Re: [Jikesrvm-researchers] One more question in OPT_StackManager.java

[David P G. <gr...@us...>]

Steve and I rewrote this code in late November.  I believe in the current
system you would not have this problem (redefining the two constants in
VM_StackFrameLayoutConstants should now be sufficient to add another header
word below the framepointer in an IA32 stackframe....let us know if in fact
it isn't).

--dave



|---------+-------------------------------------------------------->
|         |           Feng QIAN <fq...@sa...>            |
|         |           Sent by:                                     |
|         |           jik...@ww...uthbury.|
|         |           usf.ibm.com                                  |
|         |                                                        |
|         |                                                        |
|         |           01/21/2002 09:45 PM                          |
|         |           Please respond to jikesrvm-researchers       |
|         |                                                        |
|---------+-------------------------------------------------------->
  >------------------------------------------------------------------------------------------|
  |                                                                                          |
  |       To:       <jik...@ww...>                     |
  |       cc:                                                                                |
  |       Subject:  Re: [Jikesrvm-researchers] One more question in OPT_StackManager.java    |
  |                                                                                          |
  |                                                                                          |
  >------------------------------------------------------------------------------------------|



I got it, and actually I found that it is pretty tricky to add more
stuff in the prologue, the method assumes that the difference of
ESP and EBP after "mir_esp_prologue" is one word for CMID, but my code
used another slot, so in addition to increase the STACKFRAME_BODY_OFFSET,
the line in OPT_StackManager also needs adjustment. Probably that's only
my case requires such changes.

private void rewriteStackLocations(OPT_BasicBlock bb) {
  OPT_Register ESP = ir.regpool.getPhysicalRegisterSet().getESP();

  for (Enumeration e = bb.forwardInstrEnumerator(); e.hasMoreElements();) {
    OPT_Instruction s = (OPT_Instruction)e.nextElement();

    // After the prologue, ESP is set to the top of the frame.
    if (s.operator() == MIR_ESP_PROLOGUE) {

      ESPOffset = getFrameFixedSize() - STACKFRAME_ESP_FP_OFFSET;
      //           ESPOffset = getFrameFixedSize() - WORDSIZE;
      continue;
    }

New prologue:

-1          ia32_push               ebp(int)
-1          ia32_mov                ebp(int) = esp(int)
-1          ia32_push               6474
-1          ia32_push               -1
-1          mir_esp_prologue

Cheers,
====================
Feng Qian
fq...@sa...

On Mon, 21 Jan 2002, David P Grove wrote:

>
> If you change STACKFRAME_HEADER_SIZE, you also need to change
> STACKFRAME_BODY_OFFSET.  Steve and I think that is probably what is going
> wrong (STACKFRAME_BODY_OFFSET is used in VM_OptGCMapiterator, so if you
> changed one but not the other than the first GC with an opt compiled
frame
> on a thread stack is very likely to die).
>
> --dave
>
> _______________________________________________
> Jikesrvm-researchers mailing list
> Jik...@ww...
>
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-researchers

>

_______________________________________________
Jikesrvm-researchers mailing list
Jik...@ww...
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jikesrvm-researchers