oota-llvm.git
19 years agoAdd project llvm-proj to Visual Studio
Jeff Cohen [Thu, 20 Jan 2005 04:41:49 +0000 (04:41 +0000)]
Add project llvm-proj to Visual Studio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19708 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoSupport targets that do not use i8 shift amounts.
Chris Lattner [Wed, 19 Jan 2005 22:31:21 +0000 (22:31 +0000)]
Support targets that do not use i8 shift amounts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19707 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd two optimizations. The first folds (X+Y)-X -> Y
Chris Lattner [Wed, 19 Jan 2005 21:50:18 +0000 (21:50 +0000)]
Add two optimizations.  The first folds (X+Y)-X -> Y
The second folds operations into selects, e.g. (select C, (X+Y), (Y+Z))
-> (Y+(select C, X, Z)

This occurs a few times across spec, e.g.

         select    add/sub
mesa:    83        0
povray:  5         2
gcc      4         2
parser   0         22
perlbmk  13        30
twolf    0         3

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19706 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd some new tests
Chris Lattner [Wed, 19 Jan 2005 21:48:31 +0000 (21:48 +0000)]
Add some new tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19705 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd an assertion that would have made more sense to duraid
Chris Lattner [Wed, 19 Jan 2005 21:32:07 +0000 (21:32 +0000)]
Add an assertion that would have made more sense to duraid

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19704 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd support for targets that pass args in registers to calls.
Chris Lattner [Wed, 19 Jan 2005 20:24:35 +0000 (20:24 +0000)]
Add support for targets that pass args in registers to calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19703 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd an accessor for targets that pass args in regs
Chris Lattner [Wed, 19 Jan 2005 20:19:58 +0000 (20:19 +0000)]
Add an accessor for targets that pass args in regs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19702 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFold single use token factor nodes into other token factor nodes.
Chris Lattner [Wed, 19 Jan 2005 19:10:54 +0000 (19:10 +0000)]
Fold single use token factor nodes into other token factor nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19701 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRealize the individual pieces of an expanded copytoreg/store/load are
Chris Lattner [Wed, 19 Jan 2005 18:02:17 +0000 (18:02 +0000)]
Realize the individual pieces of an expanded copytoreg/store/load are
independent of each other.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19700 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoKnow some identities about tokenfactor nodes.
Chris Lattner [Wed, 19 Jan 2005 18:01:40 +0000 (18:01 +0000)]
Know some identities about tokenfactor nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19699 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoKnow some simple identities. This improves codegen for (1LL << N).
Chris Lattner [Wed, 19 Jan 2005 17:29:49 +0000 (17:29 +0000)]
Know some simple identities.  This improves codegen for (1LL << N).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19698 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix a problem where were were literally selecting for INCREASED register
Chris Lattner [Wed, 19 Jan 2005 17:24:34 +0000 (17:24 +0000)]
Fix a problem where were were literally selecting for INCREASED register
pressure, not decreases register pressure.  Fix problem where we accidentally
swapped the operands of SHLD, which caused fourinarow to fail.  This fixes
fourinarow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19697 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoJust in case, handle something that is both a use and a def.
Chris Lattner [Wed, 19 Jan 2005 17:11:51 +0000 (17:11 +0000)]
Just in case, handle something that is both a use and a def.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19696 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoWhen an instruction moves, make sure to update the VarInfo::Kills list as
Chris Lattner [Wed, 19 Jan 2005 17:09:15 +0000 (17:09 +0000)]
When an instruction moves, make sure to update the VarInfo::Kills list as
well as all of teh other stuff in livevar. This fixes the compiler crash
on fourinarow last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19695 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoWhen commuting these instructions, make sure to actually swap the operands too.
Chris Lattner [Wed, 19 Jan 2005 16:55:52 +0000 (16:55 +0000)]
When commuting these instructions, make sure to actually swap the operands too.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19694 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix 'raise' to work with packed types. Patch by Morten Ofstad.
Chris Lattner [Wed, 19 Jan 2005 16:16:35 +0000 (16:16 +0000)]
Fix 'raise' to work with packed types.  Patch by Morten Ofstad.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19693 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImplement Regression/CodeGen/X86/rotate.ll: emit rotate instructions (which
Chris Lattner [Wed, 19 Jan 2005 08:07:05 +0000 (08:07 +0000)]
Implement Regression/CodeGen/X86/rotate.ll: emit rotate instructions (which
typically cost 1 cycle) instead of shld/shrd instruction (which are typically
6 or more cycles).  This also saves code space.

