Verified Commit a5007a42 authored by Michael R. Crusoe's avatar Michael R. Crusoe 🏳️‍🌈
Browse files

Update SIMDe to for the FMA instruction as well

parent 7ede165b

debian/changelog

+8 −1
Original line number Diff line number Diff line 
raxml (8.2.12+dfsg-3) UNRELEASED; urgency=medium



  * Team upload.

  * Update SIMDe to for the FMA instruction as well.



 -- Michael R. Crusoe <michael.crusoe@gmail.com>  Fri, 20 Dec 2019 09:31:07 -0800



raxml (8.2.12+dfsg-2) unstable; urgency=medium



  * Team upload.

  * Build on non-x86 too, via SIMDE

  * Build on non-x86 too, via SIMDe



 -- Michael R. Crusoe <michael.crusoe@gmail.com>  Mon, 16 Dec 2019 17:54:43 +0100

debian/include/simde/hedley.h

+565 −187

File changed.

Preview size limit exceeded, changes collapsed.

debian/include/simde/simde-common.h

+54 −23
Original line number Diff line number Diff line
@@ -39,7 +39,7 @@ 
  HEDLEY_TINYC_VERSION_CHECK(0,9,24) || \

  HEDLEY_TI_VERSION_CHECK(8,1,0)

#  define SIMDE_ALIGN(alignment) __attribute__((aligned(alignment)))

#elif defined(_MSC_VER) && (!defined(_M_IX86) || defined(_M_AMD64))

#elif defined(_MSC_VER) && !(defined(_M_ARM) && !defined(_M_ARM64))

#  define SIMDE_ALIGN(alignment) __declspec(align(alignment))

#elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)

#  define SIMDE_ALIGN(alignment) _Alignas(alignment)
@@ -50,22 +50,22 @@ 
#endif



#define simde_assert_aligned(alignment, val) \

  simde_assert_int(((uintptr_t) (val)) % (alignment), ==, 0)

  simde_assert_int(HEDLEY_REINTERPRET_CAST(const uintptr_t, (val)) % (alignment), ==, 0)



/* TODO: this should really do something like

   HEDLEY_STATIC_CAST(T, (simde_assert_int(alignment, v), v))

   but I need to think about how to handle it in all compilers...

   may end up moving to Hedley, too. */

#if HEDLEY_HAS_BUILTIN(__builtin_assume_aligned)

#  define SIMDE_CAST_ALIGN(alignment, T, v) ((T) __builtin_assume_aligned(v, alignment))

#  define SIMDE_CAST_ALIGN(alignment, T, v) HEDLEY_REINTERPRET_CAST(T, __builtin_assume_aligned(v, alignment))

#elif HEDLEY_HAS_WARNING("-Wcast-align")

#  define SIMDE_CAST_ALIGN(alignment, T, v) \

    HEDLEY_DIAGNOSTIC_PUSH \

    _Pragma("clang diagnostic ignored \"-Wcast-align\"") \

    HEDLEY_STATIC_CAST(T, v) \

    HEDLEY_REINTERPRET_CAST(T, (v)) \

    HEDLEY_DIAGNOSTIC_POP

#else

#  define SIMDE_CAST_ALIGN(alignment, T, v) HEDLEY_STATIC_CAST(T, v)

#  define SIMDE_CAST_ALIGN(alignment, T, v) HEDLEY_REINTERPRET_CAST(T, (v))

#endif



#if HEDLEY_GCC_HAS_ATTRIBUTE(vector_size,4,6,0)
@@ -145,10 +145,21 @@ 
#  define SIMDE__END_DECLS HEDLEY_END_C_DECLS

#endif



#if HEDLEY_HAS_WARNING("-Wpedantic")

#  define SIMDE_DIAGNOSTIC_DISABLE_INT128 _Pragma("clang diagnostic ignored \"-Wpedantic\"")

#elif defined(HEDLEY_GCC_VERSION)

#  define SIMDE_DIAGNOSTIC_DISABLE_INT128 _Pragma("GCC diagnostic ignored \"-Wpedantic\"")

#else

#  define SIMDE_DIAGNOSTIC_DISABLE_INT128

#endif



#if defined(__SIZEOF_INT128__)

