| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 111.1 | Any supporters for octaword alignment? | FLYBA::BRENDER | Ron Brender | Tue Feb 18 1997 20:29 | 6 |
| Now that the dust has settled to establish quadword as the required alignment
for global variables, is there anyone who wants to argue that extended
precision (16-byte, 128-bit) global variables (both scalars and complex)
ought to have a required alignment of octaword (16-byte, 128-bit)?
If no one speaks out in favor, I'll just let this issue quietly die...
|
| 111.2 | Discussion, but not arguing in favor | WIBBIN::NOYCE | Pulling weeds, pickin' stones | Wed Feb 19 1997 13:26 | 5 |
| This would be of interest if we thought future hardware might include
load/store octaword instructions. Presumably those would work best if
the target address is octa aligned. The calling standard does already
specify octaword alignment in some other places -- stack, linkage pairs
(VMS), homed argument lists (Unix & NT -- follows from stack rule).
|
| 111.3 | | DECC::OUELLETTE | | Wed Feb 19 1997 16:48 | 1 |
| The stack on NT is Quad aligned right?
|
| 111.4 | Yes, but | WIBBIN::NOYCE | Pulling weeds, pickin' stones | Wed Feb 19 1997 17:29 | 7 |
| It's also octaword aligned.
Every fixed-size allocation is rounded up to a multiple of 16.
Every dynamic allocation is (dynamically) rounded up to a multiple of 16
(or the new stack pointer is rounded down to a multiple of 16, which is
equivalent).
|