oota-llvm.git
20 years agoMove the scev expansion code into this pass, where it belongs. There is
Chris Lattner [Fri, 23 Apr 2004 21:29:48 +0000 (21:29 +0000)]
Move the scev expansion code into this pass, where it belongs.  There is
still room for cleanup, but at least the code modification is out of the
analysis now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13135 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoEliminate all of the SCEV Expansion code which is really part of the
Chris Lattner [Fri, 23 Apr 2004 21:29:03 +0000 (21:29 +0000)]
Eliminate all of the SCEV Expansion code which is really part of the
IndVars pass, not part of SCEV *analysis*.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13134 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoRemove the SCEV::expandCodeFor method, add a new SCEVVisitor class.
Chris Lattner [Fri, 23 Apr 2004 21:28:25 +0000 (21:28 +0000)]
Remove the SCEV::expandCodeFor method, add a new SCEVVisitor class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13133 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoTeach bugpoint to be a little bit smarter and avoid repeating work
Chris Lattner [Fri, 23 Apr 2004 20:36:51 +0000 (20:36 +0000)]
Teach bugpoint to be a little bit smarter and avoid repeating work

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13132 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMerged this file into the SparcV9 target.
Brian Gaeke [Fri, 23 Apr 2004 18:17:03 +0000 (18:17 +0000)]
Merged this file into the SparcV9 target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13128 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMerge TargetRegInfo.h into SparcV9RegInfo.h, which is its only subclass.
Brian Gaeke [Fri, 23 Apr 2004 18:15:48 +0000 (18:15 +0000)]
Merge TargetRegInfo.h into SparcV9RegInfo.h, which is its only subclass.
This prepares us to be able to de-virtualize and de-abstract it, and
take the register allocator bits out and move them into the register allocator
proper...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13127 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoInclude SparcV9RegInfo.h instead of TargetRegInfo.h.
Brian Gaeke [Fri, 23 Apr 2004 18:15:47 +0000 (18:15 +0000)]
Include SparcV9RegInfo.h instead of TargetRegInfo.h.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13126 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoInclude SparcV9RegInfo.h instead of TargetRegInfo.h. This serves as a bit of
Brian Gaeke [Fri, 23 Apr 2004 18:15:46 +0000 (18:15 +0000)]
Include SparcV9RegInfo.h instead of TargetRegInfo.h. This serves as a bit of
documentation that this module needs to be made independent of the
register file description of the current target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13125 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoGet rid of the old byte-at-a-time emission code used when the Sparc JIT was
Brian Gaeke [Fri, 23 Apr 2004 18:10:38 +0000 (18:10 +0000)]
Get rid of the old byte-at-a-time emission code used when the Sparc JIT was
being tested on X86, as per Chris's request.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13124 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoGo back to the interpreter main loop after performing intrinsic lowering,
Brian Gaeke [Fri, 23 Apr 2004 18:05:28 +0000 (18:05 +0000)]
Go back to the interpreter main loop after performing intrinsic lowering,
because 1) the first instruction might not be a call site, and
2) CS and SF.Caller were not getting set to point to the new call site
anyway (resulting in a crash on e.g. call %llvm.memset).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13122 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agommap of a zero length file returns null on some platforms, so hack around it.
Brian Gaeke [Fri, 23 Apr 2004 17:38:17 +0000 (17:38 +0000)]
