Fix prolific source of 'possible loss of data' warnings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36507 91177308-0d34-0410-b5e6-96231b3b80d8

ARM TLS: implement "general dynamic", "initial exec" and "local exec" models.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36506 91177308-0d34-0410-b5e6-96231b3b80d8

Darwin runtime library does not have these.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36505 91177308-0d34-0410-b5e6-96231b3b80d8

Special handling of LEApcrel and tLEApcrel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36504 91177308-0d34-0410-b5e6-96231b3b80d8

Back out previous check-in. Incorrect.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36503 91177308-0d34-0410-b5e6-96231b3b80d8

tLEApcrel is a AddrModeTs, i.e. pc relative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36502 91177308-0d34-0410-b5e6-96231b3b80d8

Expand UINT_TO_FP in turns of SINT_TO_FP when UINTTOFP_* libcalls are not available.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36501 91177308-0d34-0410-b5e6-96231b3b80d8

Test that eh_selector lowering accepts global variables as well as bitcasts
of global variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36500 91177308-0d34-0410-b5e6-96231b3b80d8

improve EH global handling, patch by Duncan Sands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36499 91177308-0d34-0410-b5e6-96231b3b80d8

Move ~Pass() from Pass.h into Pass.cpp

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36498 91177308-0d34-0410-b5e6-96231b3b80d8

Unxfail tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36497 91177308-0d34-0410-b5e6-96231b3b80d8

enable Anton's shift/and switch lowering stuff!  It now passes ppc bootstrap
successfully! woohoo...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36496 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1339 and CodeGen/X86/dollar-name.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36495 91177308-0d34-0410-b5e6-96231b3b80d8

Have MMX registers clobbered in x86-64 too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36494 91177308-0d34-0410-b5e6-96231b3b80d8

Delete Analysis Resolver.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36493 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR1339: http://llvm.org/PR1339

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36492 91177308-0d34-0410-b5e6-96231b3b80d8

Add test to proper place. Also, XFAIL until ppc bootstrap will be ok.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36491 91177308-0d34-0410-b5e6-96231b3b80d8

Fixx off-by-one bug, which prevents llvm-gcc bootstrap on ppc32

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36490 91177308-0d34-0410-b5e6-96231b3b80d8

Update a comment to reflect recent changes in the type system.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36486 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36485 91177308-0d34-0410-b5e6-96231b3b80d8

An example for PR1362

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36484 91177308-0d34-0410-b5e6-96231b3b80d8

Rename findRegisterUseOperand to findRegisterUseOperandIdx to avoid confusion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36483 91177308-0d34-0410-b5e6-96231b3b80d8

1. Don't swap byte order in scanf. It isn't necessary and leads to
   incorrect results (canonicalization was dropped several commits ago).
2. Add support for fscanf.
3. Suppress a warning about cast to pointer from non-pointer-sized integer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36482 91177308-0d34-0410-b5e6-96231b3b80d8

Use toplevel function pass manager as OnTheFly manager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36481 91177308-0d34-0410-b5e6-96231b3b80d8

bugfix: remember that ResNode was declared.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36477 91177308-0d34-0410-b5e6-96231b3b80d8

Add two test cases to cover apintification change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36476 91177308-0d34-0410-b5e6-96231b3b80d8

Using APInt more efficiently.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36475 91177308-0d34-0410-b5e6-96231b3b80d8

Rename identifier that GCC uses as a macro, breaking llvm-gcc build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36474 91177308-0d34-0410-b5e6-96231b3b80d8

Minor bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36473 91177308-0d34-0410-b5e6-96231b3b80d8

start code for writing out instructions.  Separate BB#s from normal value
#'s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36472 91177308-0d34-0410-b5e6-96231b3b80d8

start defining codes for instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36471 91177308-0d34-0410-b5e6-96231b3b80d8

regen

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36470 91177308-0d34-0410-b5e6-96231b3b80d8

wrap long lines

Fix PR1358 and test/Assembler/2007-04-25-AssemblerFoldExternWeak.ll, by
creating forward decl globals with linkage extern_weak to defeat implicit
constant folding.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36469 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36468 91177308-0d34-0410-b5e6-96231b3b80d8

enumerate BB's separately from other function values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36467 91177308-0d34-0410-b5e6-96231b3b80d8

