source: trunk/lib/bitblock128.hpp @ 1554

Last change on this file since 1554 was 1554, checked in by ksherdy, 8 years ago

Minor fix.

File size: 5.6 KB
Line 
1/*
2    bitblock128 - Specific 128 bit IDISA implementations.
3
4    Idealized SIMD Operations with SSE versions
5    Copyright (C) 2011, Robert D. Cameron, Kenneth S. Herdy, Hua Huang and Nigel Medforth.
6    Licensed to the public under the Open Software License 3.0.
7    Licensed to International Characters Inc.
8       under the Academic Free License version 3.0.
9
10*/
11#ifndef BITBLOCK128_HPP_
12#define BITBLOCK128_HPP_
13
14#include "idisa128.hpp"
15
16union ubitblock128 {
17        bitblock128_t _128;
18        uint64_t _64[sizeof(bitblock128_t)/sizeof(uint64_t)];
19        uint32_t _32[sizeof(bitblock128_t)/sizeof(uint32_t)];
20        uint16_t _16[sizeof(bitblock128_t)/sizeof(uint16_t)];
21        uint8_t _8[sizeof(bitblock128_t)/sizeof(uint8_t)];
22};
23
24static IDISA_ALWAYS_INLINE void adc(bitblock128_t x, bitblock128_t y, bitblock128_t & carry, bitblock128_t & sum);
25static IDISA_ALWAYS_INLINE void sbb(bitblock128_t x, bitblock128_t y, bitblock128_t & borrow, bitblock128_t & difference);
26static IDISA_ALWAYS_INLINE void advance_with_carry(bitblock128_t cursor, bitblock128_t & carry, bitblock128_t & rslt);
27static IDISA_ALWAYS_INLINE bool bitblock_has_bit(bitblock128_t arg1);
28static IDISA_ALWAYS_INLINE uint64_t bitblock_bit_count(bitblock128_t arg1);
29static IDISA_ALWAYS_INLINE void signbitmask(bitblock128_t & v);
30static IDISA_ALWAYS_INLINE bitblock128_t convert (uint64_t s);
31static IDISA_ALWAYS_INLINE uint64_t convert (bitblock128_t v);
32static IDISA_ALWAYS_INLINE uint32_t count_forward_zeroes(bitblock128_t v);
33static IDISA_ALWAYS_INLINE uint32_t count_backward_zeroes(bitblock128_t v);
34static IDISA_ALWAYS_INLINE uint32_t bitstream_scan(bitblock128_t * v, uint32_t pos);
35
36IDISA_ALWAYS_INLINE void adc(bitblock128_t x, bitblock128_t y, bitblock128_t & carry, bitblock128_t & sum)
37{
38        bitblock128_t gen = simd_and(x, y);
39        bitblock128_t prop = simd_or(x, y);
40        bitblock128_t partial = simd128<64>::add(simd128<64>::add(x, y), carry);
41        bitblock128_t c1 = simd128<128>::slli<64>(simd128<64>::srli<63>(simd_or(gen, simd_andc(prop, partial))));
42        sum = simd128<64>::add(c1, partial);
43        // carry = simd_and(hibitmask<bitblock128_t>(), simd_or(gen, simd_andc(prop, sum))); // TODO - set high bit carry via hibitmask
44        carry = simd128<128>::srli<127>(simd_or(gen, simd_andc(prop, sum))); // TODO -
45}
46
47IDISA_ALWAYS_INLINE void sbb(bitblock128_t x, bitblock128_t y, bitblock128_t & borrow, bitblock128_t & difference)
48{
49        bitblock128_t gen = simd_andc(y, x);
50        bitblock128_t prop = simd_not(simd_xor(x, y));
51        bitblock128_t partial = simd128<64>::sub(simd128<64>::sub(x, y), borrow);
52        bitblock128_t b1 = simd128<128>::slli<64>(simd128<64>::srli<63>(simd_or(gen, simd_and(prop, partial))));
53        difference = simd128<64>::sub(partial, b1);
54        // borrow = simd_and(hibitmask<bitblock128_t>, simd_or(gen, simd_and(prop, difference))); // TODO - set high bit carry via hibitmask
55        borrow = simd128<128>::srli<127>(simd_or(gen, simd_and(prop, difference)));