For example, instead of emitting:

rotr:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %CL, BYTE PTR [%ESP + 8]
        shrd %EAX, %EAX, %CL
        ret
rotli:
        mov %EAX, DWORD PTR [%ESP + 4]
        shrd %EAX, %EAX, 27
        ret

Emit:

rotr32:
        mov %CL, BYTE PTR [%ESP + 8]
        mov %EAX, DWORD PTR [%ESP + 4]
        ror %EAX, %CL
        ret
rotli32:
        mov %EAX, DWORD PTR [%ESP + 4]
        ror %EAX, 27
        ret

We also emit byte rotate instructions which do not have a sh[lr]d counterpart
at all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19692 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoNew testcase for rotate instructions. Each function should codegen to a
Chris Lattner [Wed, 19 Jan 2005 08:04:08 +0000 (08:04 +0000)]
New testcase for rotate instructions.  Each function should codegen to a
rotate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19691 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd rotate instructions.
Chris Lattner [Wed, 19 Jan 2005 07:50:03 +0000 (07:50 +0000)]
Add rotate instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19690 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoMatch 16-bit shld/shrd instructions as well, implementing shift-double.llx:test5
Chris Lattner [Wed, 19 Jan 2005 07:37:26 +0000 (07:37 +0000)]
Match 16-bit shld/shrd instructions as well, implementing shift-double.llx:test5

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19689 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd a test for 16-bit sh*d.
Chris Lattner [Wed, 19 Jan 2005 07:37:01 +0000 (07:37 +0000)]
Add a test for 16-bit sh*d.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19688 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImprove coverage of the X86 instruction set by adding 16-bit shift doubles.
Chris Lattner [Wed, 19 Jan 2005 07:31:24 +0000 (07:31 +0000)]
Improve coverage of the X86 instruction set by adding 16-bit shift doubles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19687 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoTeach the code generator that shrd/shld is commutable if it has an immediate.
Chris Lattner [Wed, 19 Jan 2005 07:11:01 +0000 (07:11 +0000)]
Teach the code generator that shrd/shld is commutable if it has an immediate.
This allows us to generate this:

foo:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %EDX, DWORD PTR [%ESP + 8]
        shld %EDX, %EDX, 2
        shl %EAX, 2
        ret

instead of this:

foo:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %ECX, DWORD PTR [%ESP + 8]
        mov %EDX, %EAX
        shrd %EDX, %ECX, 30
        shl %EAX, 2
        ret

Note the magically transmogrifying immediate.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19686 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoUse the TargetInstrInfo::commuteInstruction method to commute instructions
Chris Lattner [Wed, 19 Jan 2005 07:08:42 +0000 (07:08 +0000)]
Use the TargetInstrInfo::commuteInstruction method to commute instructions
instead of doing it manually.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19685 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFinegrainify namespacification
Chris Lattner [Wed, 19 Jan 2005 06:53:34 +0000 (06:53 +0000)]
Finegrainify namespacification
Add default impl of commuteInstruction
Add notes about ugly V9 code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19684 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd a new method, described in the comment.
Chris Lattner [Wed, 19 Jan 2005 06:53:02 +0000 (06:53 +0000)]
Add a new method, described in the comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19683 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoEnsure that each these functions generates a sh[rl]d instruction.
Chris Lattner [Wed, 19 Jan 2005 06:30:36 +0000 (06:30 +0000)]
Ensure that each these functions generates a sh[rl]d instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19682 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoCodegen long >> 2 to this:
Chris Lattner [Wed, 19 Jan 2005 06:18:43 +0000 (06:18 +0000)]
Codegen long >> 2 to this:

