mirror of
https://github.com/VectorCamp/vectorscan.git
synced 2025-09-29 19:24:25 +03:00
WIP: Power VSX support almost completed
This commit is contained in:
Apostolos Tapsas
committed by
apostolos
apostolos
parent
2231f7c024
commit
7888dd4418
@@ -39,8 +39,24 @@
|
||||
#include "util/supervector/supervector.hpp"
|
||||
#include <iostream>
|
||||
|
||||
|
||||
typedef __vector uint64_t uint64x2_t;
|
||||
typedef __vector int64_t int64x2_t;
|
||||
typedef __vector uint32_t uint32x4_t;
|
||||
typedef __vector int32_t int32x4_t;
|
||||
typedef __vector uint16_t uint16x8_t;
|
||||
typedef __vector int16_t int16x8_t;
|
||||
typedef __vector uint8_t uint8x16_t;
|
||||
typedef __vector int8_t int8x16_t;
|
||||
|
||||
// 128-bit Powerpc64le implementation
|
||||
|
||||
union Tmp
|
||||
{
|
||||
uint32_t u32;
|
||||
uint16_t u16[2];
|
||||
};
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector(SuperVector const &other)
|
||||
{
|
||||
@@ -57,87 +73,69 @@ template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<int8_t>(int8_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi8(other);
|
||||
//u.v128[0] = vec_splat_s8(other);
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(other);
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<uint8_t>(uint8_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi8(static_cast<int8_t>(other));
|
||||
//u.v128[0] = vec_splat_s8(static_cast<int8_t>(other));
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(static_cast<int8_t>(other));
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<int16_t>(int16_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi16(other);
|
||||
//u.v128[0] = vec_splat_s16(other);
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(other);
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<uint16_t>(uint16_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi16(static_cast<int16_t>(other));
|
||||
//u.v128[0] = vec_splat_s16(static_cast<int8_t>(other));
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(static_cast<int8_t>(other));
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<int32_t>(int32_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi32(other);
|
||||
//u.v128[0] = vec_splat_s32(other);
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(other);
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<uint32_t>(uint32_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi32(static_cast<int32_t>(other));
|
||||
//u.v128[0] = vec_splat_s32(static_cast<int8_t>(other));
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(static_cast<int8_t>(other));
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<int64_t>(int64_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi64x(other);
|
||||
//u.v128[0] = vec_splat_u64(other);
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(other);
|
||||
}
|
||||
|
||||
template<>
|
||||
template<>
|
||||
really_inline SuperVector<16>::SuperVector<uint64_t>(uint64_t const other)
|
||||
{
|
||||
//u.v128[0] = _mm_set1_epi64x(static_cast<int64_t>(other));
|
||||
//u.v128[0] = vec_splat_u32(static_cast<int8_t>(other));
|
||||
std::cout<<other<<std::endl;
|
||||
u.v128[0] = (m128) vec_splats(static_cast<int8_t>(other));
|
||||
}
|
||||
|
||||
// Constants
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::Ones(void)
|
||||
{
|
||||
//return {_mm_set1_epi8(0xFF)};
|
||||
return {(m128) vec_splat_s8(1)};
|
||||
return {(m128) vec_splat_s8(-1)};
|
||||
}
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::Zeroes(void)
|
||||
{
|
||||
//return {_mm_set1_epi8(0)};
|
||||
return {(m128) vec_splat_s8(0)};
