RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 11:42:57 +01:00
Code Issues Projects Releases Wiki Activity
45,442 Commits 4 Branches 5 Tags 704 MiB
547cfc882e
Commit Graph

571 Commits

Author SHA1 Message Date
Chris Lattner
44e58d7816 fold fp_round(fp_round(x)) -> fp_round(x). 
llvm-svn: 46304
 2008-01-24 06:45:35 +00:00
Chris Lattner
41717f6989 This commit changes: 
1. Legalize now always promotes truncstore of i1 to i8. 
2. Remove patterns and gunk related to truncstore i1 from targets.
3. Rename the StoreXAction stuff to TruncStoreAction in TLI.
4. Make the TLI TruncStoreAction table a 2d table to handle from/to conversions.
5. Mark a wide variety of invalid truncstores as such in various targets, e.g.
   X86 currently doesn't support truncstore of any of its integer types.
6. Add legalize support for truncstores with invalid value input types.
7. Add a dag combine transform to turn store(truncate) into truncstore when
   safe.

The later allows us to compile CodeGen/X86/storetrunc-fp.ll to:

_foo:
	fldt	20(%esp)
	fldt	4(%esp)
	faddp	%st(1)
	movl	36(%esp), %eax
	fstps	(%eax)
	ret

instead of:

_foo:
	subl	$4, %esp
	fldt	24(%esp)
	fldt	8(%esp)
	faddp	%st(1)
	fstps	(%esp)
	movl	40(%esp), %eax
	movss	(%esp), %xmm0
	movss	%xmm0, (%eax)
	addl	$4, %esp
	ret

llvm-svn: 46140
 2008-01-17 19:59:44 +00:00
Chris Lattner
31081fc513 code cleanups, no functionality change. 
llvm-svn: 46126
 2008-01-17 07:20:38 +00:00
Chris Lattner
d033200a8f * Introduce a new SelectionDAG::getIntPtrConstant method 
and switch various codegen pieces and the X86 backend over
  to using it.

* Add some comments to SelectionDAGNodes.h

* Introduce a second argument to FP_ROUND, which indicates
  whether the FP_ROUND changes the value of its input. If
  not it is safe to xform things like fp_extend(fp_round(x)) -> x.

llvm-svn: 46125
 2008-01-17 07:00:52 +00:00
Evan Cheng
5be34d811c Fixes a nasty dag combiner bug that causes a bunch of tests to fail at -O0. 
It's not safe to use the two value CombineTo variant to combine away a dead load.
e.g. 
v1, chain2 = load chain1, loc
v2, chain3 = load chain2, loc
v3         = add v2, c 
Now we replace use of v1 with undef, use of chain2 with chain1.
ReplaceAllUsesWith() will iterate through uses of the first load and update operands:
v1, chain2 = load chain1, loc
v2, chain3 = load chain1, loc
v3         = add v2, c 
Now the second load is the same as the first load, SelectionDAG cse will ensure
the use of second load is replaced with the first load.
v1, chain2 = load chain1, loc
v3         = add v1, c
Then v1 is replaced with undef and bad things happen.

llvm-svn: 46099
 2008-01-16 23:11:54 +00:00
Chris Lattner
50f1b6f9a5 Factor the ReachesChainWithoutSideEffects out of dag combiner into 
a public SDOperand::reachesChainWithoutSideEffects method.  No 
functionality change.

llvm-svn: 46050
 2008-01-16 05:49:24 +00:00
Chris Lattner
c93ad7d569 Make load->store deletion a bit smarter. This allows us to compile this: 
void test(long long *P) { *P ^= 1; }

into just:

_test:
	movl	4(%esp), %eax
	xorl	$1, (%eax)
	ret

instead of code like this:

_test:
	movl	4(%esp), %ecx
        xorl    $1, (%ecx)
	movl	4(%ecx), %edx
	movl	%edx, 4(%ecx)
	ret

llvm-svn: 45762
 2008-01-08 23:08:06 +00:00
Chris Lattner
ad9a6ccb83 Remove attribution from file headers, per discussion on llvmdev. 
llvm-svn: 45418
 2007-12-29 20:36:04 +00:00
Chris Lattner
38687aa2b2 make sure not to zap volatile stores, thanks a lot to Dale for noticing this! 
llvm-svn: 45402
 2007-12-29 07:15:45 +00:00
Chris Lattner
78ae7ff876 don't fold fp_round(fp_extend(load)) -> fp_round(extload) 
llvm-svn: 45400
 2007-12-29 06:55:23 +00:00
Chris Lattner
7cb2de8e48 Delete a store whose input is a load from the same pointer: 
x = load p
  store x -> p

llvm-svn: 45398
 2007-12-29 06:26:16 +00:00
Chris Lattner
a8f6fac7a3 Tell TargetLoweringOpt whether it is running before 
or after legalize.

llvm-svn: 45321
 2007-12-22 20:56:36 +00:00
Evan Cheng
694994ba7b Don't leave newly created nodes around if it turns out they are not needed. 
llvm-svn: 45186
 2007-12-19 01:34:38 +00:00
Dale Johannesen
80dd0c5141 Redo previous patch so optimization only done for i1. 
Simpler and safer.

llvm-svn: 44663
 2007-12-06 17:53:31 +00:00
Chris Lattner
64a1a9f502 third time around: instead of disabling this completely, 
only disable it if we don't know it will be obviously profitable.
Still fixme, but less so. :)

llvm-svn: 44658
 2007-12-06 07:47:55 +00:00
Chris Lattner
bb5fb18af8 Actually, disable this code for now. More analysis and improvements to 
the X86 backend are needed before this should be enabled by default.

llvm-svn: 44657
 2007-12-06 07:44:31 +00:00
Chris Lattner
c467b49c96 implement a readme entry, compiling the code into: 
_foo:
	movl	$12, %eax
	andl	4(%esp), %eax
	movl	_array(%eax), %eax
	ret

instead of:

_foo:
	movl	4(%esp), %eax
	shrl	$2, %eax
	andl	$3, %eax
	movl	_array(,%eax,4), %eax
	ret

As it turns out, this triggers all the time, in a wide variety of
situations, for example, I see diffs like this in various programs:

-       movl    8(%eax), %eax
-       shll    $2, %eax
-       andl    $1020, %eax
-       movl    (%esi,%eax), %eax
+       movzbl  8(%eax), %eax
+       movl    (%esi,%eax,4), %eax


-       shll    $2, %edx
-       andl    $1020, %edx
-       movl    (%edi,%edx), %edx
+       andl    $255, %edx
+       movl    (%edi,%edx,4), %edx

Unfortunately, I also see stuff like this, which can be fixed in the
X86 backend:

-       andl    $85, %ebx
-       addl    _bit_count(,%ebx,4), %ebp
+       shll    $2, %ebx
+       andl    $340, %ebx
+       addl    _bit_count(%ebx), %ebp

llvm-svn: 44656
 2007-12-06 07:33:36 +00:00
Dale Johannesen
8bc5d4be6a Fix PR1842. 
llvm-svn: 44649
 2007-12-06 01:43:46 +00:00
Dan Gohman
a9f8208852 Don't lower srem/urem X%C to X-X/C*C unless the division is actually 
optimized. This avoids creating illegal divisions when the combiner is
running after legalize; this fixes PR1815. Also, it produces better
code in the included testcase by avoiding the subtract and multiply
when the division isn't optimized.

llvm-svn: 44341
 2007-11-26 23:46:11 +00:00
Duncan Sands
edf7e3b5f4 Move MinAlign to MathExtras.h. 
llvm-svn: 43944
 2007-11-09 13:41:39 +00:00
Duncan Sands
7df7c7aed1 Fix some load/store logic that would be wrong for 
apints on big-endian machines if the bitwidth is
not a multiple of 8.  Introduce a new helper,
MVT::getStoreSizeInBits, and use it.

llvm-svn: 43934
 2007-11-09 08:57:19 +00:00
Evan Cheng
d9bab93a44 If both parts of smul_lohi, etc. are used, don't simplify. If only one part is used, try simplify it. 
llvm-svn: 43888
 2007-11-08 09:25:29 +00:00
Evan Cheng
1fef4e369a Typo. 
llvm-svn: 43511
 2007-10-30 20:11:21 +00:00
Dan Gohman
02b8beff5f Fix a DAGCombiner abort on a bitcast from a scalar to a vector. 
llvm-svn: 43470
 2007-10-29 20:44:42 +00:00
Evan Cheng
5fe81cf64e Enable more fold (sext (load x)) -> (sext (truncate (sextload x))) 
transformation. Previously, it's restricted by ensuring the number of load uses
is one. Now the restriction is loosened up by allowing setcc uses to be
"extended" (e.g. setcc x, c, eq -> setcc sext(x), sext(c), eq).

llvm-svn: 43465
 2007-10-29 19:58:20 +00:00
Duncan Sands
b494fb97a4 The guaranteed alignment of ptr+offset is only the minimum of 
of offset and the alignment of ptr if these are both powers of
2.  While the ptr alignment is guaranteed to be a power of 2,
there is no reason to think that offset is.  For example, if
offset is 12 (the size of a long double on x86-32 linux) and
the alignment of ptr is 8, then the alignment of ptr+offset
will in general be 4, not 8.  Introduce a function MinAlign,
lifted from gcc, for computing the minimum guaranteed alignment.
I've tried to fix up everywhere under lib/CodeGen/SelectionDAG/.
I also changed some places that weren't wrong (because both values
were a power of 2), as a defensive change against people copying
and pasting the code.
Hopefully someone who cares about alignment will review the rest
of LLVM and fix up the remaining places.  Since I'm on x86 I'm
not very motivated to do this myself...

llvm-svn: 43421
 2007-10-28 12:59:45 +00:00
Dale Johannesen
4ae755d15c Redo "last ppc long double fix" as Chris wants. 
llvm-svn: 43189
 2007-10-19 20:29:00 +00:00
Dale Johannesen
b23b0bfa8f More ppcf128 issues (maybe the last)? 
llvm-svn: 43160
 2007-10-19 00:59:18 +00:00
Dale Johannesen
fdb488d4b5 Disable attempts to constant fold PPC f128. 
Remove the assumption that this will happen from
various places.

llvm-svn: 43053
 2007-10-16 23:38:29 +00:00
Chris Lattner
452ebc199e One mundane change: Change ReplaceAllUsesOfValueWith to *optionally* 
take a deleted nodes vector, instead of requiring it.

