refactor these interfaces a bit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22281 91177308-0d34-0410-b5e6-96231b3b80d8

improve comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22280 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22279 91177308-0d34-0410-b5e6-96231b3b80d8

add a note about variable length array

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22278 91177308-0d34-0410-b5e6-96231b3b80d8

add a debug type

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22277 91177308-0d34-0410-b5e6-96231b3b80d8

remove a pseudo instruction, make ret always right, and fix vararg chains

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22276 91177308-0d34-0410-b5e6-96231b3b80d8

finally, Working varargs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22275 91177308-0d34-0410-b5e6-96231b3b80d8

more complete Lowering for vacopy and vaarg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22274 91177308-0d34-0410-b5e6-96231b3b80d8

If we support structs as va_list, we must pass pointers to them to va_copy
See last commit for LangRef, this implements it on all targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22273 91177308-0d34-0410-b5e6-96231b3b80d8

SelectionDAG is very unhappy when the argument to an intrinsic is a struct

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22272 91177308-0d34-0410-b5e6-96231b3b80d8

Make it easier to find alpha stuff in doxygen, and fixup labeling
of memory instructions in the assembly, to allow later linking
of traces with LLVM Value*s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22271 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22270 91177308-0d34-0410-b5e6-96231b3b80d8

because some functions just use va_start and pass things to vfprintf

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22269 91177308-0d34-0410-b5e6-96231b3b80d8

the correct fix was to fix AliasAnalysis.getModRefInfo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22268 91177308-0d34-0410-b5e6-96231b3b80d8

make AliasAnalysis know that VAArg writes memory.  This is extremely conservative and should be fixed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22267 91177308-0d34-0410-b5e6-96231b3b80d8

prevent GCSE from forwarding stores to loads around vaarg.  This is uggly, and I am trying to fix the AliasInfo, as it should catch the problem instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22266 91177308-0d34-0410-b5e6-96231b3b80d8

prevent va_arg from being hoisted from a loop

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22265 91177308-0d34-0410-b5e6-96231b3b80d8

fix DCE + va_arg problem

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22264 91177308-0d34-0410-b5e6-96231b3b80d8

prevent DCE of vaarg intrinsics.  This should take care of most regressions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22263 91177308-0d34-0410-b5e6-96231b3b80d8

va_end fix

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22262 91177308-0d34-0410-b5e6-96231b3b80d8

so this doesn't crash when run.  It is hard to tell if things are right enough to work correctly with all the TmpInstructions running around

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22261 91177308-0d34-0410-b5e6-96231b3b80d8

add a check for the mixing of vaarg and vanext with va_arg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22260 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore the configure.out file generated by "make reconfigure"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22259 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore some generated files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22258 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore generated files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22257 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this test case to actually test the new va_arg instruction not the
old vaarg instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22256 91177308-0d34-0410-b5e6-96231b3b80d8

OK, at least get rid of old stuff, and mark what needs to be fixed for V9

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22255 91177308-0d34-0410-b5e6-96231b3b80d8

core changes for varargs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22254 91177308-0d34-0410-b5e6-96231b3b80d8

header file changes for varargs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22253 91177308-0d34-0410-b5e6-96231b3b80d8

new vararg test

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22252 91177308-0d34-0410-b5e6-96231b3b80d8

OK, after checking the backwards compatibility code on X86 and the new code
path on alpha, now has come the time for new vararg support.  So, with out
further ado, I revert behavior back a couple of years!

Well, ok, I lied.  A few more notes.

First, the Simple ISels cannot be expected to work any longer, but they
should still compile

Second, there are likely some bugs to track down once the nightly testers
start with this.

Third, the initial patch doesn't include sparcv9, but I'll do that today.

Forth, subsequent patches won't bother being long winded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22251 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a problem with the strcmp optimization checking the wrong string and
not casting to the correct type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22250 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up some uninitialized variables and missing return statements that
GCC 4.0.0 compiler (sometimes incorrectly) warns about under release build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22249 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure a variable is initialized before use to clean up a warning from
GCC 4.0.0 in release build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22248 91177308-0d34-0410-b5e6-96231b3b80d8

A start at a Sparc V8 Pattern ISel.  Anyone want to implement the calling
convention? ;)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22247 91177308-0d34-0410-b5e6-96231b3b80d8

silence incredibly braindead GCC 4 warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22246 91177308-0d34-0410-b5e6-96231b3b80d8

silence a bogus warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22245 91177308-0d34-0410-b5e6-96231b3b80d8

silence a huge number of bogus warnings with GCC 4 on Reid's tester.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22244 91177308-0d34-0410-b5e6-96231b3b80d8

Removed IIIi specific changes. This should be fixed to add floating point deps for the IIi.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22243 91177308-0d34-0410-b5e6-96231b3b80d8

Special dep graph for SMS for superblocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22242 91177308-0d34-0410-b5e6-96231b3b80d8

