Merge PPC instructions FMRS and FMRD into a single FMR instruction.

This is possible because F8RC is a subclass of F4RC. We keep FMRSD around so
fextend has a pattern.

Also allow folding of memory operands on FMRSD.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97275 91177308-0d34-0410-b5e6-96231b3b80d8

Fix grammaro in a comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97273 91177308-0d34-0410-b5e6-96231b3b80d8

Use the right svn:keywords value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97271 91177308-0d34-0410-b5e6-96231b3b80d8

Add Revision keywords to these files, as it's common for them to be
copied out of the source tree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97270 91177308-0d34-0410-b5e6-96231b3b80d8

Comment typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97269 91177308-0d34-0410-b5e6-96231b3b80d8

don't build edis if the x86 target isn't enabld.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97268 91177308-0d34-0410-b5e6-96231b3b80d8

Improve the vim code for highlighting trailing whitespace and lines
longer than 80 columns. This replaces the heavy-handed "textwidth"
mechanism, and makes the trailing-whitespace highlighting lazy so
that it isn't constantly jumping on the user during typing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97267 91177308-0d34-0410-b5e6-96231b3b80d8

Test autoupdate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97266 91177308-0d34-0410-b5e6-96231b3b80d8

Test auto update.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97265 91177308-0d34-0410-b5e6-96231b3b80d8

Add the alignstack keyword.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97264 91177308-0d34-0410-b5e6-96231b3b80d8

Remove bogus Updated line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97263 91177308-0d34-0410-b5e6-96231b3b80d8

Use the right floating point load/store instructions in PPCInstrInfo::foldMemoryOperandImpl().

The PowerPC floating point registers can represent both f32 and f64 via the
two register classes F4RC and F8RC. F8RC is considered a subclass of F4RC to
allow cross-class coalescing. This coalescing only affects whether registers
are spilled as f32 or f64.

Spill slots must be accessed with load/store instructions corresponding to the
class of the spilled register. PPCInstrInfo::foldMemoryOperandImpl was looking
at the instruction opcode which is wrong.

X86 has similar floating point register classes, but doesn't try to fold
memory operands, so there is no problem there.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97262 91177308-0d34-0410-b5e6-96231b3b80d8

Remove dead code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97261 91177308-0d34-0410-b5e6-96231b3b80d8

Remove REQUIRES_EH from the suggested set of make variables, since users can
build with exceptions even if LLVM is built without.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97260 91177308-0d34-0410-b5e6-96231b3b80d8

Replace a temporary std::string with SmallString.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97259 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typos in comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97257 91177308-0d34-0410-b5e6-96231b3b80d8

Delete empty directories.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97256 91177308-0d34-0410-b5e6-96231b3b80d8

Move dbg_value generation to target-independent FastISel,
as X86 is currently the only FastISel target.  Per review.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97255 91177308-0d34-0410-b5e6-96231b3b80d8

Further constify MDNode* references.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97252 91177308-0d34-0410-b5e6-96231b3b80d8

Add type printing for Metadata pseudo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97251 91177308-0d34-0410-b5e6-96231b3b80d8

Move the EnableFullLoadPRE flag from a separate command-line option to an
argument of createGVNPass and set it automatically for -O3.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97245 91177308-0d34-0410-b5e6-96231b3b80d8

disable-mem2reg and disable-gvn options should not be used by the driver.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97236 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused "NoPRE" parameter in GVN and createGVNPass().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97235 91177308-0d34-0410-b5e6-96231b3b80d8

pass in more section kinds, enough to get the .align 0x90
stuff to emit optimal nops in the right places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97233 91177308-0d34-0410-b5e6-96231b3b80d8

The cloner has nothing to do if any of the main or ISR entrypoints are not
present in the module.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97232 91177308-0d34-0410-b5e6-96231b3b80d8

fix PR6435 another bug from the MallocInst elimination work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97231 91177308-0d34-0410-b5e6-96231b3b80d8

Set svn:keywords property on docs/Packaging.html.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97230 91177308-0d34-0410-b5e6-96231b3b80d8