One more significant change:  Implement the start of a legalizer that
just works on types.  This legalizer is designed to run before the 
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.

This design/impl has the following advantages:

1. When finished, this will *significantly* reduce the amount of code in
   LegalizeDAG.cpp.  It will remove all the code related to promotion and
   expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike 
   LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
   for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of 
   deallocating and reallocating the entire graph that points to some 
   mutated node.
5. The code nicely separates out handling of operations with invalid 
   results from operations with invalid operands, making some cases
   simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :), 
   allowing you to easily understand what legalize types is doing.

This is not yet done.  Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing.  However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine.  The
biggest issues are:

1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.

Hey, at least it is a step in the right direction :).  If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it.  If
this explodes it will tell you what needs to be implemented.  Help is 
certainly appreciated.

Once this goes in, we can do three things:

1. Add a new pass of dag combine between the "type legalizer" and "operation
   legalizer" passes.  This will let us catch some long-standing isel issues
   that we miss because operation legalization often obfuscates the dag with
   target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
   making it much smaller and simpler.  When that happens we can then 
   reimplement the core functionality left in it in a much more efficient and
   non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
   selectiondags maybe...

llvm-svn: 42981
 2007-10-15 06:10:22 +00:00
Chris Lattner
c146f449b5 Enhance the truncstore optimization code to handle shifted 
values and propagate demanded bits through them in simple cases.

This allows this code:
void foo(char *P) {
   strcpy(P, "abc");
}
to compile to:

_foo:
        ldrb r3, [r1]
        ldrb r2, [r1, #+1]
        ldrb r12, [r1, #+2]!
        ldrb r1, [r1, #+1]
        strb r1, [r0, #+3]
        strb r2, [r0, #+1]
        strb r12, [r0, #+2]
        strb r3, [r0]
        bx lr

instead of:

_foo:
        ldrb r3, [r1, #+3]
        ldrb r2, [r1, #+2]
        orr r3, r2, r3, lsl #8
        ldrb r2, [r1, #+1]
        ldrb r1, [r1]
        orr r2, r1, r2, lsl #8
        orr r3, r2, r3, lsl #16
        strb r3, [r0]
        mov r2, r3, lsr #24
        strb r2, [r0, #+3]
        mov r2, r3, lsr #16
        strb r2, [r0, #+2]
        mov r3, r3, lsr #8
        strb r3, [r0, #+1]
        bx lr

testcase here: test/CodeGen/ARM/truncstore-dag-combine.ll

This also helps occasionally for X86 and other cases not involving 
unaligned load/stores.

llvm-svn: 42954
 2007-10-13 06:58:48 +00:00
Chris Lattner
133baf6012 Add a simple optimization to simplify the input to 
truncate and truncstore instructions, based on the 
knowledge that they don't demand the top bits.

llvm-svn: 42952
 2007-10-13 06:35:54 +00:00
Duncan Sands
a8baeb6dab Correct swapped arguments to getConstant. 
llvm-svn: 42824
 2007-10-10 09:54:50 +00:00
Dan Gohman
6f2a2b45fa DAGCombiner support for UDIVREM/SDIVREM and UMUL_LOHI/SMUL_LOHI. 
Check if one of the two results unneeded so see if a simpler operator
could bs used. Also check to see if each of the two computations could be
simplified if they were split into separate operators. Factor out the code
that calls visit() so that it can be used for this purpose.

llvm-svn: 42759
 2007-10-08 17:57:15 +00:00
Evan Cheng
0eed7948ce Reapply 42677. 
llvm-svn: 42692
 2007-10-06 08:19:55 +00:00
Chris Lattner
63443a5bc3 revert evan's patch until the header is committed 
llvm-svn: 42686
 2007-10-06 06:08:17 +00:00
Evan Cheng
dc95020e30 Added DAG xforms. e.g. 
(vextract (v4f32 s2v (f32 load $addr)), 0) -> (f32 load $addr) 
(vextract (v4i32 bc (v4f32 s2v (f32 load $addr))), 0) -> (i32 load $addr)
Remove x86 specific patterns.

llvm-svn: 42677
 2007-10-06 02:46:29 +00:00
Evan Cheng
2716b97b13 Fix a bogus splat xform: 
shuffle <undef, undef, x, undef>, <undef, undef, undef, undef>, <2, 2, 2, 2>
!=
<undef, undef, x, undef>

llvm-svn: 42111
 2007-09-18 21:54:37 +00:00
Dale Johannesen
83b2001d42 Prevent crash on long double. 
llvm-svn: 42103
 2007-09-18 18:36:59 +00:00
Dale Johannesen
4784ee3431 Revise previous patch per review comments. 
Next round of x87 long double stuff.
Getting close now, basically works.

llvm-svn: 41875
 2007-09-12 03:30:33 +00:00
Dale Johannesen
7bc3969cea Add APInt interfaces to APFloat (allows directly 
access to bits).  Use them in place of float and
double interfaces where appropriate.
First bits of x86 long double constants handling 
(untested, probably does not work).

llvm-svn: 41858
 2007-09-11 18:32:33 +00:00
Chris Lattner
2add65570c Emit: 
cmpl    %eax, %ecx
        setae   %al
        movzbl  %al, %eax

instead of:

        cmpl    %eax, %ecx
        setb    %al
        xorb    $1, %al
        movzbl  %al, %eax

when using logical not of a C comparison.

llvm-svn: 41807
 2007-09-10 21:39:07 +00:00
Dale Johannesen
b34e6b4898 Add mod, copysign, abs operations to APFloat. 
Implement some constant folding in SelectionDAG and
DAGCombiner using APFloat.  Remove double versions
of constructor and getValue from ConstantFPSDNode.

llvm-svn: 41664
 2007-08-31 23:34:27 +00:00
Dan Gohman
123b8effaa Make DAGCombiner's global alias analysis query more precise in the case 
where both pointers have non-zero offsets.

llvm-svn: 41491
 2007-08-27 16:32:11 +00:00
Dale Johannesen
2ceade197b Revise per review comments. 
llvm-svn: 41409
 2007-08-26 01:18:27 +00:00
Dale Johannesen
b52093236e Add APFloat interface to ConstantFPSDNode. Change 
over uses in DAGCombiner.  Fix interfaces to work
with APFloats.

llvm-svn: 41407
 2007-08-25 22:10:57 +00:00
Evan Cheng
a65c956119 Fold C ? 0 : 1 to ~C or zext(~C) or trunc(~C) depending the types. 
llvm-svn: 41163
 2007-08-18 05:57:05 +00:00
Dan Gohman
249090568d Fix the alias analysis query in DAGCombiner to not add in two 
offsets. The SrcValueOffset values are the real offsets from the
SrcValue base pointers.

llvm-svn: 40534
 2007-07-26 16:14:06 +00:00
Dan Gohman
1f4580ceb4 Don't call SimplifyVBinOp for non-vector operations, following earlier review 
feedback. This theoretically makes the common (scalar) case more efficient.

llvm-svn: 39823
 2007-07-13 20:03:40 +00:00
Dan Gohman
bd0ff28e1c Fix a bug in the folding of binary operators to undef. 
Thanks to Lauro for spotting this!

llvm-svn: 38491
 2007-07-10 15:19:29 +00:00
Dan Gohman
6b21e11f1c Fix the folding of undef in several binary operators to recognize 
undef in either the left or right operand.

llvm-svn: 38489
 2007-07-10 14:20:37 +00:00
Dan Gohman
150487b166 Preserve volatililty and alignment information when lowering or 
simplifying loads and stores.

llvm-svn: 38473
 2007-07-09 22:18:38 +00:00
Chris Lattner
aff9a84b7a Fix this warning: 
DAGCombiner.cpp: In member function 'llvm::SDOperand<unnamed>::DAGCombiner::visitOR(llvm::SDNode*)':
DAGCombiner.cpp:1608: warning: passing negative value '-0x00000000000000001' for argument 1 to 'llvm::SDOperand llvm::SelectionDAG::getConstant(uint64_t, llvm::MVT::ValueType, bool)'

oiy.

llvm-svn: 38458
 2007-07-09 16:16:34 +00:00
Dan Gohman
d2d18267e6 Fix several over-aggressive folds for undef nodes in dagcombine, to 
follow the rules for undef used in instcombine.

llvm-svn: 37851
 2007-07-03 14:03:57 +00:00
Dan Gohman
5fe8266d04 Teach GetNegatedExpression to negate 0-B to B in UnsafeFPMath mode, and 
visitFSUB to fold 0-B to -B in UnsafeFPMath mode. Also change visitFNEG
to use isNegatibleForFree/GetNegatedExpression instead of doing a subset
of the same thing manually.

This fixes test/CodeGen/X86/negative-sin.ll.

llvm-svn: 37842
 2007-07-02 15:48:56 +00:00
Dan Gohman
354f02e03d Generalize MVT::ValueType and associated functions to be able to represent 
extended vector types. Remove the special SDNode opcodes used for pre-legalize
vector operations, and the special MVT::Vector type used with them. Adjust
lowering and legalize to work with the normal SDNode kinds instead, and to
use the normal MVT functions to work with vector types instead of using the
two special operands that the pre-legalize nodes held.

