vppinfra: autogerate vector typedefs and basic inline functions

Change-Id: Ie9f611fa6a962b0937245f5cc949571ba11c5604
Signed-off-by: Damjan Marion <damarion@cisco.com>

5 files changed, 83 insertions, 379 deletions

diff --git a/src/vppinfra/vector.h b/src/vppinfra/vector.h
index 13d7ba27031..3fdffc469bd 100644
--- a/src/vppinfra/vector.h
+++ b/src/vppinfra/vector.h
@@ -73,63 +73,50 @@
 
 #define _vector_size(n) __attribute__ ((vector_size (n)))
 
-/* Signed 64 bit. */
-typedef char i8x8 _vector_size (8);
-typedef short i16x4 _vector_size (8);
-typedef int i32x2 _vector_size (8);
-
-/* Unsigned 64 bit. */
-typedef unsigned char u8x8 _vector_size (8);
-typedef unsigned short u16x4 _vector_size (8);
-typedef unsigned int u32x2 _vector_size (8);
-
-/* Floating point 64 bit. */
-typedef float f32x2 _vector_size (8);
-
-/* Signed 128 bit. */
-typedef i8 i8x16 _vector_size (16);
-typedef i16 i16x8 _vector_size (16);
-typedef i32 i32x4 _vector_size (16);
-typedef long long i64x2 _vector_size (16);
-
-/* Unsigned 128 bit. */
-typedef u8 u8x16 _vector_size (16);
-typedef u16 u16x8 _vector_size (16);
-typedef u32 u32x4 _vector_size (16);
-typedef u64 u64x2 _vector_size (16);
-
-typedef f32 f32x4 _vector_size (16);
-typedef f64 f64x2 _vector_size (16);
-
-/* Signed 256 bit. */
-typedef i8 i8x32 _vector_size (32);
-typedef i16 i16x16 _vector_size (32);
-typedef i32 i32x8 _vector_size (32);
-typedef long long i64x4 _vector_size (32);
-
-/* Unsigned 256 bit. */
-typedef u8 u8x32 _vector_size (32);
-typedef u16 u16x16 _vector_size (32);
-typedef u32 u32x8 _vector_size (32);
-typedef u64 u64x4 _vector_size (32);
-
-typedef f32 f32x8 _vector_size (32);
-typedef f64 f64x4 _vector_size (32);
-
-/* Signed 512 bit. */
-typedef i8 i8x64 _vector_size (64);
-typedef i16 i16x32 _vector_size (64);
-typedef i32 i32x16 _vector_size (64);
-typedef long long i64x8 _vector_size (64);
-
-/* Unsigned 512 bit. */
-typedef u8 u8x64 _vector_size (64);
-typedef u16 u16x32 _vector_size (64);
-typedef u32 u32x16 _vector_size (64);
-typedef u64 u64x8 _vector_size (64);
-
-typedef f32 f32x16 _vector_size (64);
-typedef f64 f64x8 _vector_size (64);
+#define foreach_vec64i  _(i,8,8)  _(i,16,4)  _(i,32,2)
+#define foreach_vec64u  _(u,8,8)  _(u,16,4)  _(u,32,2)
+#define foreach_vec64f  _(f,32,2)
+#define foreach_vec128i _(i,8,16) _(i,16,8)  _(i,32,4)  _(i,64,2)
+#define foreach_vec128u _(u,8,16) _(u,16,8)  _(u,32,4)  _(u,64,2)
+#define foreach_vec128f _(f,32,4) _(f,64,2)
+#define foreach_vec256i _(i,8,32) _(i,16,16) _(i,32,8)  _(i,64,4)
+#define foreach_vec256u _(u,8,32) _(u,16,16) _(u,32,8)  _(u,64,4)
+#define foreach_vec256f _(f,32,8) _(f,64,4)
+#define foreach_vec512i _(i,8,64) _(i,16,32) _(i,32,16) _(i,64,8)
+#define foreach_vec512u _(u,8,64) _(u,16,32) _(u,32,16) _(u,64,8)
+#define foreach_vec512f _(f,32,16) _(f,64,8)
+
+#if defined (CLIB_HAVE_VEC512)
+#define foreach_int_vec foreach_vec64i foreach_vec128i foreach_vec256i foreach_vec512i
+#define foreach_uint_vec foreach_vec64u foreach_vec128u foreach_vec256u foreach_vec512u
+#define foreach_float_vec foreach_vec64f foreach_vec128f foreach_vec256f foreach_vec512f
+#elif defined (CLIB_HAVE_VEC256)
+#define foreach_int_vec foreach_vec64i foreach_vec128i foreach_vec256i
+#define foreach_uint_vec foreach_vec64u foreach_vec128u foreach_vec256u
+#define foreach_float_vec foreach_vec64f foreach_vec128f foreach_vec256f
+#else
+#define foreach_int_vec foreach_vec64i foreach_vec128i
+#define foreach_uint_vec foreach_vec64u foreach_vec128u
+#define foreach_float_vec foreach_vec64f foreach_vec128f
+#endif
+
+#define foreach_vec foreach_int_vec foreach_uint_vec foreach_float_vec
+
+/* *INDENT-OFF* */
+
+/* Type Definitions */
+#define _(t,s,c) \
+typedef t##s t##s##x##c _vector_size (s/8*c);	\
+typedef union {	  \
+  t##s##x##c as_##t##s##x##c;	\
+  t##s as_##t##s[c];	  \
+} t##s##x##c##_union_t;
+
+  foreach_vec64i foreach_vec64u foreach_vec64f
+  foreach_vec128i foreach_vec128u foreach_vec128f
+  foreach_vec256i foreach_vec256u foreach_vec256f
+  foreach_vec512i foreach_vec512u foreach_vec512f
+#undef _
 
