Commit Graph

469 Commits

Author SHA1 Message Date
Chris Lattner
283d1ceb57 Switch pattern sorting predicate from stable sort -> sort, it
doesn't need to be stable because the patterns are fully ordered.

Add a first level sort predicate that orders patterns in this
order:  1) scalar integer operations 2) scalar floating point 
3) vector int 4) vector float.  This is a trivial sort on their
top level pattern type so it is nice and transitive.  The
benefit of doing this is that simple integer operations are
much more common than insane vector things and isel was trying
to match the big complex vector patterns before the simple
ones because the complexity of the vector operations was much
higher.  Since they can't both match, it is best (for compile
time) to try the simple integer ones first.

This cuts down the # failed match attempts on real code by
quite a bit, for example, this reduces backtracks on crafty
(as a random example) from 228285 -> 188369.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99797 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-29 02:02:45 +00:00
Chris Lattner
4810736134 revert 99795, as mentioned, it is disabled anyway.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99796 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-29 01:58:15 +00:00
Chris Lattner
a0401249e8 Check in a (disabled) failed attempt to improve the ordering of
patterns within the generated matcher.  This works great except
that the sort fails because the relation defined isn't 
transitive.  I have a much simpler solution coming next, but want
to archive the code.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99795 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-29 01:56:19 +00:00
Chris Lattner
48e86dbe29 print the complexity of the pattern being matched in the
comment in the generated table.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99794 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-29 01:40:38 +00:00
Chris Lattner
d734919465 major surgery on tblgen: generalize TreePatternNode
to maintain a list of types (one for each result of
the node) instead of a single type.  There are liberal
hacks added to emulate the old behavior in various
situations, but they can start disolving now.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98999 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-19 21:37:09 +00:00
Chris Lattner
f30187a363 look up instructions by record, not by name.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98904 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-19 00:07:20 +00:00
Chris Lattner
2cacec55f9 Completely rewrite tblgen's type inference mechanism,
changing the primary datastructure from being a 
"std::vector<unsigned char>" to being a new TypeSet class
that actually has (gasp) invariants!

This changes more things than I remember, but one major
innovation here is that it enforces that named input 
values agree in type with their output values.

This also eliminates code that transparently assumes (in 
some cases) that SDNodeXForm input/output types are the
same, because this is wrong in many case.

This also eliminates a bug which caused a lot of ambiguous
patterns to go undetected, where a register class would
sometimes pick the first possible type, causing an
ambiguous pattern to get arbitrary results.

With all the recent target changes, this causes no 
functionality change!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98534 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-15 06:00:16 +00:00
Chris Lattner
d272fee398 the sorting predicate should work for comparing an element
to itself, even though this isn't wildly useful.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97574 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-02 18:15:02 +00:00
Chris Lattner
4a3ac62332 eliminate CodeGen/DAGISelHeader.h, it is empty now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97556 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-02 06:36:28 +00:00
Chris Lattner
405f1252b9 resolve a fixme by having the .td file parser reject thigns like
(set GPR, somecomplexpattern)

if somecomplexpattern doesn't declare what it can match.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97513 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 22:29:19 +00:00
Chris Lattner
117ccb7e51 Fix PR2590 by making PatternSortingPredicate actually be
ordered correctly.  Previously it would get in trouble when
two patterns were too similar and give them nondet ordering.
We force this by using the record ID order as a fallback.

The testsuite diff is due to alpha patterns being ordered
slightly differently, the change is a semantic noop afaict:

< 	lda $0,-100($16)
---
> 	subq $16,100,$0



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97509 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 22:09:11 +00:00
Chris Lattner
adc5347b54 optimize tblgen compile time by eliminating the old isel.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97504 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 21:49:54 +00:00
Chris Lattner
459eeedbe0 don't emit the old sdnodexform stuff for the new isel.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97486 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 19:39:15 +00:00
Chris Lattner
8a01c8d00d Turn on the new isel by default. Here are some fun numbers
with a release-asserts build on x86-64-darwin10:

LLC Size:
Old: 15,426,852
New: 12,759,140  (down 2.7M)

LLI Size:
Old: 9,926,876
New: 8,864,292   (down 1.1M)

X86ISelDAGToDAG.o size:
Old: 1,401,232
New:   162,868   (down 1.3M)