foo:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %EDX, DWORD PTR [%ESP + 8]
        shrd %EAX, %EDX, 2
        sar %EDX, 2
        ret

instead of this:

test1:
        mov %ECX, DWORD PTR [%ESP + 4]
        shr %ECX, 2
        mov %EDX, DWORD PTR [%ESP + 8]
        mov %EAX, %EDX
        shl %EAX, 30
        or %EAX, %ECX
        sar %EDX, 2
        ret

and long << 2 to this:

foo:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %ECX, DWORD PTR [%ESP + 8]
***     mov %EDX, %EAX
        shrd %EDX, %ECX, 30
        shl %EAX, 2
        ret

instead of this:

foo:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %ECX, %EAX
        shr %ECX, 30
        mov %EDX, DWORD PTR [%ESP + 8]
        shl %EDX, 2
        or %EDX, %ECX
        shl %EAX, 2
        ret

The extra copy (marked ***) can be eliminated when I teach the code generator
that shrd32rri8 is really commutative.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19681 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd missing data types for VC++
Jeff Cohen [Wed, 19 Jan 2005 05:08:31 +0000 (05:08 +0000)]
Add missing data types for VC++

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19680 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImplement a way of expanding shifts. This applies to targets that offer
Chris Lattner [Wed, 19 Jan 2005 04:19:40 +0000 (04:19 +0000)]
Implement a way of expanding shifts.  This applies to targets that offer
select operations or to shifts that are by a constant.  This automatically
implements (with no special code) all of the special cases for shift by 32,
shift by < 32 and shift by > 32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19679 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoX86 shifts mask the amount.
Chris Lattner [Wed, 19 Jan 2005 03:36:30 +0000 (03:36 +0000)]
X86 shifts mask the amount.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19678 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd a hook to find out how the target handles shift amounts that are out of
Chris Lattner [Wed, 19 Jan 2005 03:36:14 +0000 (03:36 +0000)]
Add a hook to find out how the target handles shift amounts that are out of
range.  Either they are undefined (the default), they mask the shift amount
to the size of the register (X86, Alpha, etc), or they extend the shift (PPC).

This defaults to undefined, which is conservatively correct.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19677 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoMove all data members to the end of the class.
Chris Lattner [Wed, 19 Jan 2005 03:36:03 +0000 (03:36 +0000)]
Move all data members to the end of the class.

Add a hook to find out how the target handles shift amounts that are out of
range.  Either they are undefined (the default), they mask the shift amount
to the size of the register (X86, Alpha, etc), or they extend the shift (PPC).