 /* Vector word sized types. */
 #ifndef CLIB_VECTOR_WORD_BITS
@@ -167,91 +154,47 @@ typedef u64 u64x _vector_size (8);
 #define VECTOR_WORD_TYPE(t) t##x
 #define VECTOR_WORD_TYPE_LEN(t) (sizeof (VECTOR_WORD_TYPE(t)) / sizeof (t))
 
-/* Union types. */
-#if (defined(CLIB_HAVE_VEC128) || defined(CLIB_HAVE_VEC64))
-
-#define _(t)					\
-  typedef union {				\
-    t##x as_##t##x;				\
-    t as_##t[VECTOR_WORD_TYPE_LEN (t)];	\
-  } t##x##_union_t;
-
-_(u8);
-_(u16);
-_(u32);
-_(u64);
-_(i8);
-_(i16);
-_(i32);
-_(i64);
-
+/* this series of macros generate _is_equal, _is_greater, _is_zero, _add
+   and _sub inline funcitons for each vector type */
+#define _(t, s, c) \
+  static_always_inline t##s##x##c			\
+t##s##x##c##_is_equal (t##s##x##c v1, t##s##x##c v2)	\
+{ return (v1 == v2); }					\
+							\
+static_always_inline t##s##x##c				\
+t##s##x##c##_is_greater (t##s##x##c v1, t##s##x##c v2)	\
+{ return (v1 > v2); }					\
+							\
+static_always_inline t##s##x##c				\
+t##s##x##c##_is_zero (t##s##x##c v1)			\
+{ t##s##x##c z = {0}; return (v1 == z); }		\
+							\
+static_always_inline t##s##x##c				\
+t##s##x##c##_add (t##s##x##c v1, t##s##x##c v2)		\
+{ return (v1 + v2); }					\
+							\
+static_always_inline t##s##x##c				\
+t##s##x##c##_sub (t##s##x##c v1, t##s##x##c v2)		\
+{ return (v1 - v2); }
+  foreach_vec
 #undef _
 
-#endif
-
-#ifdef CLIB_HAVE_VEC64
-
-#define _(t,n)					\
-  typedef union {				\
-    t##x##n as_##t##x##n;			\
-    t as_##t[n];				\
-  } t##x##n##_union_t;				\
-
-_(u8, 8);
-_(u16, 4);
-_(u32, 2);
-_(i8, 8);
-_(i16, 4);
-_(i32, 2);
-
+/* this macro generate _splat inline funcitons for each scalar vector type */
+#define _(t, s, c) \
+  static_always_inline t##s##x##c			\
+t##s##x##c##_splat (t##s x)				\
+{							\
+    t##s##x##c r;					\
+    int i;						\
+							\
+    for (i = 0; i < c; i++)				\
+      r[i] = x;						\
+							\
+    return r;						\
+}
+  foreach_int_vec foreach_uint_vec
 #undef _
 
