Implement extract_bits_any() for 64-bit types

Author: p12tic

simdpp/core/extract_bits.h

@@ -62,6 +62,17 @@ SIMDPP_INL uint8_t extract_bits_any(const any_vec32<8, V>& a)
 {
     return detail::insn::i_extract_bits_any(uint32<8>(a));
 }
+template<class V>
+SIMDPP_INL uint8_t extract_bits_any(const any_vec64<2, V>& a)
+{
+    return detail::insn::i_extract_bits_any(uint64<2>(a));
+}
+template<class V>
+SIMDPP_INL uint8_t extract_bits_any(const any_vec64<4, V>& a)
+{
+    return detail::insn::i_extract_bits_any(uint64<4>(a));
+}
+
 
 /** Extracts specific bit from each byte of each element of a int8x16 vector.
 

simdpp/detail/insn/extract_bits.h

@@ -206,6 +206,59 @@ SIMDPP_INL uint8_t i_extract_bits_any(const uint32<8>& ca)
 #endif
 }
 
+SIMDPP_INL uint8_t i_extract_bits_any(const uint64<2>& ca)
+{
+    uint64<2> a = ca;
+#if SIMDPP_USE_NULL
+    uint8_t r = 0;
+    for (unsigned i = 0; i < a.length; i++) {
+        uint8_t x = ca.el(i);
+        x = x & 1;
+        r = (r >> 1) | (x << 1);
+    }
+    return r;
+#elif SIMDPP_USE_SSE2
+    return _mm_movemask_pd(_mm_castsi128_pd(a.native()));
+#elif SIMDPP_USE_NEON
+    uint64<2> mask = make_uint(0x1, 0x2);
+    a = bit_and(a, mask);
+    uint64x1_t r = vadd_u64(vget_low_u64(r2), vget_high_u64(r2));
+    return vget_lane_u8(vreinterpret_u8_u64(r), 0);
+#elif SIMDPP_USE_ALTIVEC
+    uint32<4> mask = make_uint(0x1, 0x0, 0x2, 0x0);
+    a = bit_and(a, mask);
+    uint32<4> zero = make_zero();
+    uint32<4> s = (int32x4)vec_sums((__vector int32_t)a.native(),
+                                    (__vector int32_t)zero.native());
+#if SIMDPP_BIG_ENDIAN
+    return extract<7>(uint16x8(s));
+#else
+    return extract<6>(uint16x8(s));
+#endif
+#elif SIMDPP_USE_MSA
+    uint32<4> mask = make_uint(0x1, 0x0, 0x2, 0x0);
+    a = bit_and(a, mask);
+    a = (v4u32) __msa_hadd_u_d(a.native(), a.native());
+    a = bit_or(a, move4_l<2>(a));
+    return extract<0>(a);
+#endif
+}
+
+SIMDPP_INL uint8_t i_extract_bits_any(const uint64<4>& ca)
+{
+#if SIMDPP_USE_AVX512DQ
+    return _mm256_movepi64_mask(ca.native());
+#elif SIMDPP_USE_AVX2
+    return _mm256_movemask_pd(_mm256_castsi256_pd(ca.native()));
+#else
+    // FIXME: can be improved
+    uint64<2> lo, hi;
+    lo = ca.vec<0>();
+    hi = ca.vec<1>();
+    return i_extract_bits_any(lo) | (i_extract_bits_any(hi) << 2);
+#endif
+}
+
 template<unsigned id> SIMDPP_INL
 uint16_t i_extract_bits(const uint8<16>& ca)
 {

test/insn/bitwise.cc

@@ -106,6 +106,39 @@ void test_extract_bits32_n(TestResultsSet& tc)
     TEST_PUSH_ALL_COMB_OP1_T(tc, std::uint32_t, uint32<B/4>, extract_bits_any, s);
 }
 
+template<unsigned B>
+void test_extract_bits64_n(TestResultsSet& tc)
+{
+    using namespace simdpp;
+
+    TestData<uint64<B/8>> s;
+
+    s.add(make_uint(0x0000000000000000, 0x0000000000000000));
+    s.add(make_uint(0xffffffffffffffff, 0x0000000000000000));
+    s.add(make_uint(0x0000000000000000, 0xffffffffffffffff));
+    s.add(make_uint(0xffffffffffffffff, 0xffffffffffffffff));
+
+    TEST_PUSH_ALL_COMB_OP1_T(tc, std::uint32_t, uint64<B/8>, extract_bits_any, s);
+}
+
+template<unsigned B>
+void test_extract_bits64_n2(TestResultsSet& tc)
+{
+    using namespace simdpp;
+
+    TestData<uint64<B/8>> s;
+
+    s.add(make_uint(0xffffffffffffffff, 0x0000000000000000, 0xffffffffffffffff, 0x0000000000000000));
+    s.add(make_uint(0x0000000000000000, 0xffffffffffffffff, 0x0000000000000000, 0xffffffffffffffff));
+    s.add(make_uint(0x0000000000000000, 0x0000000000000000, 0xffffffffffffffff, 0x0000000000000000));
+    s.add(make_uint(0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0xffffffffffffffff));
+    s.add(make_uint(0x0000000000000000, 0xffffffffffffffff, 0xffffffffffffffff, 0xffffffffffffffff));
+    s.add(make_uint(0xffffffffffffffff, 0xffffffffffffffff, 0xffffffffffffffff, 0xffffffffffffffff));
+    s.add(make_uint(0x0000000000000000, 0x0000000000000000, 0x0000000000000000, 0x0000000000000000));
+
+    TEST_PUSH_ALL_COMB_OP1_T(tc, std::uint32_t, uint64<B/8>, extract_bits_any, s);
+}
+
 template<class V, class V32I>
 void test_popcnt_type(TestResultsSet& tc)
 {
@@ -231,6 +264,8 @@ void test_bitwise(TestResults& res, TestReporter& tr)
     test_extract_bits16_n<32>(tc);
     test_extract_bits32_n<16>(tc);
     test_extract_bits32_n<32>(tc);
+    test_extract_bits64_n<16>(tc);
+    test_extract_bits64_n2<32>(tc);
 }
 
 } // namespace SIMDPP_ARCH_NAMESPACE