mmap of a zero length file returns null on some platforms, so hack around it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13121 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoUse emitWordAt() to emit forward-branch fixups.
Brian Gaeke [Fri, 23 Apr 2004 17:11:16 +0000 (17:11 +0000)]
Use emitWordAt() to emit forward-branch fixups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13120 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoEmit SPARC machine code a word at a time instead of a byte at a time.
Brian Gaeke [Fri, 23 Apr 2004 17:11:15 +0000 (17:11 +0000)]
Emit SPARC machine code a word at a time instead of a byte at a time.
Use emitWordAt() to emit forward-branch fixups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13119 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoImplement emitWordAt() for the JIT emitter.
Brian Gaeke [Fri, 23 Apr 2004 17:11:14 +0000 (17:11 +0000)]
Implement emitWordAt() for the JIT emitter.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13118 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoImplement emitWordAt() for the debug emitter and the file printer emitter. (I
Brian Gaeke [Fri, 23 Apr 2004 17:11:13 +0000 (17:11 +0000)]
Implement emitWordAt() for the debug emitter and the file printer emitter. (I
am not so sure about the file printer emitter, but the debug emitter change
should be harmless.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13117 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd emitWordAt() - a quick and dirty interface that the machine-dependent
Brian Gaeke [Fri, 23 Apr 2004 17:11:12 +0000 (17:11 +0000)]
Add emitWordAt() - a quick and dirty interface that the machine-dependent
emitters can use to emit "relocations".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13116 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd command-line option to select whether to isolate or delete function from
Misha Brukman [Thu, 22 Apr 2004 23:07:39 +0000 (23:07 +0000)]
Add command-line option to select whether to isolate or delete function from
module. Default is `isolate' as before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13113 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoClarify the logic: the flag is renamed to `deleteFn' to signify it will delete
Misha Brukman [Thu, 22 Apr 2004 23:00:51 +0000 (23:00 +0000)]
Clarify the logic: the flag is renamed to `deleteFn' to signify it will delete
the function instead of isolating it. This also means the condition is reversed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13112 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a flag to choose between isolating a function or deleting the function from
Misha Brukman [Thu, 22 Apr 2004 22:52:22 +0000 (22:52 +0000)]
Add a flag to choose between isolating a function or deleting the function from
the Module. The default behavior keeps functionality as before: the chosen
function is the one that remains.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13111 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a boolean flag to delete this function from module, leaving the rest behind.
Misha Brukman [Thu, 22 Apr 2004 22:51:37 +0000 (22:51 +0000)]
Add a boolean flag to delete this function from module, leaving the rest behind.
Useful in manual debugging when bugpoint isn't quite up to snuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13110 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a space before result for readability on the command line.
Misha Brukman [Thu, 22 Apr 2004 20:02:09 +0000 (20:02 +0000)]
Add a space before result for readability on the command line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13109 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoDisable a previous patch that was causing indvars to loop infinitely :(
Chris Lattner [Thu, 22 Apr 2004 15:12:36 +0000 (15:12 +0000)]
Disable a previous patch that was causing indvars to loop infinitely :(