-#endif
-
-#ifdef CLIB_HAVE_VEC128
-
-#define _(t,n)					\
-  typedef union {				\
-    t##x##n as_##t##x##n;			\
-    t as_##t[n];				\
-  } t##x##n##_union_t;				\
-
-_(u8, 16);
-_(u16, 8);
-_(u32, 4);
-_(u64, 2);
-_(i8, 16);
-_(i16, 8);
-_(i32, 4);
-_(i64, 2);
-_(f32, 4);
-_(f64, 2);
-
-#undef _
-
-#endif
-
-/* When we don't have vector types, still define e.g. u32x4_union_t but as an array. */
-#if !defined(CLIB_HAVE_VEC128) && !defined(CLIB_HAVE_VEC64)
-
-#define _(t,n)					\
-  typedef union {				\
-    t as_##t[n];				\
-  } t##x##n##_union_t;				\
-
-_(u8, 16);
-_(u16, 8);
-_(u32, 4);
-_(u64, 2);
-_(i8, 16);
-_(i16, 8);
-_(i32, 4);
-_(i64, 2);
-
-#undef _
-
-#endif
-
 #if defined (__SSE4_2__) && __GNUC__ >= 4
 #include <vppinfra/vector_sse42.h>
 #endif
@@ -272,8 +215,9 @@ _(i64, 2);
 #include <vppinfra/vector_funcs.h>
 #endif
 
-#endif /* included_clib_vector_h */
+/* *INDENT-ON* */
 
+#endif /* included_clib_vector_h */
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/src/vppinfra/vector_altivec.h b/src/vppinfra/vector_altivec.h
index 0e9de820ed8..c9f7facca30 100644
--- a/src/vppinfra/vector_altivec.h
+++ b/src/vppinfra/vector_altivec.h
@@ -38,26 +38,6 @@
 #ifndef included_vector_altivec_h
 #define included_vector_altivec_h
 
-/* Splats. */
-#define _(t,n,ti,fi,tr,fr)						\
-  always_inline t##x##n t##x##n##_splat (t v)				\
-  { return (t##x##n) __builtin_altivec_##fi ((ti) v); }			\
-									\
-  always_inline t##x##n t##x##n##_splat_word (t##x##n x, int word_index) \
-  { return (t##x##n) __builtin_altivec_##fr ((tr) x, word_index); }
-
-#define u16x8_splat(i) ((u16x8) __builtin_altivec_vspltish (i))
-#define i16x8_splat(i) ((i16x8) __builtin_altivec_vspltish (i))
-#define u32x4_splat(i) ((u32x4) __builtin_altivec_vspltisw (i))
-#define i32x4_splat(i) ((i32x4) __builtin_altivec_vspltisw (i))
-
-#define u16x8_splat_word(x,i) ((u16x8) __builtin_altivec_vsplth ((i16x8) (x), (i)))
-#define i16x8_splat_word(x,i) ((i16x8) __builtin_altivec_vsplth ((i16x8) (x), (i)))
-#define u32x4_splat_word(x,i) ((u32x4) __builtin_altivec_vspltw ((i32x4) (x), (i)))
-#define i32x4_splat_word(x,i) ((i32x4) __builtin_altivec_vspltw ((i32x4) (x), (i)))
-
-#undef _
-
 /* 128 bit shifts. */
 #define _(t,ti,lr,f)						\
   always_inline t t##_##lr (t x, t y)				\
@@ -149,13 +129,6 @@ _signed_binop (32, 4, is_equal, vcmpequw)
   return u16x8_is_equal (x, zero);
 }
 
-always_inline u32x4
-u32x4_is_zero (u32x4 x)
-{
-  u32x4 zero = { 0 };
-  return u32x4_is_equal (x, zero);
-}
-
 always_inline u32
 u32x4_zero_byte_mask (u32x4 x)
 {
diff --git a/src/vppinfra/vector_iwmmxt.h b/src/vppinfra/vector_iwmmxt.h
index 8e662045655..6fecb1ab71b 100644
--- a/src/vppinfra/vector_iwmmxt.h
+++ b/src/vppinfra/vector_iwmmxt.h
@@ -77,33 +77,6 @@ u32x2_interleave_lo (u32x2 a, u32x2 b)
   return __builtin_arm_wunpckilw (a, b);
 }
 