This allows pre-legalize and post-legalize DAGs, and the code that operates
on them, to be more consistent. Pre-legalize vector operators can be handled
more consistently with scalar operators. And, -view-dag-combine1-dags and
-view-legalize-dags now look prettier for vector code.

llvm-svn: 37719
 2007-06-25 16:23:39 +00:00
Dan Gohman
a62327ea40 Move ComputeMaskedBits, MaskedValueIsZero, and ComputeNumSignBits from 
TargetLowering to SelectionDAG so that they have more convenient
access to the current DAG, in preparation for the ValueType routines
being changed from standalone functions to members of SelectionDAG for
the pre-legalize vector type changes.

llvm-svn: 37704
 2007-06-22 14:59:07 +00:00
Evan Cheng
f26fc091ac Xforms: 
(add (select cc, 0, c), x) -> (select cc, x, (add, x, c))
(sub x, (select cc, 0, c)) -> (select cc, x, (sub, x, c))

llvm-svn: 37685
 2007-06-21 07:39:16 +00:00
Dan Gohman
485cb57eab Pass a SelectionDAG into SDNode::dump everywhere it's used, in prepration 
for needing the DAG node to print pre-legalize extended value types, and
to get better debug messages with target-specific nodes.

llvm-svn: 37656
 2007-06-19 14:13:56 +00:00
Dan Gohman
2fd7d26df8 Rename MVT::getVectorBaseType to MVT::getVectorElementType. 
llvm-svn: 37579
 2007-06-14 22:58:02 +00:00
Chris Lattner
5161203d52 tighten up recursion depth again 
llvm-svn: 37330
 2007-05-25 02:19:06 +00:00
Evan Cheng
439bf58dc2 Fix a typo that caused combiner to create mal-formed pre-indexed store where value store is the same as the base pointer. 
llvm-svn: 37318
 2007-05-24 02:35:39 +00:00
Chris Lattner
d540c6429f prevent exponential recursion in isNegatibleForFree 
llvm-svn: 37310
 2007-05-23 07:35:22 +00:00
Dan Gohman
b593ad9fb0 Qualify calls to getTypeForValueType with MVT:: too. 
llvm-svn: 37233
 2007-05-18 18:41:29 +00:00
Dale Johannesen
cc99a6fc32 Don't fold bitconvert(load) for preinc/postdec loads. Likewise stores. 
llvm-svn: 37130
 2007-05-16 22:45:30 +00:00
Chris Lattner
a18b36cf45 Use a ptr set instead of a linear search to unique TokenFactor operands. 
This fixes PR1423

llvm-svn: 37102
 2007-05-16 06:37:59 +00:00
Evan Cheng
a781c7788a Bug fix: should check ABI alignment, not pref. alignment. 
llvm-svn: 37094
 2007-05-16 02:04:50 +00:00
Lauro Ramos Venancio
251ea5ab65 Fix an infinite recursion in GetNegatedExpression. 
llvm-svn: 37086
 2007-05-15 17:05:43 +00:00
Chris Lattner
eba1b74df3 implement a simple fneg optimization/propagation thing. This compiles: 
CodeGen/PowerPC/fneg.ll into:

_t4:
        fmul f0, f3, f4
        fmadd f1, f1, f2, f0
        blr

instead of:

_t4:
        fneg f0, f3
        fmul f0, f0, f4
        fmsub f1, f1, f2, f0
        blr

llvm-svn: 37054
 2007-05-14 22:04:50 +00:00
Evan Cheng
649f25dad1 Can't fold the bit_convert is the store is a truncating store. 
llvm-svn: 36962
 2007-05-09 21:49:47 +00:00
Evan Cheng
e18b87868d Forgot a check. 
llvm-svn: 36910
 2007-05-07 21:36:06 +00:00
Evan Cheng
18d994d6d6 Enable a couple of xforms: 
- (store (bitconvert v)) -> (store v) if resultant store does not require
higher alignment
- (bitconvert (load v)) -> (load (bitconvert*)v) if resultant load does not
require higher alignment

llvm-svn: 36908
 2007-05-07 21:27:48 +00:00
Evan Cheng
8c8b6ce116 Don't create indexed load / store with zero offset! 
llvm-svn: 36716
 2007-05-03 23:52:19 +00:00
Evan Cheng
6dc02c2b07 Forgot about chain result; also UNDEF cannot have multiple values. 
llvm-svn: 36622
 2007-05-01 08:53:39 +00:00
Evan Cheng
fe933cd6ca * Only turn a load to UNDEF if all of its outputs have no uses (indexed loads 
produce two results.)
* Do not touch volatile loads.

llvm-svn: 36604
 2007-05-01 00:38:21 +00:00
Christopher Lamb
a157874a8a PR400 phase 2. Propagate attributed load/store information through DAGs. 
llvm-svn: 36356
 2007-04-22 23:15:30 +00:00
Reid Spencer
81070d52da Revert Christopher Lamb's load/store alignment changes. 
llvm-svn: 36309
 2007-04-21 18:36:27 +00:00
Christopher Lamb
b56b6a7ad7 add support for alignment attributes on load/store instructions 
llvm-svn: 36301
 2007-04-21 08:16:25 +00:00
Chris Lattner
ea3c945817 allow SRL to simplify its operands, as it doesn't demand all bits as input. 
llvm-svn: 36245
 2007-04-18 03:06:49 +00:00
Chris Lattner
4ce8602d58 When replacing a node in SimplifyDemandedBits, if the old node used any 
single-use nodes, they will be dead soon.  Make sure to remove them before
processing other nodes.  This implements CodeGen/X86/shl_elim.ll

llvm-svn: 36244
 2007-04-18 03:05:22 +00:00
Chris Lattner
9ad682ad80 SIGN_EXTEND_INREG does not demand its top bits. Give SimplifyDemandedBits 
a chance to hack on it.  This compiles:

int baz(long long a) { return (short)(((int)(a >>24)) >> 9); }

into:
_baz:
        slwi r2, r3, 8
        srwi r2, r2, 9
        extsh r3, r2
        blr

instead of:

_baz:
        srwi r2, r4, 24
        rlwimi r2, r3, 8, 0, 23
        srwi r2, r2, 9
        extsh r3, r2
        blr

This implements CodeGen/PowerPC/sign_ext_inreg1.ll

llvm-svn: 36212
 2007-04-17 19:03:21 +00:00
Chris Lattner
f29ad16397 fix an infinite loop compiling ldecod, notice by JeffC. 
llvm-svn: 35910
 2007-04-11 16:51:53 +00:00
Chris Lattner
1d20292190 Fix this harder. 
llvm-svn: 35888
 2007-04-11 06:50:51 +00:00
Chris Lattner
01ebc25b36 don't create shifts by zero, fix some problems with my previous patch 
llvm-svn: 35887
 2007-04-11 06:43:25 +00:00
Chris Lattner
0289490285 Teach the codegen to turn [aez]ext (setcc) -> selectcc of 1/0, which often 
allows other simplifications.  For example, this compiles:
int isnegative(unsigned int X) {
   return !(X < 2147483648U);
}

Into this code:

x86:
        movl 4(%esp), %eax
        shrl $31, %eax
        ret
arm:
        mov r0, r0, lsr #31
        bx lr
thumb:
        lsr r0, r0, #31
        bx lr

instead of:

x86:
        cmpl $0, 4(%esp)
        sets %al
        movzbl %al, %eax
        ret

arm:
        mov r3, #0
        cmp r0, #0
        movlt r3, #1
        mov r0, r3
        bx lr

thumb:
        mov r2, #1
        mov r1, #0
        cmp r0, #0
        blt LBB1_2      @entry
LBB1_1: @entry
        cpy r2, r1
LBB1_2: @entry
        cpy r0, r2
        bx lr

Testcase here: test/CodeGen/Generic/ispositive.ll

llvm-svn: 35883
 2007-04-11 05:32:27 +00:00
Chris Lattner
3f0e49403c Codegen integer abs more efficiently using the trick from the PPC CWG. This 
improves codegen on many architectures.  Tests committed as CodeGen/*/iabs.ll

X86 Old:			X86 New:
_test:				_test:
   movl 4(%esp), %ecx		   movl 4(%esp), %eax
   movl %ecx, %eax		   movl %eax, %ecx
   negl %eax			   sarl $31, %ecx
   testl %ecx, %ecx		   addl %ecx, %eax
   cmovns %ecx, %eax		   xorl %ecx, %eax
   ret				   ret

PPC Old:			PPC New:
_test:				_test:
   cmpwi cr0, r3, -1		   srawi r2, r3, 31
   neg r2, r3			   add r3, r3, r2
   bgt cr0, LBB1_2 ;		   xor r3, r3, r2
LBB1_1: ;			   blr
   mr r3, r2
LBB1_2: ;
   blr

ARM Old:			ARM New:
_test:				_test:
   rsb r3, r0, #0		   add r3, r0, r0, asr #31
   cmp r0, #0			   eor r0, r3, r0, asr #31
   movge r3, r0			   bx lr
   mov r0, r3
   bx lr

Thumb Old:			Thumb New:
_test:				_test:
   neg r2, r0			   asr r2, r0, #31
   cmp r0, #0			   add r0, r0, r2
   bge LBB1_2			   eor r0, r2
LBB1_1: @			   bx lr
   cpy r0, r2
LBB1_2: @
   bx lr