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13108 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd an ugly cast
Chris Lattner [Thu, 22 Apr 2004 15:00:36 +0000 (15:00 +0000)]
Add an ugly cast

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13107 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix an extremely serious thinko I made in revision 1.60 of this file.
Chris Lattner [Thu, 22 Apr 2004 14:59:40 +0000 (14:59 +0000)]
Fix an extremely serious thinko I made in revision 1.60 of this file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13106 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a method
Chris Lattner [Thu, 22 Apr 2004 14:56:51 +0000 (14:56 +0000)]
Add a method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13105 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoImplement a todo, rewriting all possible scev expressions inside of the
Chris Lattner [Wed, 21 Apr 2004 23:36:08 +0000 (23:36 +0000)]
Implement a todo, rewriting all possible scev expressions inside of the
loop.  This eliminates the extra add from the previous case, but it's
not clear that this will be a performance win overall.  Tommorows test
results will tell. :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13103 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoThis code really wants to iterate over the OPERANDS of an instruction, not
Chris Lattner [Wed, 21 Apr 2004 22:29:37 +0000 (22:29 +0000)]
This code really wants to iterate over the OPERANDS of an instruction, not
over its USES.  If it's dead it doesn't have any uses!  :)

Thanks to the fabulous and mysterious Bill Wendling for pointing this out.  :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13102 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoImplement a fixme. The helps loops that have induction variables of different
Chris Lattner [Wed, 21 Apr 2004 22:22:01 +0000 (22:22 +0000)]
Implement a fixme.  The helps loops that have induction variables of different
types in them.  Instead of creating an induction variable for all types, it
creates a single induction variable and casts to the other sizes.  This generates
this code:

no_exit:                ; preds = %entry, %no_exit
        %indvar = phi uint [ %indvar.next, %no_exit ], [ 0, %entry ]            ; <uint> [#uses=4]
***     %j.0.0 = cast uint %indvar to short             ; <short> [#uses=1]
        %indvar = cast uint %indvar to int              ; <int> [#uses=1]
        %tmp.7 = getelementptr short* %P, uint %indvar          ; <short*> [#uses=1]
        store short %j.0.0, short* %tmp.7
        %inc.0 = add int %indvar, 1             ; <int> [#uses=2]
        %tmp.2 = setlt int %inc.0, %N           ; <bool> [#uses=1]
        %indvar.next = add uint %indvar, 1              ; <uint> [#uses=1]
        br bool %tmp.2, label %no_exit, label %loopexit

instead of:

no_exit:                ; preds = %entry, %no_exit
        %indvar = phi ushort [ %indvar.next, %no_exit ], [ 0, %entry ]          ; <ushort> [#uses=2]
***     %indvar = phi uint [ %indvar.next, %no_exit ], [ 0, %entry ]            ; <uint> [#uses=3]
        %indvar = cast uint %indvar to int              ; <int> [#uses=1]
        %indvar = cast ushort %indvar to short          ; <short> [#uses=1]
        %tmp.7 = getelementptr short* %P, uint %indvar          ; <short*> [#uses=1]
        store short %indvar, short* %tmp.7
        %inc.0 = add int %indvar, 1             ; <int> [#uses=2]
        %tmp.2 = setlt int %inc.0, %N           ; <bool> [#uses=1]
        %indvar.next = add uint %indvar, 1
***     %indvar.next = add ushort %indvar, 1
        br bool %tmp.2, label %no_exit, label %loopexit

This is an improvement in register pressure, but probably doesn't happen that
often.

The more important fix will be to get rid of the redundant add.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13101 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix an incredibly nasty iterator invalidation problem. I am too spoiled by ilists :)
Chris Lattner [Wed, 21 Apr 2004 20:44:33 +0000 (20:44 +0000)]
Fix an incredibly nasty iterator invalidation problem.  I am too spoiled by ilists :)
Eventually it would be nice if CallGraph maintained an ilist of CallGraphNode's instead
of a vector of pointers to them, but today is not that day.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13100 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agobug fixed
Chris Lattner [Wed, 21 Apr 2004 18:52:06 +0000 (18:52 +0000)]
bug fixed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13099 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agonew testcase for PR326
Chris Lattner [Wed, 21 Apr 2004 18:49:35 +0000 (18:49 +0000)]
new testcase for PR326

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13098 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd doxygenified comments to functions.
Misha Brukman [Wed, 21 Apr 2004 18:36:43 +0000 (18:36 +0000)]
Add doxygenified comments to functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13097 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoI'm allergic to the word `stuff'.
Misha Brukman [Wed, 21 Apr 2004 18:27:56 +0000 (18:27 +0000)]
I'm allergic to the word `stuff'.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13096 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMake SparcV9RegInfo::getRegType() return the right answer for registers
Brian Gaeke [Wed, 21 Apr 2004 17:53:58 +0000 (17:53 +0000)]
Make SparcV9RegInfo::getRegType() return the right answer for registers
of IntCC, FloatCC, and Special types.

Make SparcV9RegInfo::getRegClassIDOfRegType() return the right answer
if you ask for the class corresponding to SpecialRegType.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13095 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoInclude cerrno (gcc-3.4 fix)
Alkis Evlogimenos [Wed, 21 Apr 2004 16:11:40 +0000 (16:11 +0000)]
Include cerrno (gcc-3.4 fix)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13091 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoDeclare iterator as public since it is defined as such (gcc-3.4 fix)
Alkis Evlogimenos [Wed, 21 Apr 2004 16:10:40 +0000 (16:10 +0000)]
Declare iterator as public since it is defined as such (gcc-3.4 fix)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13090 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix typeo
Chris Lattner [Wed, 21 Apr 2004 14:23:18 +0000 (14:23 +0000)]
Fix typeo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13089 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoREALLY fix PR324: don't delete linkonce functions until after the SCC traversal
Chris Lattner [Tue, 20 Apr 2004 22:06:53 +0000 (22:06 +0000)]
REALLY fix PR324: don't delete linkonce functions until after the SCC traversal
is done, which avoids invalidating iterators in the SCC traversal routines

