| 1 | Notes on i80386 string assembly routines. Author: Kees J. Bot
|
|---|
| 2 | 2 Jan 1994
|
|---|
| 3 |
|
|---|
| 4 | Remarks.
|
|---|
| 5 | All routines set up proper stack frames, so that stack traces can be
|
|---|
| 6 | derived from core dumps. String routines are often the ones that
|
|---|
| 7 | get the bad pointer.
|
|---|
| 8 |
|
|---|
| 9 | Flags are often not right in boundary cases (zero string length) on
|
|---|
| 10 | repeated string scanning or comparing instructions. This has been
|
|---|
| 11 | handled in sometimes nonobvious ways.
|
|---|
| 12 |
|
|---|
| 13 | Only the eax, edx, and ecx registers are not preserved, all other
|
|---|
| 14 | registers are. This is what GCC expects. (ACK sees ebx as scratch
|
|---|
| 15 | too.) The direction byte is assumed to be wrong, and left clear on
|
|---|
| 16 | exit.
|
|---|
| 17 |
|
|---|
| 18 | Assumptions.
|
|---|
| 19 | The average string is short, so short strings should not suffer from
|
|---|
| 20 | smart tricks to copy, compare, or search large strings fast. This
|
|---|
| 21 | means that the routines are fast on average, but not optimal for
|
|---|
| 22 | long strings.
|
|---|
| 23 |
|
|---|
| 24 | It doesn't pay to use word or longword operations on strings, the
|
|---|
| 25 | setup time hurts the average case.
|
|---|
| 26 |
|
|---|
| 27 | Memory blocks are probably large and on word or longword boundaries.
|
|---|
| 28 |
|
|---|
| 29 | No unaligned word moves are done. Again the setup time may hurt the
|
|---|
| 30 | average case. Furthermore, the author likes to enable the alignment
|
|---|
| 31 | check on a 486.
|
|---|
| 32 |
|
|---|
| 33 | String routines.
|
|---|
| 34 | They have been implemented using byte string instructions. The
|
|---|
| 35 | length of a string it usually determined first, followed by the
|
|---|
| 36 | actual operation.
|
|---|
| 37 |
|
|---|
| 38 | Strcmp.
|
|---|
| 39 | This is the only string routine that uses a loop, and not
|
|---|
| 40 | instructions with a repeat prefix. Problem is that we don't know
|
|---|
| 41 | how long the string is. Scanning for the end costs if the strings
|
|---|
| 42 | are unequal in the first few bytes.
|
|---|
| 43 |
|
|---|
| 44 | Strchr.
|
|---|
| 45 | The character we look for is often not there, or at some distance
|
|---|
| 46 | from the start. The string is scanned twice, for the terminating
|
|---|
| 47 | zero and the character searched, in chunks of increasing length.
|
|---|
| 48 |
|
|---|
| 49 | Memory routines.
|
|---|
| 50 | Memmove, memcpy, and memset use word or longword instructions if the
|
|---|
| 51 | address(es) are at word or longword boundaries. No tricks to get
|
|---|
| 52 | alignment after doing a few bytes. No unaligned operations.
|
|---|