|
||||
return {(m128) vec_splat_s8(0)};
|
||||
}
|
||||
|
||||
// Methods
|
||||
@@ -151,28 +149,24 @@ really_inline void SuperVector<16>::operator=(SuperVector<16> const &other)
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator&(SuperVector<16> const &b) const
|
||||
{
|
||||
//return {_mm_and_si128(u.v128[0], b.u.v128[0])};
|
||||
return {vec_and(u.v128[0], b.u.v128[0])};
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator|(SuperVector<16> const &b) const
|
||||
{
|
||||
//return {_mm_or_si128(u.v128[0], b.u.v128[0])};
|
||||
return {vec_or(u.v128[0], b.u.v128[0])};
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator^(SuperVector<16> const &b) const
|
||||
{
|
||||
//return {_mm_xor_si128(u.v128[0], b.u.v128[0])};
|
||||
return {vec_xor(u.v128[0], b.u.v128[0])};
|
||||
return {(m128) vec_xor(u.v128[0], b.u.v128[0])};
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::opandnot(SuperVector<16> const &b) const
|
||||
{
|
||||
//return {_mm_andnot_si128(u.v128[0], b.u.v128[0])};
|
||||
m128 and_res = vec_and(u.v128[0], b.u.v128[0]);
|
||||
return vec_xor(and_res,and_res);
|
||||
}
|
||||
@@ -180,215 +174,156 @@ really_inline SuperVector<16> SuperVector<16>::opandnot(SuperVector<16> const &b
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::eq(SuperVector<16> const &b) const
|
||||
{
|
||||
//return {_mm_cmpeq_epi8(u.v128[0], b.u.v128[0])};
|
||||
return { vec_all_eq(u.v128[0], b.u.v128[0])};
|
||||
return {(m128) vec_cmpeq((int8x16_t)u.v128[0], (int8x16_t)b.u.v128[0])};
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline typename SuperVector<16>::movemask_type SuperVector<16>::movemask(void)const
|
||||
{
|
||||
//return _mm_movemask_epi8(u.v128[0]);
|
||||
// Compute the mask from the input
|
||||
//uint64x2_t mask = vpaddlq_u32(vpaddlq_u16(vpaddlq_u8(vandq_u8((uint8x16_t)u.v128[0], 0))));
|
||||
//uint64x2_t mask1 = (m128)vextq_s8(mask, Zeroes(), 7);
|
||||
//mask = vorrq_u8(mask, mask1);
|
||||
{
|
||||
uint8x16_t s1 = vec_sr((uint8x16_t)u.v128[0], vec_splat_u8(7));
|
||||
uint16x8_t ss = vec_sr((uint16x8_t)s1, vec_splat_u16(7));
|
||||
uint16x8_t res_and = vec_and((uint16x8_t)s1, vec_splats((uint16_t)0xff));
|
||||
uint16x8_t s2 = vec_or((uint16x8_t)ss, res_and);
|
||||
|
||||
// Get the resulting bytes
|
||||
//uint16_t output;
|
||||
//vst1q_lane_u16((uint16_t*)&output, (uint16x8_t)mask, 0);
|
||||
//return output;
|
||||
//#warning FIXME
|
||||
return 0;
|
||||
uint32x4_t ss2 = vec_sr((uint32x4_t)s2 , vec_splat_u32(14));
|
||||
uint32x4_t res_and2 = vec_and((uint32x4_t)s2, vec_splats((uint32_t)0xff));
|
||||
uint32x4_t s3 = vec_or((uint32x4_t)ss2, res_and2);
|
||||
|
||||
uint64x2_t ss3 = vec_sr((uint64x2_t)s3, (uint64x2_t)vec_splats(28));
|
||||
uint64x2_t res_and3 = vec_and((uint64x2_t)s3, vec_splats((uint64_t)0xff));
|
||||
uint64x2_t s4 = vec_or((uint64x2_t)ss3, res_and3);
|
||||
|
||||
uint64x2_t ss4 = vec_sld((uint64x2_t) vec_splats(0), s4, 9);
|
||||
uint64x2_t res_and4 = vec_and((uint64x2_t)s4, vec_splats((uint64_t)0xff));
|
||||
uint64x2_t s5 = vec_or((uint64x2_t)ss4, res_and4);
|
||||
|
||||
return s5[0];
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline typename SuperVector<16>::movemask_type SuperVector<16>::eqmask(SuperVector<16> const b) const
|
||||
{
|
||||
return eq(b).movemask();
|
||||
return eq(b).movemask();
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::rshift128_var(uint8_t const N) const
|
||||
{
|
||||
/*
|
||||
{
|
||||
switch(N) {
|
||||
case 1: return {vec_srl(u.v128[0], Zeroes(), 1)}; break;
|
||||
case 2: return {vec_srl(u.v128[0], Zeroes(), 2)}; break;