This defaults to undefined, which is conservatively correct.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19676 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoZero is cheaper than sign extend.
Chris Lattner [Tue, 18 Jan 2005 21:57:59 +0000 (21:57 +0000)]
Zero is cheaper than sign extend.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19675 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoCode to handle FP_EXTEND is dead now. X86 doesn't support any data types to
Chris Lattner [Tue, 18 Jan 2005 20:05:56 +0000 (20:05 +0000)]
Code to handle FP_EXTEND is dead now.  X86 doesn't support any data types to
FP_EXTEND from!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19674 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRemove more dead code.
Chris Lattner [Tue, 18 Jan 2005 19:50:08 +0000 (19:50 +0000)]
Remove more dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19673 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoThe selection dag code handles the promotions from F32 to F64 for us, so we
Chris Lattner [Tue, 18 Jan 2005 19:46:54 +0000 (19:46 +0000)]
The selection dag code handles the promotions from F32 to F64 for us, so we
don't need to even think about F32 in the X86 code anymore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19672 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix some fixmes (promoting bools for select and brcond), fix promotion
Chris Lattner [Tue, 18 Jan 2005 19:27:06 +0000 (19:27 +0000)]
Fix some fixmes (promoting bools for select and brcond), fix promotion
of zero and sign extends.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19671 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoKeep track of the retval type as well.
Chris Lattner [Tue, 18 Jan 2005 19:26:36 +0000 (19:26 +0000)]
Keep track of the retval type as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19670 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoKeep track of the returned value type as well.
Chris Lattner [Tue, 18 Jan 2005 19:26:18 +0000 (19:26 +0000)]
Keep track of the returned value type as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19669 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoTeach legalize to promote copy(from|to)reg, instead of making the isel pass
Chris Lattner [Tue, 18 Jan 2005 17:54:55 +0000 (17:54 +0000)]
Teach legalize to promote copy(from|to)reg, instead of making the isel pass
do it.  This results in better code on X86 for floats (because if strict
precision is not required, we can elide some more expensive double -> float
conversions like the old isel did), and allows other targets to emit
CopyFromRegs that are not legal for arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19668 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix 124.m88ksim.
Chris Lattner [Tue, 18 Jan 2005 17:35:28 +0000 (17:35 +0000)]
Fix 124.m88ksim.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19667 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd project llvm-ld to Visual Studio
Jeff Cohen [Tue, 18 Jan 2005 05:44:50 +0000 (05:44 +0000)]
Add project llvm-ld to Visual Studio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19665 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd project llvm-nm to Visual Studio
Jeff Cohen [Tue, 18 Jan 2005 05:44:25 +0000 (05:44 +0000)]
Add project llvm-nm to Visual Studio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19664 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd project llvm-ld to Visual Studio
Jeff Cohen [Tue, 18 Jan 2005 05:39:37 +0000 (05:39 +0000)]
Add project llvm-ld to Visual Studio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19663 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd llvm-bcanalyzer project to Visual Studio
Jeff Cohen [Tue, 18 Jan 2005 05:31:34 +0000 (05:31 +0000)]
Add llvm-bcanalyzer project to Visual Studio

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19662 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDo not emit loads multiple times, potentially in the wrong places.
Chris Lattner [Tue, 18 Jan 2005 04:18:32 +0000 (04:18 +0000)]
Do not emit loads multiple times, potentially in the wrong places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19661 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoMinor changes.
Tanya Lattner [Tue, 18 Jan 2005 04:15:41 +0000 (04:15 +0000)]
Minor changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19660 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoEliminate bad assertions.
Chris Lattner [Tue, 18 Jan 2005 04:00:54 +0000 (04:00 +0000)]
Eliminate bad assertions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19659 91177308-0d34-0410-b5e6-96231b3b80d8

19 years ago* Eliminate the TokenSet and just use the ExprMap for both tokens and values.
Chris Lattner [Tue, 18 Jan 2005 03:51:59 +0000 (03:51 +0000)]
* Eliminate the TokenSet and just use the ExprMap for both tokens and values.
* Insert some really pedantic assertions that will notice when we emit the
  same loads more than one time, exposing bugs.  This turns a miscompilation in
  bzip2 into a compile-fail.  yaay.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19658 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoTeach legalize to promote SetCC results.
Chris Lattner [Tue, 18 Jan 2005 02:59:52 +0000 (02:59 +0000)]
Teach legalize to promote SetCC results.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19657 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAllow setcc operations to have nonbool types.
Chris Lattner [Tue, 18 Jan 2005 02:52:03 +0000 (02:52 +0000)]
Allow setcc operations to have nonbool types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19656 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAllow setcc operations to have non-bool types.
Chris Lattner [Tue, 18 Jan 2005 02:51:41 +0000 (02:51 +0000)]
Allow setcc operations to have non-bool types.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19655 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRely on the code in MatchAddress to do this work. Otherwise we fail to
Chris Lattner [Tue, 18 Jan 2005 02:25:52 +0000 (02:25 +0000)]
Rely on the code in MatchAddress to do this work.  Otherwise we fail to
match (X+Y)+(Z << 1), because we match the X+Y first, consuming the index
register, then there is no place to put the Z.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19652 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix the completely broken FP constant folds for setcc's.
Chris Lattner [Tue, 18 Jan 2005 02:11:55 +0000 (02:11 +0000)]
Fix the completely broken FP constant folds for setcc's.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19651 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix a problem where probing for addressing modes caused expressions to be
Chris Lattner [Tue, 18 Jan 2005 01:06:26 +0000 (01:06 +0000)]
Fix a problem where probing for addressing modes caused expressions to be
emitted too early.  In particular, this fixes
Regression/CodeGen/X86/regpressure.ll:regpressure3.