Sparc Old:			Sparc New:
test:				test:
   save -96, %o6, %o6		   save -96, %o6, %o6
   sethi 0, %l0			   sra %i0, 31, %l0
   sub %l0, %i0, %l0		   add %i0, %l0, %l1
   subcc %i0, -1, %l1		   xor %l1, %l0, %i0
   bg .BB1_2			   restore %g0, %g0, %g0
   nop				   retl
.BB1_1:				   nop
   or %g0, %l0, %i0
.BB1_2:
   restore %g0, %g0, %g0
   retl
   nop

It also helps alpha/ia64 :)

llvm-svn: 35881
 2007-04-11 05:11:38 +00:00
Scott Michel
d6b3d3d6ab 1. Insert custom lowering hooks for ISD::ROTR and ISD::ROTL. 
2. Help DAGCombiner recognize zero/sign/any-extended versions of ROTR and ROTL
patterns. This was motivated by the X86/rotate.ll testcase, which should now
generate code for other platforms (and soon-to-come platforms.) Rewrote code
slightly to make it easier to read.

llvm-svn: 35605
 2007-04-02 21:36:32 +00:00
Dale Johannesen
d4ab7d28e9 Fix incorrect combination of different loads. Reenable zext-over-truncate 
combination.

llvm-svn: 35517
 2007-03-30 21:38:07 +00:00
Evan Cheng
2d09850760 Disable load width reduction xform of variant (zext (truncate load x)) for 
big endian targets until llvm-gcc build issue has been resolved.

llvm-svn: 35449
 2007-03-29 07:56:46 +00:00
Evan Cheng
5b1c21d27b SIGN_EXTEND_INREG requires one extra operand, a ValueType node. 
llvm-svn: 35350
 2007-03-26 07:12:51 +00:00
Evan Cheng
a484f31d4b Adjust offset to compensate for big endian machines. 
llvm-svn: 35293
 2007-03-24 00:02:43 +00:00
Evan Cheng
84aecc56e7 Make sure SEXTLOAD of the specific type is supported on the target. 
llvm-svn: 35289
 2007-03-23 22:13:36 +00:00
Evan Cheng
7dd7666120 Also replace uses of SRL if that's also folded during ReduceLoadWidth(). 
llvm-svn: 35286
 2007-03-23 20:55:21 +00:00
Evan Cheng
62ccdaea67 A couple of bug fixes for reducing load width xform: 
1. Address offset is in bytes.
2. Make sure truncate node uses are replaced with new load.

llvm-svn: 35274
 2007-03-23 02:16:52 +00:00
Evan Cheng
d7be4893f4 More opportunities to reduce load size. 
llvm-svn: 35254
 2007-03-22 01:54:19 +00:00
Evan Cheng
9867632e64 fold (truncate (srl (load x), c)) -> (smaller load (x+c/vt bits)) 
llvm-svn: 35239
 2007-03-21 20:14:05 +00:00
Evan Cheng
2f55532e84 Avoid combining indexed load further. 
llvm-svn: 35005
 2007-03-07 08:07:03 +00:00
Chris Lattner
8c7d418eaf fold away addc nodes when we know there cannot be a carry-out. 
llvm-svn: 34913
 2007-03-04 20:40:38 +00:00
Chris Lattner
7021449d8b generalize 
llvm-svn: 34910
 2007-03-04 20:08:45 +00:00
Chris Lattner
06e4ea2b21 canonicalize constants to the RHS of addc/adde. If nothing uses the carry out of 
addc, turn it into add.

This allows us to compile:

long long test(long long A, unsigned B) {
  return (A + ((long long)B << 32)) & 123;
}

into:

_test:
        movl $123, %eax
        andl 4(%esp), %eax
        xorl %edx, %edx
        ret

instead of:
_test:
        xorl %edx, %edx
        movl %edx, %eax
        addl 4(%esp), %eax   ;; add of zero
        andl $123, %eax
        ret

llvm-svn: 34909
 2007-03-04 20:03:15 +00:00
Chris Lattner
65de507797 Fold (sext (truncate x)) more aggressively, by avoiding creation of a 
sextinreg if not needed.   This is useful in two cases: before legalize,
it avoids creating a sextinreg that will be trivially removed.  After legalize
if the target doesn't support sextinreg, the trunc/sext would not have been
removed before.

llvm-svn: 34621
 2007-02-26 03:13:59 +00:00
Evan Cheng
1b155ac243 Move SimplifySetCC to TargetLowering and allow it to be shared with legalizer. 
llvm-svn: 34065
 2007-02-08 22:13:59 +00:00
Evan Cheng
ae02dfb090 Fix for PR1108: type of insert_vector_elt index operand is PtrVT, not MVT::i32. 
llvm-svn: 33398
 2007-01-20 10:10:26 +00:00
Evan Cheng
ced4fcb608 Remove this xform: 
(shl (add x, c1), c2) -> (add (shl x, c2), c1<<c2)
Replace it with:
(add (shl (add x, c1), c2), ) -> (add (add (shl x, c2), c1<<c2), )

This fixes test/CodeGen/ARM/smul.ll

llvm-svn: 33361
 2007-01-19 17:51:44 +00:00
Chris Lattner
3af776359d Fix PR1114 and CodeGen/Generic/2007-01-15-LoadSelectCycle.ll by being 
careful when folding "c ? load p : load q" that C doesn't reach either load.
If so, folding this into load (c ? p : q) will induce a cycle in the graph.

llvm-svn: 33251
 2007-01-16 05:59:59 +00:00
Chris Lattner
7560c5f913 add options to view the dags before the first or second pass of dag combine. 
llvm-svn: 33249
 2007-01-16 04:55:25 +00:00
Chris Lattner
0f765fae89 Implement some trivial FP foldings when -enable-unsafe-fp-math is specified. 
This implements CodeGen/PowerPC/unsafe-math.ll

llvm-svn: 33024
 2007-01-08 23:04:05 +00:00
Chris Lattner
a975b95adb Eliminate static ctors from Statistics 
llvm-svn: 32698
 2006-12-19 22:41:21 +00:00
Evan Cheng
7775b33a32 Cannot combine an indexed load / store any further. 
llvm-svn: 32629
 2006-12-16 06:25:23 +00:00
Jim Laskey
0809c79c60 This code was usurping the sextload expand in teh legalizer. Just make 
sure the right conditions are checked.

llvm-svn: 32611
 2006-12-15 21:38:30 +00:00
Chris Lattner
df2345fcbe make this code more aggressive about turning store fpimm into store int imm. 
This is not sufficient to fix X86/store-fp-constant.ll

llvm-svn: 32465
 2006-12-12 04:16:14 +00:00
Evan Cheng
93c75d4cfb Don't convert store double C, Ptr to store long C, Ptr if i64 is not a legal type. 
llvm-svn: 32434
 2006-12-11 17:25:19 +00:00
Nate Begeman
2566f75e7a Move something that should be in the dag combiner from the legalizer to the 
dag combiner.

llvm-svn: 32431
 2006-12-11 02:23:46 +00:00
Chris Lattner
db346e68a9 Fix CodeGen/PowerPC/2006-12-07-SelectCrash.ll on PPC64 
llvm-svn: 32336
 2006-12-07 22:36:47 +00:00
Bill Wendling
23b8b13c9d Removing even more <iostream> includes. 
llvm-svn: 32320
 2006-12-07 20:04:42 +00:00
Chris Lattner
a531ce882e Detemplatize the Statistic class. The only type it is instantiated with 
is 'unsigned'.

llvm-svn: 32279
 2006-12-06 17:46:33 +00:00
Chris Lattner
aa8f67c2b5 For better or worse, load from i1 is assumed to be zero extended. Do not 
form a load from i1 from larger loads that may not be zext'd.

llvm-svn: 31933
 2006-11-27 04:40:53 +00:00
Chris Lattner
a2bbd246e0 Fix PR1011 and CodeGen/Generic/2006-11-20-DAGCombineCrash.ll 
llvm-svn: 31878
 2006-11-20 18:05:46 +00:00
Evan Cheng
43ba439abd Fix an incorrectly inverted condition. 
llvm-svn: 31773
 2006-11-16 00:08:20 +00:00
Chris Lattner
edfc824673 disallow preinc of a frameindex. This is not profitable and causes 2-addr 
pass to explode.  This fixes a bunch of llc-beta failures on ppc last night.

llvm-svn: 31661
 2006-11-11 01:00:15 +00:00
Chris Lattner
bf0c3e3a02 reduce indentation by using early exits. No functionality change. 
llvm-svn: 31660
 2006-11-11 00:56:29 +00:00
Chris Lattner
671ea7a93b move big chunks of code out-of-line, no functionality change. 
llvm-svn: 31658
 2006-11-11 00:39:41 +00:00
Chris Lattner
1f61c6c84d Fix a dag combiner bug exposed by my recent instcombine patch. This fixes 
CodeGen/Generic/2006-11-10-DAGCombineMiscompile.ll and PPC gsm/toast

llvm-svn: 31644
 2006-11-10 21:37:15 +00:00
Evan Cheng
78ba99caa5 When forming a pre-indexed store, make sure ptr isn't the same or is a pred of value being stored. It would cause a cycle. 
llvm-svn: 31631
 2006-11-10 08:28:11 +00:00
Evan Cheng
17dd6dd46c Don't attempt expensive pre-/post- indexed dag combine if target does not support them. 
llvm-svn: 31598
 2006-11-09 19:10:46 +00:00
Evan Cheng
89ee587963 Rename ISD::MemOpAddrMode to ISD::MemIndexedMode 
llvm-svn: 31595
 2006-11-09 17:55:04 +00:00
Evan Cheng
6b7d127df9 getPostIndexedAddressParts change: passes in load/store instead of its loaded / stored VT. 
llvm-svn: 31584
 2006-11-09 04:29:46 +00:00
Evan Cheng
fcef896f6e Match more post-indexed ops. 
llvm-svn: 31569
 2006-11-08 20:27:27 +00:00
Jim Laskey
28fec74f1b Remove redundant <cmath>. 
llvm-svn: 31561
 2006-11-08 19:16:44 +00:00
Evan Cheng
1f5c4a6c43 - When performing pre-/post- indexed load/store transformation, do not worry 
about whether the new base ptr would be live below the load/store. Let two
  address pass split it back to non-indexed ops.
- Minor tweaks / fixes.