|
||||
case 3: return {vec_srl(u.v128[0], Zeroes(),3)}; break;
|
||||
case 4: return {vec_srl(u.v128[0], Zeroes(),4)}; break;
|
||||
case 5: return {vec_srl(u.v128[0], Zeroes(),5)}; break;
|
||||
case 6: return {vec_srl(u.v128[0], Zeroes(),6)}; break;
|
||||
case 7: return {vec_srl(u.v128[0], Zeroes(),7)}; break;
|
||||
case 8: return {vec_srl(u.v128[0], Zeroes(),8)}; break;
|
||||
case 9: return {vec_srl(u.v128[0], Zeroes(),9)}; break;
|
||||
case 10: return {vec_srl(u.v128[0], Zeroes(),10)}; break;
|
||||
case 11: return {vec_srl(u.v128[0], Zeroes(),11)}; break;
|
||||
case 12: return {vec_srl(u.v128[0], Zeroes(),12)}; break;
|
||||
case 13: return {vec_srl(u.v128[0], Zeroes(),13)}; break;
|
||||
case 14: return {vec_srl(u.v128[0], Zeroes(),14)}; break;
|
||||
case 15: return {vec_srl(u.v128[0], Zeroes(),15)}; break;
|
||||
case 1: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 15)}; break;
|
||||
case 2: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 14)}; break;
|
||||
case 3: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 13)}; break;
|
||||
case 4: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 12)}; break;
|
||||
case 5: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 11)}; break;
|
||||
case 6: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 10)}; break;
|
||||
case 7: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 9)}; break;
|
||||
case 8: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 8)}; break;
|
||||
case 9: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 7)}; break;
|
||||
case 10: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 6)}; break;
|
||||
case 11: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 5)}; break;
|
||||
case 12: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 4)}; break;
|
||||
case 13: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 3)}; break;
|
||||
case 14: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 2)}; break;
|
||||
case 15: return {(m128) vec_sld((int16x8_t) vec_splat_s8(0), (int16x8_t) u.v128[0], 1)}; break;
|
||||
case 16: return Zeroes(); break;
|
||||
default: break;
|
||||
}
|
||||
return *this;
|
||||
*/
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
|
||||
#ifdef HS_OPTIMIZE
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator>>(uint8_t const N) const
|
||||
{
|
||||
//return {vec_srl(u.v128[0], N)};
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
return rshift128_var(N);
|
||||
}
|
||||
#else
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator>>(uint8_t const N) const
|
||||
{
|
||||
//return rshift128_var(N);
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
#endif
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::lshift128_var(uint8_t const N) const
|
||||
{
|
||||
/*
|
||||
switch(N) {
|
||||
case 1: return {vec_sll(u.v128[0], 1)}; break;
|
||||
case 2: return {vec_sll(u.v128[0], 2)}; break;
|
||||
case 3: return {vec_sll(u.v128[0], 3)}; break;
|
||||
case 4: return {vec_sll(u.v128[0], 4)}; break;
|
||||
case 5: return {vec_sll(u.v128[0], 5)}; break;
|
||||
case 6: return {vec_sll(u.v128[0], 6)}; break;
|
||||
case 7: return {vec_sll(u.v128[0], 7)}; break;
|
||||
case 8: return {vec_sll(u.v128[0], 8)}; break;
|
||||
case 9: return {vec_sll(u.v128[0], 9)}; break;
|
||||
case 10: return {vec_sll(u.v128[0], 10)}; break;
|
||||
case 11: return {vec_sll(u.v128[0], 11)}; break;
|
||||
case 12: return {vec_sll(u.v128[0], 12)}; break;
|
||||
case 13: return {vec_sll(u.v128[0], 13)}; break;
|
||||
case 14: return {vec_sll(u.v128[0], 14)}; break;
|
||||
