Bit expand byte array

Question

I have a situation where I need to upscale a dynamic sized byte array by 3.

Example:

10101010 11001100

to

11100011 10001110 00111000 11111100 00001111 11000000

I've used the algorithm here to generate a lookup table.

https://stackoverflow.com/a/9044057/280980

static const uint32_t bitExpandTable[256] = {
00000000, 0x000007, 0x000038, 0x00003f, 0x0001c0, 0x0001c7, 0x0001f8, 0x0001ff,
0x000e00, 0x000e07, 0x000e38, 0x000e3f, 0x000fc0, 0x000fc7, 0x000ff8, 0x000fff,
0x007000, 0x007007, 0x007038, 0x00703f, 0x0071c0, 0x0071c7, 0x0071f8, 0x0071ff,
0x007e00, 0x007e07, 0x007e38, 0x007e3f, 0x007fc0, 0x007fc7, 0x007ff8, 0x007fff,
0x038000, 0x038007, 0x038038, 0x03803f, 0x0381c0, 0x0381c7, 0x0381f8, 0x0381ff,
0x038e00, 0x038e07, 0x038e38, 0x038e3f, 0x038fc0, 0x038fc7, 0x038ff8, 0x038fff,
0x03f000, 0x03f007, 0x03f038, 0x03f03f, 0x03f1c0, 0x03f1c7, 0x03f1f8, 0x03f1ff,
0x03fe00, 0x03fe07, 0x03fe38, 0x03fe3f, 0x03ffc0, 0x03ffc7, 0x03fff8, 0x03ffff,
0x1c0000, 0x1c0007, 0x1c0038, 0x1c003f, 0x1c01c0, 0x1c01c7, 0x1c01f8, 0x1c01ff,
0x1c0e00, 0x1c0e07, 0x1c0e38, 0x1c0e3f, 0x1c0fc0, 0x1c0fc7, 0x1c0ff8, 0x1c0fff,
0x1c7000, 0x1c7007, 0x1c7038, 0x1c703f, 0x1c71c0, 0x1c71c7, 0x1c71f8, 0x1c71ff,
0x1c7e00, 0x1c7e07, 0x1c7e38, 0x1c7e3f, 0x1c7fc0, 0x1c7fc7, 0x1c7ff8, 0x1c7fff,
0x1f8000, 0x1f8007, 0x1f8038, 0x1f803f, 0x1f81c0, 0x1f81c7, 0x1f81f8, 0x1f81ff,
0x1f8e00, 0x1f8e07, 0x1f8e38, 0x1f8e3f, 0x1f8fc0, 0x1f8fc7, 0x1f8ff8, 0x1f8fff,
0x1ff000, 0x1ff007, 0x1ff038, 0x1ff03f, 0x1ff1c0, 0x1ff1c7, 0x1ff1f8, 0x1ff1ff,
0x1ffe00, 0x1ffe07, 0x1ffe38, 0x1ffe3f, 0x1fffc0, 0x1fffc7, 0x1ffff8, 0x1fffff,
0xe00000, 0xe00007, 0xe00038, 0xe0003f, 0xe001c0, 0xe001c7, 0xe001f8, 0xe001ff,
0xe00e00, 0xe00e07, 0xe00e38, 0xe00e3f, 0xe00fc0, 0xe00fc7, 0xe00ff8, 0xe00fff,
0xe07000, 0xe07007, 0xe07038, 0xe0703f, 0xe071c0, 0xe071c7, 0xe071f8, 0xe071ff,
0xe07e00, 0xe07e07, 0xe07e38, 0xe07e3f, 0xe07fc0, 0xe07fc7, 0xe07ff8, 0xe07fff,
0xe38000, 0xe38007, 0xe38038, 0xe3803f, 0xe381c0, 0xe381c7, 0xe381f8, 0xe381ff,
0xe38e00, 0xe38e07, 0xe38e38, 0xe38e3f, 0xe38fc0, 0xe38fc7, 0xe38ff8, 0xe38fff,
0xe3f000, 0xe3f007, 0xe3f038, 0xe3f03f, 0xe3f1c0, 0xe3f1c7, 0xe3f1f8, 0xe3f1ff,
0xe3fe00, 0xe3fe07, 0xe3fe38, 0xe3fe3f, 0xe3ffc0, 0xe3ffc7, 0xe3fff8, 0xe3ffff,
0xfc0000, 0xfc0007, 0xfc0038, 0xfc003f, 0xfc01c0, 0xfc01c7, 0xfc01f8, 0xfc01ff,
0xfc0e00, 0xfc0e07, 0xfc0e38, 0xfc0e3f, 0xfc0fc0, 0xfc0fc7, 0xfc0ff8, 0xfc0fff,
0xfc7000, 0xfc7007, 0xfc7038, 0xfc703f, 0xfc71c0, 0xfc71c7, 0xfc71f8, 0xfc71ff,
0xfc7e00, 0xfc7e07, 0xfc7e38, 0xfc7e3f, 0xfc7fc0, 0xfc7fc7, 0xfc7ff8, 0xfc7fff,
0xff8000, 0xff8007, 0xff8038, 0xff803f, 0xff81c0, 0xff81c7, 0xff81f8, 0xff81ff,
0xff8e00, 0xff8e07, 0xff8e38, 0xff8e3f, 0xff8fc0, 0xff8fc7, 0xff8ff8, 0xff8fff,
0xfff000, 0xfff007, 0xfff038, 0xfff03f, 0xfff1c0, 0xfff1c7, 0xfff1f8, 0xfff1ff,
0xfffe00, 0xfffe07, 0xfffe38, 0xfffe3f, 0xffffc0, 0xffffc7, 0xfffff8, 0xffffff,
};

