Redesigned SIMD module and unit tests.

[alexxy/gromacs.git] / src / gromacs / gmxlib / nonbonded / nb_kernel_avx_256_single / kernelutil_x86_avx_256_single.h

diff --git a/src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/kernelutil_x86_avx_256_single.h b/src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/kernelutil_x86_avx_256_single.h
index 59c9e053b7139c8d171d3466a0eee9bf2d776226..d4f041b119e5755a7ccf02075c82f7d892735d9b 100644 (file)
--- a/src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/kernelutil_x86_avx_256_single.h
+++ b/src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/kernelutil_x86_avx_256_single.h
@@ -1,7 +1,7 @@
 /*
  * This file is part of the GROMACS molecular simulation package.
  *
- * Copyright (c) 2012,2013, by the GROMACS development team, led by
+ * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
  * and including many others, as listed in the AUTHORS file in the
  * top-level source directory and at http://www.gromacs.org.
@@ -35,7 +35,38 @@
 #ifndef _kernelutil_x86_avx_256_single_h_
 #define _kernelutil_x86_avx_256_single_h_
 
-#include "gromacs/simd/general_x86_avx_256.h"
+#define gmx_mm_castsi128_ps(a) _mm_castsi128_ps(a)
+
+static gmx_inline __m256
+gmx_mm256_unpack128lo_ps(__m256 xmm1, __m256 xmm2)
+{
+    return _mm256_permute2f128_ps(xmm1, xmm2, 0x20);
+}
+
+static gmx_inline __m256
+gmx_mm256_unpack128hi_ps(__m256 xmm1, __m256 xmm2)
+{
+    return _mm256_permute2f128_ps(xmm1, xmm2, 0x31);
+}
+
+static gmx_inline __m256
+gmx_mm256_set_m128(__m128 hi, __m128 lo)
+{
+    return _mm256_insertf128_ps(_mm256_castps128_ps256(lo), hi, 0x1);
+}
+
+/* Work around gcc bug with wrong type for mask formal parameter to maskload/maskstore */
+#ifdef GMX_SIMD_X86_AVX_GCC_MASKLOAD_BUG
+#    define gmx_mm_maskload_ps(mem, mask)       _mm_maskload_ps((mem), _mm_castsi128_ps(mask))
+#    define gmx_mm_maskstore_ps(mem, mask, x)    _mm_maskstore_ps((mem), _mm_castsi128_ps(mask), (x))
+#    define gmx_mm256_maskload_ps(mem, mask)    _mm256_maskload_ps((mem), _mm256_castsi256_ps(mask))
+#    define gmx_mm256_maskstore_ps(mem, mask, x) _mm256_maskstore_ps((mem), _mm256_castsi256_ps(mask), (x))
+#else
+#    define gmx_mm_maskload_ps(mem, mask)       _mm_maskload_ps((mem), (mask))
+#    define gmx_mm_maskstore_ps(mem, mask, x)    _mm_maskstore_ps((mem), (mask), (x))
+#    define gmx_mm256_maskload_ps(mem, mask)    _mm256_maskload_ps((mem), (mask))
+#    define gmx_mm256_maskstore_ps(mem, mask, x) _mm256_maskstore_ps((mem), (mask), (x))
+#endif
 
 /* Transpose lower/upper half of 256-bit registers separately */
 #define GMX_MM256_HALFTRANSPOSE4_PS(ymm0, ymm1, ymm2, ymm3) {            \