Add a restore folder, which shaves a dozen or so machineinstrs off oggenc.  Update a testcase to check this.

llvm-svn: 66029

 - Add Options.def file, collects option information.

 - Actual option instantiation is handled lazily by OptTable to allow
   the driver to not need to instantiate all options.

 - cast<> support for Option, other minor tweaks.

llvm-svn: 66028

llvm-svn: 66027

because less bytes are allocated and subobject construction is gone.
For reference how it works, see BasicBlock.h.
Btw. it is very assuring to see that somebody has invented
this ilist-embedded sentinel technique before me :-)

llvm-svn: 66026

preexpanded macro arguments from the preprocessor's bump pointer.
This reduces # mallocs from 12444 to 11792.

llvm-svn: 66025

llvm-svn: 66024

decls.  This reduces the number of calls to malloc on cocoa.h with pth and
-disable-free from 15958 to 12444 times (down ~3500).

llvm-svn: 66023

llvm-svn: 66022

llvm-svn: 66021

this, make DeclBase::Destroy destroy attributes instead of the DeclBase dtor.

llvm-svn: 66020

Linux.

llvm-svn: 66019

chosen sub-expression, rather than just evaluating the condition.

llvm-svn: 66018

This has two advantages 1) no more leaking them, 2) fewer calls to malloc.

This changes us from calling malloc 3685/1390/883/2974/1185 times respectively on
16/20/24/28/32 byte objects when parsing cocoa.h with pth and -disable-free to
calling it 2816/1020/702/2903/1168 times each respectively.

llvm-svn: 66017

llvm-svn: 66016

multiple sequential cases to a) not create tons of fall-through basic blocks
and b) not recurse deeply.  This fixes codegen on 100K deep cases, and improves
codegen on moderate cases from this:

        switch i32 %tmp, label %sw.epilog [
                i32 1000, label %sw.bb
                i32 1001, label %sw.bb1
                i32 1002, label %sw.bb2
                i32 1003, label %sw.bb3
                i32 1004, label %sw.bb4
...
sw.bb:          ; preds = %entry
        br label %sw.bb1

sw.bb1:         ; preds = %entry, %sw.bb
        br label %sw.bb2

sw.bb2:         ; preds = %entry, %sw.bb1
        br label %sw.bb3

sw.bb3:         ; preds = %entry, %sw.bb2
        br label %sw.bb4

to:

        switch i32 %tmp, label %sw.epilog [
                i32 1000, label %sw.bb
                i32 1001, label %sw.bb
                i32 1002, label %sw.bb
                i32 1003, label %sw.bb
                i32 1004, label %sw.bb
sw.bb:  ;; many preds

llvm-svn: 66015

llvm-svn: 66014

nested cases instead of recursive.

llvm-svn: 66013

booleans.

llvm-svn: 66012

multiple sequential case statements instead of doing it with recursion.  This
fixes a problem where we run out of stack space parsing 100K directly nested
cases.

There are a couple other problems that prevent this from being useful in 
practice (right now the example only parses correctly with -disable-free and
doesn't work with -emit-llvm), but this is a start.

I'm not including a testcase because it is large and uninteresting for 
regtesting.

Sebastian, I would appreciate it if you could scrutinize the smart pointer 
gymnastics I do.

llvm-svn: 66011

llvm-svn: 66010

first extend the integer to the correct width.

llvm-svn: 66009

llvm-svn: 66008

still give an unsupported error for them due to the fact this is a
work in progress.

llvm-svn: 66007

llvm-svn: 66006

Added the notion of a "boundable region", which is a region that can have a direct binding in the StoreManager.

llvm-svn: 66005

result from add, sub, inc, and dec instructions in simple cases.

llvm-svn: 66004

llvm-svn: 66003

reduces allocated Decl size by ~8% on Cocoa.h (~700k).

llvm-svn: 66002

llvm-svn: 66001

llvm-svn: 66000

per Chris' suggestion.  Slightly faster.

llvm-svn: 65999

Fix PR3701. 1. X86 target renamed eflags register to flags. This matches what llvm-gcc generates so codegen knows flags register is being clobbered by inline asm. 2. BURR scheduler should also check if inline asm nodes can clobber "live" physical registers. Previously it was only checking target nodes with implicit defs.

llvm-svn: 65996

If a global constant is dead then global's debug info should not prevent the optimizer in deleting the global. And while deleting global, delete global's debug info also.

llvm-svn: 65994

when counting work; it was only off by 1.

llvm-svn: 65993

While the patch is clearly correct in itself, it's become
apparent other places are assuming debug intrinsics are
marked as touching memory...this needs more testing.

llvm-svn: 65992

llvm-svn: 65988

llvm-svn: 65987

so it changed it into a 31 via the TLO.ShrinkDemandedConstant() call. Then it
would go through the DAG combiner again. This time it had a value of 31, which
was turned into a -1 by TLI.SimplifyDemandedBits(). This would ping pong
forever.

Teach the TLO.ShrinkDemandedConstant() call not to lower a value if the demanded
value is an XOR of all ones.

llvm-svn: 65985

pointer-to-pointer casts involving symbolic locations.

llvm-svn: 65984

llvm-svn: 65983