This also improves the 2nd basic block in 164.gzip:flush_block, which went from

.LBBflush_block_1:      # loopentry.1.i
        movzx %EAX, WORD PTR [dyn_ltree + 20]
        movzx %ECX, WORD PTR [dyn_ltree + 16]
        mov DWORD PTR [%ESP + 32], %ECX
        movzx %ECX, WORD PTR [dyn_ltree + 12]
        movzx %EDX, WORD PTR [dyn_ltree + 8]
        movzx %EBX, WORD PTR [dyn_ltree + 4]
        mov DWORD PTR [%ESP + 36], %EBX
        movzx %EBX, WORD PTR [dyn_ltree]
        add DWORD PTR [%ESP + 36], %EBX
        add %EDX, DWORD PTR [%ESP + 36]
        add %ECX, %EDX
        add DWORD PTR [%ESP + 32], %ECX
        add %EAX, DWORD PTR [%ESP + 32]
        movzx %ECX, WORD PTR [dyn_ltree + 24]
        add %EAX, %ECX
        mov %ECX, 0
        mov %EDX, %ECX

to

.LBBflush_block_1:      # loopentry.1.i
        movzx %EAX, WORD PTR [dyn_ltree]
        movzx %ECX, WORD PTR [dyn_ltree + 4]
        add %ECX, %EAX
        movzx %EAX, WORD PTR [dyn_ltree + 8]
        add %EAX, %ECX
        movzx %ECX, WORD PTR [dyn_ltree + 12]
        add %ECX, %EAX
        movzx %EAX, WORD PTR [dyn_ltree + 16]
        add %EAX, %ECX
        movzx %ECX, WORD PTR [dyn_ltree + 20]
        add %ECX, %EAX
        movzx %EAX, WORD PTR [dyn_ltree + 24]
        add %ECX, %EAX
        mov %EAX, 0
        mov %EDX, %EAX

... which results in less spilling in the function.

This change alone speeds up 164.gzip from 37.23s to 36.24s on apoc.  The
default isel takes 37.31s.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19650 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix indentation.
Chris Lattner [Mon, 17 Jan 2005 23:25:45 +0000 (23:25 +0000)]
Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19649 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoThis is a carefully contrived testcase where the X86 ISel is emitting all loads
Chris Lattner [Mon, 17 Jan 2005 23:16:01 +0000 (23:16 +0000)]
This is a carefully contrived testcase where the X86 ISel is emitting all loads
before other ops, causing it to spill like mad.  This occurs in
164.gzip:flush_block.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19648 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDon't bother using max here.
Chris Lattner [Mon, 17 Jan 2005 23:02:13 +0000 (23:02 +0000)]
Don't bother using max here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19647 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDo not give token factor nodes outrageous weights
Chris Lattner [Mon, 17 Jan 2005 22:56:09 +0000 (22:56 +0000)]
Do not give token factor nodes outrageous weights

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19645 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoNon-volatile loads can be freely reordered against each other. This fixes
Chris Lattner [Mon, 17 Jan 2005 22:19:26 +0000 (22:19 +0000)]
Non-volatile loads can be freely reordered against each other.  This fixes
X86/reg-pressure.ll again, and allows us to do nice things in other cases.
For example, we now codegen this sort of thing:

int %loadload(int *%X, int* %Y) {
  %Z = load int* %Y
  %Y = load int* %X      ;; load between %Z and store
  %Q = add int %Z, 1
  store int %Q, int* %Y
  ret int %Y
}