I've tried looping through the byte array using the LUT to memcpy the first 3 bytes to the new array. However my output never looks correct.

Does anyone have any suggestions on how to efficiently implement this? This will be running on an embedded ARM processor.

Edit

LUT test

uint8_t msg[] = { 0xaa, 0x02, 0x43, 0x5a, 0x8d, 0x06, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xd0, 0x84, 0xc6, 0x2d, 0x00, 0xb9 };

uint8_t expanded_msg[63] = { 0 };
uint8_t tmp_val = 0;
uint32_t lut_val;
for (int i = 0; i < 21; i++)
{
    tmp_val = *(uint8_t*)(&msg + i);
    lut_val = bitExpandTable[tmp_val];
    memcpy(&expanded_msg[(i * 3)], &lut_val, 3);
}

print_binary(&msg, 21);
print_binary(&expanded_msg, sizeof(expanded_msg));

Output

[ 10101010 00000010 01000011 01011010 10001101 00000110 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 11010000 10000100 11000110 00101101 00000000 10111001 ]

[ 00111000 10001110 11100011 00000000 00000000 00000000 11000111 10001111 00000011 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000111 01110000 00011100 00000111 11110000 00000011 11111111 11111111 00011111 11000111 11110001 11100011 00000000 00000000 00000000 11111111 11111111 11111111 11000000 10000001 11100011 00000000 00000000 00000000 00000111 01110000 00011100 00111000 10000000 00011111 11111111 11111111 00011111 11000111 11110001 11100011 00000000 00000000 00000000 11111111 11111111 11111111 ]

If I understand correctly, you want to repeat every bit in the original array three times? — Willem Van Onsem, Feb 24 '16 at 21:34
Yes, every bit in the original array needs repeated 3 times. — Jared, Feb 24 '16 at 21:36
Show us what you tried, as well as input, expected output, and actual output. — dbush, Feb 24 '16 at 21:37
Push each bit through a function that writes to output when it has 8, 32 whatever in its buffer. Then flush anything not written. You don't need a lookup table. — Weather Vane, Feb 24 '16 at 21:40
On x86 with BMI2 instruction-set extension, a `pdep+(shl+or)*2` could do this easily. — EOF, Feb 24 '16 at 22:00
lets start with defining a string of bits for a byte is by using a leading 0B — user3629249, Feb 26 '16 at 04:14
you do not need that look up table. a simple loop of shift + and for the 8 bits to select a bit, then output that bit 3 times would do the job very nicely. — user3629249, Feb 26 '16 at 04:17

Stefan Haustein · Accepted Answer · 2016-02-24T22:23:56.710

4

There is a byte order issue with your table (or how you copy the data):

For the first input byte

10101010

your output is

a        b        c
00111000 10001110 11100011

but should be

c        b        a
11100011 10001110 00111000

So the 1st and 3rd byte need to be swapped (and so on)

Instead of changing the table to suit memcpy, I'd just do something like

int j = 0;
for (int i = 0; i < 21; i++) {
  lut_val = bitExpandTable[msg[i]];
  expanded_msg[j++] = (uint8_t) (lut_val >> 16);
  expanded_msg[j++] = (uint8_t) (lut_val >> 8);
  expanded_msg[j++] = (uint8_t) lut_val;
}

edited Feb 24 '16 at 22:23

answered Feb 24 '16 at 22:07

Stefan Haustein

17,024
3
30
45

The single bit at the start of `b` is right when `a` and `c` are swapped – Stefan Haustein Feb 24 '16 at 22:16
1

As written that doesn't work. Reversing the order of appending the bytes does work (16, 8, 0) – Jared Feb 24 '16 at 22:23
Yeah sorry I initially had the values swapped, too... Should be fixed now.. Or do you see a different problem? – Stefan Haustein Feb 24 '16 at 22:25
@WeatherVane 8 bits expand to 24 bits = 3 bytes. The lookup table expands each possible byte value to the corresponding 3 bytes, stored in an int. – Stefan Haustein Feb 24 '16 at 22:28
Thanks for the help. The system now works as expected. – Jared Feb 24 '16 at 23:06

Bit expand byte array

Edit

1 Answers1