RFR (M): 8003853: specify offset of IC load in java_to_interp stub

[Dean Long]

On 4/23/2013 11:00 AM, Christian Thalinger wrote:
> The change looks good.  I have some minor comments though:
>
>
>     src/share/vm/code/compiledIC.hpp:
>
> +  static uint to_interp_stub_size();
> +  static uint reloc_to_interp_stub();
> +            stub_size  += CompiledStaticCall::to_interp_stub_size();
> +            reloc_size += CompiledStaticCall::reloc_to_interp_stub();
> stub_size and reloc_size are ints.  We should change the return type 
> from uint to int.
>
> src/cpu/x86/vm/compiledIC_x86.cpp:
>   108 uint CompiledStaticCall::to_interp_stub_size() {
>   109   return 10              // movl; jmp
>   110          LP64_ONLY(+5);  // movq (1+1+8); jmp (1+4)
>   111 }
> In expectancy of future changes we should do:
>   108 uint CompiledStaticCall::to_interp_stub_size() {
>   109   return NOT_LP64(10)    // movl; jmp
>   110          LP64_ONLY(15);  // movq (1+1+8); jmp (1+4)
>   111 }
> And one question unrelated to this change (because it was there before):
>
> src/cpu/sparc/vm/compiledIC_sparc.cpp:
>   126 // Relocation entries for call stub, compiled java to interpreter.
>   127 uint CompiledStaticCall::reloc_to_interp_stub() {
>   128   return 10;  // 4 in emit_java_to_interp + 1 in Java_Static_Call
>   129 }
> Why does it say 5 relocations but returns 10?
>
Archeology reveals that it has been like that since the initial 
implementation in 1998.  Apparently it's better to over-estimate rather 
than under-estimate this value.

dl

> Did someone try a Zero build?
>
> -- Chris
>
> On Apr 23, 2013, at 1:09 AM, "Lindenmaier, Goetz" 
> <goetz.lindenmaier at sap.com <mailto:goetz.lindenmaier at sap.com>> wrote:
>
>> Hi all,
>> In the java_to_interp stub of CompiledStaticCalls one relocation is 
>> used to find
>> two code locations:
>> -the beginning of the stub
>> -the IC load instruction.
>> On x86 and sparc, the IC load instruction is the first instruction of 
>> the stub,
>> thus both locations have the same address.
>> On PPC, the IC load instruction is not the first instruction in the 
>> stub, thus
>> we must add an offset to the relocation to access it.
>> As the manipulation of  CompiledStaticCalls is in shared code, we 
>> would have
>> to introduce platform dependencies there.
>> This change moves the platform dependent code from CompiledStaticCall
>> to a new file compiledIC_<cpu>.cpp.  It also moves the code issuing the
>> stub from the ad file to this new file into the class CompiledStaticCall.
>> This removes redundancies between x86_64.ad and x86_32.ad.  Also,
>> ‘extern’ declarations in compile.cpp can be replaced by properly typed
>> Calls to methods in CompiledStaticCall.
>> Dean Long ran this change through JPRT with status OK (thanks!).
>> Please review and push this change.
>> http://cr.openjdk.java.net/~goetz/webrevs/8003853-3/ 
>> <http://cr.openjdk.java.net/%7Egoetz/webrevs/8003853-3/>
>> Best regards,
>>   Goetz Lindenmaier
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20130423/74d04957/attachment-0001.html