bmi.go

Documentation: github.com/apache/arrow/go/v15/parquet/internal/bmi

     1  // Licensed to the Apache Software Foundation (ASF) under one
     2  // or more contributor license agreements.  See the NOTICE file
     3  // distributed with this work for additional information
     4  // regarding copyright ownership.  The ASF licenses this file
     5  // to you under the Apache License, Version 2.0 (the
     6  // "License"); you may not use this file except in compliance
     7  // with the License.  You may obtain a copy of the License at
     8  //
     9  // http://www.apache.org/licenses/LICENSE-2.0
    10  //
    11  // Unless required by applicable law or agreed to in writing, software
    12  // distributed under the License is distributed on an "AS IS" BASIS,
    13  // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    14  // See the License for the specific language governing permissions and
    15  // limitations under the License.
    16  
    17  // Package bmi contains helpers for manipulating bitmaps via BMI2 extensions
    18  // properly falling back to pure go implementations if the CPU doesn't support
    19  // BMI2.
    20  package bmi
    21  
    22  import "math/bits"
    23  
    24  type funcs struct {
    25  	extractBits func(uint64, uint64) uint64
    26  	gtbitmap    func([]int16, int16) uint64
    27  }
    28  
    29  var funclist funcs
    30  
    31  // ExtractBits performs a Parallel Bit extract as per the PEXT instruction for
    32  // x86/x86-64 cpus to use the second parameter as a mask to extract the bits from
    33  // the first argument into a new bitmap.
    34  //
    35  // For each bit Set in selectBitmap, the corresponding bits are extracted from bitmap
    36  // and written to contiguous lower bits of the result, the remaining upper bits are zeroed.
    37  func ExtractBits(bitmap, selectBitmap uint64) uint64 {
    38  	return funclist.extractBits(bitmap, selectBitmap)
    39  }
    40  
    41  // GreaterThanBitmap builds a bitmap where each bit corresponds to whether or not
    42  // the level in that index is greater than the value of rhs.
    43  func GreaterThanBitmap(levels []int16, rhs int16) uint64 {
    44  	return funclist.gtbitmap(levels, rhs)
    45  }
    46  
    47  /* Python code to generate lookup table:
    48  kLookupBits = 5
    49  count = 0
    50  print('constexpr int kLookupBits = {};'.format(kLookupBits))
    51  print('constexpr uint8_t kPextTable[1 << kLookupBits][1 << kLookupBits] = {')
    52  print(' ', end = '')
    53  for mask in range(1 << kLookupBits):
    54      for data in range(1 << kLookupBits):
    55          bit_value = 0
    56          bit_len = 0
    57          for i in range(kLookupBits):
    58              if mask & (1 << i):
    59                  bit_value |= (((data >> i) & 1) << bit_len)
    60                  bit_len += 1
    61          out = '0x{:02X},'.format(bit_value)
    62          count += 1
    63          if count % (1 << kLookupBits) == 1:
    64              print(' {')
    65          if count % 8 == 1:
    66              print('    ', end = '')
    67          if count % 8 == 0:
    68              print(out, end = '\n')
    69          else:
    70              print(out, end = ' ')
    71          if count % (1 << kLookupBits) == 0:
    72              print('  },', end = '')
    73  print('\n};')
    74  */
    75  
    76  const lookupBits = 5
    77  
    78  var pextTable = [1 << lookupBits][1 << lookupBits]uint8{
    79  	{
    80  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    81  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    82  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
    83  	},
    84  	{
    85  		0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00,
    86  		0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01,
    87  		0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01,
    88  	},
    89  	{
    90  		0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01,
    91  		0x01, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00,
    92  		0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01,
    93  	},