For PR1355:
Make the script process multiple files and describe its usage properly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36466 91177308-0d34-0410-b5e6-96231b3b80d8

add support for incorporating and purging functions to the value enumerator

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36465 91177308-0d34-0410-b5e6-96231b3b80d8

improve a comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36464 91177308-0d34-0410-b5e6-96231b3b80d8

move some code around, fix a bug in the reader reading globalinits (which
I just introduced), stub out function reading, purge aggregate values from
the value table before reading functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36463 91177308-0d34-0410-b5e6-96231b3b80d8

add a new code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36462 91177308-0d34-0410-b5e6-96231b3b80d8

add bitcode alias support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36461 91177308-0d34-0410-b5e6-96231b3b80d8

add alias support to bitcode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36460 91177308-0d34-0410-b5e6-96231b3b80d8

print aliases before functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36459 91177308-0d34-0410-b5e6-96231b3b80d8

Be careful when to add implicit kill / dead operands. Don't add them during / post reg-allocation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36458 91177308-0d34-0410-b5e6-96231b3b80d8

PR1348 test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36457 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR1348. If stack inc / dec amount is > 32-bits, issue a series of add / sub instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36456 91177308-0d34-0410-b5e6-96231b3b80d8

New test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36455 91177308-0d34-0410-b5e6-96231b3b80d8

Quiet a compiler warning about assignment in an if-then conditional.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36454 91177308-0d34-0410-b5e6-96231b3b80d8

&& no longer needed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36453 91177308-0d34-0410-b5e6-96231b3b80d8

Match MachineFunction::UsedPhysRegs changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36452 91177308-0d34-0410-b5e6-96231b3b80d8

Change UsedPhysRegs from array bool to BitVector to save some space. Setting / getting its states now go through MachineFunction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36451 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for codegen bug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36450 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36449 91177308-0d34-0410-b5e6-96231b3b80d8

Add SSSE3 as a feature of Core2. Add MMX registers to the list of registers
clobbered by a call.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36448 91177308-0d34-0410-b5e6-96231b3b80d8

Data structure change to improve compile time (especially in debug mode).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36447 91177308-0d34-0410-b5e6-96231b3b80d8

Mem2Reg does not need TargetData.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36446 91177308-0d34-0410-b5e6-96231b3b80d8

This was lefted out. Fixed sumarray-dbl.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36445 91177308-0d34-0410-b5e6-96231b3b80d8

Mem2Reg does not need TargetData.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36444 91177308-0d34-0410-b5e6-96231b3b80d8

Add missed semicolon. Resotre compatibility with older bisons

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36443 91177308-0d34-0410-b5e6-96231b3b80d8

do the multiplication as signed, so that 2*-2 == -4 instead of 4294967292
when promoted to 64-bits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36442 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused function argument.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36441 91177308-0d34-0410-b5e6-96231b3b80d8

Note new header file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36440 91177308-0d34-0410-b5e6-96231b3b80d8

Add missed file

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36439 91177308-0d34-0410-b5e6-96231b3b80d8

remember to emit weak reference in one more case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36438 91177308-0d34-0410-b5e6-96231b3b80d8

Regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36437 91177308-0d34-0410-b5e6-96231b3b80d8

Implement aliases. This fixes PR1017 and it's dependent bugs. CFE part
will follow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36435 91177308-0d34-0410-b5e6-96231b3b80d8

Fix for PR1306.
- A register def / use now implicitly affects sub-register liveness but does
not affect liveness information of super-registers.
- Def of a larger register (if followed by a use later) is treated as
read/mod/write of a smaller register.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36434 91177308-0d34-0410-b5e6-96231b3b80d8

Test if a register is sub- or super-register of another.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36433 91177308-0d34-0410-b5e6-96231b3b80d8

Minor speed tweak.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36432 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36431 91177308-0d34-0410-b5e6-96231b3b80d8

Relex assertions to account for additional implicit def / use operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36430 91177308-0d34-0410-b5e6-96231b3b80d8

If an alloca only has two types of uses: 1) reads 2) a memcpy/memmove that
copies from a constant global, then we can change the reads to read from the
global instead of from the alloca.  This eliminates the alloca and the memcpy,
and promotes secondary optimizations (because the loads are now loads from
a constant global).

