Introduce field "alignment" describing the necessary alignment for vector types

11 files changed, 39 insertions, 6 deletions

diff --git a/vec_double_avx.h b/vec_double_avx.h
index cb6345a..cfee0fc 100644
--- a/vec_double_avx.h
+++ b/vec_double_avx.h
@@ -29,6 +29,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef bool scalar_t;
     typedef __m256d bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -126,6 +127,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef int_t scalar_t;
     typedef __m256i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -349,6 +351,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef real_t scalar_t;
     typedef __m256d vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<AVX:4*double>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_double_sse2.h b/vec_double_sse2.h
index 7387bff..97f4e91 100644
--- a/vec_double_sse2.h
+++ b/vec_double_sse2.h
@@ -41,6 +41,7 @@ namespace vecmathlib {
     static int const size = 2;
     typedef bool scalar_t;
     typedef __m128d bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -131,6 +132,7 @@ namespace vecmathlib {
     static int const size = 2;
     typedef int_t scalar_t;
     typedef __m128i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -305,6 +307,7 @@ namespace vecmathlib {
     static int const size = 2;
     typedef real_t scalar_t;
     typedef __m128d vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<SSE2:2*double>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_double_sse2_scalar.h b/vec_double_sse2_scalar.h
index ee7dd02..d8e7955 100644
--- a/vec_double_sse2_scalar.h
+++ b/vec_double_sse2_scalar.h
@@ -41,6 +41,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef bool scalar_t;
     typedef uint_t bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -112,6 +113,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef int_t scalar_t;
     typedef int_t ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -217,6 +219,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef real_t scalar_t;
     typedef double vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<SSE2:1*double>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_float_altivec.h b/vec_float_altivec.h
index c4f48f7..a1fca6c 100644
--- a/vec_float_altivec.h
+++ b/vec_float_altivec.h
@@ -32,6 +32,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef bool scalar_t;
     typedef __vector __bool int bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -111,6 +112,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef int_t scalar_t;
     typedef __vector int ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -218,6 +220,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef real_t scalar_t;
     typedef __vector float vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<Altivec:4*float>"; }
     void barrier() { __asm__("": "+v" (v)); }
diff --git a/vec_float_avx.h b/vec_float_avx.h
index aed972a..8d9cf82 100644
--- a/vec_float_avx.h
+++ b/vec_float_avx.h
@@ -29,6 +29,7 @@ namespace vecmathlib {
     static int const size = 8;
     typedef bool scalar_t;
     typedef __m256 bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -136,6 +137,7 @@ namespace vecmathlib {
     static int const size = 8;
     typedef int_t scalar_t;
     typedef __m256i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -321,6 +323,7 @@ namespace vecmathlib {
     static int const size = 8;
     typedef real_t scalar_t;
     typedef __m256 vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<AVX:8*float>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_float_sse2.h b/vec_float_sse2.h
index 8902305..f8d7156 100644
--- a/vec_float_sse2.h
+++ b/vec_float_sse2.h
@@ -41,6 +41,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef bool scalar_t;
     typedef __m128 bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -136,6 +137,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef int_t scalar_t;
     typedef __m128i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -275,6 +277,7 @@ namespace vecmathlib {
     static int const size = 4;
     typedef real_t scalar_t;
     typedef __m128 vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<SSE2:4*float>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_float_sse2_scalar.h b/vec_float_sse2_scalar.h
index f8786c6..0aba3b2 100644
--- a/vec_float_sse2_scalar.h
+++ b/vec_float_sse2_scalar.h
@@ -41,6 +41,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef bool scalar_t;
     typedef uint_t bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -112,6 +113,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef int_t scalar_t;
     typedef int_t ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -217,6 +219,7 @@ namespace vecmathlib {
     static int const size = 1;
     typedef real_t scalar_t;
     typedef float vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<SSE2:1*float>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_fp16_avx.h b/vec_fp16_avx.h
index bd5e494..d0139bb 100644
--- a/vec_fp16_avx.h
+++ b/vec_fp16_avx.h
@@ -29,6 +29,7 @@ namespace vecmathlib {
     static int const size = 16;
     typedef bool scalar_t;
     typedef __m256i bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -143,6 +144,7 @@ namespace vecmathlib {
     static int const size = 16;
     typedef int_t scalar_t;
     typedef __m256i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -347,6 +349,7 @@ namespace vecmathlib {
     static int const size = 16;
     typedef real_t scalar_t;
     typedef __m256i vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<AVX:16*fp16>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_fp8_avx.h b/vec_fp8_avx.h
index 1068f64..2660269 100644
--- a/vec_fp8_avx.h
+++ b/vec_fp8_avx.h
@@ -29,6 +29,7 @@ namespace vecmathlib {
     static int const size = 32;
     typedef bool scalar_t;
     typedef __m256i bvector_t;
+    static int const alignment = sizeof(bvector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                   "vector size is wrong");
@@ -159,6 +160,7 @@ namespace vecmathlib {
     static int const size = 32;
     typedef int_t scalar_t;
     typedef __m256i ivector_t;
+    static int const alignment = sizeof(ivector_t);
     