#  define SIMDE__HAVE_INT128

HEDLEY_DIAGNOSTIC_PUSH

SIMDE_DIAGNOSTIC_DISABLE_INT128

typedef __int128 simde_int128;

typedef unsigned __int128 simde_uint128;

HEDLEY_DIAGNOSTIC_POP

#endif



/* TODO: we should at least make an attempt to detect the correct
@@ -211,7 +222,10 @@ HEDLEY_STATIC_ASSERT(sizeof(simde_float64) == 8, "Unable to find 64-bit floating 
#if HEDLEY_HAS_BUILTIN(__builtin_shufflevector)

#  define SIMDE__SHUFFLE_VECTOR(elem_size, vec_size, a, b, ...) __builtin_shufflevector(a, b, __VA_ARGS__)

#elif HEDLEY_GCC_HAS_BUILTIN(__builtin_shuffle,4,7,0) && !defined(__INTEL_COMPILER)

#  define SIMDE__SHUFFLE_VECTOR(elem_size, vec_size, a, b, ...) __builtin_shuffle(a, b, (int##elem_size##_t __attribute__((__vector_size__(vec_size)))) { __VA_ARGS__ })

#  define SIMDE__SHUFFLE_VECTOR(elem_size, vec_size, a, b, ...) (__extension__ ({ \

       int##elem_size##_t __attribute__((__vector_size__(vec_size))) simde_shuffle_ = { __VA_ARGS__ }; \

       __builtin_shuffle(a, b, simde_shuffle_); \

     }))

#endif



#if HEDLEY_GCC_HAS_BUILTIN(__builtin_convertvector,9,0,0)
@@ -222,12 +236,21 @@ HEDLEY_STATIC_ASSERT(sizeof(simde_float64) == 8, "Unable to find 64-bit floating 
#  define SIMDE_CONVERT_FTOI(T,v) \

    HEDLEY_DIAGNOSTIC_PUSH \

    _Pragma("clang diagnostic ignored \"-Wbad-function-cast\"") \

    ((T) (v)) \

    HEDLEY_STATIC_CAST(T, (v)) \

    HEDLEY_DIAGNOSTIC_POP

#else

#  define SIMDE_CONVERT_FTOI(T,v) ((T) (v))

#endif





#if HEDLEY_HAS_WARNING("-Wfloat-equal")

#  define SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL _Pragma("clang diagnostic ignored \"-Wfloat-equal\"")

#elif HEDLEY_GCC_VERSION_CHECK(3,0,0)

#  define SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL _Pragma("GCC diagnostic ignored \"-Wfloat-equal\"")

#else

#  define SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL

#endif



/* Some algorithms are iterative, and fewer iterations means less

   accuracy.  Lower values here will result in faster, but less

   accurate, calculations for some functions. */
@@ -305,21 +328,29 @@ HEDLEY_STATIC_ASSERT(sizeof(simde_float64) == 8, "Unable to find 64-bit floating 
#  endif

#endif



#if !defined(__cplusplus)

#define SIMDE_F64_ALL_SET   (((union { uint64_t u64; simde_float64 f64; }) { .u64 = ~UINT64_C(0x0) }).f64)

#define SIMDE_F64_ALL_UNSET (((union { uint64_t u64; simde_float64 f64; }) { .u64 =  UINT64_C(0x0) }).f64)

#define SIMDE_F32_ALL_SET   (((union { uint32_t u32; simde_float32 f32; }) { .u32 = ~UINT32_C(0x0) }).f32)

#define SIMDE_F32_ALL_UNSET (((union { uint32_t u32; simde_float32 f32; }) { .u32 =  UINT32_C(0x0) }).f32)

#else

static const union { uint64_t u64; simde_float64 f64; } simde_f64_all_set   = { .u64 = ~UINT64_C(0) };

static const union { uint64_t u64; simde_float64 f64; } simde_f64_all_unset = { .u64 =  UINT64_C(0) };

static const union { uint64_t u32; simde_float64 f32; } simde_f32_all_set   = { .u32 = ~UINT32_C(0) };

static const union { uint64_t u32; simde_float64 f32; } simde_f32_all_unset = { .u32 =  UINT32_C(0) };