case 15: return {vec_sll(u.v128[0], 15)}; break;
|
||||
case 1: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 1)}; break;
|
||||
case 2: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 2)}; break;
|
||||
case 3: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 3)}; break;
|
||||
case 4: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 4)}; break;
|
||||
case 5: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 5)}; break;
|
||||
case 6: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 6)}; break;
|
||||
case 7: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 7)}; break;
|
||||
case 8: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 8)}; break;
|
||||
case 9: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 9)}; break;
|
||||
case 10: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 10)}; break;
|
||||
case 11: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 11)}; break;
|
||||
case 12: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 12)}; break;
|
||||
case 13: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 13)}; break;
|
||||
case 14: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 14)}; break;
|
||||
case 15: return {(m128) vec_sld((int16x8_t) u.v128[0], (int16x8_t) vec_splat_s8(0), 15)}; break;
|
||||
case 16: return Zeroes(); break;
|
||||
default: break;
|
||||
}
|
||||
return *this;
|
||||
*/
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
|
||||
#ifdef HS_OPTIMIZE
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator<<(uint8_t const N) const
|
||||
{
|
||||
//return {vec_sll(u.v128[0], N)};
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
return lshift128_var(N);
|
||||
}
|
||||
#else
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::operator<<(uint8_t const N) const
|
||||
{
|
||||
//return lshift128_var(N);
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::loadu(void const *ptr)
|
||||
{
|
||||
//return _mm_loadu_si128((const m128 *)ptr);
|
||||
//#warning FIXME
|
||||
std::cout<<ptr<<std::endl;
|
||||
return Zeroes();
|
||||
return (m128) vec_xl(0, (const int64_t*)ptr);
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::load(void const *ptr)
|
||||
{
|
||||
//assert(ISALIGNED_N(ptr, alignof(SuperVector::size)));
|
||||
//ptr = assume_aligned(ptr, SuperVector::size);
|
||||
//return _mm_load_si128((const m128 *)ptr);
|
||||
//assert(ISALIGNED_N(ptr, alignof(m128)));
|
||||
//return vld1q_s32((const int32_t *)ptr);
|
||||
//#warning FIXME
|
||||
std::cout<<ptr<<std::endl;
|
||||
return Zeroes();
|
||||
assert(ISALIGNED_N(ptr, alignof(SuperVector::size)));
|
||||
return (m128) vec_xl(0, (const int64_t*)ptr);
|
||||
}
|
||||
|
||||
template <>
|
||||
really_inline SuperVector<16> SuperVector<16>::loadu_maskz(void const *ptr, uint8_t const len)
|
||||
{
|
||||
//SuperVector<16> mask = Ones().rshift128_var(16 -len);
|
||||
//mask.print8("mask");
|
||||
//SuperVector<16> v = vld1q_s32((const int32_t *)ptr);
|
||||
//v.print8("v");
|
||||
//return mask & v;
|
||||
//#warning FIXME
|
||||
std::cout<<len<<std::endl;
|
||||
std::cout<<ptr<<std::endl;
|
||||
return Zeroes();
|
||||
SuperVector<16> mask = Ones().rshift128_var(16 -len);
|
||||
mask.print8("mask");
|
||||
SuperVector<16> v = loadu(ptr);
|
||||
v.print8("v");
|
||||
return mask & v;
|
||||
}
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::pshufb(SuperVector<16> b)
|
||||
{
|
||||
//return {_mm_shuffle_epi8(u.v128[0], b.u.v128[0])};
|
||||
//int8x16_t btranslated = vandq_s8((int8x16_t)b.u.v128[0],vdupq_n_s8(0x8f));
|
||||
//return (m128)vqtbl1q_s8((int8x16_t)u.v128[0], (uint8x16_t)btranslated);
