source: trunk/lib/bitblock128.hpp @ 1815

Last change on this file since 1815 was 1815, checked in by ksherdy, 8 years ago

Updated union type declaration/defintion.

File size: 5.0 KB
RevLine 
[1503]1/*
[1550]2    bitblock128 - Specific 128 bit IDISA implementations.
[1503]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
[1550]14#include "idisa128.hpp"
[1555]15#include "builtins.hpp"
[1550]16
[1815]17union ubitblock {
[1550]18        bitblock128_t _128;
19        uint64_t _64[sizeof(bitblock128_t)/sizeof(uint64_t)];
20        uint32_t _32[sizeof(bitblock128_t)/sizeof(uint32_t)];
21        uint16_t _16[sizeof(bitblock128_t)/sizeof(uint16_t)];
22        uint8_t _8[sizeof(bitblock128_t)/sizeof(uint8_t)];
23};
24
[1542]25static IDISA_ALWAYS_INLINE void adc(bitblock128_t x, bitblock128_t y, bitblock128_t & carry, bitblock128_t & sum);
26static IDISA_ALWAYS_INLINE void sbb(bitblock128_t x, bitblock128_t y, bitblock128_t & borrow, bitblock128_t & difference);
27static IDISA_ALWAYS_INLINE void advance_with_carry(bitblock128_t cursor, bitblock128_t & carry, bitblock128_t & rslt);
28static IDISA_ALWAYS_INLINE bool bitblock_has_bit(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);
[1550]32static IDISA_ALWAYS_INLINE uint32_t count_forward_zeroes(bitblock128_t v);
[1622]33static IDISA_ALWAYS_INLINE uint32_t count_reverse_zeroes(bitblock128_t v);
[1554]34static IDISA_ALWAYS_INLINE uint32_t bitstream_scan(bitblock128_t * v, uint32_t pos);
[1527]35
[1542]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);
[1550]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);
[1542]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));
[1550]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);
[1542]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{
[1550]60bitblock128_t shift_out = simd128<64>::srli<63>(cursor);
61bitblock128_t low_bits = esimd128<64>::mergel(shift_out, carry);
[1542]62// carry = simd_and(hibitmask<bitblock128_t>, cursor);
63carry = simd128<128>::srli<64>(shift_out);                                              // carry - accumlated 127 shift right locical shift of cursor
[1550]64rslt = simd_or(simd128<64>::add(cursor, cursor), low_bits);
[1542]65}
66
67IDISA_ALWAYS_INLINE bool bitblock_has_bit(bitblock128_t arg1)
68{
69        return bitblock::any(arg1);
70}
71
72IDISA_ALWAYS_INLINE void signbitmask(bitblock128_t & v)
73{
74        v =  simd128<128>::slli<127>(simd128<128>::constant<1>());
75}
76
77IDISA_ALWAYS_INLINE bitblock128_t convert(uint64_t s)
78{
[1815]79        ubitblock b = {b._128 = simd128<128>::constant<0>()}; // = {0};
[1542]80        b._64[0] = s;
81        return b._128;
82}
83
84IDISA_ALWAYS_INLINE uint64_t convert (bitblock128_t v)
85{
[1550]86        return (uint64_t) mvmd128<64>::extract<0>(v);
[1542]87}
88
[1550]89IDISA_ALWAYS_INLINE uint32_t count_forward_zeroes(bitblock128_t v) {
90        union {bitblock128_t bitblock; unsigned long elems[sizeof(bitblock128_t)/sizeof(long)];} u;
91        u.bitblock = v;
[1578]92  uint32_t so_far = 0;
93  for (int i = 0; i < sizeof(bitblock128_t)/sizeof(long); i++) {
94    if (u.elems[i] != 0) return so_far + cfzl(u.elems[i]);
95    so_far += 8 * sizeof(long);
96  }
97  return so_far;
[1550]98}
99
[1622]100IDISA_ALWAYS_INLINE uint32_t count_reverse_zeroes(bitblock128_t v) {
[1550]101        union {bitblock128_t bitblock; unsigned long elems[sizeof(bitblock128_t)/sizeof(long)];} u;
102        u.bitblock = v;
[1578]103  uint32_t so_far = 0;
104  for (int i = sizeof(bitblock128_t)/sizeof(long) - 1; i >= 0; i--) {
105    if (u.elems[i] != 0) return so_far + cbzl(u.elems[i]);
106    so_far += 8 * sizeof(long);
107  }
108  return so_far;
[1550]109}
110
[1553]111IDISA_ALWAYS_INLINE uint32_t bitstream_scan(bitblock128_t * v, uint32_t pos) {
[1552]112  unsigned long * bitstream_ptr = (unsigned long *) (((intptr_t) v) + pos/8);
113  unsigned long bitstream_slice = ((*bitstream_ptr) >> (pos % 8));
114  if (bitstream_slice != 0) return pos + cfzl(bitstream_slice);
115  else {
116    do {
117      bitstream_ptr++;
118      bitstream_slice = *bitstream_ptr;
119    } while (bitstream_slice == 0);
120    uint32_t base_posn = 8*((intptr_t) bitstream_ptr - (intptr_t) v);
121    return base_posn + cfzl(bitstream_slice);
122  }
123}
124
125
[1503]126#endif /* BITBLOCK128_HPP_ */
Note: See TracBrowser for help on using the repository browser.