llvm-svn: 31544
 2006-11-08 08:30:28 +00:00
Evan Cheng
acc6a98286 Fixed a minor bug preventing some pre-indexed load / store transformation. 
llvm-svn: 31543
 2006-11-08 06:56:05 +00:00
Evan Cheng
e50f5e4c05 Fix a obscure post-indexed load / store dag combine bug. 
llvm-svn: 31537
 2006-11-08 02:38:55 +00:00
Evan Cheng
3db2b3aab9 Add post-indexed load / store transformations. 
llvm-svn: 31498
 2006-11-07 09:03:05 +00:00
Evan Cheng
f191d53a9a Add comment. 
llvm-svn: 31473
 2006-11-06 08:14:30 +00:00
Jeff Cohen
e1003da1a2 Unbreak VC++ build. 
llvm-svn: 31464
 2006-11-05 19:31:28 +00:00
Evan Cheng
bf7db95159 Added pre-indexed store support. 
llvm-svn: 31459
 2006-11-05 09:31:14 +00:00
Evan Cheng
466e20fca2 Rename 
llvm-svn: 31413
 2006-11-03 07:21:16 +00:00
Reid Spencer
8cac48e619 Remove dead variable. Fix 80 column violations. 
llvm-svn: 31412
 2006-11-03 03:30:34 +00:00
Evan Cheng
a0133317f7 Added DAG combiner transformation to generate pre-indexed loads. 
llvm-svn: 31410
 2006-11-03 03:06:21 +00:00
Reid Spencer
4bafa71dc1 For PR786: 
Turn on -Wunused and -Wno-unused-parameter. Clean up most of the resulting
fall out by removing unused variables. Remaining warnings have to do with
unused functions (I didn't want to delete code without review) and unused
variables in generated code. Maintainers should clean up the remaining
issues when they see them. All changes pass DejaGnu tests and Olden.

llvm-svn: 31380
 2006-11-02 20:25:50 +00:00
Jim Laskey
c06a0bc853 Add option for controlling inclusion of global AA. 
llvm-svn: 31040
 2006-10-18 19:08:31 +00:00
Jim Laskey
288c230cbd Use global info for alias analysis. 
llvm-svn: 31035
 2006-10-18 12:29:57 +00:00
Chris Lattner
bfbb76e103 Fix CodeGen/PowerPC/2006-10-17-brcc-miscompile.ll 
llvm-svn: 31019
 2006-10-17 21:24:15 +00:00
Jim Laskey
7f16ca4872 Make it simplier to dump DAGs while in DAGCombiner. Remove a nasty optimization. 
llvm-svn: 31009
 2006-10-17 19:33:52 +00:00
Evan Cheng
d9bec725a2 Make sure operand does have size and element type operands. 
llvm-svn: 30999
 2006-10-17 17:06:35 +00:00
Evan Cheng
2d9318cff1 Be careful when looking through a vbit_convert. Optimizing this: 
(vector_shuffle
  (vbitconvert (vbuildvector (copyfromreg v4f32), 1, v4f32), 4, f32),
  (undef, undef, undef, undef), (0, 0, 0, 0), 4, f32)
to the
  vbitconvert
is a very bad idea.

llvm-svn: 30989
 2006-10-16 22:49:37 +00:00
Jim Laskey
06f4428abc Pass AliasAnalysis thru to DAGCombiner. 
llvm-svn: 30984
 2006-10-16 20:52:31 +00:00
Jim Laskey
1070dfefba Tidy up after truncstore changes. 
llvm-svn: 30961
 2006-10-14 12:14:27 +00:00
Chris Lattner
08aa96b824 Make sure that the node returned by SimplifySetCC is added to the worklist 
so that it can be deleted if unused.

llvm-svn: 30955
 2006-10-14 03:52:46 +00:00
Chris Lattner
a515f322f3 fold setcc of a setcc. 
llvm-svn: 30953
 2006-10-14 01:02:29 +00:00
Chris Lattner
25ad62d132 When SimplifySetCC was moved to the DAGCombiner, it was never removed from 
SelectionDAG and it has since bitrotted.  Remove the copy from SelectionDAG.
Next, remove the constant folding piece of DAGCombiner::SimplifySetCC into
a new FoldSetCC method which can be used by getNode() and SimplifySetCC.

This fixes obscure bugs.

llvm-svn: 30952
 2006-10-14 00:41:01 +00:00
Jim Laskey
bf50140aac Reduce the workload by not adding chain users to work list. 
llvm-svn: 30948
 2006-10-13 23:32:28 +00:00
Evan Cheng
fe5bb5dbe6 Merge ISD::TRUNCSTORE to ISD::STORE. Switch to using StoreSDNode. 
llvm-svn: 30945
 2006-10-13 21:14:26 +00:00
Chris Lattner
70444d5663 Lower X%C into X/C+stuff. This allows the 'division by a constant' logic to 
apply to rems as well as divs.  This fixes PR945 and speeds up ReedSolomon
from 14.57s to 10.90s (which is now faster than gcc).

It compiles CodeGen/X86/rem.ll into:

_test1:
        subl $4, %esp
        movl %esi, (%esp)
        movl $2155905153, %ecx
        movl 8(%esp), %esi
        movl %esi, %eax
        imull %ecx
        addl %esi, %edx
        movl %edx, %eax
        shrl $31, %eax
        sarl $7, %edx
        addl %eax, %edx
        imull $255, %edx, %eax
        subl %eax, %esi
        movl %esi, %eax
        movl (%esp), %esi
        addl $4, %esp
        ret
_test2:
        movl 4(%esp), %eax
        movl %eax, %ecx
        sarl $31, %ecx
        shrl $24, %ecx
        addl %eax, %ecx
        andl $4294967040, %ecx
        subl %ecx, %eax
        ret
_test3:
        subl $4, %esp
        movl %esi, (%esp)
        movl $2155905153, %ecx
        movl 8(%esp), %esi
        movl %esi, %eax
        mull %ecx
        shrl $7, %edx
        imull $255, %edx, %eax
        subl %eax, %esi
        movl %esi, %eax
        movl (%esp), %esi
        addl $4, %esp
        ret

instead of div/idiv instructions.

llvm-svn: 30920
 2006-10-12 20:58:32 +00:00
Chris Lattner
e38ce54cc9 add a minor dag combine noticed when looking at PR945 
llvm-svn: 30915
 2006-10-12 20:23:19 +00:00
Jim Laskey
388c9681ef D'oh - need to use the rigth kind of store. 
llvm-svn: 30903
 2006-10-12 15:22:24 +00:00
Jim Laskey
eba756c1a7 Alias analysis of TRUNCSTORE. 
llvm-svn: 30889
 2006-10-11 18:55:16 +00:00
Jim Laskey
4791a4ad14 Handle aliasing of loadext. 
llvm-svn: 30883
 2006-10-11 17:47:52 +00:00
Jim Laskey
fd6218f8f5 Fix regression in combiner alias analysis. 
llvm-svn: 30880
 2006-10-11 13:47:09 +00:00
Evan Cheng
9b31a4d4ed Naming consistency. 
llvm-svn: 30878
 2006-10-11 07:10:22 +00:00
Evan Cheng
d22f3dd3ed Reflects ISD::LOAD / ISD::LOADX / LoadSDNode changes. 
llvm-svn: 30844
 2006-10-09 20:57:25 +00:00
Chris Lattner
b0e0a23959 Eliminate more token factors by taking advantage of transitivity: 
if TF depends on A and B, and A depends on B, TF just needs to depend on
A.  With Jim's alias-analysis stuff enabled, this compiles the testcase in
PR892 into:

__Z4test3Val:
        subl $44, %esp
        call L__Z3foov$stub
        movl %edx, 28(%esp)
        movl %eax, 32(%esp)
        movl %eax, 24(%esp)
        movl %edx, 36(%esp)
        movl 52(%esp), %ecx
        movl %ecx, 4(%esp)
        movl %eax, 8(%esp)
        movl %edx, 12(%esp)
        movl 48(%esp), %eax
        movl %eax, (%esp)
        call L__Z3bar3ValS_$stub
        addl $44, %esp
        ret

instead of:

__Z4test3Val:
        subl $44, %esp
        call L__Z3foov$stub
        movl %eax, 24(%esp)
        movl %edx, 28(%esp)
        movl 24(%esp), %eax
        movl %eax, 32(%esp)
        movl 28(%esp), %eax
        movl %eax, 36(%esp)
        movl 32(%esp), %eax
        movl 36(%esp), %ecx
        movl 52(%esp), %edx
        movl %edx, 4(%esp)
        movl %eax, 8(%esp)
        movl %ecx, 12(%esp)
        movl 48(%esp), %eax
        movl %eax, (%esp)
        call L__Z3bar3ValS_$stub
        addl $44, %esp
        ret

llvm-svn: 30821
 2006-10-08 22:57:01 +00:00
Jim Laskey
9260b2f86e Combiner alias analysis passes Multisource (release-asserts.) 
llvm-svn: 30818
 2006-10-07 23:37:56 +00:00
Evan Cheng
275825195a Make use of getStore(). 
llvm-svn: 30759
 2006-10-05 23:01:46 +00:00
Jim Laskey
3f9f064fd1 Alias analysis code clean ups. 
llvm-svn: 30753
 2006-10-05 15:07:25 +00:00
Jim Laskey
dd74085b55 More extensive alias analysis. 
llvm-svn: 30721
 2006-10-04 16:53:27 +00:00
Evan Cheng
494e8e6971 Combine ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD into ISD::LOADX. Add an 
extra operand to LOADX to specify the exact value extension type.