56}
57
58IDISA_ALWAYS_INLINE void advance_with_carry(bitblock128_t cursor, bitblock128_t & carry, bitblock128_t & rslt)
59{
60bitblock128_t shift_out = simd128<64>::srli<63>(cursor);
61bitblock128_t low_bits = esimd128<64>::mergel(shift_out, carry);
62// carry = simd_and(hibitmask<bitblock128_t>, cursor);
63carry = simd128<128>::srli<64>(shift_out);                                              // carry - accumlated 127 shift right locical shift of cursor
64rslt = simd_or(simd128<64>::add(cursor, cursor), low_bits);
65}
66
67IDISA_ALWAYS_INLINE bool bitblock_has_bit(bitblock128_t arg1)
68{
69        return bitblock::any(arg1);
70}
71
72IDISA_ALWAYS_INLINE uint64_t bitblock_bit_count(bitblock128_t arg1)
73{
74        return mvmd128<64>::extract<0>(simd128<128>::popcount(arg1));
75}
76
77IDISA_ALWAYS_INLINE void signbitmask(bitblock128_t & v)
78{
79        v =  simd128<128>::slli<127>(simd128<128>::constant<1>());
80}
81
82IDISA_ALWAYS_INLINE bitblock128_t convert(uint64_t s)
83{
84        ubitblock128 b = {b._128 = simd128<128>::constant<0>()}; // = {0};
85        b._64[0] = s;
86        return b._128;
87}
88
89IDISA_ALWAYS_INLINE uint64_t convert (bitblock128_t v)
90{
91        return (uint64_t) mvmd128<64>::extract<0>(v);
92}
93
94IDISA_ALWAYS_INLINE uint32_t count_forward_zeroes(bitblock128_t v) {
95        union {bitblock128_t bitblock; unsigned long elems[sizeof(bitblock128_t)/sizeof(long)];} u;
96        u.bitblock = v;
97
98#if LONG_BIT == 64
99  if (u.elems[0] != 0) {return cfzll(u.elems[0]);}
100  else if (u.elems[1] != 0) {return 64 + cfzll(u.elems[1]);}
101  else {return 128;}
102#endif
103
104#if LONG_BIT < 64
105  if (u.elems[0] != 0) {return cfzl(u.elems[0]);}
106  else if (u.elems[1] != 0) {return 32 + cfzl(u.elems[1]);}
107  else if (u.elems[2] != 0) {return 64 + cfzl(u.elems[2]);}
108  else if (u.elems[3] != 0) {return 96 + cfzl(u.elems[3]);}
109#endif
110  else {return 128;}
111}
112
113IDISA_ALWAYS_INLINE uint32_t count_backward_zeroes(bitblock128_t v) {
114        union {bitblock128_t bitblock; unsigned long elems[sizeof(bitblock128_t)/sizeof(long)];} u;
115        u.bitblock = v;
116
117#if LONG_BIT == 64
118  if (u.elems[1] != 0) return cbzll(u.elems[1]);
119  else if (u.elems[0] != 0) return LONG_BIT + cbzll(u.elems[0]);
120#endif
121#if LONG_BIT < 64
122  if (u.elems[3] != 0) return cbzl(u.elems[3]);
123  else if (u.elems[2] != 0) return 32 + cbzl(u.elems[2]);
124  else if (u.elems[1] != 0) return 64 + cbzl(u.elems[1]);
125  else if (u.elems[0] != 0) return 96 + cbzl(u.elems[0]);
126#endif
127  else {return 128;}
128}
129
130IDISA_ALWAYS_INLINE uint32_t bitstream_scan(bitblock128_t * v, uint32_t pos) {
131  unsigned long * bitstream_ptr = (unsigned long *) (((intptr_t) v) + pos/8);
132  unsigned long bitstream_slice = ((*bitstream_ptr) >> (pos % 8));
133  if (bitstream_slice != 0) return pos + cfzl(bitstream_slice);
134  else {
135    do {
136      bitstream_ptr++;
137      bitstream_slice = *bitstream_ptr;
138    } while (bitstream_slice == 0);
139    uint32_t base_posn = 8*((intptr_t) bitstream_ptr - (intptr_t) v);
140    return base_posn + cfzl(bitstream_slice);
141  }
142}
143
144
145#endif /* BITBLOCK128_HPP_ */
Note: See TracBrowser for help on using the repository browser.