unbreak the build

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24339 91177308-0d34-0410-b5e6-96231b3b80d8

Separate the type-became-concrete case from the type-is-resolved case, the
former of which takes much less work than the later.  This speeds up linking
eon from 3.749 to 3.637s with a release build (about 3%).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24338 91177308-0d34-0410-b5e6-96231b3b80d8

enable LSR by default on alpha

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24337 91177308-0d34-0410-b5e6-96231b3b80d8

fix more regressions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24335 91177308-0d34-0410-b5e6-96231b3b80d8

fix READCYCLECOUNTER

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24334 91177308-0d34-0410-b5e6-96231b3b80d8

this is no apparently passing on alpha

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24331 91177308-0d34-0410-b5e6-96231b3b80d8

trivial cleanup

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24330 91177308-0d34-0410-b5e6-96231b3b80d8

Fix reading LLVM 1.3 bc files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24329 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24328 91177308-0d34-0410-b5e6-96231b3b80d8

refactor grammar to eliminate shift-reduce conflict.  Move alignment checking
code out of all of the clients and into OptAlign/OptCAlign

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24327 91177308-0d34-0410-b5e6-96231b3b80d8

Shrink the Type class from 32 to 28 bytes on Darwin (which has silly 32-bit
bools).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24326 91177308-0d34-0410-b5e6-96231b3b80d8

remove a dead method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24325 91177308-0d34-0410-b5e6-96231b3b80d8

Unswitch a loop, add more assertions, eliminate use of removeUserFromConcrete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24324 91177308-0d34-0410-b5e6-96231b3b80d8

Tiny speedup, improve comments and code organization.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24323 91177308-0d34-0410-b5e6-96231b3b80d8

Describe section name encoding

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24322 91177308-0d34-0410-b5e6-96231b3b80d8

Read and write section info from/to .bc files

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24321 91177308-0d34-0410-b5e6-96231b3b80d8

Wrap long line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24320 91177308-0d34-0410-b5e6-96231b3b80d8

Fix VC++ warning that bools cannot be ORed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24319 91177308-0d34-0410-b5e6-96231b3b80d8

Fix operator precedence bug caught by VC++.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24318 91177308-0d34-0410-b5e6-96231b3b80d8

document sections

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24317 91177308-0d34-0410-b5e6-96231b3b80d8

regenerate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24316 91177308-0d34-0410-b5e6-96231b3b80d8

add a token

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24315 91177308-0d34-0410-b5e6-96231b3b80d8

Parse section info

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24314 91177308-0d34-0410-b5e6-96231b3b80d8

print section info

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24313 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24310 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for putting globals in a particular section

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24309 91177308-0d34-0410-b5e6-96231b3b80d8

fix yet more regressions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24308 91177308-0d34-0410-b5e6-96231b3b80d8

generate chain result

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24307 91177308-0d34-0410-b5e6-96231b3b80d8

added a chain output

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24306 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a bunch more alpha regressions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24304 91177308-0d34-0410-b5e6-96231b3b80d8

prevent cse of readcyclecounter

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24303 91177308-0d34-0410-b5e6-96231b3b80d8

continued readcyclecounter support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24300 91177308-0d34-0410-b5e6-96231b3b80d8

Add support for a cycle counter intrinsic.  As basically all processors have
this and have it in about the same form, I think this makes sense.
on X86, you do a RDTSC (64bit result, from any ring since the P5MMX)
on Alpha, you do a RDCC
on PPC, there is a sequence which may or may not work depending on how things
are setup by the OS.  Or something like that.  Maybe someone who knows PPC
can add support.  Something about the time base register.
on Sparc, you read %tick, which in some solaris versions (>=8) is readable by
userspace
on IA64 read ar.itc

So I think the ulong is justified since all of those are 64bit.
Support is slighly flaky on old chips (P5 and lower) and sometimes
depends on OS (PPC, Sparc).  But for modern OS/Hardware (aka this decade),
we should be ok.

I am still not sure what to do about lowering.  I can either see a lower to 0, to
gettimeofday (or the target os equivalent), or loudly complaining and refusing to
continue.

I am commiting an Alpha implementation.  I will add the X86 implementation if I
have to (I have use of it in the near future), but if someone who knows that
backend (and the funky multi-register results) better wants to add it, it would
take them a lot less time ;)

TODO: better lowering and legalizing, and support more platforms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24299 91177308-0d34-0410-b5e6-96231b3b80d8

Fix some typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24298 91177308-0d34-0410-b5e6-96231b3b80d8

Add new file to Visual Studio.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24297 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the optimized code handling of user asm strings

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24296 91177308-0d34-0410-b5e6-96231b3b80d8

Make BB and CPI labels use the function number, not the function name as a
uniquing id.  This makes things happy when the function name is quoted,
preventing labels like LBB"foo"_2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24295 91177308-0d34-0410-b5e6-96231b3b80d8

speedup the common case where nothing needs to be quoted

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24294 91177308-0d34-0410-b5e6-96231b3b80d8

Allow per-character control over what target assemblers allow in symbol
names.  This also changes the default to allow all of "$_." in addition
to letters and numbers as symbol names.  If you don't want this, use
markCharUnacceptable to remove one of these or markCharAcceptable to add
to the set.  This corresponds with what GAS accepts by default.