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13088 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoPass the callgraph not the module
Chris Lattner [Tue, 20 Apr 2004 21:52:26 +0000 (21:52 +0000)]
Pass the callgraph not the module

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13087 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAllow getting the module from a call graph
Chris Lattner [Tue, 20 Apr 2004 21:52:12 +0000 (21:52 +0000)]
Allow getting the module from a call graph

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13086 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoChange it to take a callgraph, from which we can get a module
Chris Lattner [Tue, 20 Apr 2004 21:52:07 +0000 (21:52 +0000)]
Change it to take a callgraph, from which we can get a module

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13085 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd the ability for SCC passes to initialize and finalize themselves
Chris Lattner [Tue, 20 Apr 2004 21:30:06 +0000 (21:30 +0000)]
Add the ability for SCC passes to initialize and finalize themselves

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13084 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix PR325
Chris Lattner [Tue, 20 Apr 2004 20:26:03 +0000 (20:26 +0000)]
Fix PR325

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13081 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix PR324 and testcase: Inline/2004-04-20-InlineLinkOnce.llx
Chris Lattner [Tue, 20 Apr 2004 20:20:59 +0000 (20:20 +0000)]
Fix PR324 and testcase: Inline/2004-04-20-InlineLinkOnce.llx

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13080 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoNew testcase, distilled from PR324
Chris Lattner [Tue, 20 Apr 2004 20:15:01 +0000 (20:15 +0000)]
New testcase, distilled from PR324

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13079 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMake it legal to ask for the type of a specialreg
Brian Gaeke [Tue, 20 Apr 2004 20:12:57 +0000 (20:12 +0000)]
Make it legal to ask for the type of a specialreg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13078 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd some PRs which were fixed, but not previously mentioned in the rel. notes.
Brian Gaeke [Tue, 20 Apr 2004 18:18:46 +0000 (18:18 +0000)]
Add some PRs which were fixed, but not previously mentioned in the rel. notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13077 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd support for the select instruction
Chris Lattner [Tue, 20 Apr 2004 16:43:21 +0000 (16:43 +0000)]
Add support for the select instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13076 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMake it legal to request a load or store of %fsr.
Brian Gaeke [Mon, 19 Apr 2004 19:12:12 +0000 (19:12 +0000)]
Make it legal to request a load or store of %fsr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13073 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix typo.
Brian Gaeke [Mon, 19 Apr 2004 19:09:24 +0000 (19:09 +0000)]
Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13072 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoRegularize include guards and remove some excess whitespace.
Brian Gaeke [Mon, 19 Apr 2004 18:53:44 +0000 (18:53 +0000)]
Regularize include guards and remove some excess whitespace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13071 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoTighten up SparcV9FloatCCRegClass::getRegName()'s assertion - if you ask it for
Brian Gaeke [Mon, 19 Apr 2004 18:53:43 +0000 (18:53 +0000)]
Tighten up SparcV9FloatCCRegClass::getRegName()'s assertion - if you ask it for
the name of %fsr (as the comment in SparcV9RegClassInfo.h used to suggest)
you would walk off the end of the FloatCCRegName array.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13070 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoRegularize include guards, remove some excess whitespace and fix some comments.
Brian Gaeke [Mon, 19 Apr 2004 18:53:42 +0000 (18:53 +0000)]
Regularize include guards, remove some excess whitespace and fix some comments.
Remove the extra %fsr register from SparcV9FloatCCRegClass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13069 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoInitial checkin of a simple loop unswitching pass. It still needs work,
Chris Lattner [Mon, 19 Apr 2004 18:07:02 +0000 (18:07 +0000)]
Initial checkin of a simple loop unswitching pass.  It still needs work,
but it's a start, and seems to do it's basic job.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13068 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoNew testcase
Chris Lattner [Mon, 19 Apr 2004 18:06:34 +0000 (18:06 +0000)]
New testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13067 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd accessor for a Loop Unswitching pass
Chris Lattner [Mon, 19 Apr 2004 06:28:37 +0000 (06:28 +0000)]
Add accessor for a Loop Unswitching pass