|
||||
//#warning FIXM
|
||||
return eq(b).movemask();
|
||||
return (m128) vec_permxor((int8x16_t)vec_splat_s8(0), (int8x16_t)u.v128[0], (int8x16_t) b.u.v128[0]);
|
||||
}
|
||||
|
||||
#ifdef HS_OPTIMIZE
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset)
|
||||
{
|
||||
//return {vextq_s8(u.v128[0], other.u.v128[0], offset)};
|
||||
//#warning FIXME
|
||||
std::cout<<offset<<std::endl;
|
||||
SuperVector<16> mask = Ones().rshift128_var(16 - 0);
|
||||
return mask & pshufb(other);
|
||||
}
|
||||
#else
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::alignr(SuperVector<16> &other, int8_t offset)
|
||||
{
|
||||
/*
|
||||
|
||||
switch(offset) {
|
||||
case 0: return other; break;
|
||||
case 1: return {vextq_s8(u.v128[0], other.u.v128[0], 1)}; break;
|
||||
case 2: return {vextq_s8(u.v128[0], other.u.v128[0], 2)}; break;
|
||||
case 3: return {vextq_s8(u.v128[0], other.u.v128[0], 3)}; break;
|
||||
case 4: return {vextq_s8(u.v128[0], other.u.v128[0], 4)}; break;
|
||||
case 5: return {vextq_s8(u.v128[0], other.u.v128[0], 5)}; break;
|
||||
case 6: return {vextq_s8(u.v128[0], other.u.v128[0], 6)}; break;
|
||||
case 7: return {vextq_s8(u.v128[0], other.u.v128[0], 7)}; break;
|
||||
case 8: return {vextq_s8(u.v128[0], other.u.v128[0], 8)}; break;
|
||||
case 9: return {vextq_s8(u.v128[0], other.u.v128[0], 9)}; break;
|
||||
case 10: return {vextq_s8(u.v128[0], other.u.v128[0], 10)}; break;
|
||||
case 11: return {vextq_s8(u.v128[0], other.u.v128[0], 11)}; break;
|
||||
case 12: return {vextq_s8(u.v128[0], other.u.v128[0], 12)}; break;
|
||||
case 13: return {vextq_s8(u.v128[0], other.u.v128[0], 13)}; break;
|
||||
case 14: return {vextq_s8(u.v128[0], other.u.v128[0], 14)}; break;
|
||||
case 15: return {vextq_s8(u.v128[0], other.u.v128[0], 15)}; break;
|
||||
case 1: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 15)}; break;
|
||||
case 2: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 14)}; break;
|
||||
case 3: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 13)}; break;
|
||||
case 4: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 12)}; break;
|
||||
case 5: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 11)}; break;
|
||||
case 6: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 10)}; break;
|
||||
case 7: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 9)}; break;
|
||||
case 8: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 8)}; break;
|
||||
case 9: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 7)}; break;
|
||||
case 10: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 6)}; break;
|
||||
case 11: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 5)}; break;
|
||||
case 12: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 4)}; break;
|
||||
case 13: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 3)}; break;
|
||||
case 14: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 2)}; break;
|
||||
case 15: return {(m128) vec_sld((int8x16_t) u.v128[0], (int8x16_t) other.u.v128[0], 1)}; break;
|
||||
default: break;
|
||||
}
|
||||
return *this;
|
||||
*/
|
||||
//#warning FIXME
|
||||
SuperVector<16> mask = Ones().rshift128_var(16 - 0);
|
||||
std::cout<<offset<<std::endl;
|
||||
return mask & pshufb(other);
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::pshufb_maskz(SuperVector<16> b, uint8_t const len)
|
||||
@@ -397,88 +332,21 @@ really_inline SuperVector<16> SuperVector<16>::pshufb_maskz(SuperVector<16> b, u
|
||||
return mask & pshufb(b);
|
||||
}
|
||||
|
||||
#ifdef HS_OPTIMIZE
