Another batch of added config.h

[alexxy/gromacs.git] / src / gromacs / simd / impl_x86_sse2 / impl_x86_sse2.h

diff --git a/src/gromacs/simd/impl_x86_sse2/impl_x86_sse2.h b/src/gromacs/simd/impl_x86_sse2/impl_x86_sse2.h
index 5481de1186eb67cff96840b724ff3f0b6d69eb09..c16d3d4d576793cf9f01115e4657026d210b18b0 100644 (file)
--- a/src/gromacs/simd/impl_x86_sse2/impl_x86_sse2.h
+++ b/src/gromacs/simd/impl_x86_sse2/impl_x86_sse2.h
@@ -39,6 +39,8 @@
 #include <math.h>
 #include <emmintrin.h>
 
+#include "config.h"
+
 /* Set capabilities that can be inherited */
 #define GMX_SIMD_X86_SSE2_OR_HIGHER
 
@@ -99,8 +101,8 @@
 #define gmx_simd_xor_f            _mm_xor_ps
 #define gmx_simd_rsqrt_f          _mm_rsqrt_ps
 #define gmx_simd_rcp_f            _mm_rcp_ps
-#define gmx_simd_fabs_f(x)        _mm_andnot_ps(_mm_set1_ps(-0.0), x)
-#define gmx_simd_fneg_f(x)        _mm_xor_ps(x, _mm_set1_ps(-0.0))
+#define gmx_simd_fabs_f(x)        _mm_andnot_ps(_mm_set1_ps(GMX_FLOAT_NEGZERO), x)
+#define gmx_simd_fneg_f(x)        _mm_xor_ps(x, _mm_set1_ps(GMX_FLOAT_NEGZERO))
 #define gmx_simd_max_f            _mm_max_ps
 #define gmx_simd_min_f            _mm_min_ps
 #define gmx_simd_round_f(x)       _mm_cvtepi32_ps(_mm_cvtps_epi32(x))
@@ -120,7 +122,7 @@
 #define gmx_simd_cvt_f2i          _mm_cvtps_epi32
 #define gmx_simd_cvtt_f2i         _mm_cvttps_epi32
 #define gmx_simd_cvt_i2f          _mm_cvtepi32_ps
-#define gmx_simd_extract_fi(x, i)  _mm_cvtsi128_si32(_mm_srli_si128((x), sizeof(int) * (i)))
+#define gmx_simd_extract_fi(x, i)  _mm_cvtsi128_si32(_mm_srli_si128((x), 4 * (i)))
 /* Integer logical ops on gmx_simd_fint32_t */
 #define gmx_simd_slli_fi          _mm_slli_epi32
 #define gmx_simd_srli_fi          _mm_srli_epi32
@@ -183,8 +185,8 @@
 #define gmx_simd_rsqrt_d(x)        _mm_cvtps_pd(_mm_rsqrt_ps(_mm_cvtpd_ps(x)))
 /* Don't use FMA for sqrt N-R iterations - this saves 1 instruction without FMA hardware */
 #define gmx_simd_rcp_d(x)          _mm_cvtps_pd(_mm_rcp_ps(_mm_cvtpd_ps(x)))
-#define gmx_simd_fabs_d(x)         _mm_andnot_pd(_mm_set1_pd(-0.0), x)
-#define gmx_simd_fneg_d(x)         _mm_xor_pd(x, _mm_set1_pd(-0.0))
+#define gmx_simd_fabs_d(x)         _mm_andnot_pd(_mm_set1_pd(GMX_DOUBLE_NEGZERO), x)
+#define gmx_simd_fneg_d(x)         _mm_xor_pd(x, _mm_set1_pd(GMX_DOUBLE_NEGZERO))
 #define gmx_simd_max_d             _mm_max_pd
 #define gmx_simd_min_d             _mm_min_pd
 #define gmx_simd_round_d(x)        _mm_cvtepi32_pd(_mm_cvtpd_epi32(x))
@@ -204,7 +206,7 @@
 #define gmx_simd_cvt_d2i           _mm_cvtpd_epi32
 #define gmx_simd_cvtt_d2i          _mm_cvttpd_epi32
 #define gmx_simd_cvt_i2d           _mm_cvtepi32_pd
-#define gmx_simd_extract_di(x, i)   _mm_cvtsi128_si32(_mm_srli_si128((x), sizeof(int) * (i)))
+#define gmx_simd_extract_di(x, i)   _mm_cvtsi128_si32(_mm_srli_si128((x), 4 * (i)))
 /* Integer logical ops on gmx_simd_dint32_t */
 #define gmx_simd_slli_di           _mm_slli_epi32
 #define gmx_simd_srli_di           _mm_srli_epi32
@@ -249,7 +251,7 @@
 /****************************************************
  * SINGLE PRECISION IMPLEMENTATION HELPER FUNCTIONS *
  ****************************************************/
-static gmx_inline __m128
+static gmx_inline __m128 gmx_simdcall
 gmx_simd_get_exponent_f_sse2(__m128 x)
 {
     const __m128  expmask      = _mm_castsi128_ps(_mm_set1_epi32(0x7F800000));
@@ -261,7 +263,7 @@ gmx_simd_get_exponent_f_sse2(__m128 x)
     return _mm_cvtepi32_ps(iexp);
 }
 
