Commit Graph

25984 Commits

Author SHA1 Message Date
Duncan Sands
5de8739964 There are no longer any places that require a
MERGE_VALUES node with only one operand, so get
rid of special code that only existed to handle
that possibility.

llvm-svn: 60349
2008-12-01 11:41:29 +00:00
Duncan Sands
1fae2ea219 Change the interface to the type legalization method
ReplaceNodeResults: rather than returning a node which
must have the same number of results as the original
node (which means mucking around with MERGE_VALUES,
and which is also easy to get wrong since SelectionDAG
folding may mean you don't get the node you expect),
return the results in a vector.

llvm-svn: 60348
2008-12-01 11:39:25 +00:00
Bill Wendling
48b7cbbc01 Generalize the FoldOrWithConstant method to fold for any two constants which
don't have overlapping bits.

llvm-svn: 60344
2008-12-01 08:32:40 +00:00
Bill Wendling
2a182b838d Reduce copy-and-paste code by splitting out the code into its own function.
llvm-svn: 60343
2008-12-01 08:23:25 +00:00
Bill Wendling
a6e7dd2299 Use m_Specific() instead of double matching.
llvm-svn: 60341
2008-12-01 08:09:47 +00:00
Bill Wendling
8e484e9556 Move pattern check outside of the if-then statement. This prevents us from fiddling with constants unless we have to.
llvm-svn: 60340
2008-12-01 07:47:02 +00:00
Chris Lattner
3b908483b7 Rename some variables, only increment BI once at the start of the loop instead of throughout it.
llvm-svn: 60339
2008-12-01 07:35:54 +00:00
Chris Lattner
c6e6eaf6d3 pull the predMap densemap out of the inner loop of performPRE, so
that it isn't reallocated all the time.  This is a tiny speedup for
GVN: 3.90->3.88s

llvm-svn: 60338
2008-12-01 07:29:03 +00:00
Chris Lattner
f72f8e3b74 switch a couple more calls to use array_pod_sort.
llvm-svn: 60337
2008-12-01 06:52:57 +00:00
Chris Lattner
80d0eff786 Introduce a new array_pod_sort function and switch LSR to use it
instead of std::sort.  This shrinks the release-asserts LSR.o file
by 1100 bytes of code on my system.

We should start using array_pod_sort where possible.

llvm-svn: 60335
2008-12-01 06:49:59 +00:00
Chris Lattner
74f1e6d3ec Eliminate use of setvector for the DeadInsts set, just use a smallvector.
This is a lot cheaper and conceptually simpler.

llvm-svn: 60332
2008-12-01 06:27:41 +00:00
Chris Lattner
db86ff62f9 DeleteTriviallyDeadInstructions is always passed the
DeadInsts ivar, just use it directly.

llvm-svn: 60330
2008-12-01 06:14:28 +00:00
Chris Lattner
d6be279b4d simplify DeleteTriviallyDeadInstructions again, unlike my previous
buggy rewrite, this notifies ScalarEvolution of a pending instruction
about to be removed and then erases it, instead of erasing it then 
notifying.

llvm-svn: 60329
2008-12-01 06:11:32 +00:00
Chris Lattner
e6c7ed156f simplify these patterns using m_Specific. No need to grep for
xor in testcase (or is a substring).

llvm-svn: 60328
2008-12-01 05:16:26 +00:00
Chris Lattner
c92e1e104b Teach jump threading to clean up after itself, DCE and constfolding the
new instructions it simplifies.  Because we're threading jumps on edges
with constants coming in from PHI's, we inherently are exposing a lot more
constants to the new block.  Folding them and deleting dead conditions
allows the cost model in jump threading to be more accurate as it iterates.

llvm-svn: 60327
2008-12-01 04:48:07 +00:00
Chris Lattner
ceeb559995 The PreVerifier pass preserves everything. In practice, this
prevents the passmgr from adding yet-another domtree invocation
for Verifier if there is already one live.