Add to the packaging advice.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97229 91177308-0d34-0410-b5e6-96231b3b80d8

Reapply things reverted back in 97220, with the fixed test case.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97228 91177308-0d34-0410-b5e6-96231b3b80d8

movl is a cheaper way to materialize 0 without clobbering EFLAGS than movabsq.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97227 91177308-0d34-0410-b5e6-96231b3b80d8

Fix XCoreTargetLowering::isLegalAddressingMode() to handle VoidTy.
Previously LoopStrengthReduce would sometimes be unable to find
a legal formula, causing an assertion failure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97226 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r97211 and r97213 to get the build green again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97220 91177308-0d34-0410-b5e6-96231b3b80d8

add some helpful comments to the emitter

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97219 91177308-0d34-0410-b5e6-96231b3b80d8

switch from my nice hashtable based merging solution to a
gross little neighbor merging implementation.  This one has
the benefit of not violating the ordering of patterns, so it
generates code that passes tests again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97218 91177308-0d34-0410-b5e6-96231b3b80d8

fix same bug in CheckChainCompatibleMatcher::isEqualImpl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97217 91177308-0d34-0410-b5e6-96231b3b80d8

fix a nasty bug in CheckTypeMatcher::isEqualImpl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97216 91177308-0d34-0410-b5e6-96231b3b80d8

finish off the factoring optimization along the lines of the
current design.  This generates a matcher that successfully
runs, but it turns out that the factoring we're doing violates
the ordering of patterns, so we end up matching (e.g.) movups
where we want movaps.  This won't due, but I'll address this in
a follow on patch.  It's nice to not be on by default yet! :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97215 91177308-0d34-0410-b5e6-96231b3b80d8

add a new setNumChildren method for resizing scopes.  Tweak getHash() so
that we never return a tombstone value, which (thankfully) triggers an
assert in densemap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97214 91177308-0d34-0410-b5e6-96231b3b80d8

Currently in LLVM, names of libcalls are assigned during TargetLowering
object construction. There is no provision to change them when the
code for a function generated.
So we have to change these names while printing assembly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97213 91177308-0d34-0410-b5e6-96231b3b80d8

fix the matcher in the presence of multiple scopes: we need to save
and restore the entire matcher stack by value.  This is because children
we're testing could do moveparent or other things besides just
scribbling on additions to the stack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97212 91177308-0d34-0410-b5e6-96231b3b80d8

Before converting an operand to mem, check if it is legal to do so.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97211 91177308-0d34-0410-b5e6-96231b3b80d8

this file lacks a run line!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97208 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97206 91177308-0d34-0410-b5e6-96231b3b80d8

Delete a bunch of redundant predicates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97201 91177308-0d34-0410-b5e6-96231b3b80d8

Fix comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97200 91177308-0d34-0410-b5e6-96231b3b80d8

Give packagers some advice about how to build LLVM so it's useful to
the most users.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97198 91177308-0d34-0410-b5e6-96231b3b80d8

Add another (and hopefully the last) exception case, where once we recalculate
the alignment requirement, if it no longer makes the TType base offset overflow
into extra bytes, then we need to pad to those bytes ourselves.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97196 91177308-0d34-0410-b5e6-96231b3b80d8

And should use the correct variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97193 91177308-0d34-0410-b5e6-96231b3b80d8

Got assertion check backwards.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97192 91177308-0d34-0410-b5e6-96231b3b80d8

Catch a corner case where adding the padding to the "TType base offset" field
will eliminate the need for padding in the "Call site table length". E.g., if
we have this:

    GCC_except_table1:
    Lexception1:
        .byte   0xff  ## @LPStart Encoding = omit
        .byte   0x9b  ## @TType Encoding = indirect pcrel sdata4
        .byte   0x7f  ## @TType base offset
        .byte   0x03  ## Call site Encoding = udata4
        .byte   0x89  ## Call site table length

with padding of 1. We want to emit the padding like this:

    GCC_except_table1:
    Lexception1:
        .byte   0xff  ## @LPStart Encoding = omit
        .byte   0x9b  ## @TType Encoding = indirect pcrel sdata4
        .byte   0xff  ## @TType base offset
        .space  1,0   ## Padding
        .byte   0x03  ## Call site Encoding = udata4
        .byte   0x89  ## Call site table length

and not with padding on the "Call site table length" entry.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97183 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a few more typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97182 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97181 91177308-0d34-0410-b5e6-96231b3b80d8

Added SAFECode (safecode) to the list of projects to automatically configure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97179 91177308-0d34-0410-b5e6-96231b3b80d8

rewrite OptimizeGlobalAddressOfMalloc to fix PR6422, some bugs
introduced when mallocinst was eliminated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97178 91177308-0d34-0410-b5e6-96231b3b80d8

tests: Propogate the HOME environment variable through to tests. I'm ambivalent
about this, but it can be useful for users who use ccache, since the LLVMC tests
are fond of calling gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97171 91177308-0d34-0410-b5e6-96231b3b80d8

Fix HTML.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97170 91177308-0d34-0410-b5e6-96231b3b80d8

Make comment more meaningful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97169 91177308-0d34-0410-b5e6-96231b3b80d8

Add svn:ignore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97167 91177308-0d34-0410-b5e6-96231b3b80d8

Fix ExpandVectorBuildThroughStack for the case where the
operands are themselves vectors. Based on a patch by
Micah Villmow for PR6338.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97165 91177308-0d34-0410-b5e6-96231b3b80d8

Added the following 32-bit Thumb instructions for disassembly only: SMC, RFE,
and SRS.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97164 91177308-0d34-0410-b5e6-96231b3b80d8

Added the 32-bit Thumb instructions (BXJ) for disassembly only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97163 91177308-0d34-0410-b5e6-96231b3b80d8

change the scope node to include a list of children to be checked
instead of to have a chained series of scope nodes.  This makes
the generated table smaller, improves the efficiency of the
interpreter, and make the factoring optimization much more
reasonable to implement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97160 91177308-0d34-0410-b5e6-96231b3b80d8

Added the 32-bit Thumb instructions (MRS and MSR) for disassembly only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97159 91177308-0d34-0410-b5e6-96231b3b80d8

This is a patch to the assembler frontend to detect when aligning a text
section with TextAlignFillValue and calls EmitCodeAlignment() instead of
calling EmitValueToAlignment().  This allows x86 assembly code to be aligned
with optimal nops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97158 91177308-0d34-0410-b5e6-96231b3b80d8

stkrc is gone.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97156 91177308-0d34-0410-b5e6-96231b3b80d8

Add the union keyword.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97155 91177308-0d34-0410-b5e6-96231b3b80d8

Merge the advanced getelementptr FAQ into the regular
getelementptr FAQ.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97154 91177308-0d34-0410-b5e6-96231b3b80d8

Fix TextAlignFillValue in a few places

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97151 91177308-0d34-0410-b5e6-96231b3b80d8

Add simple script for finding most-recent-rev-before-N in a git-svn repo; useful
when bisecting multiple repos in sync.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97150 91177308-0d34-0410-b5e6-96231b3b80d8

Added the following 16-bit Thumb instructions for disassembly only: YIELD, WFE,
WFI, SEV, SETEND.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97149 91177308-0d34-0410-b5e6-96231b3b80d8

remove a dead PatLeaf, I previously changed all uses to use -1 instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97148 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97144 91177308-0d34-0410-b5e6-96231b3b80d8

Clarify the description of pointer types, and move the
address space content to its own paragraph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97143 91177308-0d34-0410-b5e6-96231b3b80d8

Teach the constant folder about union types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97142 91177308-0d34-0410-b5e6-96231b3b80d8

Remove code which assumes it knows how vectors are stored in memory.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97141 91177308-0d34-0410-b5e6-96231b3b80d8

Add more information to the getSizeOf comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97140 91177308-0d34-0410-b5e6-96231b3b80d8

Remove this paragraph. Vectors may not always have the same layout as
arrays now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97139 91177308-0d34-0410-b5e6-96231b3b80d8

Revert r97064. Duncan pointed out that bitcasts are defined in
terms of store and load, which means bitcasting between scalar
integer and vector has endian-specific results, which undermines
this whole approach.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97137 91177308-0d34-0410-b5e6-96231b3b80d8

Make the side-numbering of instructions used by metadata (which is needed to
keep track of instructions that return void) per-function. This fixes PR5278.