-always_inline u32x2
-u32x2_splat (u32 a)
-{
-  u32x2 x = { a };
-  x = u32x2_interleave_lo (x, x);
-  return x;
-}
-
-always_inline u16x4
-u16x4_splat (u16 a)
-{
-  u32 t = (u32) a | ((u32) a << 16);
-  return u32x2_splat (t);
-}
-
-always_inline u8x8
-u8x8_splat (u8 a)
-{
-  u32 t = (u32) a | ((u32) a << 8);
-  t |= t << 16;
-  return u32x2_splat (t);
-}
-
-#define i32x2_splat u32x2_splat
-#define i16x4_splat u16x4_splat
-#define i8x8_splat u8x8_splat
-
 /* 64 bit shifts. */
 
 /* As of July 2008 the __builtin_arm shifts cause gcc-4.3.1 to crash
diff --git a/src/vppinfra/vector_neon.h b/src/vppinfra/vector_neon.h
index 6357d68fabb..6a3e64e946d 100644
--- a/src/vppinfra/vector_neon.h
+++ b/src/vppinfra/vector_neon.h
@@ -17,53 +17,10 @@
 #define included_vector_neon_h
 #include <arm_neon.h>
 
-/* Splats. */
-
-#define u8x16_splat(i) vdupq_n_u8(i)
-#define u16x8_splat(i) vdupq_n_u16(i)
-#define i16x8_splat(i) vdupq_n_s16(i)
-#define u32x4_splat(i) vdupq_n_u32(i)
-#define i32x4_splat(i) vdupq_n_s32(i)
-
 /* Arithmetic */
-#define u16x8_add(a,b) vaddq_u16(a,b)
-#define i16x8_add(a,b) vaddq_s16(a,b)
 #define u16x8_sub_saturate(a,b) vsubq_u16(a,b)
 #define i16x8_sub_saturate(a,b) vsubq_s16(a,b)
 
-
-/* Compare operations. */
-#define u8x16_is_equal(a,b) vceqq_u8(a,b)
-#define i8x16_is_equal(a,b) vceqq_s8(a,b)
-#define u16x8_is_equal(a,b) vceqq_u16(a,b)
-#define i16x8_is_equal(a,b) vceqq_i16(a,b)
-#define u32x4_is_equal(a,b) vceqq_u32(a,b)
-#define i32x4_is_equal(a,b) vceqq_s32(a,b)
-#define i8x16_is_greater(a,b) vcgtq_s8(a,b)
-#define i16x8_is_greater(a,b) vcgtq_u8(a,b)
-#define i32x4_is_greater(a,b) vcgtq_s32(a,b)
-
-always_inline u8x16
-u8x16_is_zero (u8x16 x)
-{
-  u8x16 zero = { 0 };
-  return u8x16_is_equal (x, zero);
-}
-
-always_inline u16x8
-u16x8_is_zero (u16x8 x)
-{
-  u16x8 zero = { 0 };
-  return u16x8_is_equal (x, zero);
-}
-
-always_inline u32x4
-u32x4_is_zero (u32x4 x)
-{
-  u32x4 zero = { 0 };
-  return u32x4_is_equal (x, zero);
-}
-
 /* Converts all ones/zeros compare mask to bitmap. */
 always_inline u32
 u8x16_compare_byte_mask (u8x16 x)
diff --git a/src/vppinfra/vector_sse42.h b/src/vppinfra/vector_sse42.h
index ce1bb81c93f..dab22deff7c 100644
--- a/src/vppinfra/vector_sse42.h
+++ b/src/vppinfra/vector_sse42.h
@@ -171,64 +171,6 @@ i32x2_pack (i32x2 lo, i32x2 hi)
   return (i16x4) _m_packssdw ((__m64) lo, (__m64) hi);
 }
 