     static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                   "vector size is wrong");
@@ -397,6 +399,7 @@ namespace vecmathlib {
     static int const size = 32;
     typedef real_t scalar_t;
     typedef __m256i vector_t;
+    static int const alignment = sizeof(vector_t);
     
     static char const* name() { return "<AVX:32*fp8>"; }
     void barrier() { __asm__("": "+x" (v)); }
diff --git a/vec_pseudo.h b/vec_pseudo.h
index 95894fd..a319502 100644
--- a/vec_pseudo.h
+++ b/vec_pseudo.h
@@ -32,6 +32,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef bool scalar_t;
     typedef uint_t bvector_t[size];
+    static int const alignment = sizeof(bvector_t);
     
     typedef boolpseudovec boolvec_t;
     typedef intpseudovec<real_t, size> intvec_t;
@@ -49,7 +50,7 @@ namespace vecmathlib {
     
     
     
-    bvector_t v __attribute__((__aligned__(sizeof(bvector_t))));
+    bvector_t v __attribute__((__aligned__(alignment)));
     
     boolpseudovec() {}
     // Can't have a non-trivial copy constructor; if so, objects won't
@@ -133,6 +134,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef int_t scalar_t;
     typedef int_t ivector_t[size];
+    static int const alignment = sizeof(ivector_t);
     
     typedef boolpseudovec<real_t, size> boolvec_t;
     typedef intpseudovec intvec_t;
@@ -150,7 +152,7 @@ namespace vecmathlib {
     
     
     
-    ivector_t v __attribute__((__aligned__(sizeof(ivector_t))));
+    ivector_t v __attribute__((__aligned__(alignment)));
     
     intpseudovec() {}
     // Can't have a non-trivial copy constructor; if so, objects won't
@@ -403,6 +405,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef real_t scalar_t;
     typedef real_t vector_t[size];
+    static int const alignment = sizeof(vector_t);
     
     static char const* name()
     {
@@ -471,7 +474,7 @@ namespace vecmathlib {
     
     
     
-    vector_t v __attribute__((__aligned__(sizeof(vector_t))));
+    vector_t v __attribute__((__aligned__(alignment)));
     
     realpseudovec() {}
     // Can't have a non-trivial copy constructor; if so, objects won't
diff --git a/vec_test.h b/vec_test.h
index 694217d..4a10824 100644
--- a/vec_test.h
+++ b/vec_test.h
@@ -29,6 +29,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef bool scalar_t;
     typedef bool bvector_t[size];
+    static int const alignment = sizeof(bvector_t);
     
     typedef booltestvec boolvec_t;
     typedef inttestvec<real_t, size> intvec_t;
@@ -46,7 +47,7 @@ namespace vecmathlib {
     
     
     
-    bvector_t v __attribute__((__aligned__(sizeof(bvector_t))));
+    bvector_t v __attribute__((__aligned__(alignment)));
     
     booltestvec() {}
     // can't have a non-trivial copy constructor; if so, objects won't
@@ -131,6 +132,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef int_t scalar_t;
     typedef int_t ivector_t[size];
+    static int const alignment = sizeof(ivector_t);
     
     typedef booltestvec<real_t, size> boolvec_t;
     typedef inttestvec intvec_t;
@@ -148,7 +150,7 @@ namespace vecmathlib {
     
     
     
-    ivector_t v __attribute__((__aligned__(sizeof(ivector_t))));
+    ivector_t v __attribute__((__aligned__(alignment)));
     
     inttestvec() {}
     // can't have a non-trivial copy constructor; if so, objects won't
@@ -373,6 +375,7 @@ namespace vecmathlib {
     static int const size = N;
     typedef real_t scalar_t;
     typedef real_t vector_t[size];
+    static int const alignment = sizeof(vector_t);
     
     static char const* name()
     {
@@ -414,7 +417,7 @@ namespace vecmathlib {
     
     
     
-    vector_t v __attribute__((__aligned__(sizeof(vector_t))));
+    vector_t v __attribute__((__aligned__(alignment)));
     
     realtestvec() {}
     // can't have a non-trivial copy constructor; if so, objects won't