#1074: Identify and fix performance bottlenecks in our app_server
---------------------------------+----------------------
Reporter: axeld | Owner: jua
Type: enhancement | Status: assigned
Priority: normal | Milestone: R1
Component: Servers/app_server | Version:
Resolution: | Keywords:
Blocked By: | Blocking:
Platform: All |
---------------------------------+----------------------
Comment (by waddlesplash):
system_time with int128 as it currently exists:
{{{
0000000000000010 <system_time>:
10: 0f 31 rdtsc
12: 48 c1 e2 20 shl $0x20,%rdx
16: 48 09 d0 or %rdx,%rax
19: 48 f7 25 00 00 00 00 mulq 0x0(%rip) # 20
<system_time+0x10>
20: 48 89 d0 mov %rdx,%rax
23: c3 retq
}}}
system_time with the logic as used in the bootloader involving uint64:
{{{
0000000000000010 <system_time>:
10: 0f 31 rdtsc
12: 48 8b 0d 00 00 00 00 mov 0x0(%rip),%rcx # 19
<system_time+0x9>
19: 48 c1 e2 20 shl $0x20,%rdx
1d: 48 09 d0 or %rdx,%rax
20: 89 c2 mov %eax,%edx
22: 48 c1 e8 20 shr $0x20,%rax
26: 48 0f af d1 imul %rcx,%rdx
2a: 48 0f af c1 imul %rcx,%rax
2e: 48 c1 ea 20 shr $0x20,%rdx
32: 48 01 d0 add %rdx,%rax
35: c3 retq
}}}
So the uint128 version is actually fewer instructions, so unless imul is
somehow significantly faster than mulq, the current version is faster.
--
Ticket URL: <https://dev.haiku-os.org/ticket/1074#comment:29>
Haiku <https://dev.haiku-os.org>
The Haiku operating system.