#  define SIMDE_F64_ALL_SET   (simde_f64_all_set.f64)

#  define SIMDE_F64_ALL_UNSET (simde_f64_all_unset.f64)

#  define SIMDE_F32_ALL_SET   (simde_f32_all_set.f32)

#  define SIMDE_F32_ALL_UNSET (simde_f32_all_unset.f32)

#endif

HEDLEY_ALWAYS_INLINE static

simde_float32 simde_u32_to_f32(uint32_t val) {

  union {

    uint32_t u32;

    simde_float32 f32;

  } u;

  u.u32 = val;

  return u.f32;

}



HEDLEY_ALWAYS_INLINE static

simde_float64 simde_u64_to_f64(uint64_t val) {

  union {

    uint64_t u64;

    simde_float64 f64;

  } u;

  u.u64 = val;

  return u.f64;

}



#define SIMDE_F32_ALL_SET   (simde_u32_to_f32(~UINT32_C(0)))

#define SIMDE_F32_ALL_UNSET (simde_u32_to_f32( UINT32_C(0)))

#define SIMDE_F64_ALL_SET   (simde_u64_to_f64(~UINT64_C(0)))

#define SIMDE_F64_ALL_UNSET (simde_u64_to_f64( UINT64_C(0)))



#endif /* !defined(SIMDE_COMMON_H) */

debian/include/simde/x86/avx.h

+34 −29
Original line number Diff line number Diff line
@@ -1100,7 +1100,7 @@ simde_mm256_castpd_ps (simde__m256d a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256_FROM_NATIVE(_mm256_castpd_ps(a.n));

#else

  return *((simde__m256*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1113,7 +1113,7 @@ simde_mm256_castpd_si256 (simde__m256d a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256I_FROM_NATIVE(_mm256_castpd_si256(a.n));

#else

  return *((simde__m256i*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256i*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1162,7 +1162,7 @@ simde_mm256_castps_pd (simde__m256 a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256D_FROM_NATIVE(_mm256_castps_pd(a.n));

#else

  return *((simde__m256d*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256d*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1175,7 +1175,7 @@ simde_mm256_castps_si256 (simde__m256 a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256I_FROM_NATIVE(_mm256_castps_si256(a.n));

#else

  return *((simde__m256i*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256i*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1260,7 +1260,7 @@ simde_mm256_castsi256_pd (simde__m256i a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256D_FROM_NATIVE(_mm256_castsi256_pd(a.n));

#else

  return *((simde__m256d*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256d*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1273,7 +1273,7 @@ simde_mm256_castsi256_ps (simde__m256i a) { 
#if defined(SIMDE_AVX_NATIVE)

  return SIMDE__M256_FROM_NATIVE(_mm256_castsi256_ps(a.n));

#else

  return *((simde__m256*) &a);

  return *HEDLEY_REINTERPRET_CAST(simde__m256*, &a);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -1320,6 +1320,9 @@ simde_mm256_ceil_ps (simde__m256 a) { 
#  define _mm256_ceil_ps(a) SIMDE__M256_TO_NATIVE(simde_mm256_ceil_ps(SIMDE__M256_FROM_NATIVE(a)))

#endif



HEDLEY_DIAGNOSTIC_PUSH

SIMDE_DIAGNOSTIC_DISABLE_FLOAT_EQUAL



/* This implementation does not support signaling NaNs (yet?) */

SIMDE__FUNCTION_ATTRIBUTES

simde__m128d
@@ -2457,6 +2460,8 @@ simde_mm256_cmp_ps (simde__m256 a, simde__m256 b, const int imm8) 
#  define _mm256_cmp_ps(a, b, imm8) SIMDE__M256_TO_NATIVE(simde_mm256_cmp_ps(SIMDE__M256_FROM_NATIVE(a), SIMDE__M256_FROM_NATIVE(b), imm8))

#endif



HEDLEY_DIAGNOSTIC_POP /* -Wfloat-equal */



SIMDE__FUNCTION_ATTRIBUTES

simde__m256d

simde_mm256_cvtepi32_pd (simde__m128i a) {
@@ -2467,7 +2472,7 @@ simde_mm256_cvtepi32_pd (simde__m128i a) { 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f64) / sizeof(r.f64[0])) ; i++) {

    r.f64[i] = (simde_float64) a.i32[i];

    r.f64[i] = HEDLEY_STATIC_CAST(simde_float64, a.i32[i]);

