
+
+#ifdef _WIN32
+#define UNLIKELY(x) ((x))
+#else
+#define UNLIKELY(x) (__builtin_expect((x), 0))
+#endif
+
+#endif

randomgen/src/common/randomgen_config_numpy.h

@@ -0,0 +1,25 @@
+#ifndef _RANDOMDGEN__CONFIG_NUMPY_H_
+#define _RANDOMDGEN__CONFIG_NUMPY_H_
+
+#ifndef RANDOMGEN_STANDALONE
+
+#include "Python.h"
+#include "numpy/npy_common.h"
+#include "numpy/npy_math.h"
+
+#else
+
+#include "standalone/aligned_malloc.h"
+#include "standalone/npy_common.h"
+#include "standalone/npy_math.h"
+#include "standalone/python.h"
+
+#endif
+
+#ifndef NPY_MEMALIGN
+
+#define NPY_MEMALIGN 64 /* 16 for SSE2, 32 for AVX, 64 for Xeon Phi */
+
+#endif
+
+#endif

randomgen/src/common/randomgen_immintrin.h

@@ -0,0 +1,17 @@
+#ifndef _RANDOMGEN__IMMINTRIN_H_
+#define _RANDOMGEN__IMMINTRIN_H_
+
+#undef HAVE_IMMINTRIN
+#if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86)
+#if defined(_MSC_VER) && defined(_WIN32)
+#if _MSC_VER >= 1900
+#include <immintrin.h>
+#define HAVE_IMMINTRIN 1
+#endif
+#else
+#include <immintrin.h>
+#define HAVE_IMMINTRIN 1
+#endif
+#endif
+
+#endif /*  _RANDOMGEN__IMMINTRIN_H_ */

randomgen/src/common/standalone/Python.h

@@ -0,0 +1,9 @@
+#ifndef Py_PYTHON_H
+#define Py_PYTHON_H
+
+#include <inttypes.h>
+
+typedef uintptr_t       Py_uintptr_t;
+typedef intptr_t        Py_intptr_t;
+
+#endif

randomgen/src/common/standalone/aligned_malloc.h

@@ -0,0 +1,28 @@
+#ifndef _RANDOMDGEN__ALIGNED_MALLOC_H_
+#define _RANDOMDGEN__ALIGNED_MALLOC_H_
+
+#include <stdlib.h>
+#include <malloc.h>
+#include <string.h>
+
+static inline void *_aligned_calloc(size_t n, size_t size, size_t alignment) {
+
+    void *p = 0;
+
+    size_t asize = n * size;
+
+    p = _aligned_malloc(asize, alignment);
+
+    if (p) {
+        memset(p, 0, asize);
+    }
+
+    return p;
+
+}
+
+#define malloc_aligned(a) _aligned_calloc(1,(a),NPY_MEMALIGN)
+#define calloc_aligned(a, b) _aligned_calloc((a),(b),NPY_MEMALIGN)
+#define free_aligned(a) _aligned_free(a)
+
+#endif

randomgen/src/common/standalone/npy_common.h

@@ -0,0 +1,31 @@
+#ifndef _NPY_COMMON_H_
+#define _NPY_COMMON_H_
+
+#include "Python.h"
+
+#if defined(_MSC_VER)
+    #define NPY_INLINE __inline
+#elif defined(__GNUC__)
+    #if defined(__STRICT_ANSI__)
+         #define NPY_INLINE __inline__
+    #else
+         #define NPY_INLINE inline
+    #endif
+#else
+    #define NPY_INLINE
+#endif
+
+#define NPY_SIZEOF_LONG SIZEOF_LONG
+
+typedef double npy_double;
+
+typedef unsigned long npy_uint32;
+
+typedef unsigned char npy_bool;
+#define NPY_FALSE 0
+#define NPY_TRUE 1
+
+typedef Py_intptr_t npy_intp;
+typedef Py_uintptr_t npy_uintp;
+
+#endif

randomgen/src/common/standalone/npy_math.h

@@ -0,0 +1,25 @@
+#ifndef __NPY_MATH_C99_H_
+#define __NPY_MATH_C99_H_
+
+#include <math.h>
+
+#include "npy_common.h"
+
+#if defined(_MSC_VER) && (_MSC_VER < 1900)
+    #define npy_isnan(x) _isnan((x))
+#else
+    #define npy_isnan(x) isnan(x)
+#endif
+
+NPY_INLINE static float __npy_nanf(void)
+{
+    const union { npy_uint32 __i; float __f;} __bint = {0x7fc00000UL};
+    return __bint.__f;
+}
+
+#define NPY_NANF __npy_nanf()
+
+#define NPY_NAN ((npy_double)NPY_NANF)
+
+#endif
+

randomgen/src/common/stdbool.h

@@ -0,0 +1,13 @@
+#ifndef _RANDOMDGEN__STDBOOL_H
+#define _RANDOMDGEN__STDBOOL_H
+
+#if defined(_WIN32) && defined(_MSC_VER)
+#ifndef __bool_true_false_are_defined
+#define __bool_true_false_are_defined
+typedef int bool;
+#define false 0
+#define true 1
+#endif
+#endif
+
+#endif