Updated the test case to use clang cc1 instead of just clang.

@lattner
Added Chris Lattner as a reviewer in case I did something silly in the target indep code.

Fixed some of the style issues.

Hi Eric,

Modifications based on review comments.

Ran a series of tests on this patch.
Overall the results are positive. I'm going to list the degradations.

Address comments from previous review.

I've updated the negative frame indices to use isFixedObjectIndex which is cleaner.

Committed in rL330117.
Closing this item.

After some code cleanup that was required for requiresFrameIndexScavenging here is the new version of this patch.
It should take into consideration all of the reviewer's comments.

@echristo
Hi Eric,
What is your opinion on this? Should I be adding the #if .. and #endif? Does it matter?

Implemented the cleanup for the loads as well.
Addressed comments from the reviewer.

Added the XFormMemOp flag to the td files for the instructions the require it. This flag indicates that this instructions is a memory instruction and that it is an X-Form instruction.
Used a script to compare the modifications against the ISA to make sure that the correct instructions have the flag set.

I've reverted the change in rL327143.

@echristo
Hi Eric,
Would you like me to pull this change out and fix the bug before I put it back in?

With the patch rL327021 we should now be able to mark Power 9 as complete.

Thank you for the very quick review! :)
I'll add more comments to the files as I work on them over the next couple of days.

I also learned that we accept regexes here when I started doing this a couple of days ago. :)

Addressed comments from reviewer and removed work that was not really related to this patch.
Will split out the work that was not really related into a separate patch.

Fixed the issue where we could have assigned a VSR register to an instruction that requires a VR register.

Modified the tuning formula to take into consideration the idea that we shouldn’t be enlarging the class of equal items.
After looking into the LSR algorithm we discovered that if we enlarge the space of equal items we do not favor unchanged code as we had initially desired.
Therefore, this change does almost exactly what the default isLSRCostLess function does (from the base class) except that number of instructions has been added as the most important category to compare.

Fixed the issues mentioned in previous comments.

@hfinkel
Hi Hal,
Yes that is correct. We are increasing the size of the class of equal elements.
The idea is that LSR will only replace one formula with another if it finds something with a lower cost. By returning false more often LSR will do fewer replacements and that is what we mean by more conservative. Maybe that's not really clear in the comment and I can update the comment to make it more clear.

I'm going to start looking at this patch now.

Yes, we are hand scheduling the prologue and epilogue.
However, this is a little more than moving instructions around. Since we are moving the stack pointer update instruction, spills that use the stack pointer may have to be updated with new offsets.

Fixed a typo and a missing set of brackets.

@hfinkel
Hi Hal,
Thank you for the approval. After doing a bit more investigation I realized that the condition should always be true. Before calling this function we always check that this is a return block. Therefore, I've change it slightly so that the guard is now an assert.

Hi Everyone,
Thank you for your comments.

Hi Eric,

Sorry for the late reply Eric. I was on vacation Thursday and Friday last week.