  }

#endif
@@ -2487,7 +2492,7 @@ simde_mm256_cvtepi32_ps (simde__m256i a) { 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f32) / sizeof(r.f32[0])) ; i++) {

    r.f32[i] = (simde_float32) a.i32[i];

    r.f32[i] = HEDLEY_STATIC_CAST(simde_float32, a.i32[i]);

  }

#endif
@@ -2527,7 +2532,7 @@ simde_mm256_cvtpd_ps (simde__m256d a) { 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f32) / sizeof(r.f32[0])) ; i++) {

    r.f32[i] = (simde_float32) a.f64[i];

    r.f32[i] = HEDLEY_STATIC_CAST(simde_float32, a.f64[i]);

  }

#endif
@@ -2984,7 +2989,7 @@ simde_mm256_load_pd (const double a[HEDLEY_ARRAY_PARAM(4)]) { 
#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_load_pd(a);

#else

  r = *SIMDE_CAST_ALIGN(32, simde__m256d*, a);

  r = *SIMDE_CAST_ALIGN(32, simde__m256d const*, a);

#endif



  return r;
@@ -3003,7 +3008,7 @@ simde_mm256_load_ps (const float a[HEDLEY_ARRAY_PARAM(8)]) { 
#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_load_ps(a);

#else

  r = *SIMDE_CAST_ALIGN(32, simde__m256*, a);

  r = *SIMDE_CAST_ALIGN(32, simde__m256 const*, a);

#endif



  return r;
@@ -3125,7 +3130,7 @@ simde_mm_maskload_pd (const simde_float64 mem_addr[HEDLEY_ARRAY_PARAM(4)], simde 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f64) / sizeof(r.f64[0])) ; i++) {

    r.i64[i] = ((int64_t*) mem_addr)[i] & (mask.i64[i] >> 63);

    r.i64[i] = HEDLEY_REINTERPRET_CAST(int64_t const*, mem_addr)[i] & (mask.i64[i] >> 63);

  }

#endif
@@ -3145,7 +3150,7 @@ simde_mm256_maskload_pd (const simde_float64 mem_addr[HEDLEY_ARRAY_PARAM(4)], si 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f64) / sizeof(r.f64[0])) ; i++) {

    r.i64[i] = ((int64_t*) mem_addr)[i] & (mask.i64[i] >> 63);

    r.i64[i] = HEDLEY_REINTERPRET_CAST(int64_t const*, mem_addr)[i] & (mask.i64[i] >> 63);

  }

#endif
@@ -3165,7 +3170,7 @@ simde_mm_maskload_ps (const simde_float32 mem_addr[HEDLEY_ARRAY_PARAM(4)], simde 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f32) / sizeof(r.f32[0])) ; i++) {

    r.i32[i] = ((int32_t*) mem_addr)[i] & (mask.i32[i] >> 31);

    r.i32[i] = HEDLEY_REINTERPRET_CAST(int32_t const*, mem_addr)[i] & (mask.i32[i] >> 31);

  }

#endif
@@ -3185,7 +3190,7 @@ simde_mm256_maskload_ps (const simde_float32 mem_addr[HEDLEY_ARRAY_PARAM(4)], si 
#else

  SIMDE__VECTORIZE

  for (size_t i = 0 ; i < (sizeof(r.f32) / sizeof(r.f32[0])) ; i++) {

    r.i32[i] = ((int32_t*) mem_addr)[i] & (mask.i32[i] >> 31);

    r.i32[i] = HEDLEY_REINTERPRET_CAST(int32_t const*, mem_addr)[i] & (mask.i32[i] >> 31);

  }

#endif
@@ -3293,7 +3298,7 @@ simde_mm256_min_pd (simde__m256d a, simde__m256d b) { 


#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_min_pd(a.n,b.n);

#elif defined(SIMDE_SSE_NATIVE)

#elif defined(SIMDE_SSE2_NATIVE)

  r.m128d[0].n = _mm_min_pd(a.m128d[0].n, b.m128d[0].n);

  r.m128d[1].n = _mm_min_pd(a.m128d[1].n, b.m128d[1].n);

#else
@@ -3339,7 +3344,7 @@ simde_mm256_max_pd (simde__m256d a, simde__m256d b) { 