Into this:

loadload:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %EAX, DWORD PTR [%EAX]
        mov %ECX, DWORD PTR [%ESP + 8]
        inc DWORD PTR [%ECX]
        ret

where we weren't able to form the 'inc [mem]' before.  This also lets the
instruction selector emit loads in any order it wants to, which can be good
for register pressure as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19644 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoTwo changes:
Chris Lattner [Mon, 17 Jan 2005 22:10:42 +0000 (22:10 +0000)]
Two changes:
 1. Fold  [mem] += (1|-1) into inc [mem]/dec [mem] to save some icache space.
 2. Do not let token factor nodes prevent forming '[mem] op= val' folds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19643 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDon't call SelectionDAG.getRoot() directly, go through a forwarding method.
Chris Lattner [Mon, 17 Jan 2005 19:43:36 +0000 (19:43 +0000)]
Don't call SelectionDAG.getRoot() directly, go through a forwarding method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19642 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRefactor load/op/store folding into it's own method, no functionality changes.
Chris Lattner [Mon, 17 Jan 2005 19:25:26 +0000 (19:25 +0000)]
Refactor load/op/store folding into it's own method, no functionality changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19641 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImplement a target independent optimization to codegen arguments only into
Chris Lattner [Mon, 17 Jan 2005 17:55:19 +0000 (17:55 +0000)]
Implement a target independent optimization to codegen arguments only into
the basic block that uses them if possible.  This is a big win on X86, as it
lets us fold the argument loads into instructions and reduce register pressure
(by not loading all of the arguments in the entry block).

For this (contrived to show the optimization) testcase:

int %argtest(int %A, int %B) {
        %X = sub int 12345, %A
        br label %L
L:
        %Y = add int %X, %B
        ret int %Y
}

we used to produce:

argtest:
        mov %ECX, DWORD PTR [%ESP + 4]
        mov %EAX, 12345
        sub %EAX, %ECX
        mov %EDX, DWORD PTR [%ESP + 8]
.LBBargtest_1:  # L
        add %EAX, %EDX
        ret

now we produce:

argtest:
        mov %EAX, 12345
        sub %EAX, DWORD PTR [%ESP + 4]
.LBBargtest_1:  # L
        add %EAX, DWORD PTR [%ESP + 8]
        ret

This also fixes the FIXME in the code.

BTW, this occurs in real code.  164.gzip shrinks from 8623 to 8608 lines of
.s file.  The stack frame in huft_build shrinks from 1644->1628 bytes,
inflate_codes shrinks from 116->108 bytes, and inflate_block from 2620->2612,
due to fewer spills.

Take that alkis. :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19639 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix a major regression last night that prevented us from producing [mem] op= reg
Chris Lattner [Mon, 17 Jan 2005 17:49:14 +0000 (17:49 +0000)]
Fix a major regression last night that prevented us from producing [mem] op= reg
operations.

The body of the if is less indented but unmodified in this patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19638 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRefactor code into a new method.
Chris Lattner [Mon, 17 Jan 2005 17:15:02 +0000 (17:15 +0000)]
Refactor code into a new method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19635 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoMake methods private, add a method.
Chris Lattner [Mon, 17 Jan 2005 17:14:43 +0000 (17:14 +0000)]
Make methods private, add a method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19634 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoCodegen this:
Chris Lattner [Mon, 17 Jan 2005 06:48:02 +0000 (06:48 +0000)]
Codegen this:

int %foo(int %X) {
        %T = add int %X, 13
        %S = mul int %T, 3
        ret int %S
}

as this:

        mov %ECX, DWORD PTR [%ESP + 4]
        lea %EAX, DWORD PTR [%ECX + 2*%ECX + 39]
        ret

instead of this:

        mov %ECX, DWORD PTR [%ESP + 4]
        mov %EAX, %ECX
        add %EAX, 13
        imul %EAX, %EAX, 3
        ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19633 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdded tmp instructions to preserve ssa.
Tanya Lattner [Mon, 17 Jan 2005 06:47:26 +0000 (06:47 +0000)]
Added tmp instructions to preserve ssa.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19632 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix test/Regression/CodeGen/X86/2005-01-17-CycleInDAG.ll and 132.ijpeg.
Chris Lattner [Mon, 17 Jan 2005 06:26:58 +0000 (06:26 +0000)]
Fix test/Regression/CodeGen/X86/2005-01-17-CycleInDAG.ll and 132.ijpeg.
Do not fold a load into an operation if it will induce a cycle in the DAG.