llvm-svn: 60326
2008-12-01 03:58:38 +00:00
Chris Lattner
13942f82c4 Change instcombine to use FoldPHIArgGEPIntoPHI to fold two operand PHIs
instead of using FoldPHIArgBinOpIntoPHI.  In addition to being more
obvious, this also fixes a problem where instcombine wouldn't merge two
phis that had different variable indices.  This prevented instcombine
from factoring big chunks of code in 403.gcc.  For example:

 insn_cuid.exit:                
-       %tmp336 = load i32** @uid_cuid, align 4      
-       %tmp337 = getelementptr %struct.rtx_def* %insn_addr.0.ph.i, i32 0, i32 3    
-       %tmp338 = bitcast [1 x %struct.rtunion]* %tmp337 to i32*               
-       %tmp339 = load i32* %tmp338, align 4           
-       %tmp340 = getelementptr i32* %tmp336, i32 %tmp339     
        br label %bb62
 
 bb61:       
-       %tmp341 = load i32** @uid_cuid, align 4     
-       %tmp342 = getelementptr %struct.rtx_def* %insn, i32 0, i32 3        
-       %tmp343 = bitcast [1 x %struct.rtunion]* %tmp342 to i32*           
-       %tmp344 = load i32* %tmp343, align 4        
-       %tmp345 = getelementptr i32* %tmp341, i32 %tmp344          
        br label %bb62
 
 bb62:      
-       %iftmp.62.0.in = phi i32* [ %tmp345, %bb61 ], [ %tmp340, %insn_cuid.exit ]         
+       %insn.pn2 = phi %struct.rtx_def* [ %insn, %bb61 ], [ %insn_addr.0.ph.i, %insn_cuid.exit ]         
+       %tmp344.pn.in.in = getelementptr %struct.rtx_def* %insn.pn2, i32 0, i32 3     
+       %tmp344.pn.in = bitcast [1 x %struct.rtunion]* %tmp344.pn.in.in to i32*  
+       %tmp341.pn = load i32** @uid_cuid     
+       %tmp344.pn = load i32* %tmp344.pn.in 
+       %iftmp.62.0.in = getelementptr i32* %tmp341.pn, i32 %tmp344.pn   
        %iftmp.62.0 = load i32* %iftmp.62.0.in     

llvm-svn: 60325
2008-12-01 03:42:51 +00:00
Chris Lattner
0e03e40a76 Teach inst combine to merge GEPs through PHIs. This is really
important because it is sinking the loads using the GEPs, but
not the GEPs themselves.  This triggers 647 times on 403.gcc
and makes the .s file much much nicer.  For example before:

        je      LBB1_87 ## bb78
LBB1_62:        ## bb77
        leal    84(%esi), %eax
LBB1_63:        ## bb79
        movl    (%eax), %eax
...
LBB1_87:        ## bb78
        movl    $0, 4(%esp)
        movl    %esi, (%esp)
        call    L_make_decl_rtl$stub
        jmp     LBB1_62 ## bb77


after:

        jne     LBB1_63 ## bb79
LBB1_62:        ## bb78
        movl    $0, 4(%esp)
        movl    %esi, (%esp)
        call    L_make_decl_rtl$stub
LBB1_63:        ## bb79
        movl    84(%esi), %eax

The input code was (and the GEPs are merged and
the PHI is now eliminated by instcombine):

        br i1 %tmp233, label %bb78, label %bb77
bb77:           
        %tmp234 = getelementptr %struct.tree_node* %t_addr.3, i32 0, i32 0, i32 22              
        br label %bb79
bb78:           
        call void @make_decl_rtl(%struct.tree_node* %t_addr.3, i8* null) nounwind
        %tmp235 = getelementptr %struct.tree_node* %t_addr.3, i32 0, i32 0, i32 22              
        br label %bb79
bb79:           
        %iftmp.12.0.in = phi %struct.rtx_def** [ %tmp235, %bb78 ], [ %tmp234, %bb77 ]           
        %iftmp.12.0 = load %struct.rtx_def** %iftmp.12.0.in             

llvm-svn: 60322
2008-12-01 02:34:36 +00:00
Chris Lattner
c1adf6fc51 Make GVN be more intelligent about redundant load
elimination: when finding dependent load/stores, realize that
they are the same if aliasing claims must alias instead of relying
on the pointers to be exactly equal.  This makes load elimination
more aggressive.  For example, on 403.gcc, we had:

<     68 gvn    - Number of instructions PRE'd
< 152718 gvn    - Number of instructions deleted
<  49699 gvn    - Number of loads deleted
<   6153 memdep - Number of dirty cached non-local responses
< 169336 memdep - Number of fully cached non-local responses
< 162428 memdep - Number of uncached non-local responses

now we have:

>     64 gvn    - Number of instructions PRE'd
> 153623 gvn    - Number of instructions deleted
>  49856 gvn    - Number of loads deleted
>   5022 memdep - Number of dirty cached non-local responses
> 159030 memdep - Number of fully cached non-local responses
> 162443 memdep - Number of uncached non-local responses

That's an extra 157 loads deleted and extra 905 other instructions nuked.