This is important for a common C idiom:

void foo() {
   int A[] = {1,2,3,4,5,6,7,8,9...};
   ... only reads of A ...
}

For some reason, people forget to mark the array static or const.

This triggers on these multisource benchmarks:
JM/ldecode: block_pos, [3 x [4 x [4 x i32]]]
FreeBench/mason: m, [18 x i32], inlined 4 times
MiBench/office-stringsearch: search_strings, [1332 x i8*]
MiBench/office-stringsearch: find_strings, [1333 x i8*]
Prolangs-C++/city: dirs, [9 x i8*], inlined 4 places

and these spec benchmarks:
177.mesa: message, [8 x [32 x i8]]
186.crafty: bias_rl45, [64 x i32]
186.crafty: diag_sq, [64 x i32]
186.crafty: empty, [9 x i8]
186.crafty: xlate, [15 x i8]
186.crafty: status, [13 x i8]
186.crafty: bdinfo, [25 x i8]
445.gobmk: routines, [16 x i8*]
458.sjeng: piece_rep, [14 x i8*]
458.sjeng: t, [13 x i32], inlined 4 places.
464.h264ref: block8x8_idx, [3 x [4 x [4 x i32]]]
464.h264ref: block_pos, [3 x [4 x [4 x i32]]]
464.h264ref: j_off_tab, [12 x i32]

This implements Transforms/ScalarRepl/memcpy-from-global.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36429 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase, should be able to eliminate the alloca and memcpy

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36428 91177308-0d34-0410-b5e6-96231b3b80d8

simplify some code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36427 91177308-0d34-0410-b5e6-96231b3b80d8

refactor the SROA code out into its own method, no functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36426 91177308-0d34-0410-b5e6-96231b3b80d8

support for >4G stack frames

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36425 91177308-0d34-0410-b5e6-96231b3b80d8

support for >4G frames

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36424 91177308-0d34-0410-b5e6-96231b3b80d8

support >4G stack frames

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36423 91177308-0d34-0410-b5e6-96231b3b80d8

support > 4G stack objects

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36422 91177308-0d34-0410-b5e6-96231b3b80d8

Undo my previous changes.  Since my approach to this problem is being revised,
this approach is no longer appropriate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36421 91177308-0d34-0410-b5e6-96231b3b80d8

allow support for 64-bit stack objects

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36420 91177308-0d34-0410-b5e6-96231b3b80d8

allow 64-bit stack objects

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36419 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36418 91177308-0d34-0410-b5e6-96231b3b80d8

Fix
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070423/048376.html

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36417 91177308-0d34-0410-b5e6-96231b3b80d8

New test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36416 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR1352

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36415 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36414 91177308-0d34-0410-b5e6-96231b3b80d8

Be more careful about folding op(x, undef) when we have vector operands.
This fixes CodeGen/X86/2007-04-24-VectorCrash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36413 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase that crashes llc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36412 91177308-0d34-0410-b5e6-96231b3b80d8

Add me

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36411 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1351 and CodeGen/PowerPC/2007-04-24-InlineAsm-I-Modifier.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36410 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR1351

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36409 91177308-0d34-0410-b5e6-96231b3b80d8

Testcases for MMX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36408 91177308-0d34-0410-b5e6-96231b3b80d8

Update.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36407 91177308-0d34-0410-b5e6-96231b3b80d8

Add the PADDQ to the list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36406 91177308-0d34-0410-b5e6-96231b3b80d8

Add the final MMX instructions. Correct a few wrong patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36405 91177308-0d34-0410-b5e6-96231b3b80d8

Remove some invalid instructions from this check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36404 91177308-0d34-0410-b5e6-96231b3b80d8

Support for the special case of a vector with the canonical form:

        vector_shuffle v1, v2, <2, 6, 3, 7>

I.e.

         vector_shuffle v, undef, <2, 2, 3, 3>

MMX only has a shuffle for v4i16 vectors. It needs to use the unpackh for
this type of operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36403 91177308-0d34-0410-b5e6-96231b3b80d8

Assertion when using a 1-element vector for an add operation. Get the
real vector type in this case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36402 91177308-0d34-0410-b5e6-96231b3b80d8

New test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@36401 91177308-0d34-0410-b5e6-96231b3b80d8