Special versions of the dep graph and scheduled for SMS for superblocks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22241 91177308-0d34-0410-b5e6-96231b3b80d8

Added statistic to count number of spills.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22240 91177308-0d34-0410-b5e6-96231b3b80d8

Numerous bug fixes and the completed modschedSB algorithm (minor bugs still exist for course).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22239 91177308-0d34-0410-b5e6-96231b3b80d8

Added SMS for superblocks as an option (experimental)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22238 91177308-0d34-0410-b5e6-96231b3b80d8

Added ModuloSchedSB as a friend class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22237 91177308-0d34-0410-b5e6-96231b3b80d8

This is not true: (X != 13 | X < 15) -> X < 15

It is actually always true.  This fixes PR586 and
Transforms/InstCombine/2005-06-16-SetCCOrSetCCMiscompile.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22236 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR586

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22235 91177308-0d34-0410-b5e6-96231b3b80d8

Don't crash when dealing with INTMIN.  This fixes PR585 and
Transforms/InstCombine/2005-06-16-RangeCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22234 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR585

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22233 91177308-0d34-0410-b5e6-96231b3b80d8

Don't crash on:  X = phi (X, X).

This fixes PR584 and Transforms/SimplifyCFG/2005-06-16-PHICrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22232 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR584

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22231 91177308-0d34-0410-b5e6-96231b3b80d8

avoid constructing out of range shift amounts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22230 91177308-0d34-0410-b5e6-96231b3b80d8

Fix bug 537 test 2, which checks to make sure that we fold A+(B-A) -> B for
integer types.  Add a couple checks to not perform these kinds of transform
on floating point values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22228 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR583 and testcase Transforms/InstCombine/2005-06-15-DivSelectCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22227 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR583

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22226 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR571, removing code that does just the WRONG thing :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22225 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bug in my previous patch.  Do not get the shift amount type (which
is always ubyte, get the type being shifted).  This unbreaks espresso

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22224 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR575, patch provided by John Mellor-Crummey.  Thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22223 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR580

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22222 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR582.  The rewriter can move casts around, which invalidated the
BB iterator.  This fixes Transforms/IndVarsSimplify/2005-06-15-InstMoveCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22221 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR582

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22220 91177308-0d34-0410-b5e6-96231b3b80d8

Do not promote globals only used by main to locals if there are constantexprs
or other uses hanging off of them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22219 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR579

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22218 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR577 and testcase InstCombine/2005-06-15-ShiftSetCCCrash.ll.
Do not perform undefined out of range shifts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22217 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for PR577

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22216 91177308-0d34-0410-b5e6-96231b3b80d8

Don't lie to the casual reader about Value subclassing from Annotable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22215 91177308-0d34-0410-b5e6-96231b3b80d8

Add some operators the PowerPC backend needs to efficiently and correctly
generate conditional branches.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22214 91177308-0d34-0410-b5e6-96231b3b80d8

Put the hack back in that removes features, causes regressions to fail, but
allows test programs to succeed. Actual fix for this is forthcoming.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22213 91177308-0d34-0410-b5e6-96231b3b80d8

Commit fix for generating conditional branch pseudo instructions that
avoids dereferencing the end() iterator when selecting the fallthrough
block.  This requires an ilist change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22212 91177308-0d34-0410-b5e6-96231b3b80d8

nothing funny about building the CFE on Linux/Itanium

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22211 91177308-0d34-0410-b5e6-96231b3b80d8

Commit a small improvement that is already in the x86 and ia64 backends to
not generate unnecessary register copies.  This improves compile time by
2-5% depending on the test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22210 91177308-0d34-0410-b5e6-96231b3b80d8

Unbreak several InstCombine regression checks introduced by a hack to
fix the bzip2 test. A better hack is needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22209 91177308-0d34-0410-b5e6-96231b3b80d8

When compiled with GCC 4.0, a latent bug was exposed where both SparcV9
and the target independant register allocator were both using a class named
'LiveRange'.  This lead to the target independant code calling code in the
SparcV9 backend, which crashed.  Fixed by renaming SparcV9's LiveRange to
V9LiveRange.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22208 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a memory smasher caught by Mac OS X's debug malloc library.  We were
incorrectly using an iterator after it was invalid.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22207 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a 64-bit problem, passing (int)0 through ... instead of (void*)0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22206 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a problem on 64-bit targets where we passed (int)0 through ... instead of
(void*)0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22205 91177308-0d34-0410-b5e6-96231b3b80d8

make SparcV8 and V9 seperately configurable

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22204 91177308-0d34-0410-b5e6-96231b3b80d8

fix BranchCC with a setCC with an arg of 0

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22203 91177308-0d34-0410-b5e6-96231b3b80d8

For PR572:
Undefine the PPC symbol which is defined by Linux/PPC (erroneously) so it
doesn't pollute the user namespace and clash with our namespace declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22202 91177308-0d34-0410-b5e6-96231b3b80d8

Handle some more real world cases of rlwimi.  These don't come up that
regularly in "normal" code, but for things like software graphics, they
make a big difference.