This slows down GVN very slightly, from 3.91 to 3.96s.

llvm-svn: 60314
2008-12-01 01:31:36 +00:00
Chris Lattner
bd1bc4a75e Reimplement the non-local dependency data structure in terms of a sorted
vector instead of a densemap.  This shrinks the memory usage of this thing
substantially (the high water mark) as well as making operations like
scanning it faster.  This speeds up memdep slightly, gvn goes from
3.9376 to 3.9118s on 403.gcc

This also splits out the statistics for the cached non-local case to
differentiate between the dirty and clean cached case.  Here's the stats
for 403.gcc:

  6153 memdep - Number of dirty cached non-local responses
169336 memdep - Number of fully cached non-local responses
162428 memdep - Number of uncached non-local responses

yay for caching :)

llvm-svn: 60313
2008-12-01 01:15:42 +00:00
Bill Wendling
23684a026c Implement ((A|B)&1)|(B&-2) -> (A&1) | B transformation. This also takes care of
permutations of this pattern.

llvm-svn: 60312
2008-12-01 01:07:11 +00:00
Chris Lattner
1f8482ffc8 Cache analyses in ivars and add some useful DEBUG output.
This speeds up GVN from 4.0386s to 3.9376s.

llvm-svn: 60310
2008-12-01 00:40:32 +00:00
Chris Lattner
77908d9ccf improve indentation, do cheap checks before expensive ones,
remove some fixme's.  This speeds up GVN very slightly on 403.gcc 
(4.06->4.03s)

llvm-svn: 60309
2008-11-30 23:39:23 +00:00
Chris Lattner
36257aabe4 Eliminate the DepResultTy abstraction. It is now completely
redundant with MemDepResult, and MemDepResult has a nicer interface.

llvm-svn: 60308
2008-11-30 23:17:19 +00:00
Eli Friedman
052df7e062 Minor cleanup: use getTrue and getFalse where appropriate. No
functional change.

llvm-svn: 60307
2008-11-30 22:48:49 +00:00
Eli Friedman
8da9f2f8d3 Some minor cleanups to instcombine; no functionality change.
Note that the FoldOpIntoPhi call is dead because it's impossible for the 
first operand of a subtraction to be both a ConstantInt and a PHINode.

llvm-svn: 60306
2008-11-30 21:09:11 +00:00
Chris Lattner
9f7facc8eb Cache TargetData/AliasAnalysis in the pass instead of calling
getAnalysis<>.  getAnalysis<> is apparently extremely expensive.
Doing this speeds up GVN on 403.gcc by 16%!

llvm-svn: 60304
2008-11-30 19:24:31 +00:00
Bill Wendling
66a7442059 Add instruction combining for ((A&~B)|(~A&B)) -> A^B and all permutations.
llvm-svn: 60291
2008-11-30 13:52:49 +00:00
Bill Wendling
3e27ac16a6 Implement (A&((~A)|B)) -> A&B transformation in the instruction combiner. This
takes care of all permutations of this pattern.

llvm-svn: 60290
2008-11-30 13:08:13 +00:00
Bill Wendling
92ebd6902d Forgot one remaining call to getSExtValue().
llvm-svn: 60289
2008-11-30 12:41:09 +00:00
Bill Wendling
97ad688c1b getSExtValue() doesn't work for ConstantInts with bitwidth > 64 bits. Use all
APInt calls instead.

This fixes PR3144.

llvm-svn: 60288
2008-11-30 12:38:24 +00:00
Eli Friedman
2bc3921ce2 Optimize memmove and memset into the LLVM builtins. Note that these
only show up in code from front-ends besides llvm-gcc, like clang.

llvm-svn: 60287
2008-11-30 08:32:11 +00:00
Eli Friedman
3b8efd50d7 A couple small cleanups, plus a new potential optimization.
llvm-svn: 60286
2008-11-30 07:52:27 +00:00
Eli Friedman
97d37825f1 Moving potential optimizations out of PR2330 into lib/Target/README.txt.
Hopefully this isn't too much stuff to dump into this file.

llvm-svn: 60285
2008-11-30 07:36:04 +00:00
Eli Friedman
ccdfdbfc99 Followup to r60283: optimize arbitrary width signed divisions as well
as unsigned divisions.  Same caveats as before.

llvm-svn: 60284
2008-11-30 06:35:39 +00:00
Eli Friedman
d7a261120f Fix for PR2164: allow transforming arbitrary-width unsigned divides into
multiplies.