Time to build X86ISelDAGToDAG.o:
Old: 67.147u 2.060s 1:09.78
New:  4.234u 0.387s 0:04.77



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97475 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 19:05:25 +00:00
Chris Lattner
fa342faef9 eliminate the CheckMultiOpcodeMatcher code and have each
ComplexPattern at the root be generated multiple times, once
for each opcode they are part of.  This encourages factoring
because the opcode checks get treated just like everything
else in the matcher.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97439 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 07:17:40 +00:00
Chris Lattner
4d0c931ba7 inline the node transforms and node predicates into the generated
dispatcher method.  This eliminates the dependence of the new isel's
generated code on the old isel's predicates, however some random
hand written isel code still uses them.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97431 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-01 01:54:19 +00:00
Chris Lattner
c78f2a3994 enhance the new isel to use SelectNodeTo for most patterns,
even some the old isel didn't.  There are several parts of
this that make me feel dirty, but it's no worse than the
old isel.  I'll clean up the parts I can do without ripping
out the old one next.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97415 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-28 20:49:53 +00:00
Chris Lattner
d6c84720df 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
2010-02-25 19:00:39 +00:00
Chris Lattner
b21ba71045 rename fooMatcherNode to fooMatcher.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97096 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-25 02:04:40 +00:00
Chris Lattner
60df53e30a 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
2010-02-25 01:56:48 +00:00
Chris Lattner
a86509d436 lets not break the old isel.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97034 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-24 07:41:11 +00:00
Chris Lattner
c84edb7bdd Since the new instruction selector now works, I don't need to keep
the old one around for comparative purposes: have the
ENABLE_NEW_ISEL #define (which is not enabled on mainline) stop
emitting the old isel at all, yay for build time win.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97033 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-24 07:35:09 +00:00
Chris Lattner
19b5a7590b implement a simple proof-of-concept optimization for
the new isel: fold movechild+record+moveparent into a
single recordchild N node.  This shrinks the X86 table
from 125443 to 117502 bytes.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97031 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-24 07:31:45 +00:00
Chris Lattner
91c6a822ba The new isel passes all tests, time to start making it go fast.
Also add an easy macro at the top of DAGISelEmitter.cpp to enable
it.  Lets see if I can avoid accidentally turning it on :)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@97029 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-24 07:06:50 +00:00
Chris Lattner
99ce6e8aff Sort the patterns before adding them to the FA so that we get the
least cost matches.  This gets us from 195 -> 208 passes on the ppc codegen tests.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96747 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-21 19:22:06 +00:00
Chris Lattner
0ab1c5f8f8 oops don't turn this on for everyone yet.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96725 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-21 06:03:56 +00:00
Chris Lattner
77f2e2724d implement the last known missing feature: updating uses of results
of the matched pattern to use the newly created node results.  Onto
the "making it actually work" phase!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96724 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-21 06:03:07 +00:00
Chris Lattner
8e946bea14 Lots of improvements to the new dagisel emitter. This gets it to
the point where it is to the 95% feature complete mark, it just
needs result updating to be done (then testing, optimization 
etc).

More specificallly, this adds support for chain and flag handling
on the result nodes, support for sdnodexforms, support for variadic
nodes, memrefs, pinned physreg inputs, and probably lots of other
stuff.

In the old DAGISelEmitter, this deletes the dead code related to
OperatorMap, cleans up a variety of dead stuff handling "implicit
remapping" from things like globaladdr -> targetglobaladdr (which
is no longer used because globaladdr always needs to be legalized),
and some minor formatting fixes.