Repeat after me: dAg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19631 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoNew testcase for a problem that occurred in 132.ijpeg
Chris Lattner [Mon, 17 Jan 2005 06:25:59 +0000 (06:25 +0000)]
New testcase for a problem that occurred in 132.ijpeg

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19630 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDelete PHI nodes that are not dead but are locked in a cycle of single
Chris Lattner [Mon, 17 Jan 2005 05:10:15 +0000 (05:10 +0000)]
Delete PHI nodes that are not dead but are locked in a cycle of single
useness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19629 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoMove code out of indentation one level to make it easier to read.
Chris Lattner [Mon, 17 Jan 2005 03:20:02 +0000 (03:20 +0000)]
Move code out of indentation one level to make it easier to read.

Disable the xform for < > cases.  It turns out that the following is being
miscompiled:

bool %test(sbyte %S) {
        %T = cast sbyte %S to uint
        %V = setgt uint %T, 255
        ret bool %V
}

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19628 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd methods
Chris Lattner [Mon, 17 Jan 2005 02:24:59 +0000 (02:24 +0000)]
Add methods

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19627 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDo not fold a load into a comparison that is used by more than one place.
Chris Lattner [Mon, 17 Jan 2005 01:34:14 +0000 (01:34 +0000)]
Do not fold a load into a comparison that is used by more than one place.
The comparison will probably be folded, so this is not ok to do.
This fixed 197.parser.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19624 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agosysconfdir -> PROJ_etcdir
Reid Spencer [Mon, 17 Jan 2005 00:42:31 +0000 (00:42 +0000)]
sysconfdir -> PROJ_etcdir

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19623 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDo not codegen 'xor bool, true' as 'not reg'. not reg inverts the upper bits
Chris Lattner [Mon, 17 Jan 2005 00:23:16 +0000 (00:23 +0000)]
Do not codegen 'xor bool, true' as 'not reg'.  not reg inverts the upper bits
of the bytereg.  This fixes yacr2, 300.twolf and probably others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19622 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix typos noticed by Gabor Greif, thanks Gabor!
Chris Lattner [Mon, 17 Jan 2005 00:12:04 +0000 (00:12 +0000)]
Fix typos noticed by Gabor Greif, thanks Gabor!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19621 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoSet up the shift and setcc types.
Chris Lattner [Mon, 17 Jan 2005 00:00:33 +0000 (00:00 +0000)]
Set up the shift and setcc types.
If we emit a load because we followed a token chain to get to it, try to
fold it into its single user if possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19620 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoShift and setcc types default to the pointer type.
Chris Lattner [Sun, 16 Jan 2005 23:59:48 +0000 (23:59 +0000)]
Shift and setcc types default to the pointer type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19619 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd comments
Chris Lattner [Sun, 16 Jan 2005 23:59:30 +0000 (23:59 +0000)]
Add comments
Add fields to hold the result type of setcc operations and shift amounts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19618 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImplement legalize of call nodes.
Chris Lattner [Sun, 16 Jan 2005 19:46:48 +0000 (19:46 +0000)]
Implement legalize of call nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19617 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix llvm-java project autconfiguration.
Alkis Evlogimenos [Sun, 16 Jan 2005 09:44:58 +0000 (09:44 +0000)]
Fix llvm-java project autconfiguration.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19616 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdded paramters to a few functions in order to allow me to change the functions to...
Tanya Lattner [Sun, 16 Jan 2005 08:51:10 +0000 (08:51 +0000)]
Added paramters to a few functions in order to allow me to change the functions to preserve SSA

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19615 91177308-0d34-0410-b5e6-96231b3b80d8

19 years ago* Adjust to changes in TargetLowering interfaces.
Chris Lattner [Sun, 16 Jan 2005 07:34:08 +0000 (07:34 +0000)]
* Adjust to changes in TargetLowering interfaces.
* Remove custom promotion for bool and byte select ops.  Legalize now
  promotes them for us.
* Allow folding ConstantPoolIndexes into EXTLOAD's, useful for float immediates.
* Declare which operations are not supported better.
* Add some hacky code for TRUNCSTORE to pretend that we have truncstore
  for i16 types.  This is useful for testing promotion code because I can
  just remove 16-bit registers all together and verify that programs work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19614 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRevamp supported ops. Instead of just being supported or not, we now keep
Chris Lattner [Sun, 16 Jan 2005 07:29:19 +0000 (07:29 +0000)]
Revamp supported ops.  Instead of just being supported or not, we now keep
track of how to deal with it, and provide the target with a hook that they
can use to legalize arbitrary operations in arbitrary ways.