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13066 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoEliminate a poorly conceived method
Chris Lattner [Mon, 19 Apr 2004 06:26:46 +0000 (06:26 +0000)]
Eliminate a poorly conceived method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13065 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoIt's not just a printer, it's actually an analysis too
Chris Lattner [Mon, 19 Apr 2004 03:42:32 +0000 (03:42 +0000)]
It's not just a printer, it's actually an analysis too

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13064 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAs per Chris, greatly simplify handling of external functions by using the
Misha Brukman [Mon, 19 Apr 2004 03:36:47 +0000 (03:36 +0000)]
As per Chris, greatly simplify handling of external functions by using the
wrapper idea uniformly: we can use Value::replaceAllUsesWith() instead of
special-casing by class of user.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13063 91177308-0d34-0410-b5e6-96231b3b80d8

20 years ago* Remove note about "UNSUPPORTED: " feature that just became supported
Misha Brukman [Mon, 19 Apr 2004 03:28:39 +0000 (03:28 +0000)]
* Remove note about "UNSUPPORTED: " feature that just became supported
* Fix usage of <p> tag and some grammar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13062 91177308-0d34-0410-b5e6-96231b3b80d8

20 years ago* Reorder #includes
Misha Brukman [Mon, 19 Apr 2004 03:12:35 +0000 (03:12 +0000)]
* Reorder #includes
* Wrap a long line

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13061 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoDisplay sample bugpoint output with the `code' stylesheet class.
Misha Brukman [Mon, 19 Apr 2004 03:07:29 +0000 (03:07 +0000)]
Display sample bugpoint output with the `code' stylesheet class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13060 91177308-0d34-0410-b5e6-96231b3b80d8

20 years ago* Improve file header comment
Chris Lattner [Mon, 19 Apr 2004 03:03:19 +0000 (03:03 +0000)]
* Improve file header comment
* Remove #include
* Add some methods to update loop/loopinfo
* Stop explicitly holding the loop depth in the Loop class.  Instead, just
  dynamically calculate it.  This makes it easier to update LoopInfo as a
  result of transformations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13059 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoRemove code to update loop depths
Chris Lattner [Mon, 19 Apr 2004 03:02:09 +0000 (03:02 +0000)]
Remove code to update loop depths

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13058 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd #include
Chris Lattner [Mon, 19 Apr 2004 03:01:23 +0000 (03:01 +0000)]
Add #include

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13057 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd notes about PRs 38 and 290 being fixed.
Misha Brukman [Mon, 19 Apr 2004 03:01:01 +0000 (03:01 +0000)]
Add notes about PRs 38 and 290 being fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13056 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoWrap (some) lines to 80 cols.
Misha Brukman [Mon, 19 Apr 2004 02:56:49 +0000 (02:56 +0000)]
Wrap (some) lines to 80 cols.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13055 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFinally implement rewriting global initializers which use external functions
Misha Brukman [Mon, 19 Apr 2004 01:12:01 +0000 (01:12 +0000)]
Finally implement rewriting global initializers which use external functions
by creating an internal wrapper function with same signature as the external
function, and use it instead of the "real" function.

The wrapper then calls the external function using the same JIT function
resolution API that has been used before for rewriting instructions, since the
wrapper has an explicit call instruction which we can rewrite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13054 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoProvide an interface that is more convenient for iterating over the blocks
Chris Lattner [Sun, 18 Apr 2004 23:37:03 +0000 (23:37 +0000)]
Provide an interface that is more convenient for iterating over the blocks
in a loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13052 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMove isLoopInvariant to the Loop class
Chris Lattner [Sun, 18 Apr 2004 22:46:08 +0000 (22:46 +0000)]
Move isLoopInvariant to the Loop class