Some more cleverness would be nice, though. It would be nice if we 
could do this transformation on illegal types.  Also, we would 
prefer a narrower constant when possible so that we can use a narrower
multiply, which can be cheaper.

llvm-svn: 60283
2008-11-30 06:02:26 +00:00
Bill Wendling
115290ddd3 Don't make TwoToExp signed by default.
llvm-svn: 60279
2008-11-30 05:29:33 +00:00
Bill Wendling
4e018f4c22 From Hacker's Delight:
"For signed integers, the determination of overflow of x*y is not so simple. If
x and y have the same sign, then overflow occurs iff xy > 2**31 - 1. If they
have opposite signs, then overflow occurs iff xy < -2**31."

In this case, x == -1.

llvm-svn: 60278
2008-11-30 05:01:05 +00:00
Eli Friedman
0ef5e1dc82 APIntify a test which is potentially unsafe otherwise, and fix the
nearby FIXME.

I'm not sure what the right way to fix the Cell test was; if the 
approach I used isn't okay, please let me know.

llvm-svn: 60277
2008-11-30 04:59:26 +00:00
Bill Wendling
ac11f7d37e Instcombine was illegally transforming -X/C into X/-C when either X or C
overflowed on negation. This commit checks to make sure that neithe C nor X
overflows. This requires that the RHS of X (a subtract instruction) be a
constant integer.

llvm-svn: 60275
2008-11-30 03:42:12 +00:00
Chris Lattner
baf3cdd3a1 Two changes: Make getDependency remove QueryInst for a dirty record's
ReverseLocalDeps when we update it.  This fixes a regression test
failure from my last commit.

Second, for each non-local cached information structure, keep a bit that
indicates whether it is dirty or not.  This saves us a scan over the whole
thing in the common case when it isn't dirty.

llvm-svn: 60274
2008-11-30 02:52:26 +00:00
Chris Lattner
90904fda3c introduce a typedef, no functionality change.
llvm-svn: 60272
2008-11-30 02:30:50 +00:00
Chris Lattner
a772cad6ff Change NonLocalDeps to be a densemap of pointers to densemap
instead of containing them by value.  This increases the density
(!) of NonLocalDeps as well as making the reallocation case 
faster.  This speeds up gvn on 403.gcc by 2% and makes room for
future improvements.

I'm not super thrilled with having to explicitly manage the new/delete
of the map, but it is necesary for the next change.

llvm-svn: 60271
2008-11-30 02:28:25 +00:00
Chris Lattner
2e3c1d37d9 calls never depend on allocations.
llvm-svn: 60268
2008-11-30 01:44:00 +00:00
Chris Lattner
2f7da36732 Fix a fixme by making memdep's handling of allocations more logical.
If we see that a load depends on the allocation of its memory with no
intervening stores, we now return a 'None' depedency instead of "Normal".
This tweaks GVN to do its optimization with the new result.

llvm-svn: 60267
2008-11-30 01:39:32 +00:00
Chris Lattner
ae0e214d84 implement a fixme by introducing a new getDependencyFromInternal
method that returns its result as a DepResultTy instead of as a
MemDepResult.  This reduces conversion back and forth.

llvm-svn: 60266
2008-11-30 01:26:32 +00:00
Chris Lattner
20e7b9bf7e Move the getNonLocalDependency method to a more logical place in
the file, no functionality change.

llvm-svn: 60265
2008-11-30 01:18:27 +00:00
Chris Lattner
9be5c5c763 REmove an old fixme, resolve another fixme by adding liberal
comments about what this class does.

llvm-svn: 60264
2008-11-30 01:17:08 +00:00
Chris Lattner
214e6decde remove a bit of incorrect code that tried to be tricky about speeding up
dependencies.  The basic situation was this: consider if we had:

  store1
  ...
  store2
  ...
  store3

Where memdep thinks that store3 depends on store2 and store2 depends 
on store1.  The problem happens when we delete store2: The code in 
question was updating dep info for store3 to be store1.  This is a
spiffy optimization, but is not safe at all, because aliasing isn't
transitive.  This bug isn't exposed today with DSE because DSE will only
zap store2 if it is identifical to store 3, and in this case, it is 
safe to update it to depend on store1.  However, memcpyopt is not so
fortunate, which is presumably why the "dropInstruction" code used to
exist.

Since this doesn't actually provide a speedup in practice, just rip the
code out.

llvm-svn: 60263
2008-11-30 01:09:30 +00:00
Chris Lattner
adf33d42ed Eliminate the dropInstruction method, which is not needed any more.
Fix a subtle iterator invalidation bug I introduced in the last commit.

llvm-svn: 60258
2008-11-29 23:30:39 +00:00