Implement custom lowering for a couple of ops, implement promotion for select
operations (which x86 needs).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19613 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoadd method stub
Chris Lattner [Sun, 16 Jan 2005 07:28:41 +0000 (07:28 +0000)]
add method stub

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19612 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDon't mash stuff together.
Chris Lattner [Sun, 16 Jan 2005 07:28:31 +0000 (07:28 +0000)]
Don't mash stuff together.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19611 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoUse enums, move virtual dtor out of line.
Chris Lattner [Sun, 16 Jan 2005 07:28:11 +0000 (07:28 +0000)]
Use enums, move virtual dtor out of line.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19610 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoRevamp supported ops. Instead of just being supported or not, we now keep
Chris Lattner [Sun, 16 Jan 2005 07:27:49 +0000 (07:27 +0000)]
Revamp supported ops.  Instead of just being supported or not, we now keep
track of how to deal with it, and provide the target with a hook that they
can use to legalize arbitrary operations in arbitrary ways.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19609 91177308-0d34-0410-b5e6-96231b3b80d8

19 years ago* Revise the projects section and make reference to Projects.html
Reid Spencer [Sun, 16 Jan 2005 07:18:31 +0000 (07:18 +0000)]
* Revise the projects section and make reference to Projects.html
* the dist-clean target no longer implies a check

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19608 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoDon't confuse the LLVM_OBJ_DIR and the PROJ_OBJ_DIR because there might be
Reid Spencer [Sun, 16 Jan 2005 06:53:48 +0000 (06:53 +0000)]
Don't confuse the LLVM_OBJ_DIR and the PROJ_OBJ_DIR because there might be
a symbolic link making the autoconf name for the directory (LLVM_OBJ_ROOT)
and the "make" name for the directory (PROJ_OBJ_ROOT) different.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19607 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoImplement some more missing promotions.
Chris Lattner [Sun, 16 Jan 2005 05:06:12 +0000 (05:06 +0000)]
Implement some more missing promotions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19606 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix bugpoint
Chris Lattner [Sun, 16 Jan 2005 04:23:22 +0000 (04:23 +0000)]
Fix bugpoint

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19605 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agocycles_t -> CycleCount_t
Chris Lattner [Sun, 16 Jan 2005 04:20:30 +0000 (04:20 +0000)]
cycles_t -> CycleCount_t

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19604 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoAdd some .cvsignores to the win32 hierarchy to account for generated files
Jeff Cohen [Sun, 16 Jan 2005 03:18:23 +0000 (03:18 +0000)]
Add some .cvsignores to the win32 hierarchy to account for generated files

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19603 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoProvide support for HP/UX aCC compiler's variant of hash_map and hash_set
Reid Spencer [Sun, 16 Jan 2005 02:58:39 +0000 (02:58 +0000)]
Provide support for HP/UX aCC compiler's variant of hash_map and hash_set
(RogueWave). These are implemented in rw/stdex/hash_map.h and
rw/stdex/hash_set.h on HP/UX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19600 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoFix locations of libraries and executables to match makefiles.
Reid Spencer [Sun, 16 Jan 2005 02:38:06 +0000 (02:38 +0000)]
Fix locations of libraries and executables to match makefiles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19599 91177308-0d34-0410-b5e6-96231b3b80d8

19 years agoBUILD_* to PROJ_*
Reid Spencer [Sun, 16 Jan 2005 02:35:47 +0000 (02:35 +0000)]
BUILD_* to PROJ_*

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@19598 91177308-0d34-0410-b5e6-96231b3b80d8