llvm-svn: 30714
 2006-10-04 00:56:09 +00:00
Jim Laskey
74ba822f79 Load chain check is not needed 
llvm-svn: 30613
 2006-09-26 17:44:58 +00:00
Jim Laskey
2a8d8270eb Chain can be any operand 
llvm-svn: 30611
 2006-09-26 09:32:41 +00:00
Jim Laskey
ae81857cba Wrong size for load 
llvm-svn: 30610
 2006-09-26 08:14:06 +00:00
Jim Laskey
d72f4cfe04 Can't move a load node if it's chain is not used. 
llvm-svn: 30609
 2006-09-26 07:37:42 +00:00
Jim Laskey
6ae9f53d2c Accidental enable of bad code 
llvm-svn: 30601
 2006-09-25 21:11:32 +00:00
Jim Laskey
640b7dbed5 Fix chain dropping in load and drop unused stores in ret blocks. 
llvm-svn: 30600
 2006-09-25 19:32:58 +00:00
Jim Laskey
ba2f6127b2 Core antialiasing for load and store. 
llvm-svn: 30597
 2006-09-25 16:29:54 +00:00
Evan Cheng
ce6a660148 Make it work for DAG combine of multi-value nodes. 
llvm-svn: 30573
 2006-09-21 19:04:05 +00:00
Jim Laskey
231343018b core corrections 
llvm-svn: 30570
 2006-09-21 17:35:47 +00:00
Jim Laskey
50750cf500 Basic "in frame" alias analysis. 
llvm-svn: 30568
 2006-09-21 16:28:59 +00:00
Chris Lattner
c17b86ef22 fold (aext (and (trunc x), cst)) -> (and x, cst). 
llvm-svn: 30561
 2006-09-21 06:40:43 +00:00
Chris Lattner
d9fca453f1 Check the right value type. This fixes 186.crafty on x86 
llvm-svn: 30560
 2006-09-21 06:17:39 +00:00
Chris Lattner
34768d5361 Compile: 
int %test(ulong *%tmp) {
        %tmp = load ulong* %tmp         ; <ulong> [#uses=1]
        %tmp.mask = shr ulong %tmp, ubyte 50            ; <ulong> [#uses=1]
        %tmp.mask = cast ulong %tmp.mask to ubyte
        %tmp2 = and ubyte %tmp.mask, 3          ; <ubyte> [#uses=1]
        %tmp2 = cast ubyte %tmp2 to int         ; <int> [#uses=1]
        ret int %tmp2
}

to:

_test:
        movl 4(%esp), %eax
        movl 4(%eax), %eax
        shrl $18, %eax
        andl $3, %eax
        ret

instead of:

_test:
        movl 4(%esp), %eax
        movl 4(%eax), %eax
        shrl $18, %eax
        # TRUNCATE movb %al, %al
        andb $3, %al
        movzbl %al, %eax
        ret

llvm-svn: 30558
 2006-09-21 06:14:31 +00:00
Chris Lattner
eb12877970 Generalize (zext (truncate x)) and (sext (truncate x)) folding to work when 
the src/dst are not the same size.  This catches things like "truncate
32-bit X to 8 bits, then zext to 16", which happens a bit on X86.

llvm-svn: 30557
 2006-09-21 06:00:20 +00:00
Chris Lattner
a0243b3ad3 Compile: 
int test3(int a, int b) { return (a < 0) ? a : 0; }

to:

_test3:
        srawi r2, r3, 31
        and r3, r2, r3
        blr

instead of:

_test3:
        cmpwi cr0, r3, 1
        li r2, 0
        blt cr0, LBB2_2 ;entry
LBB2_1: ;entry
        mr r3, r2
LBB2_2: ;entry
        blr


This implements: PowerPC/select_lt0.ll:seli32_a_a

llvm-svn: 30517
 2006-09-20 06:41:35 +00:00
Chris Lattner
e78d019082 Fold the full generality of (any_extend (truncate x)) 
llvm-svn: 30514
 2006-09-20 06:29:17 +00:00
Chris Lattner
6440707b6f Two things: 
1. teach SimplifySetCC that '(srl (ctlz x), 5) == 0' is really x != 0.
2. Teach visitSELECT_CC to use SimplifySetCC instead of calling it and
   ignoring the result.  This allows us to compile:

bool %test(ulong %x) {
  %tmp = setlt ulong %x, 4294967296
  ret bool %tmp
}

to:

_test:
        cntlzw r2, r3
        cmplwi cr0, r3, 1
        srwi r2, r2, 5
        li r3, 0
        beq cr0, LBB1_2 ;
LBB1_1: ;
        mr r3, r2
LBB1_2: ;
        blr

instead of:

_test:
        addi r2, r3, -1
        cntlzw r2, r2
        cntlzw r3, r3
        srwi r2, r2, 5
        cmplwi cr0, r2, 0
        srwi r2, r3, 5
        li r3, 0
        bne cr0, LBB1_2 ;
LBB1_1: ;
        mr r3, r2
LBB1_2: ;
        blr

This isn't wonderful, but it's an improvement.

llvm-svn: 30513
 2006-09-20 06:19:26 +00:00
Chris Lattner
c3f56368db Fold (X & C1) | (Y & C2) -> (X|Y) & C3 when possible. 
This implements CodeGen/X86/and-or-fold.ll

llvm-svn: 30379
 2006-09-14 21:11:37 +00:00
Chris Lattner
dbe8078c76 Split rotate matching code out to its own function. Make it stronger, by 
matching things like ((x >> c1) & c2) | ((x << c3) & c4) to (rot x, c5) & c6

llvm-svn: 30376
 2006-09-14 20:50:57 +00:00
Evan Cheng
b2933f3f52 DAG combiner fix for rotates. Previously the outer-most condition checks 
for ROTL availability. This prevents it from forming ROTR for targets that
has ROTR only.

llvm-svn: 29997
 2006-08-31 07:41:12 +00:00
Evan Cheng
2335c819cd Move isCommutativeBinOp from SelectionDAG.cpp and DAGCombiner.cpp out. Make it a static method of SelectionDAG. 
llvm-svn: 29951
 2006-08-29 06:42:35 +00:00
Chris Lattner
33bd5dcfb7 s|llvm/Support/Visibility.h|llvm/Support/Compiler.h| 
llvm-svn: 29911
 2006-08-27 12:54:02 +00:00
Chris Lattner
0d57396628 change internal impl of dag combiner so that calls to CombineTo never have to 
make a temporary vector.

llvm-svn: 29618
 2006-08-11 17:56:38 +00:00
Chris Lattner
a47d3dd2cc Change one ReplaceAllUsesWith method to take an array of operands to replace 
instead of a vector of operands.

llvm-svn: 29616
 2006-08-11 17:46:28 +00:00
Chris Lattner
7b1362fa52 Start eliminating temporary vectors used to create DAG nodes. Instead, pass 
in the start of an array and a count of operands where applicable.  In many
cases, the number of operands is known, so this static array can be allocated
on the stack, avoiding the heap.  In many other cases, a SmallVector can be
used, which has the same benefit in the common cases.

I updated a lot of code calling getNode that takes a vector, but ran out of
time.  The rest of the code should be updated, and these methods should be
removed.

We should also do the same thing to eliminate the methods that take a
vector of MVT::ValueTypes.

It would be extra nice to convert the dagiselemitter to avoid creating vectors
for operands when calling getTargetNode.

llvm-svn: 29566
 2006-08-08 02:23:42 +00:00
Reid Spencer
fb0feb79f0 Initialize some variables the compiler warns about. 
llvm-svn: 29277
 2006-07-25 20:44:41 +00:00
Evan Cheng
3b2a8d2749 If a shuffle is a splat, check if the argument is a build_vector with all elements being the same. If so, return the argument. 
llvm-svn: 29242
 2006-07-21 08:25:53 +00:00
Evan Cheng
fe4cf8c64a If a shuffle is unary, i.e. one of the vector argument is not needed, turn the 
operand into a undef and adjust mask accordingly.

llvm-svn: 29232
 2006-07-20 22:44:41 +00:00
Andrew Lenharth
bf871a2ad5 80 cols 
llvm-svn: 29221
 2006-07-20 17:43:27 +00:00
Andrew Lenharth
c1074954fb Reduce number of exported symbols 
llvm-svn: 29220
 2006-07-20 17:28:38 +00:00
Chris Lattner
601a416d22 Mark these two classes as hidden, shrinking libllbmgcc.dylib by 25K 
llvm-svn: 28970
 2006-06-28 21:58:30 +00:00
Andrew Lenharth
a2bda5b0e1 Start on my todo list 
llvm-svn: 28752
 2006-06-12 16:07:18 +00:00
Evan Cheng
bbc183c90e visitVBinOp: Can't fold divide by zero! 
llvm-svn: 28584
 2006-05-31 06:08:35 +00:00
Chris Lattner
d8bd52bfd2 Fix a nasty dag combiner bug that caused nondeterminstic crashes (MY FAVORITE!): 
SimplifySelectOps would eliminate a Select, delete it, then return true.

The clients would see that it did something and return null.

The top level would see a null return, and decide that nothing happened,
proceeding to process the node in other ways: boom.

The fix is simple: clients of SimplifySelectOps should return the select
node itself.