Also, this includes some minor speedups

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24293 91177308-0d34-0410-b5e6-96231b3b80d8

do not allow '.' in symbol names

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24292 91177308-0d34-0410-b5e6-96231b3b80d8

Allow per-character control over what target assemblers allow in symbol
names.  This also changes the default to allow all of "$_." in addition
to letters and numbers as symbol names.  If you don't want this, use
markCharUnacceptable to remove one of these or markCharAcceptable to add
to the set.  This corresponds with what GAS accepts by default.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24291 91177308-0d34-0410-b5e6-96231b3b80d8

the pain isn't gone unless the phinodes are spilled too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24288 91177308-0d34-0410-b5e6-96231b3b80d8

Darwin supports quoted labels.  This implements:
test/Regression/CodeGen/PowerPC/darwin-labels.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24287 91177308-0d34-0410-b5e6-96231b3b80d8

Check that this compiles to a label with a space in it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24284 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new option for targets that accept quoted labels.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24283 91177308-0d34-0410-b5e6-96231b3b80d8

Remove M, which is dead.  Eliminate a dead typedef.  Add comments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24282 91177308-0d34-0410-b5e6-96231b3b80d8

remove the M instance var

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24281 91177308-0d34-0410-b5e6-96231b3b80d8

This method is no longer static

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24280 91177308-0d34-0410-b5e6-96231b3b80d8

Call this method with the object we have

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24279 91177308-0d34-0410-b5e6-96231b3b80d8

nuke blank line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24278 91177308-0d34-0410-b5e6-96231b3b80d8

Make this more efficient of the common case where we are only mangling globals.
Do not mangle internal global names that do not collide with anything.

This gives us strings that now look like this:

__2E_str_1:                             ; '.str_1'
        .asciz  "foo"

instead of this:

l1__2E_str_1:                           ; '.str_1'
        .asciz  "foo"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24277 91177308-0d34-0410-b5e6-96231b3b80d8

minor interface changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24276 91177308-0d34-0410-b5e6-96231b3b80d8

Get rid of casts by #including the right header

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24275 91177308-0d34-0410-b5e6-96231b3b80d8

Make the aix asm printer interface properly with the parent class

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24274 91177308-0d34-0410-b5e6-96231b3b80d8

Compile C strings to:

l1__2E_str_1:                           ; '.str_1'
        .asciz  "foo"

not:

        .align  0
l1__2E_str_1:                           ; '.str_1'
        .asciz  "foo"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24273 91177308-0d34-0410-b5e6-96231b3b80d8

add support for .asciz, and enable it by default.  If your target assemblerdoesn't support .asciz, just set AscizDirective to null in your asmprinter.
This compiles C strings to:

l1__2E_str_1:                           ; '.str_1'
        .asciz  "foo"

instead of:

l1__2E_str_1:                           ; '.str_1'
        .ascii  "foo\000"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24272 91177308-0d34-0410-b5e6-96231b3b80d8

add support for .asciz, and enable it by default.  If your target assembler
doesn't support .asciz, just set AscizDirective to null in your asmprinter.

This compiles C strings to:

l1__2E_str_1:                           ; '.str_1'
        .asciz  "foo"

instead of:

l1__2E_str_1:                           ; '.str_1'
        .ascii  "foo\000"

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24271 91177308-0d34-0410-b5e6-96231b3b80d8

this works with backedges to the existing entry block alot better

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24270 91177308-0d34-0410-b5e6-96231b3b80d8

fix a bunch of regressions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24269 91177308-0d34-0410-b5e6-96231b3b80d8

needs to go here to apparently.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24268 91177308-0d34-0410-b5e6-96231b3b80d8

The pass everyone has been waiting for!
Reg2Mem

for fun you can opt -reg2mem -mem2reg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24267 91177308-0d34-0410-b5e6-96231b3b80d8

Packed elements must be a power of two in size

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24266 91177308-0d34-0410-b5e6-96231b3b80d8

Force vectors to be a power of two in size

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24265 91177308-0d34-0410-b5e6-96231b3b80d8

Force packed vectors to be a power of two in length.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24264 91177308-0d34-0410-b5e6-96231b3b80d8

Switch the allnodes list from a vector of pointers to an ilist of nodes.This eliminates the vector, allows constant time removal of a node froma graph, and makes iteration over the all nodes list stable when adding
nodes to the graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24263 91177308-0d34-0410-b5e6-96231b3b80d8