For the following code:
unsigned short Trans16Bit(unsigned srcA,unsigned srcB,unsigned alpha)
{
unsigned tmpA,tmpB,mixed;
tmpA = ((srcA & 0x03E0) << 15) | (srcA & 0x7C1F);
tmpB = ((srcB & 0x03E0) << 15) | (srcB & 0x7C1F);
mixed = (tmpA * alpha) + (tmpB * (32 - alpha));
return ((mixed >> 5) & 0x7C1F) | ((mixed >> 20) & 0x03E0);
}

We now generate:
_Trans16Bit:
.LBB_Trans16Bit_0:      ; entry
        andi. r2, r4, 31775
        rlwimi r2, r4, 15, 7, 11
        subfic r4, r5, 32
        mullw r2, r2, r4
        andi. r4, r3, 31775
        rlwimi r4, r3, 15, 7, 11
        mullw r3, r4, r5
        add r2, r2, r3
        srwi r3, r2, 5
        andi. r3, r3, 31775
        rlwimi r3, r2, 12, 22, 26
        blr

Instead of:
_Trans16Bit:
.LBB_Trans16Bit_0:      ; entry
        slwi r2, r4, 15
        rlwinm r2, r2, 0, 7, 11
        andi. r4, r4, 31775
        or r2, r2, r4
        subfic r4, r5, 32
        mullw r2, r2, r4
        slwi r4, r3, 15
        rlwinm r4, r4, 0, 7, 11
        andi. r3, r3, 31775
        or r3, r4, r3
        mullw r3, r3, r5
        add r2, r2, r3
        srwi r3, r2, 5
        andi. r3, r3, 31775
        srwi r2, r2, 20
        rlwimi r3, r2, 0, 22, 26
        blr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22201 91177308-0d34-0410-b5e6-96231b3b80d8

Fix lli linking on Mac OS X 10.4.1 for Intel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22200 91177308-0d34-0410-b5e6-96231b3b80d8

Remove code for conversion from old style va_args.  Preparing the way for
returning to the old style :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22199 91177308-0d34-0410-b5e6-96231b3b80d8

We also have V8, so do not force every Sparc to be treated like a V9.  It is
still possible to force V9 (even if configure doesn't think it's one) via
``./configure --target=sparcv9-sun-solaris2.8'' so nothing is lost.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22198 91177308-0d34-0410-b5e6-96231b3b80d8

* Add ability to specify the target LLVM will compile for via configure
* Minor whitespace cleanups

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22197 91177308-0d34-0410-b5e6-96231b3b80d8

Fix spelling of `correlate'

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22196 91177308-0d34-0410-b5e6-96231b3b80d8

allow marking of loads and stores in the instruction stream with enough information to reconstruct the Value* if it existed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22195 91177308-0d34-0410-b5e6-96231b3b80d8

hide basic block labels.  The utility of these for debuging is long since passed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22194 91177308-0d34-0410-b5e6-96231b3b80d8

* Replace block of commented-out lines with #if 0
* Remove warning "control reaches end of non-void function"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22193 91177308-0d34-0410-b5e6-96231b3b80d8

hack to fix bzip2 (bug 571)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22192 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the definitions of LLVMGCC and LLVMGXX to include the EXEEXT (the
.exe extension) on Cygwin. This fixes the last few remaining Cygwin
issues. Thanks to Aaron Gray for tracking this down.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22191 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure that Cygwin assembly includes _ as part of function names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22190 91177308-0d34-0410-b5e6-96231b3b80d8

Put in a hack for Cygwin that prevents mkdtemp from being used since
configure seems to find it on Cygwin but linking against it fails.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22189 91177308-0d34-0410-b5e6-96231b3b80d8

try custom expanders, doesn't seem to expand yet, so disabled

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22188 91177308-0d34-0410-b5e6-96231b3b80d8

switch to the new live in thing.  Really, this time it works

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22187 91177308-0d34-0410-b5e6-96231b3b80d8

switch to the new live in thing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22186 91177308-0d34-0410-b5e6-96231b3b80d8

switch to the new live in thing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22185 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure the tools get built if no build target is specified.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22184 91177308-0d34-0410-b5e6-96231b3b80d8

match gcc, makes diff easier

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22179 91177308-0d34-0410-b5e6-96231b3b80d8

Fix 2005-05-12-Int64ToFP

The issue is there is no unsigned -> double conversion, only signed.  So I
need to test the sign and do a different thing depending on it.  Ideally
this should be in a different BB, but in the mean time, I use a branch
free method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22177 91177308-0d34-0410-b5e6-96231b3b80d8

Add a -cvstag option for testing specific branches, such as release candidates

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22176 91177308-0d34-0410-b5e6-96231b3b80d8