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13051 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd new method
Chris Lattner [Sun, 18 Apr 2004 22:45:27 +0000 (22:45 +0000)]
Add new method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13050 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd new method.
Chris Lattner [Sun, 18 Apr 2004 22:45:16 +0000 (22:45 +0000)]
Add new method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13049 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoCorrect rewriting of exit blocks after my last patch
Chris Lattner [Sun, 18 Apr 2004 22:27:10 +0000 (22:27 +0000)]
Correct rewriting of exit blocks after my last patch

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13048 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix computation of exit blocks
Chris Lattner [Sun, 18 Apr 2004 22:21:41 +0000 (22:21 +0000)]
Fix computation of exit blocks

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13047 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoLoop exit sets are no longer explicitly held, they are dynamically computed on demand.
Chris Lattner [Sun, 18 Apr 2004 22:15:13 +0000 (22:15 +0000)]
Loop exit sets are no longer explicitly held, they are dynamically computed on demand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13046 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoChange the ExitBlocks list from being explicitly contained in the Loop
Chris Lattner [Sun, 18 Apr 2004 22:14:10 +0000 (22:14 +0000)]
Change the ExitBlocks list from being explicitly contained in the Loop
structure to being dynamically computed on demand.  This makes updating
loop information MUCH easier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13045 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoSpiff is no longer in our tree
Chris Lattner [Sun, 18 Apr 2004 18:26:38 +0000 (18:26 +0000)]
Spiff is no longer in our tree

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13041 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoReduce the unrolling limit
Chris Lattner [Sun, 18 Apr 2004 18:06:14 +0000 (18:06 +0000)]
Reduce the unrolling limit

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13040 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoIf the preheader of the loop was the entry block of the function, make sure
Chris Lattner [Sun, 18 Apr 2004 17:38:42 +0000 (17:38 +0000)]
If the preheader of the loop was the entry block of the function, make sure
that the exit block of the loop becomes the new entry block of the function.

This was causing a verifier assertion on 252.eon.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13039 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoBe much more careful about how we update instructions outside of the loop
Chris Lattner [Sun, 18 Apr 2004 17:32:39 +0000 (17:32 +0000)]
Be much more careful about how we update instructions outside of the loop
using instructions inside of the loop.  This should fix the MishaTest failure
from last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13038 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAnother testcase
Chris Lattner [Sun, 18 Apr 2004 06:55:57 +0000 (06:55 +0000)]
Another testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13037 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoImplement method
Chris Lattner [Sun, 18 Apr 2004 06:54:48 +0000 (06:54 +0000)]
Implement method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13036 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a new method
Chris Lattner [Sun, 18 Apr 2004 06:54:34 +0000 (06:54 +0000)]
Add a new method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13035 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAfter unrolling our single basic block loop, fold it into the preheader and exit
Chris Lattner [Sun, 18 Apr 2004 06:27:43 +0000 (06:27 +0000)]
After unrolling our single basic block loop, fold it into the preheader and exit
block.  The primary motivation for doing this is that we can now unroll nested loops.

This makes a pretty big difference in some cases.  For example, in 183.equake,
we are now beating the native compiler with the CBE, and we are a lot closer
with LLC.

I'm now going to play around a bit with the unroll factor and see what effect
it really has.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13034 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix a bug: this does not preserve the CFG!
Chris Lattner [Sun, 18 Apr 2004 05:38:37 +0000 (05:38 +0000)]
Fix a bug: this does not preserve the CFG!

While we're at it, add support for updating loop information correctly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13033 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd a new method, add a check missing that caused a segfault if a loop didn't
Chris Lattner [Sun, 18 Apr 2004 05:38:05 +0000 (05:38 +0000)]
Add a new method, add a check missing that caused a segfault if a loop didn't
have a canonical indvar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13032 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAllow clients to delete loops, add a new method
Chris Lattner [Sun, 18 Apr 2004 05:37:42 +0000 (05:37 +0000)]
Allow clients to delete loops, add a new method

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13031 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMove loop optimization passes up, add loop unroller
Chris Lattner [Sun, 18 Apr 2004 05:21:01 +0000 (05:21 +0000)]
Move loop optimization passes up, add loop unroller