Switch the allnodes list from a vector of pointers to an ilist of nodes.
This eliminates the vector, allows constant time removal of a node from
a graph, and makes iteration over the all nodes list stable when adding
nodes to the graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24262 91177308-0d34-0410-b5e6-96231b3b80d8

Refactor intrinsic lowering stuff out of visitCall

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24261 91177308-0d34-0410-b5e6-96231b3b80d8

whatever.  Intermediate patch to see what breaks.  Seems ok.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24260 91177308-0d34-0410-b5e6-96231b3b80d8

Handle the trivial (but common) two-op case more efficiently

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24259 91177308-0d34-0410-b5e6-96231b3b80d8

Nuke noop copies.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24258 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/X86/shift-folding.ll:test3 on X86

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24256 91177308-0d34-0410-b5e6-96231b3b80d8

Disable some overly-aggressive checking code.  This speeds up the local
allocator from 23s to 11s on kc++ in debug mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24255 91177308-0d34-0410-b5e6-96231b3b80d8

Avoid creating a token factor node in trivially redundant cases.  This
eliminates almost one node per block in common cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24254 91177308-0d34-0410-b5e6-96231b3b80d8

Handle GEP's a bit more intelligently.  Fold constant indices early and
turn power-of-two multiplies into shifts early to improve compile time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24253 91177308-0d34-0410-b5e6-96231b3b80d8

Allocate the right amount of memory for this vector up front.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24252 91177308-0d34-0410-b5e6-96231b3b80d8

Change the ValueList array for each node to be shared instead of individuallyallocated.  Further, in the common case where a node has a single value, justreference an element from a small array.  This is a small compile-time win.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24251 91177308-0d34-0410-b5e6-96231b3b80d8

Change the ValueList array for each node to be shared instead of individually
allocated.  Further, in the common case where a node has a single value, just
reference an element from a small array.  This is a small compile-time wi.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24250 91177308-0d34-0410-b5e6-96231b3b80d8

Switch the operandlist/valuelist from being vectors to being just an array.This saves 12 bytes from SDNode, but doesn't speed things up substantially
(our graphs apparently already fit within the cache on my g5).  In any case
this reduces memory usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24249 91177308-0d34-0410-b5e6-96231b3b80d8

Switch the operandlist/valuelist from being vectors to being just an array.
This saves 12 bytes from SDNode, but doesn't speed things up substantially
(our graphs apparently already fit within the cache on my g5).  In any case
this reduces memory usage.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24248 91177308-0d34-0410-b5e6-96231b3b80d8

Explicitly initialize some instance vars

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24247 91177308-0d34-0410-b5e6-96231b3b80d8

Rip out 1.6ness, bump version # to 1.7cvs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24246 91177308-0d34-0410-b5e6-96231b3b80d8

Updated version to 1.7cvs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24244 91177308-0d34-0410-b5e6-96231b3b80d8

Merged from RELEASE_16.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24243 91177308-0d34-0410-b5e6-96231b3b80d8

adjust itf

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24242 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up RemoveDeadNodes significantly, by eliminating the need for a temporary
set and eliminating the need to iterate whenever something is removed (which
can be really slow in some cases).  Thx to Jim for pointing out something silly
I was getting stuck on. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24241 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new option to indicate we want the code generator to emit code quickly,
not spending tons of time microoptimizing it.  This is useful for an -O0
style of build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24235 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new -fast option, which generates code quickly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24234 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new option to indicate we want the code generator to emit code quickly,not spending tons of time microoptimizing it.  This is useful for an -O0style of build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24233 91177308-0d34-0410-b5e6-96231b3b80d8

Let's try ignoring resource utilization on the backward pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24231 91177308-0d34-0410-b5e6-96231b3b80d8

add support for storing and returning bools

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24228 91177308-0d34-0410-b5e6-96231b3b80d8

Always compute max align.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24227 91177308-0d34-0410-b5e6-96231b3b80d8

Change a comment slightly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24226 91177308-0d34-0410-b5e6-96231b3b80d8

just some random hacking - calls (particularly indirect) need a lot of
love (especially with -sched=simple)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24225 91177308-0d34-0410-b5e6-96231b3b80d8

Add the necessary support to the ISel to allow targets to codegen the new
alignment information appropriately.  Includes code for PowerPC to support
fixed-size allocas with alignment larger than the stack.  Support for
arbitrarily aligned dynamic allocas coming soon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24224 91177308-0d34-0410-b5e6-96231b3b80d8

minor clarity changes, no functionality difference.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24223 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure to initialize the alignment field

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24222 91177308-0d34-0410-b5e6-96231b3b80d8

document alignment on globals, functions, and allocation instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24221 91177308-0d34-0410-b5e6-96231b3b80d8

Minor correction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24220 91177308-0d34-0410-b5e6-96231b3b80d8

don't misencode CC#'s

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24219 91177308-0d34-0410-b5e6-96231b3b80d8

encode/decode function alignment in bc files

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@24218 91177308-0d34-0410-b5e6-96231b3b80d8