#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_max_pd(a.n,b.n);

#elif defined(SIMDE_SSE_NATIVE)

#elif defined(SIMDE_SSE2_NATIVE)

  r.m128d[0].n = _mm_max_pd(a.m128d[0].n, b.m128d[0].n);

  r.m128d[1].n = _mm_max_pd(a.m128d[1].n, b.m128d[1].n);

#else
@@ -3493,7 +3498,7 @@ simde_mm256_mul_pd (simde__m256d a, simde__m256d b) { 


#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_mul_pd(a.n,b.n);

#elif defined(SIMDE_SSE_NATIVE)

#elif defined(SIMDE_SSE2_NATIVE)

  r.m128d[0].n = _mm_mul_pd(a.m128d[0].n, b.m128d[0].n);

  r.m128d[1].n = _mm_mul_pd(a.m128d[1].n, b.m128d[1].n);

#elif defined(SIMDE__ENABLE_GCC_VEC_EXT)
@@ -3543,7 +3548,7 @@ simde_mm256_or_pd (simde__m256d a, simde__m256d b) { 


#if defined(SIMDE_AVX_NATIVE)

  r.n = _mm256_or_pd(a.n,b.n);

#elif defined(SIMDE_SSE_NATIVE)

#elif defined(SIMDE_SSE2_NATIVE)

  r.m128d[0].n = _mm_or_pd(a.m128d[0].n, b.m128d[0].n);

  r.m128d[1].n = _mm_or_pd(a.m128d[1].n, b.m128d[1].n);

#elif defined(SIMDE__ENABLE_GCC_VEC_EXT)
@@ -4688,7 +4693,7 @@ simde_mm_testc_ps (simde__m128 a, simde__m128 b) { 
    r |= ~a.u32[i] & b.u32[i];

  }



  return (int) ((~r >> 31) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 31) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4708,7 +4713,7 @@ simde_mm_testc_pd (simde__m128d a, simde__m128d b) { 
    r |= ~a.u64[i] & b.u64[i];

  }



  return (int) ((~r >> 63) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 63) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4728,7 +4733,7 @@ simde_mm256_testc_ps (simde__m256 a, simde__m256 b) { 
    r |= ~a.u32[i] & b.u32[i];

  }



  return (int) ((~r >> 31) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 31) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4748,7 +4753,7 @@ simde_mm256_testc_pd (simde__m256d a, simde__m256d b) { 
    r |= ~a.u64[i] & b.u64[i];

  }



  return (int) ((~r >> 63) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 63) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4768,7 +4773,7 @@ simde_mm256_testc_si256 (simde__m256i a, simde__m256i b) { 
    r |= ~a.i32f[i] & b.i32f[i];

  }



  return (int) !r;

  return HEDLEY_STATIC_CAST(int, !r);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4788,7 +4793,7 @@ simde_mm_testz_ps (simde__m128 a, simde__m128 b) { 
    r |= a.u32[i] & b.u32[i];

  }



  return (int) ((~r >> 31) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 31) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4808,7 +4813,7 @@ simde_mm_testz_pd (simde__m128d a, simde__m128d b) { 
    r |= a.u64[i] & b.u64[i];

  }



  return (int) ((~r >> 63) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 63) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4828,7 +4833,7 @@ simde_mm256_testz_ps (simde__m256 a, simde__m256 b) { 
    r |= a.u32[i] & b.u32[i];

  }



  return (int) ((~r >> 31) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 31) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4848,7 +4853,7 @@ simde_mm256_testz_pd (simde__m256d a, simde__m256d b) { 
    r |= a.u64[i] & b.u64[i];

  }



  return (int) ((~r >> 63) & 1);

  return HEDLEY_STATIC_CAST(int, ((~r >> 63) & 1));

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)
@@ -4868,7 +4873,7 @@ simde_mm256_testz_si256 (simde__m256i a, simde__m256i b) { 
    r |= a.i32f[i] & b.i32f[i];

  }



  return (int) !r;

  return HEDLEY_STATIC_CAST(int, !r);

#endif

}

#if defined(SIMDE_AVX_ENABLE_NATIVE_ALIASES)

debian/include/simde/x86/avx2.h

0 → 100644
+596 −0

File added.

Preview size limit exceeded, changes collapsed.