-/* Splats: replicate scalar value into vector. */
-always_inline u64x2
-u64x2_splat (u64 a)
-{
-  u64x2 x = { a, a };
-  return x;
-}
-
-always_inline u32x4
-u32x4_splat (u32 a)
-{
-  u32x4 x = { a, a, a, a };
-  return x;
-}
-
-always_inline u16x8
-u16x8_splat (u16 a)
-{
-  u16x8 x = { a, a, a, a, a, a, a, a };
-  return x;
-}
-
-always_inline u8x16
-u8x16_splat (u8 a)
-{
-  u8x16 x = { a, a, a, a, a, a, a, a, a, a, a, a, a, a, a, a };
-  return x;
-}
-
-always_inline u32x2
-u32x2_splat (u32 a)
-{
-  u32x2 x = { a, a };
-  return x;
-}
-
-always_inline u16x4
-u16x4_splat (u16 a)
-{
-  u16x4 x = { a, a, a, a };
-  return x;
-}
-
-always_inline u8x8
-u8x8_splat (u8 a)
-{
-  u8x8 x = { a, a, a, a, a, a, a, a };
-  return x;
-}
-
-#define i64x2_splat u64x2_splat
-#define i32x4_splat u32x4_splat
-#define i16x8_splat u16x8_splat
-#define i8x16_splat u8x16_splat
-#define i32x2_splat u32x2_splat
-#define i16x4_splat u16x4_splat
-#define i8x8_splat u8x8_splat
-
 #ifndef __ICC
 always_inline u64x2
 u64x2_read_lo (u64x2 x, u64 * a)
@@ -275,14 +217,6 @@ _(u8x16) _(u16x8) _(u32x4) _(u64x2) _(i8x16) _(i16x8) _(i32x4) _(i64x2)
   always_inline i##n##x##m                                             \
   i##n##x##m##_##f (i##n##x##m x, i##n##x##m y)                        \
   { return (i##n##x##m) _mm_##g##n ((__m128i) x, (__m128i) y); }
-/* Addition/subtraction. */
-  _signed_binop (8, 16, add, add_epi)
-_signed_binop (16, 8, add, add_epi)
-_signed_binop (32, 4, add, add_epi)
-_signed_binop (64, 2, add, add_epi)
-_signed_binop (8, 16, sub, sub_epi)
-_signed_binop (16, 8, sub, sub_epi)
-_signed_binop (32, 4, sub, sub_epi) _signed_binop (64, 2, sub, sub_epi)
 /* Addition/subtraction with saturation. */
   _signed_binop (8, 16, add_saturate, adds_epu)
 _signed_binop (16, 8, add_saturate, adds_epu)
@@ -469,83 +403,6 @@ _(u64, 2, right, left);
 #undef _
 #endif
 
-/* Compare operations. */
-always_inline u8x16
-u8x16_is_equal (u8x16 x, u8x16 y)
-{
-  return (u8x16) _mm_cmpeq_epi8 ((__m128i) x, (__m128i) y);
-}
-
-always_inline i8x16
-i8x16_is_equal (i8x16 x, i8x16 y)
-{
-  return (i8x16) _mm_cmpeq_epi8 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u16x8
-u16x8_is_equal (u16x8 x, u16x8 y)
-{
-  return (u16x8) _mm_cmpeq_epi16 ((__m128i) x, (__m128i) y);
-}
-
-always_inline i16x8
-i16x8_is_equal (i16x8 x, i16x8 y)
-{
-  return (i16x8) _mm_cmpeq_epi16 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u32x4
-u32x4_is_equal (u32x4 x, u32x4 y)
-{
-  return (u32x4) _mm_cmpeq_epi32 ((__m128i) x, (__m128i) y);
-}
-
-always_inline i32x4
-i32x4_is_equal (i32x4 x, i32x4 y)
-{
-  return (i32x4) _mm_cmpeq_epi32 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u8x16
-i8x16_is_greater (i8x16 x, i8x16 y)
-{
-  return (u8x16) _mm_cmpgt_epi8 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u16x8
-i16x8_is_greater (i16x8 x, i16x8 y)
-{
-  return (u16x8) _mm_cmpgt_epi16 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u32x4
-i32x4_is_greater (i32x4 x, i32x4 y)
-{
-  return (u32x4) _mm_cmpgt_epi32 ((__m128i) x, (__m128i) y);
-}
-
-always_inline u8x16
-u8x16_is_zero (u8x16 x)
-{
-  u8x16 zero = { 0 };
-  return u8x16_is_equal (x, zero);
-}
-
-always_inline u16x8
-u16x8_is_zero (u16x8 x)
-{
-  u16x8 zero = { 0 };
-  return u16x8_is_equal (x, zero);
-}
-
-always_inline u32x4
-u32x4_is_zero (u32x4 x)
-{
-  u32x4 zero = { 0 };
-  return u32x4_is_equal (x, zero);
-}
-
-
 always_inline int
 u8x16_is_all_zero (u8x16 x)
 {