-static gmx_inline __m128
+static gmx_inline __m128 gmx_simdcall
 gmx_simd_get_mantissa_f_sse2(__m128 x)
 {
     const __m128 mantmask = _mm_castsi128_ps(_mm_set1_epi32(0x007FFFFF));
@@ -271,7 +273,7 @@ gmx_simd_get_mantissa_f_sse2(__m128 x)
     return _mm_or_ps(x, one);
 }
 
-static gmx_inline __m128
+static gmx_inline __m128 gmx_simdcall
 gmx_simd_set_exponent_f_sse2(__m128 x)
 {
     const __m128i expbias      = _mm_set1_epi32(127);
@@ -281,7 +283,7 @@ gmx_simd_set_exponent_f_sse2(__m128 x)
     return _mm_castsi128_ps(iexp);
 }
 
-static gmx_inline __m128i
+static gmx_inline __m128i gmx_simdcall
 gmx_simd_mul_fi_sse2(__m128i a, __m128i b)
 {
     __m128i a1 = _mm_srli_si128(a, 4); /* - a[3] a[2] a[1] */
@@ -295,7 +297,7 @@ gmx_simd_mul_fi_sse2(__m128i a, __m128i b)
     return _mm_unpacklo_epi32(c, c1);
 }
 
-static gmx_inline float
+static gmx_inline float gmx_simdcall
 gmx_simd_reduce_f_sse2(__m128 a)
 {
     __m128 b;
@@ -309,7 +311,7 @@ gmx_simd_reduce_f_sse2(__m128 a)
 /****************************************************
  * DOUBLE PRECISION IMPLEMENTATION HELPER FUNCTIONS *
  ****************************************************/
-static gmx_inline __m128d
+static gmx_inline __m128d gmx_simdcall
 gmx_simd_get_exponent_d_sse2(__m128d x)
 {
     /* Don't use _mm_set1_epi64x() - on MSVC it is only supported for 64-bit builds */
@@ -323,7 +325,7 @@ gmx_simd_get_exponent_d_sse2(__m128d x)
     return _mm_cvtepi32_pd(iexp);
 }
 
-static gmx_inline __m128d
+static gmx_inline __m128d gmx_simdcall
 gmx_simd_get_mantissa_d_sse2(__m128d x)
 {
     /* Don't use _mm_set1_epi64x() - on MSVC it is only supported for 64-bit builds */
@@ -334,7 +336,7 @@ gmx_simd_get_mantissa_d_sse2(__m128d x)
     return _mm_or_pd(x, one);
 }
 
-static gmx_inline __m128d
+static gmx_inline __m128d gmx_simdcall
 gmx_simd_set_exponent_d_sse2(__m128d x)
 {
     const __m128i  expbias      = _mm_set1_epi32(1023);
@@ -347,7 +349,7 @@ gmx_simd_set_exponent_d_sse2(__m128d x)
     return _mm_castsi128_pd(iexp);
 }
 
-static gmx_inline __m128i
+static gmx_inline __m128i gmx_simdcall
 gmx_simd_mul_di_sse2(__m128i a, __m128i b)
 {
     __m128i c;
@@ -359,7 +361,7 @@ gmx_simd_mul_di_sse2(__m128i a, __m128i b)
     return _mm_shuffle_epi32(c, _MM_SHUFFLE(3, 1, 2, 0)); /* 0 0 a[1]*b[1] a[0]*b[0] */
 }
 
-static gmx_inline double
+static gmx_inline double gmx_simdcall
 gmx_simd_reduce_d_sse2(__m128d a)
 {
     __m128d b;
@@ -436,7 +438,7 @@ gmx_simd_check_and_reset_overflow(void)
 #define gmx_simd4_reduce_f               gmx_simd_reduce_f
 
 /* SIMD4 Dotproduct helper function */
-static gmx_inline float
+static gmx_inline float gmx_simdcall
 gmx_simd4_dotproduct3_f_sse2(__m128 a, __m128 b)
 {
     float  f;