    94  	{
    95  		0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02,
    96  		0x03, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01,
    97  		0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03,
    98  	},
    99  	{
   100  		0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00,
   101  		0x00, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
   102  		0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01,
   103  	},
   104  	{
   105  		0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x00, 0x01, 0x00,
   106  		0x01, 0x02, 0x03, 0x02, 0x03, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03,
   107  		0x02, 0x03, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03,
   108  	},
   109  	{
   110  		0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x00, 0x00, 0x01,
   111  		0x01, 0x02, 0x02, 0x03, 0x03, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02,
   112  		0x03, 0x03, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03,
   113  	},
   114  	{
   115  		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02,
   116  		0x03, 0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05,
   117  		0x06, 0x07, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
   118  	},
   119  	{
   120  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01,
   121  		0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
   122  		0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
   123  	},
   124  	{
   125  		0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02,
   126  		0x03, 0x02, 0x03, 0x02, 0x03, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01,
   127  		0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x02, 0x03, 0x02, 0x03,
   128  	},
   129  	{
   130  		0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03,
   131  		0x03, 0x02, 0x02, 0x03, 0x03, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00,
   132  		0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
   133  	},
   134  	{
   135  		0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06,
   136  		0x07, 0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01,
   137  		0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x04, 0x05, 0x06, 0x07,
   138  	},
   139  	{
   140  		0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x02, 0x02, 0x02,
   141  		0x02, 0x03, 0x03, 0x03, 0x03, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
   142  		0x01, 0x01, 0x02, 0x02, 0x02, 0x02, 0x03, 0x03, 0x03, 0x03,
   143  	},
   144  	{
   145  		0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x04, 0x05, 0x04,
   146  		0x05, 0x06, 0x07, 0x06, 0x07, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03,
   147  		0x02, 0x03, 0x04, 0x05, 0x04, 0x05, 0x06, 0x07, 0x06, 0x07,
   148  	},
   149  	{
   150  		0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05,
   151  		0x05, 0x06, 0x06, 0x07, 0x07, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02,
   152  		0x03, 0x03, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06, 0x07, 0x07,
   153  	},
   154  	{
   155  		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A,
   156  		0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05,
   157  		0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
   158  	},
   159  	{
   160  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
   161  		0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
   162  		0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
   163  	},
   164  	{
   165  		0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00,
   166  		0x01, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x02, 0x03,
   167  		0x02, 0x03, 0x02, 0x03, 0x02, 0x03, 0x02, 0x03, 0x02, 0x03,
   168  	},
   169  	{
   170  		0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01,
   171  		0x01, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x02, 0x02,
   172  		0x03, 0x03, 0x02, 0x02, 0x03, 0x03, 0x02, 0x02, 0x03, 0x03,
   173  	},
   174  	{
   175  		0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02,
   176  		0x03, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x04, 0x05,
   177  		0x06, 0x07, 0x04, 0x05, 0x06, 0x07, 0x04, 0x05, 0x06, 0x07,
