What are the sizes of tword, oword and yword opera

Question

What are the sizes of tword, oword and yword operands, as used in the NASM/YASM manual? And on a related note, is there a trick or underlying idea to these names? Is there a way by which bigger word sizes are given logical names?

I know that while word sizes may differ between systems, a NASM word is 2 bytes, dword is double that (4 bytes), qword is a quad word (8 bytes), but... is tword a triple word (6 bytes)? And for oword and yword I can't even think of a plausible meaning.

Note that it is probably an easy question, but I couldn't find an answer. In the NASM and YASM manuals these sizes are not explained, not even at the DQ, DT, DY, RESQ, REST, RESY pseudo-instructions. I read somewhere that MASM uses a similar system, but could not find anything on that either.

---

Edit: Based on the answers, this is the complete list:

• 1 byte (8 bit): byte, DB, RESB
• 2 bytes (16 bit): word, DW, RESW
• 4 bytes (32 bit): dword, DD, RESD
• 8 bytes (64 bit): qword, DQ, RESQ
• 10 bytes (80 bit): tword, DT, REST
• 16 bytes (128 bit): oword, DO, RESO, DDQ, RESDQ
• 32 bytes (256 bit): yword, DY, RESY
• 64 bytes (512 bit): zword, DZ, RESZ

Answer 1

I have checked it with two approaches for NASM: source code and empirical.

Source code

Source at: http://repo.or.cz/w/nasm.git

Then:

git grep -C2 tword

And we fall upon:

switch (size) {
case 1:
    return "byte";
case 2:
    return "word";
case 4:
    return "dword";
case 8:
    return "qword";
case 10:
    return "tword";
case 16:
    return "oword";
case 32:
    return "yword";
case 64:
    return "zword";
default:
    return "???";
}

Empirical

git log -p and git tag --contains tell me that zword was added in 2.11, and since I'm on 2.10 and lazy, I'll omit that one.

On our .asm file:

section .bss
resb1 resb 1
resw1 resw 1
resq1 resq 1
rest1 rest 1
reso1 reso 1
resy1 resy 1
; Just to read the objdump better.
resb2 resb 1

Then compile and:

objdump -D -j .bss main.o

gives:

00000000 <resb1>:
    ...

00000001 <resw1>:
    ...

00000003 <resd1>:
3:  00 00                   add    %al,(%eax)
    ...

00000007 <resq1>:
    ...

0000000f <rest1>:
    ...

00000019 <reso1>:
    ...

00000029 <resy1>:
    ...

00000049 <resb2>:
    ...

If we take the differences between each position, we reach the same conclusion as before.

zword menemonic

For the ZMM registers added by AVX-512: https://en.wikipedia.org/wiki/AVX-512

I wonder what Intel will do when the letters of the alphabet end.

Answer 2

Looking at TFS, it looks like:

• 'oword'/'DO' is 8 times as big as "word" (O for octoword), synonymous with dqword ("double-quad"); that would be 128 bits, corresponding to the size of an SSE vector register.
• 'tword'/'DT' is 80 bits (T for ten bytes), the full size of an Intel x87 floating point register.
• 'yword'/'DY' is 256 bits, and the Y is presumably mnemonic for the YMM names of the 256-bit vector registers in the Intel AVX extensions.
• 'zword'/'DZ' is 512 bits, Z for the ZMM names of the 512-bit vector registers in the Intel AVX-512 extensions.

So, not exactly a logical naming convention.