github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

[VSX] optimize shifting methods, replace template Unroller

Browse Source
This commit is contained in:
Konstantinos Margaritis 2022-09-07 11:52:08 +03:00
parent dc6b8ae92d
commit be20c2c519
1 changed files with 21 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
src/util/supervector/arch/ppc64el/impl.cpp
View File

@ -396,50 +396,40 @@ template <>
really_inline SuperVector<16> SuperVector<16>::vshl_8 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshl_8 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint8x16_t shift_indices = vec_splats((uint8_t) N);
SuperVector result; return { vec_sl(u.u8x16[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sl(v->u.s8x16[0], vec_splats((uint8_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshl_16 (uint8_t const UNUSED N) const really_inline SuperVector<16> SuperVector<16>::vshl_16 (uint8_t const UNUSED N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint16x8_t shift_indices = vec_splats((uint16_t) N);
SuperVector result; return { vec_sl(u.u16x8[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sl(v->u.s16x8[0], vec_splats((uint16_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshl_32 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshl_32 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint32x4_t shift_indices = vec_splats((uint32_t) N);
SuperVector result; return { vec_sl(u.u32x4[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sl(v->u.s32x4[0], vec_splats((uint32_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshl_64 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshl_64 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint64x2_t shift_indices = vec_splats((ulong64_t) N);
SuperVector result; return { vec_sl(u.u64x2[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sl(v->u.s64x2[0], vec_splats((ulong64_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshl_128(uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshl_128(uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); SuperVector sl{N << 3};
SuperVector result; return { vec_slo(u.u8x16[0], sl.u.u8x16[0]) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sld(v->u.s8x16[0], (int8x16_t)vec_splat_s8(0), n)}; });
return result;
} }
template <> template <>
@ -452,50 +442,40 @@ template <>
really_inline SuperVector<16> SuperVector<16>::vshr_8 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_8 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint8x16_t shift_indices = vec_splats((uint8_t) N);
SuperVector result; return { vec_sr(u.u8x16[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sr(v->u.s8x16[0], vec_splats((uint8_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_16 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_16 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint16x8_t shift_indices = vec_splats((uint16_t) N);
SuperVector result; return { vec_sr(u.u16x8[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sr(v->u.s16x8[0], vec_splats((uint16_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_32 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_32 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint32x4_t shift_indices = vec_splats((uint32_t) N);
SuperVector result; return { vec_sr(u.u32x4[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sr(v->u.s32x4[0], vec_splats((uint32_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_64 (uint8_t const N) const really_inline SuperVector<16> SuperVector<16>::vshr_64 (uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); uint64x2_t shift_indices = vec_splats((ulong64_t) N);
SuperVector result; return { vec_sr(u.u64x2[0], shift_indices) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sr(v->u.s64x2[0], vec_splats((ulong64_t)n))}; });
return result;
} }
template <> template <>
really_inline SuperVector<16> SuperVector<16>::vshr_128(uint8_t const UNUSED N) const really_inline SuperVector<16> SuperVector<16>::vshr_128(uint8_t const N) const
{ {
if (N == 0) return *this; if (N == 0) return *this;
if (N == 16) return Zeroes(); SuperVector sr{N << 3};
SuperVector result; return { vec_sro(u.u8x16[0], sr.u.u8x16[0]) };
Unroller<1, 16>::iterator([&,v=this](auto const i) { constexpr uint8_t n = i.value; if (N == n) result = {(m128) vec_sld((int8x16_t)vec_splat_u8(0), v->u.s8x16[0], 16 - n)}; });
return result;
} }
template <> template <>