In order to catch really obnoxious boogs like this in the future, add an
assert that nodes are not deleted.  We do this by checking for a sentry node
type that the SDNode dtor sets when a node is destroyed.

llvm-svn: 28514
 2006-05-27 00:43:02 +00:00
Andrew Lenharth
14504c85ed Move this code to a common place 
llvm-svn: 28329
 2006-05-16 17:42:15 +00:00
Chris Lattner
474e1b7ef3 Comment out dead variables 
llvm-svn: 28252
 2006-05-12 17:57:54 +00:00
Chris Lattner
a500852895 Two simplifications for token factor nodes: simplify tf(x,x) -> x. 
simplify tf(x,y,y,z) -> tf(x,y,z).

llvm-svn: 28233
 2006-05-12 05:01:37 +00:00
Evan Cheng
3f72d2121b Debugging info 
llvm-svn: 28200
 2006-05-09 06:55:15 +00:00
Chris Lattner
eed10e837c Make the case I just checked in stronger. Now we compile this: 
short test2(short X, short x) {
  int Y = (short)(X+x);
  return Y >> 1;
}

to:

_test2:
        add r2, r3, r4
        extsh r2, r2
        srawi r3, r2, 1
        blr

instead of:

_test2:
        add r2, r3, r4
        extsh r2, r2
        srwi r2, r2, 1
        extsh r3, r2
        blr

llvm-svn: 28175
 2006-05-08 21:18:59 +00:00
Chris Lattner
7b8a0cfff3 Implement and_sext.ll:test3, generating: 
_test4:
        srawi r3, r3, 16
        blr

instead of:

_test4:
        srwi r2, r3, 16
        extsh r3, r2
        blr

for:

short test4(unsigned X) {
  return (X >> 16);
}

llvm-svn: 28174
 2006-05-08 20:59:41 +00:00
Chris Lattner
4f66de151c Compile this: 
short test4(unsigned X) {
  return (X >> 16);
}

to:

_test4:
        movl 4(%esp), %eax
        sarl $16, %eax
        ret

instead of:

_test4:
        movl $-65536, %eax
        andl 4(%esp), %eax
        sarl $16, %eax
        ret

llvm-svn: 28171
 2006-05-08 20:51:54 +00:00
Nate Begeman
b8e351aec5 Fix PR772 
llvm-svn: 28161
 2006-05-08 01:35:01 +00:00
Chris Lattner
f76c0b6662 Simplify some code, add a couple minor missed folds 
llvm-svn: 28152
 2006-05-06 23:06:26 +00:00
Chris Lattner
3d5d01a74b remove cases handled elsewhere 
llvm-svn: 28150
 2006-05-06 22:43:44 +00:00
Chris Lattner
ca06e2522e Use the new TargetLowering::ComputeNumSignBits method to eliminate 
sign_extend_inreg operations.  Though ComputeNumSignBits is still rudimentary,
this is enough to compile this:

short test(short X, short x) {
  int Y = X+x;
  return (Y >> 1);
}
short test2(short X, short x) {
  int Y = (short)(X+x);
  return Y >> 1;
}

into:

_test:
        add r2, r3, r4
        srawi r3, r2, 1
        blr
_test2:
        add r2, r3, r4
        extsh r2, r2
        srawi r3, r2, 1
        blr

instead of:

_test:
        add r2, r3, r4
        srawi r2, r2, 1
        extsh r3, r2
        blr
_test2:
        add r2, r3, r4
        extsh r2, r2
        srawi r2, r2, 1
        extsh r3, r2
        blr

llvm-svn: 28146
 2006-05-06 09:30:03 +00:00
Chris Lattner
32e96402c0 Fold trunc(any_ext). This gives stuff like: 
27,28c27
<       movzwl %di, %edi
<       movl %edi, %ebx
---
>       movw %di, %bx

llvm-svn: 28137
 2006-05-05 22:56:26 +00:00
Chris Lattner
c912cf0b07 Shrink shifts when possible. 
llvm-svn: 28136
 2006-05-05 22:53:17 +00:00
Chris Lattner
4b581e4167 Fold (fpext (load x)) -> (extload x) 
llvm-svn: 28130
 2006-05-05 21:34:35 +00:00
Chris Lattner
d7637651b6 Fold some common code. 
llvm-svn: 28124
 2006-05-05 06:32:04 +00:00
Chris Lattner
584874682a Implement: 
// fold (and (sext x), (sext y)) -> (sext (and x, y))
  // fold (or  (sext x), (sext y)) -> (sext (or  x, y))
  // fold (xor (sext x), (sext y)) -> (sext (xor x, y))
  // fold (and (aext x), (aext y)) -> (aext (and x, y))
  // fold (or  (aext x), (aext y)) -> (aext (or  x, y))
  // fold (xor (aext x), (aext y)) -> (aext (xor x, y))

llvm-svn: 28123
 2006-05-05 06:31:05 +00:00
Chris Lattner
bec98440f4 Pull and through and/or/xor. This compiles some bitfield code to: 
mov EAX, DWORD PTR [ESP + 4]
        mov ECX, DWORD PTR [EAX]
        mov EDX, ECX
        add EDX, EDX
        or EDX, ECX
        and EDX, -2147483648
        and ECX, 2147483647
        or EDX, ECX
        mov DWORD PTR [EAX], EDX
        ret

instead of:

        sub ESP, 4
        mov DWORD PTR [ESP], ESI
        mov EAX, DWORD PTR [ESP + 8]
        mov ECX, DWORD PTR [EAX]
        mov EDX, ECX
        add EDX, EDX
        mov ESI, ECX
        and ESI, -2147483648
        and EDX, -2147483648
        or EDX, ESI
        and ECX, 2147483647
        or EDX, ECX
        mov DWORD PTR [EAX], EDX
        mov ESI, DWORD PTR [ESP]
        add ESP, 4
        ret

llvm-svn: 28122
 2006-05-05 06:10:43 +00:00
Chris Lattner
02bb78abd3 Implement a variety of simplifications for ANY_EXTEND. 
llvm-svn: 28121
 2006-05-05 05:58:59 +00:00
Chris Lattner
53e8cbbb83 Factor some code, add these transformations: 
// fold (and (trunc x), (trunc y)) -> (trunc (and x, y))
  // fold (or  (trunc x), (trunc y)) -> (trunc (or  x, y))
  // fold (xor (trunc x), (trunc y)) -> (trunc (xor x, y))

llvm-svn: 28120
 2006-05-05 05:51:50 +00:00
Chris Lattner
6ce6942d21 Remove a bogus transformation. This fixes SingleSource/UnitTests/2006-01-23-InitializedBitField.c 
with some changes I have to the new CFE.

llvm-svn: 28022
 2006-04-28 23:33:20 +00:00
Chris Lattner
3b9a7570cb Fix a couple more memory issues 
llvm-svn: 27930
 2006-04-21 15:32:26 +00:00
Chris Lattner
2ae3ed5e1a Fix a really subtle and obnoxious memory bug that caused issues with an 
llvm-gcc4 boostrap.  Whenever a node is deleted by the dag combiner, it
*must* be returned by the visit function, or the dag combiner will not
know that the node has been processed (and will, e.g., send it to the
target dag combine xforms).

llvm-svn: 27922
 2006-04-20 23:55:59 +00:00
Evan Cheng
618314ed2f Turn a VAND into a VECTOR_SHUFFLE is applicable. 
DAG combiner can turn a VAND V, <-1, 0, -1, -1>, i.e. vector clear elements,
into a vector shuffle with a zero vector. It only does so when TLI tells it
the xform is profitable.

llvm-svn: 27874
 2006-04-20 08:56:16 +00:00
Chris Lattner
9e4a289fae Canonicalize vvector_shuffle(x,x) -> vvector_shuffle(x,undef) to enable patterns 
to match again :)

llvm-svn: 27533
 2006-04-08 05:34:25 +00:00
Chris Lattner
fc546e1780 Codegen shufflevector as VVECTOR_SHUFFLE 
llvm-svn: 27529
 2006-04-08 04:15:24 +00:00
Evan Cheng
e5eefd369a 1. If both vector operands of a vector_shuffle are undef, turn it into an undef. 
2. A shuffle mask element can also be an undef.

llvm-svn: 27472
 2006-04-06 23:20:43 +00:00
Chris Lattner
cad2bfa3d7 Do not create ZEXTLOAD's unless we are before legalize or the operation is 
legal.

llvm-svn: 27402
 2006-04-04 17:39:18 +00:00
Chris Lattner
d13dd8ef5c Add a missing check, this fixes UnitTests/Vector/sumarray.c 
llvm-svn: 27375
 2006-04-03 17:29:28 +00:00
Chris Lattner
d9902c3de0 Add a missing check, which broke a bunch of vector tests. 
llvm-svn: 27374
 2006-04-03 17:21:50 +00:00
Andrew Lenharth
b133e47444 back this out 
llvm-svn: 27367
 2006-04-03 03:16:50 +00:00
Andrew Lenharth
91c6f28ad6 This should be a win of every arch 
llvm-svn: 27364
 2006-04-02 21:42:45 +00:00
Chris Lattner
dbdc830c83 Add a little dag combine to compile this: 
int %AreSecondAndThirdElementsBothNegative(<4 x float>* %in) {
entry:
        %tmp1 = load <4 x float>* %in           ; <<4 x float>> [#uses=1]
        %tmp = tail call int %llvm.ppc.altivec.vcmpgefp.p( int 1, <4 x float> < float 0x7FF8000000000000, float 0.000000e+00, float 0.000000e+00, float 0x7FF8000000000000 >, <4 x float> %tmp1 )           ; <int> [#uses=1]
        %tmp = seteq int %tmp, 0                ; <bool> [#uses=1]
        %tmp3 = cast bool %tmp to int           ; <int> [#uses=1]
        ret int %tmp3
}

into this:

