add single precision, add aa-vec-sl-soa kernel, updated doc

[LbmBenchmarkKernelsPublic.git] / src / Vector.h

diff --git a/src/Vector.h b/src/Vector.h
index 41b9a7984f6e31c704f1089797bcb1c616e5b1a1..af12f77868f6da4d3796cdf1f4e8654e3fcfdb3b 100644 (file)
--- a/src/Vector.h
+++ b/src/Vector.h
@@ -36,48 +36,104 @@
        #error Only VECTOR_AVX or VECTOR_SSE can be defined at the same time.
 #endif
 
-#ifdef VECTOR_AVX
+#if !defined(PRECISION_DP) && !defined(PRECISION_SP)
+       #error PRECISION_DP or PRECISION_SP must be defined.
+#endif
 
-       #include <immintrin.h>
-       // Vector size in double-precision floatin-point numbers.
-       #define VSIZE   4
+#if defined(PRECISION_DP) && defined(PRECISION_SP)
+       #error Only PRECISION_DP or PRECISION_SP can be defined at the same time.
+#endif
 
-       #define VPDFT                           __m256d
+#ifdef PRECISION_DP
 
-       #define VSET(scalar)            _mm256_set1_pd(scalar)
+       #ifdef VECTOR_AVX
 
-       #define VLD(expr)                       _mm256_load_pd(expr)
-       #define VLDU(expr)                      _mm256_loadu_pd(expr)
+               #include <immintrin.h>
+               // Vector size in double-precision floating-point numbers.
+               #define VSIZE   4
 
-       #define VST(dst, src)           _mm256_store_pd(dst, src)
-       #define VSTU(dst, src)          _mm256_storeu_pd(dst, src)
-       #define VSTNT(dst, src)         _mm256_stream_pd(dst, src)
+               #define VPDFT                           __m256d
 
-       #define VMUL(a, b)                      _mm256_mul_pd(a, b)
-       #define VADD(a, b)                      _mm256_add_pd(a, b)
-       #define VSUB(a, b)                      _mm256_sub_pd(a, b)
-#endif
+               #define VSET(scalar)            _mm256_set1_pd(scalar)
 
-#ifdef VECTOR_SSE
-       #include <emmintrin.h>
-       // Vector size in double-precision floatin-point numbers.
-       #define VSIZE 2
+               #define VLD(expr)                       _mm256_load_pd(expr)
+               #define VLDU(expr)                      _mm256_loadu_pd(expr)
 
-       #define VPDFT                           __m128d
+               #define VST(dst, src)           _mm256_store_pd(dst, src)
+               #define VSTU(dst, src)          _mm256_storeu_pd(dst, src)
+               #define VSTNT(dst, src)         _mm256_stream_pd(dst, src)
 
-       #define VSET(scalar)            _mm_set1_pd(scalar)
+               #define VMUL(a, b)                      _mm256_mul_pd(a, b)
+               #define VADD(a, b)                      _mm256_add_pd(a, b)
+               #define VSUB(a, b)                      _mm256_sub_pd(a, b)
+       #endif
 
-       #define VLD(expr)                       _mm_load_pd(expr)
-       #define VLDU(expr)                      _mm_loadu_pd(expr)
+       #ifdef VECTOR_SSE
+               #include <emmintrin.h>
+               // Vector size in double-precision floating-point numbers.
+               #define VSIZE 2
 
-       #define VST(dst, src)           _mm_store_pd(dst, src)
-       #define VSTU(dst, src)          _mm_storeu_pd(dst, src)
-       #define VSTNT(dst, src)         _mm_stream_pd(dst, src)
+               #define VPDFT                           __m128d
 
-       #define VMUL(a, b)                      _mm_mul_pd(a, b)
-       #define VADD(a, b)                      _mm_add_pd(a, b)
-       #define VSUB(a, b)                      _mm_sub_pd(a, b)
-#endif
+               #define VSET(scalar)            _mm_set1_pd(scalar)
+
+               #define VLD(expr)                       _mm_load_pd(expr)
+               #define VLDU(expr)                      _mm_loadu_pd(expr)
+
+               #define VST(dst, src)           _mm_store_pd(dst, src)
+               #define VSTU(dst, src)          _mm_storeu_pd(dst, src)
+               #define VSTNT(dst, src)         _mm_stream_pd(dst, src)
+
+               #define VMUL(a, b)                      _mm_mul_pd(a, b)
+               #define VADD(a, b)                      _mm_add_pd(a, b)
+               #define VSUB(a, b)                      _mm_sub_pd(a, b)
+       #endif
+
+#elif defined(PRECISION_SP)
+
+       #ifdef VECTOR_AVX
+
+               #include <immintrin.h>
+               // Vector size in double-precision floating-point numbers.
+               #define VSIZE   8
+
+               #define VPDFT                           __m256
+
+               #define VSET(scalar)            _mm256_set1_ps(scalar)
+
+               #define VLD(expr)                       _mm256_load_ps(expr)
+               #define VLDU(expr)                      _mm256_loadu_ps(expr)
+
+               #define VST(dst, src)           _mm256_store_ps(dst, src)
+               #define VSTU(dst, src)          _mm256_storeu_ps(dst, src)
+               #define VSTNT(dst, src)         _mm256_stream_ps(dst, src)
+
+               #define VMUL(a, b)                      _mm256_mul_ps(a, b)
+               #define VADD(a, b)                      _mm256_add_ps(a, b)
+               #define VSUB(a, b)                      _mm256_sub_ps(a, b)
+       #endif
+
+       #ifdef VECTOR_SSE
+               #include <emmintrin.h>
+               // Vector size in double-precision floating-point numbers.
+               #define VSIZE   4
+
+               #define VPDFT                           __m128
+
+               #define VSET(scalar)            _mm_set1_ps(scalar)
+
+               #define VLD(expr)                       _mm_load_ps(expr)
+               #define VLDU(expr)                      _mm_loadu_ps(expr)
+
+               #define VST(dst, src)           _mm_store_ps(dst, src)
+               #define VSTU(dst, src)          _mm_storeu_ps(dst, src)
+               #define VSTNT(dst, src)         _mm_stream_ps(dst, src)
+
+               #define VMUL(a, b)                      _mm_mul_ps(a, b)
+               #define VADD(a, b)                      _mm_add_ps(a, b)
+               #define VSUB(a, b)                      _mm_sub_ps(a, b)
+       #endif
 
+#endif  // PRECISION
 
 #endif // __VECTOR_H__