This breaks backwards compatibility with the metadata format. That's okay
because we haven't released the metadata bitcode yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97132 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the first half of redundancy factoring: efficiently
splitting all the patterns under scope nodes into equality sets
based on their first node.  The second step is to rewrite the
graph info a form that exposes the sharing.  Before I do this,
I want to redesign the Scope node.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97130 91177308-0d34-0410-b5e6-96231b3b80d8

Make LoopSimplify change conditional branches in loop exiting blocks
which branch on undef to branch on a boolean constant for the edge
exiting the loop. This helps ScalarEvolution compute trip counts for
loops.

Teach ScalarEvolution to recognize single-value PHIs, when safe, and
ForgetSymbolicName to forget such single-value PHI nodes as apprpriate
in ForgetSymbolicName.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97126 91177308-0d34-0410-b5e6-96231b3b80d8

factor the print method better.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97125 91177308-0d34-0410-b5e6-96231b3b80d8

Dump the presence of attached metadata even if we don't know what it is. This
format is not parsable, even if the module is legal. To get parsable output,
dump the module instead of the function or smaller, since metadata kind are
attached to the module (not the context).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97124 91177308-0d34-0410-b5e6-96231b3b80d8

add methods to do equality checks and get hashes of Matchers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97123 91177308-0d34-0410-b5e6-96231b3b80d8

Modernize comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97121 91177308-0d34-0410-b5e6-96231b3b80d8

Correct whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97120 91177308-0d34-0410-b5e6-96231b3b80d8

Try r96559 for the third time.  This time the shared library is only built if
--enable-shared is passed to configure.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97119 91177308-0d34-0410-b5e6-96231b3b80d8

Each field of auxiliary debug entry is only 1 byte long.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97108 91177308-0d34-0410-b5e6-96231b3b80d8

Added tNOP for disassembly only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97105 91177308-0d34-0410-b5e6-96231b3b80d8

Truncate from i64 to i32 is "free" on x86-32, because it involves
just discarding one of the registers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97100 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this patch for the time being. Needs more testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97099 91177308-0d34-0410-b5e6-96231b3b80d8

Added tSVC and tTRAP for disassembly only.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97098 91177308-0d34-0410-b5e6-96231b3b80d8

formatting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97097 91177308-0d34-0410-b5e6-96231b3b80d8

rename fooMatcherNode to fooMatcher.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97096 91177308-0d34-0410-b5e6-96231b3b80d8

add some noop code to push it out of my tree.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97094 91177308-0d34-0410-b5e6-96231b3b80d8

rename PushMatcherNode -> ScopeMatcherNode to more accurately
reflect what it does.  Switch the sense of the Next and the Check
arms to be more logical.  No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97093 91177308-0d34-0410-b5e6-96231b3b80d8

Large stack frame patch for the CellSPU: handle stack frames that exceed 8176
(511*16) bytes register displacement (D-form).

NOTE: This is a potential headache, given the SPU's local core limitations,
allowing the software developer to commit stack overrun suicide unknowingly.
Also, large SPU stack frames will cause code size explosion. But, one presumes
that the software developer knows what they're doing...

Contributed by Kalle.Raiskila@nokia.com, edited slightly before commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97091 91177308-0d34-0410-b5e6-96231b3b80d8

Filter the future all-of-llvm shared library out of the llvm-config
results.  I'm checking this in before the shared library so that I can
tell if it breaks anything on its own.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97089 91177308-0d34-0410-b5e6-96231b3b80d8

MC'ize padding when padding the ULEB128 value.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97087 91177308-0d34-0410-b5e6-96231b3b80d8

add a fixme for an experiment that defeated me for the time being.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97081 91177308-0d34-0410-b5e6-96231b3b80d8