_AreSecondAndThirdElementsBothNegative:
        mfspr r2, 256
        oris r4, r2, 49152
        mtspr 256, r4
        li r4, lo16(LCPI1_0)
        lis r5, ha16(LCPI1_0)
        lvx v0, 0, r3
        lvx v1, r5, r4
        vcmpgefp. v0, v1, v0
        mfcr r3, 2
        rlwinm r3, r3, 27, 31, 31
        mtspr 256, r2
        blr

instead of this:

_AreSecondAndThirdElementsBothNegative:
        mfspr r2, 256
        oris r4, r2, 49152
        mtspr 256, r4
        li r4, lo16(LCPI1_0)
        lis r5, ha16(LCPI1_0)
        lvx v0, 0, r3
        lvx v1, r5, r4
        vcmpgefp. v0, v1, v0
        mfcr r3, 2
        rlwinm r3, r3, 27, 31, 31
        xori r3, r3, 1
        cntlzw r3, r3
        srwi r3, r3, 5
        mtspr 256, r2
        blr

llvm-svn: 27356
 2006-04-02 06:11:11 +00:00
Chris Lattner
104db817c8 Constant fold all of the vector binops. This allows us to compile this: 
"vector unsigned char mergeLowHigh = (vector unsigned char)
( 8, 9, 10, 11, 16, 17, 18, 19, 12, 13, 14, 15, 20, 21, 22, 23 );
vector unsigned char mergeHighLow = vec_xor( mergeLowHigh, vec_splat_u8(8));"

aka:

void %test2(<16 x sbyte>* %P) {
  store <16 x sbyte> cast (<4 x int> xor (<4 x int> cast (<16 x ubyte> < ubyte 8, ubyte 9, ubyte 10, ubyte 11, ubyte 16, ubyte 17, ubyte 18, ubyte 19, ubyte 12, ubyte 13, ubyte 14, ubyte 15, ubyte 20, ubyte 21, ubyte 22, ubyte 23 > to <4 x int>), <4 x int> cast (<16 x sbyte> < sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8, sbyte 8 > to <4 x int>)) to <16 x sbyte>), <16 x sbyte> * %P
  ret void
}

into this:

_test2:
        mfspr r2, 256
        oris r4, r2, 32768
        mtspr 256, r4
        li r4, lo16(LCPI2_0)
        lis r5, ha16(LCPI2_0)
        lvx v0, r5, r4
        stvx v0, 0, r3
        mtspr 256, r2
        blr

instead of this:

_test2:
        mfspr r2, 256
        oris r4, r2, 49152
        mtspr 256, r4
        li r4, lo16(LCPI2_0)
        lis r5, ha16(LCPI2_0)
        vspltisb v0, 8
        lvx v1, r5, r4
        vxor v0, v1, v0
        stvx v0, 0, r3
        mtspr 256, r2
        blr

... which occurs here:
http://developer.apple.com/hardware/ve/calcspeed.html

llvm-svn: 27343
 2006-04-02 03:25:57 +00:00
Chris Lattner
52732a272f Implement constant folding of bit_convert of arbitrary constant vbuild_vector nodes. 
llvm-svn: 27341
 2006-04-02 02:53:43 +00:00
Chris Lattner
b088cfc01a Delete identity shuffles, implementing CodeGen/Generic/vector-identity-shuffle.ll 
llvm-svn: 27317
 2006-03-31 22:16:43 +00:00
Chris Lattner
0e7da656a7 Remove dead *extloads. This allows us to codegen vector.ll:test_extract_elt 
to:

test_extract_elt:
        alloc r3 = ar.pfs,0,1,0,0
        adds r8 = 12, r32
        ;;
        ldfs f8 = [r8]
        mov ar.pfs = r3
        br.ret.sptk.many rp

instead of:

test_extract_elt:
        alloc r3 = ar.pfs,0,1,0,0
        adds r8 = 28, r32
        adds r9 = 24, r32
        adds r10 = 20, r32
        adds r11 = 16, r32
        ;;
        ldfs f6 = [r8]
        ;;
        ldfs f6 = [r9]
        adds r8 = 12, r32
        adds r9 = 8, r32
        adds r14 = 4, r32
        ;;
        ldfs f6 = [r10]
        ;;
        ldfs f6 = [r11]
        ldfs f8 = [r8]
        ;;
        ldfs f6 = [r9]
        ;;
        ldfs f6 = [r14]
        ;;
        ldfs f6 = [r32]
        mov ar.pfs = r3
        br.ret.sptk.many rp

llvm-svn: 27297
 2006-03-31 18:10:41 +00:00
Chris Lattner
c3be332547 Delete dead loads in the dag. This allows us to compile 
vector.ll:test_extract_elt2 into:

_test_extract_elt2:
        lfd f1, 32(r3)
        blr

instead of:

_test_extract_elt2:
        lfd f0, 56(r3)
        lfd f0, 48(r3)
        lfd f0, 40(r3)
        lfd f1, 32(r3)
        lfd f0, 24(r3)
        lfd f0, 16(r3)
        lfd f0, 8(r3)
        lfd f0, 0(r3)
        blr

llvm-svn: 27296
 2006-03-31 18:06:18 +00:00
Chris Lattner
95a8c4fb11 When building a VVECTOR_SHUFFLE node from extract_element operations, make 
sure to build it as SHUFFLE(X, undef, mask), not SHUFFLE(X, X, mask).

The later is not canonical form, and prevents the PPC splat pattern from
matching.  For a particular splat, we go from generating this:

	li r10, lo16(LCPI1_0)
	lis r11, ha16(LCPI1_0)
	lvx v3, r11, r10
	vperm v3, v2, v2, v3

to generating:

	vspltw v3, v2, 3

llvm-svn: 27236
 2006-03-28 22:19:47 +00:00
Chris Lattner
017e8f1798 Canonicalize VECTOR_SHUFFLE(X, X, Y) -> VECTOR_SHUFFLE(X,undef,Y') 
llvm-svn: 27235
 2006-03-28 22:11:53 +00:00
Chris Lattner
a623f6f696 Turn a series of extract_element's feeding a build_vector into a 
vector_shuffle node.  For this:

void test(__m128 *res, __m128 *A, __m128 *B) {
  *res = _mm_unpacklo_ps(*A, *B);
}

we now produce this code:

_test:
        movl 8(%esp), %eax
        movaps (%eax), %xmm0
        movl 12(%esp), %eax
        unpcklps (%eax), %xmm0
        movl 4(%esp), %eax
        movaps %xmm0, (%eax)
        ret

instead of this:

_test:
        subl $76, %esp
        movl 88(%esp), %eax
        movaps (%eax), %xmm0
        movaps %xmm0, (%esp)
        movaps %xmm0, 32(%esp)
        movss 4(%esp), %xmm0
        movss 32(%esp), %xmm1
        unpcklps %xmm0, %xmm1
        movl 84(%esp), %eax
        movaps (%eax), %xmm0
        movaps %xmm0, 16(%esp)
        movaps %xmm0, 48(%esp)
        movss 20(%esp), %xmm0
        movss 48(%esp), %xmm2
        unpcklps %xmm0, %xmm2
        unpcklps %xmm1, %xmm2
        movl 80(%esp), %eax
        movaps %xmm2, (%eax)
        addl $76, %esp
        ret

GCC produces this (with -fomit-frame-pointer):

_test:
        subl    $12, %esp
        movl    20(%esp), %eax
        movaps  (%eax), %xmm0
        movl    24(%esp), %eax
        unpcklps        (%eax), %xmm0
        movl    16(%esp), %eax
        movaps  %xmm0, (%eax)
        addl    $12, %esp
        ret

llvm-svn: 27233
 2006-03-28 20:28:38 +00:00
Chris Lattner
cad173698d Don't crash on X^X if X is a vector. Instead, produce a vector of zeros. 
llvm-svn: 27229
 2006-03-28 19:11:05 +00:00
Chris Lattner
62185c0496 Don't call SimplifyDemandedBits on vectors 
llvm-svn: 27128
 2006-03-25 22:19:00 +00:00
Chris Lattner
c9c081fc40 fold insertelement(buildvector) -> buildvector if the inserted element # is 
a constant.  This implements test_constant_insert in CodeGen/Generic/vector.ll

llvm-svn: 26851
 2006-03-19 01:27:56 +00:00
Nate Begeman
42736d46b2 Remove BRTWOWAY* 
Make the PPC backend not dependent on BRTWOWAY_CC and make the branch
selector smarter about the code it generates, fixing a case in the
readme.

llvm-svn: 26814
 2006-03-17 01:40:33 +00:00
Chris Lattner
3b4ceba4ff make sure dead token factor nodes are removed by the dag combiner. 
llvm-svn: 26731
 2006-03-13 18:37:30 +00:00
Chris Lattner
3a3c8682b5 Fold X+Y -> X|Y when safe. This implements: 
Regression/CodeGen/PowerPC/and_add.ll

a case that occurs with dynamic allocas of constant size.

llvm-svn: 26727
 2006-03-13 06:51:27 +00:00
Chris Lattner
9d0ebb55a6 add a couple of missing folds 
llvm-svn: 26724
 2006-03-13 06:26:26 +00:00
Chris Lattner
f2bed5f46e Reinstate this now that the offending opposite xform has been removed. 
llvm-svn: 26548
 2006-03-05 19:53:55 +00:00
Evan Cheng
e0a6cf78f8 Back out fold (shl (add x, c1), c2) -> (add (shl x, c2), c1<<c2) for now. 
It's causing an infinite loop compiling ldecod on x86 / Darwin.

llvm-svn: 26544
 2006-03-05 07:30:16 +00:00