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13030 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd prototype
Chris Lattner [Sun, 18 Apr 2004 05:20:32 +0000 (05:20 +0000)]
Add prototype

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13029 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoInitial checkin of a simple loop unroller. This pass is extremely basic and
Chris Lattner [Sun, 18 Apr 2004 05:20:17 +0000 (05:20 +0000)]
Initial checkin of a simple loop unroller.  This pass is extremely basic and
limited.  Even in it's extremely simple state (it can only *fully* unroll single
basic block loops that execute a constant number of times), it already helps improve
performance a LOT on some benchmarks, particularly with the native code generators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13028 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoMake the tail duplication threshold accessible from the command line instead of hardcoded
Chris Lattner [Sun, 18 Apr 2004 00:52:43 +0000 (00:52 +0000)]
Make the tail duplication threshold accessible from the command line instead of hardcoded

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13025 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix a memory leak. We leaked the vector holding the entries in switch tables.
Chris Lattner [Sat, 17 Apr 2004 23:49:15 +0000 (23:49 +0000)]
Fix a memory leak.  We leaked the vector holding the entries in switch tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13023 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoNew testcase
Chris Lattner [Sat, 17 Apr 2004 23:00:51 +0000 (23:00 +0000)]
New testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13020 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd the ability to compute exit values for complex loop using unanalyzable
Chris Lattner [Sat, 17 Apr 2004 22:58:41 +0000 (22:58 +0000)]
Add the ability to compute exit values for complex loop using unanalyzable
operations.  This allows us to compile this testcase:

int main() {
        int h = 1;
         do h = 3 * h + 1; while (h <= 256);
        printf("%d\n", h);
        return 0;
}

into this:

int %main() {
entry:
        call void %__main( )
        %tmp.6 = call int (sbyte*, ...)* %printf( sbyte* getelementptr ([4 x sbyte]*  %.str_1, long 0, long 0), int 364 )        ; <int> [#uses=0]
        ret int 0
}

This testcase was taken directly from 256.bzip2, believe it or not.

This code is not as general as I would like.  Next up is to refactor it
a bit to handle more cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13019 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoIf the loop executes a constant number of times, try a bit harder to replace
Chris Lattner [Sat, 17 Apr 2004 18:44:09 +0000 (18:44 +0000)]
If the loop executes a constant number of times, try a bit harder to replace
exit values.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13018 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd the ability to compute trip counts that are only controlled by constants
Chris Lattner [Sat, 17 Apr 2004 18:36:24 +0000 (18:36 +0000)]
Add the ability to compute trip counts that are only controlled by constants
even if the loop is using expressions that we can't compute as a closed-form.
This allows us to calculate that this function always returns 55:

int test() {
  double X;
  int Count = 0;
  for (X = 100; X > 1; X = sqrt(X), ++Count)
    /*empty*/;
  return Count;
}

And allows us to compute trip counts for loops like:

        int h = 1;
         do h = 3 * h + 1; while (h <= 256);

(which occurs in bzip2), and for this function, which occurs after inlining
and other optimizations:

int popcount()
{
   int x = 666;
  int result = 0;
  while (x != 0) {
    result = result + (x & 0x1);
    x = x >> 1;
  }
  return result;
}

We still cannot compute the exit values of result or h in the two loops above,
which means we cannot delete the loop, but we are getting closer.  Being able to
compute a constant trip count for these two loops will allow us to unroll them
completely though.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13017 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix a HUGE pessimization on X86. The indvars pass was taking this
Chris Lattner [Sat, 17 Apr 2004 18:16:10 +0000 (18:16 +0000)]
Fix a HUGE pessimization on X86.  The indvars pass was taking this
(familiar) function:

int _strlen(const char *str) {
    int len = 0;
    while (*str++) len++;
    return len;
}

And transforming it to use a ulong induction variable, because the type of
the pointer index was left as a constant long.  This is obviously very bad.

The fix is to shrink long constants in getelementptr instructions to intptr_t,
making the indvars pass insert a uint induction variable, which is much more
efficient.