   178  	},
   179  	{
   180  		0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00,
   181  		0x00, 0x01, 0x01, 0x01, 0x01, 0x02, 0x02, 0x02, 0x02, 0x03, 0x03,
   182  		0x03, 0x03, 0x02, 0x02, 0x02, 0x02, 0x03, 0x03, 0x03, 0x03,
   183  	},
   184  	{
   185  		0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x00, 0x01, 0x00,
   186  		0x01, 0x02, 0x03, 0x02, 0x03, 0x04, 0x05, 0x04, 0x05, 0x06, 0x07,
   187  		0x06, 0x07, 0x04, 0x05, 0x04, 0x05, 0x06, 0x07, 0x06, 0x07,
   188  	},
   189  	{
   190  		0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x00, 0x00, 0x01,
   191  		0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06,
   192  		0x07, 0x07, 0x04, 0x04, 0x05, 0x05, 0x06, 0x06, 0x07, 0x07,
   193  	},
   194  	{
   195  		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x00, 0x01, 0x02,
   196  		0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D,
   197  		0x0E, 0x0F, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
   198  	},
   199  	{
   200  		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01,
   201  		0x01, 0x01, 0x01, 0x01, 0x01, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
   202  		0x02, 0x02, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03, 0x03,
   203  	},
   204  	{
   205  		0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02,
   206  		0x03, 0x02, 0x03, 0x02, 0x03, 0x04, 0x05, 0x04, 0x05, 0x04, 0x05,
   207  		0x04, 0x05, 0x06, 0x07, 0x06, 0x07, 0x06, 0x07, 0x06, 0x07,
   208  	},
   209  	{
   210  		0x00, 0x00, 0x01, 0x01, 0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03,
   211  		0x03, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05, 0x05, 0x04, 0x04,
   212  		0x05, 0x05, 0x06, 0x06, 0x07, 0x07, 0x06, 0x06, 0x07, 0x07,
   213  	},
   214  	{
   215  		0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06,
   216  		0x07, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x08, 0x09,
   217  		0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x0C, 0x0D, 0x0E, 0x0F,
   218  	},
   219  	{
   220  		0x00, 0x00, 0x00, 0x00, 0x01, 0x01, 0x01, 0x01, 0x02, 0x02, 0x02,
   221  		0x02, 0x03, 0x03, 0x03, 0x03, 0x04, 0x04, 0x04, 0x04, 0x05, 0x05,
   222  		0x05, 0x05, 0x06, 0x06, 0x06, 0x06, 0x07, 0x07, 0x07, 0x07,
   223  	},
   224  	{
   225  		0x00, 0x01, 0x00, 0x01, 0x02, 0x03, 0x02, 0x03, 0x04, 0x05, 0x04,
   226  		0x05, 0x06, 0x07, 0x06, 0x07, 0x08, 0x09, 0x08, 0x09, 0x0A, 0x0B,
   227  		0x0A, 0x0B, 0x0C, 0x0D, 0x0C, 0x0D, 0x0E, 0x0F, 0x0E, 0x0F,
   228  	},
   229  	{
   230  		0x00, 0x00, 0x01, 0x01, 0x02, 0x02, 0x03, 0x03, 0x04, 0x04, 0x05,
   231  		0x05, 0x06, 0x06, 0x07, 0x07, 0x08, 0x08, 0x09, 0x09, 0x0A, 0x0A,
   232  		0x0B, 0x0B, 0x0C, 0x0C, 0x0D, 0x0D, 0x0E, 0x0E, 0x0F, 0x0F,
   233  	},
   234  	{
   235  		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A,
   236  		0x0B, 0x0C, 0x0D, 0x0E, 0x0F, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15,
   237  		0x16, 0x17, 0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F,
   238  	},
   239  }
   240  
   241  // software emulation of _pext_u64
   242  func extractBitsGo(bitmap, selectBitmap uint64) uint64 {
   243  	if selectBitmap == ^uint64(0) {
   244  		return bitmap
   245  	} else if selectBitmap == 0 {
   246  		return 0
   247  	}
   248  
   249  	// fallback to lookup table method
   250  	bitValue := uint64(0)
   251  	bitLen := int(0)
   252  	const lookupMask = uint64((uint(1) << lookupBits) - 1)
   253  
   254  	for selectBitmap != 0 {
   255  		maskLen := bits.OnesCount32(uint32(selectBitmap & lookupMask))
   256  		value := pextTable[selectBitmap&lookupMask][bitmap&lookupMask]
   257  		bitValue |= uint64(value) << bitLen
   258  		bitLen += maskLen
   259  		bitmap >>= lookupBits
   260  		selectBitmap >>= lookupBits
   261  	}
   262  	return bitValue
   263  }
   264  
   265  func greaterThanBitmapGo(levels []int16, rhs int16) uint64 {
   266  	mask := uint64(0)
   267  	for idx, lvl := range levels {
   268  		if lvl > rhs {
   269  			mask |= uint64(1) << idx
   270  		} else {
   271  			mask |= uint64(0) << idx
   272  		}
   273  	}
   274  	return mask
   275  }
   276
View as plain text