[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-serviceability-dev
Subject:    Re: RFR: 8152538: UL doesn't inline the LogTagSet::is_level check
From:       Bengt Rutisson <bengt.rutisson () oracle ! com>
Date:       2016-03-23 12:37:59
Message-ID: 56F28E27.6030600 () oracle ! com
[Download RAW message or body]


Hi StefanK,

Looks good!

Bengt

On 2016-03-23 11:28, Stefan Karlsson wrote:
> Hi all,
>
> Please, review this patch to optimize the log_is_enabled check in the 
> Unified Logging framework.
>
> http://cr.openjdk.java.net/~stefank/8152538/webrev.01/
> https://bugs.openjdk.java.net/browse/JDK-8152538
>
> The following code:
>   void logLine() {
>     log_debug(gc)("logLine");
>   }
>
> expands into:
> (!log_is_enabled(Debug, __VA_ARGS__)) ? ... : ... )
>
> The log_is_enabled macro further expands and inlines functions, but 
> then end up calling LogTagSet::is_level.
>
> Currently, the generated code looks like this:
> (gdb) disass logLine
> Dump of assembler code for function logLine():
>    0x0000000000ae3bc0 <+0>: push %rbp
>    0x0000000000ae3bc1 <+1>: lea 0x755fb8(%rip),%rdi # 0x1239b80 
> <_ZN16LogTagSetMappingILN6LogTag4typeE22ELS1_0ELS1_0ELS1_0ELS1_0EE7_tagsetE>
>    0x0000000000ae3bc8 <+8>: mov $0x2,%esi
>    0x0000000000ae3bcd <+13>: mov %rsp,%rbp
>    0x0000000000ae3bd0 <+16>: callq 0x969c30 
> <LogTagSet::is_level(LogLevel::type) const>
>    0x0000000000ae3bd5 <+21>: test %al,%al
>    0x0000000000ae3bd7 <+23>: jne 0xae3be0 <logLine()+32>
>    0x0000000000ae3bd9 <+25>: pop %rbp
>    0x0000000000ae3bda <+26>: retq
>    0x0000000000ae3bdb <+27>: nopl 0x0(%rax,%rax,1)
>    0x0000000000ae3be0 <+32>: pop %rbp
>    0x0000000000ae3be1 <+33>: lea 0x265133(%rip),%rdi # 0xd48d1b
>    0x0000000000ae3be8 <+40>: xor %eax,%eax
>    0x0000000000ae3bea <+42>:
>     jmpq 0x5d9e00 <LogImpl<(LogTag::type)22, (LogTag::type)0, 
> (LogTag::type)0, (LogTag::type)0, (LogTag::type)0, 
> (LogTag::type)0>::write<(LogLevel::type)2>(char const*, ...)>
>
> By making LogTagSet::is_level inlinable, we get:
> (gdb) disass logLine
> Dump of assembler code for function logLine():
>    0x0000000000ae1690 <+0>: cmpq $0x0,0x756520(%rip) # 0x1237bb8 
> <_ZN16LogTagSetMappingILN6LogTag4typeE22ELS1_0ELS1_0ELS1_0ELS1_0EE7_tagsetE+56>
>    0x0000000000ae1698 <+8>: push %rbp
>    0x0000000000ae1699 <+9>: mov %rsp,%rbp
>    0x0000000000ae169c <+12>: je 0xae16b0 <logLine()+32>
>    0x0000000000ae169e <+14>: pop %rbp
>    0x0000000000ae169f <+15>: lea 0x264d75(%rip),%rdi # 0xd4641b
>    0x0000000000ae16a6 <+22>: xor %eax,%eax
>    0x0000000000ae16a8 <+24>:
>     jmpq 0x5d8f60 <LogImpl<(LogTag::type)22, (LogTag::type)0, 
> (LogTag::type)0, (LogTag::type)0, (LogTag::type)0, 
> (LogTag::type)0>::write<(LogLevel::type)2>(char const*, ...)>
>    0x0000000000ae16ad <+29>: nopl (%rax)
>    0x0000000000ae16b0 <+32>: pop %rbp
>    0x0000000000ae16b1 <+33>: retq
> End of assembler dump.
>
> Thanks,
> StefanK

[prev in list] [next in list] [prev in thread] [next in thread]