Here's the before code for this function:

int %_strlen(sbyte* %str) {
entry:
        %tmp.13 = load sbyte* %str              ; <sbyte> [#uses=1]
        %tmp.24 = seteq sbyte %tmp.13, 0                ; <bool> [#uses=1]
        br bool %tmp.24, label %loopexit, label %no_exit

no_exit:                ; preds = %entry, %no_exit
***     %indvar = phi uint [ %indvar.next, %no_exit ], [ 0, %entry ]            ; <uint> [#uses=2]
***     %indvar = phi ulong [ %indvar.next, %no_exit ], [ 0, %entry ]           ; <ulong> [#uses=2]
        %indvar1 = cast ulong %indvar to uint           ; <uint> [#uses=1]
        %inc.02.sum = add uint %indvar1, 1              ; <uint> [#uses=1]
        %inc.0.0 = getelementptr sbyte* %str, uint %inc.02.sum          ; <sbyte*> [#uses=1]
        %tmp.1 = load sbyte* %inc.0.0           ; <sbyte> [#uses=1]
        %tmp.2 = seteq sbyte %tmp.1, 0          ; <bool> [#uses=1]
        %indvar.next = add ulong %indvar, 1             ; <ulong> [#uses=1]
        %indvar.next = add uint %indvar, 1              ; <uint> [#uses=1]
        br bool %tmp.2, label %loopexit.loopexit, label %no_exit

loopexit.loopexit:              ; preds = %no_exit
        %indvar = cast uint %indvar to int              ; <int> [#uses=1]
        %inc.1 = add int %indvar, 1             ; <int> [#uses=1]
        ret int %inc.1

loopexit:               ; preds = %entry
        ret int 0
}

Here's the after code:

int %_strlen(sbyte* %str) {
entry:
        %inc.02 = getelementptr sbyte* %str, uint 1             ; <sbyte*> [#uses=1]
        %tmp.13 = load sbyte* %str              ; <sbyte> [#uses=1]
        %tmp.24 = seteq sbyte %tmp.13, 0                ; <bool> [#uses=1]
        br bool %tmp.24, label %loopexit, label %no_exit

no_exit:                ; preds = %entry, %no_exit
***     %indvar = phi uint [ %indvar.next, %no_exit ], [ 0, %entry ]            ; <uint> [#uses=3]
        %indvar = cast uint %indvar to int              ; <int> [#uses=1]
        %inc.0.0 = getelementptr sbyte* %inc.02, uint %indvar           ; <sbyte*> [#uses=1]
        %inc.1 = add int %indvar, 1             ; <int> [#uses=1]
        %tmp.1 = load sbyte* %inc.0.0           ; <sbyte> [#uses=1]
        %tmp.2 = seteq sbyte %tmp.1, 0          ; <bool> [#uses=1]
        %indvar.next = add uint %indvar, 1              ; <uint> [#uses=1]
        br bool %tmp.2, label %loopexit, label %no_exit

loopexit:               ; preds = %entry, %no_exit
        %len.0.1 = phi int [ 0, %entry ], [ %inc.1, %no_exit ]          ; <int> [#uses=1]
        ret int %len.0.1
}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13016 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoEven if there are not any induction variables in the loop, if we can compute
Chris Lattner [Sat, 17 Apr 2004 18:08:33 +0000 (18:08 +0000)]
Even if there are not any induction variables in the loop, if we can compute
the trip count for the loop, insert one so that we can canonicalize the exit
condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13015 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoAdd support for evaluation of exp/log/log10/pow
Chris Lattner [Fri, 16 Apr 2004 22:35:33 +0000 (22:35 +0000)]
Add support for evaluation of exp/log/log10/pow

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13011 91177308-0d34-0410-b5e6-96231b3b80d8

20 years agoFix some really nasty dominance bugs that were exposed by my patch to
Chris Lattner [Fri, 16 Apr 2004 18:08:07 +0000 (18:08 +0000)]
Fix some really nasty dominance bugs that were exposed by my patch to
make the verifier more strict.  This fixes building zlib

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@13002 91177308-0d34-0410-b5e6-96231b3b80d8