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::lshift64(uint8_t const N)
|
||||
{
|
||||
//return {vshlq_n_s64(u.v128[0], N)};
|
||||
//return {vec_sldw((int64x2_t)u.v128[0], N, 8)};
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();;
|
||||
}
|
||||
#else
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::lshift64(uint8_t const N)
|
||||
{
|
||||
/*
|
||||
switch(N) {
|
||||
case 0: return *this; break;
|
||||
case 1: return {vec_sldw((int64x2_t)u.v128[0], 1, 8)}; break;
|
||||
case 2: return {vec_sldw((int64x2_t)u.v128[0], 2, 8)}; break;
|
||||
case 3: return {vec_sldw((int64x2_t)u.v128[0], 3, 8)}; break;
|
||||
case 4: return {vec_sldw((int64x2_t)u.v128[0], 4, 8)}; break;
|
||||
case 5: return {vec_sldw((int64x2_t)u.v128[0], 5, 8)}; break;
|
||||
case 6: return {vec_sldw((int64x2_t)u.v128[0], 6, 8)}; break;
|
||||
case 7: return {vec_sldw((int64x2_t)u.v128[0], 7, 8)}; break;
|
||||
case 8: return {vec_sldw((int64x2_t)u.v128[0], 8, 8)}; break;
|
||||
case 9: return {vec_sldw((int64x2_t)u.v128[0], 9, 8)}; break;
|
||||
case 10: return {vec_sldw((int64x2_t)u.v128[0], 10, 8)}; break;
|
||||
case 11: return {vec_sldw((int64x2_t)u.v128[0], 11, 8)}; break;
|
||||
case 12: return {vec_sldw((int64x2_t)u.v128[0], 12, 8)}; break;
|
||||
case 13: return {vec_sldw((int64x2_t)u.v128[0], 13, 8)}; break;
|
||||
case 14: return {vec_sldw((int64x2_t)u.v128[0], 14, 8)}; break;
|
||||
case 15: return {vec_sldw((int64x2_t)u.v128[0], 15, 8)}; break;
|
||||
case 16: return Zeroes();
|
||||
default: break;
|
||||
}
|
||||
return *this;
|
||||
*/
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef HS_OPTIMIZE
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::lshift64(uint8_t const N)
|
||||
{
|
||||
uint64x2_t shift_indices = vec_splats((uint64_t)N);
|
||||
return (m128) vec_sl((int64x2_t)u.v128[0] , shift_indices);
|
||||
}
|
||||
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::rshift64(uint8_t const N)
|
||||
{
|
||||
//return {vshrq_n_s64(u.v128[0], N)};
|
||||
//#warning FIXME
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
uint64x2_t shift_indices = vec_splats((uint64_t)N);
|
||||
return (m128) vec_sr((int64x2_t)u.v128[0] , shift_indices);
|
||||
}
|
||||
#else
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::rshift64(uint8_t const N)
|
||||
{
|
||||
/*
|
||||
switch(N) {
|
||||
case 0: return {vshrq_n_s64(u.v128[0], 0)}; break;
|
||||
case 1: return {vshrq_n_s64(u.v128[0], 1)}; break;
|
||||
case 2: return {vshrq_n_s64(u.v128[0], 2)}; break;
|
||||
case 3: return {vshrq_n_s64(u.v128[0], 3)}; break;
|
||||
case 4: return {vshrq_n_s64(u.v128[0], 4)}; break;
|
||||
case 5: return {vshrq_n_s64(u.v128[0], 5)}; break;
|
||||
case 6: return {vshrq_n_s64(u.v128[0], 6)}; break;
|
||||
case 7: return {vshrq_n_s64(u.v128[0], 7)}; break;
|
||||
case 8: return {vshrq_n_s64(u.v128[0], 8)}; break;
|
||||
case 9: return {vshrq_n_s64(u.v128[0], 9)}; break;
|
||||
case 10: return {vshrq_n_s64(u.v128[0], 10)}; break;
|
||||
case 11: return {vshrq_n_s64(u.v128[0], 11)}; break;
|
||||
case 12: return {vshrq_n_s64(u.v128[0], 12)}; break;
|
||||
case 13: return {vshrq_n_s64(u.v128[0], 13)}; break;
|
||||
case 14: return {vshrq_n_s64(u.v128[0], 14)}; break;
|
||||
case 15: return {vshrq_n_s64(u.v128[0], 15)}; break;
|
||||
case 16: return Zeroes();
|
||||
default: break;
|
||||
}
|
||||
return *this;
|
||||
*/
|
||||
//#warning FIXME
|
||||
std::cout<<N<<std::endl;
|
||||
return Zeroes();
|
||||
}
|
||||
#endif
|
||||
|
||||
template<>
|
||||
really_inline SuperVector<16> SuperVector<16>::lshift128(uint8_t const N)
|
||||
|
||||
Reference in New Issue
Block a user