git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96716 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-21 03:22:59 +00:00
Chris Lattner
b49985a849 start sketching out the structure of code for result emission generation.
Nothing real here yet.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96575 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-18 06:47:49 +00:00
Chris Lattner
dcdcef2cb3 fit in 80 cols
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96541 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-18 00:23:27 +00:00
Chris Lattner
03ddb20bfd redisable this to save people a small amount of build time.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96497 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-17 19:19:50 +00:00
Chris Lattner
0a3f39985b move isOnlyReachableByFallthrough out of MachineBasicBlock into AsmPrinter,
and add a sparc implementation that knows about delay slots.  Patch by
Nathan Keynes!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96492 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-17 18:52:56 +00:00
Chris Lattner
409ac586e8 sink special case "cannotyetselect" for intrinsics out of the
tblgen splatted code into the implementation.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96460 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-17 06:28:22 +00:00
Chris Lattner
d9c1a344a0 simplify IsChainCompatible codegen, add comments. no
functionality change.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96453 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-17 05:35:28 +00:00
Chris Lattner
e609a513f3 make the new isel generator plop out a CheckComplexPattern function
for evaluating complex patterns.  Some cleanup has to happen before
this can be used though.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96419 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-17 00:31:50 +00:00
Chris Lattner
05446e7d47 clean up some code, eliminate NodeIsComplexPattern, which
does the same thing as getComplexPatternInfo.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96411 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 23:13:59 +00:00
Chris Lattner
5b08f77096 fix indentation
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96409 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 22:38:31 +00:00
Chris Lattner
92d3ada814 fix rdar://7653908, a crash on a case where we would fold a load
into a roundss intrinsic, producing a cyclic dag.  The root cause
of this is badness handling ComplexPattern nodes in the old dagisel
that I noticed through inspection.  Eliminate a copy of the of the
code that handled ComplexPatterns by making EmitChildMatchCode call
into EmitMatchCode.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96408 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 22:35:06 +00:00
Chris Lattner
29c6270328 change dag isel emitter to only call 'IsProfitableToFold' on nodes
with chains.  On interior nodes that lead up to them, we just directly
check that there is a single use.  This generates slightly more
efficient code.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96366 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 19:03:34 +00:00
Chris Lattner
ccba15f914 mark all the generated node predicates 'const'.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96337 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 07:26:36 +00:00
Chris Lattner
e2de49d9a1 remove now dead code and fixme.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96333 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 06:15:00 +00:00
Chris Lattner
6e6975d419 remove dead code. This is never generated for any targets in mainline.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96332 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 06:14:22 +00:00
Chris Lattner
e39650a805 add support for the new isel matcher to generate
(isprofitable|islegal)tofold checks.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96331 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-16 06:10:58 +00:00
Evan Cheng
014bf215c3 Split SelectionDAGISel::IsLegalAndProfitableToFold to
IsLegalToFold and IsProfitableToFold. The generic version of the later simply checks whether the folding candidate has a single use.

This allows the target isel routines more flexibility in deciding whether folding makes sense. The specific case we are interested in is folding constant pool loads with multiple uses.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96255 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-15 19:41:07 +00:00
Chris Lattner
da272d1a70 Check in the first big step of rewriting DAGISelEmitter to
produce a table based matcher instead of gobs of C++ Code.

Though it's not done yet, the shrinkage seems promising,
the table for the X86 ISel is 75K and still has a lot of 
optimization to come (compare to the ~1.5M of .o generated
the old way, much of which will go away).

The code is currently disabled by default (the #if 0 in
DAGISelEmitter.cpp).  When enabled it generates a dead
SelectCode2 function in the DAGISel Header which will
eventually replace SelectCode.

There is still a lot of stuff left to do, which are
documented with a trail of FIXMEs.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96215 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-15 08:04:42 +00:00
Chris Lattner
4766132036 clean up a bunch of code, move some random predicates
on TreePatternNode to be methods on TreePatternNode.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96197 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-14 22:22:58 +00:00
Chris Lattner
c87bf389dc remove the DisablePatternForFastISel predicate, which is a check
that predated -fast-isel which attempted to speed up the dag pattern
matchers at -O0.  Since fast-isel is around, this is basically 
obsolete and removing it shrinks the generated dag isels.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96188 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-14 21:11:53 +00:00
Chris Lattner
a0cdf17f37 pull a bunch of huge inline methods in the PatternCodeEmitter
class out of line.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96113 91177308-0d34-0410-b5e6-96231b3b80d8
2010-02-13 20:06:50 +00:00
Dan Gohman
eeb3a00b84 Change SelectCode's argument from SDValue to SDNode *, to make it more
clear what information these functions are actually using.

This is also a micro-optimization, as passing a SDNode * around is
simpler than passing a { SDNode *, int } by value or reference.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@92564 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-05 01:24:18 +00:00
Dan Gohman
ba7a662f54 Replace some special-case code which probably was buggy with an assertion
verifying that the special case does not occur.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@92504 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-04 20:36:57 +00:00